Proceedings of the Royal Society B: Biological Sciences
You have accessResearch article

An avian seed dispersal paradox: New Zealand's extinct megafaunal birds did not disperse large seeds

Joanna K. Carpenter

Joanna K. Carpenter

Centre for Integrative Ecology, School of Biological Sciences, University of Canterbury, Christchurch 8140, New Zealand

[email protected]

Google Scholar

Find this author on PubMed

, ,
Janet M. Wilmshurst

Janet M. Wilmshurst

Landcare Research, Lincoln, Canterbury 7640, New Zealand

School of Environment, University of Auckland, Auckland 1142, New Zealand

Google Scholar

Find this author on PubMed

and
Dave Kelly

Dave Kelly

Centre for Integrative Ecology, School of Biological Sciences, University of Canterbury, Christchurch 8140, New Zealand

Google Scholar

Find this author on PubMed

    Abstract

    Often the mutualistic roles of extinct species are inferred based on plausible assumptions, but sometimes palaeoecological evidence can overturn such inferences. We present an example from New Zealand, where it has been widely assumed that some of the largest-seeded plants were dispersed by the giant extinct herbivorous moa (Dinornithiformes). The presence of large seeds in preserved moa gizzard contents supported this hypothesis, and five slow-germinating plant species (Elaeocarpus dentatus, E. hookerianus, Prumnopitys ferruginea, P. taxifolia, Vitex lucens) with thick seedcoats prompted speculation about whether these plants were adapted for moa dispersal. However, we demonstrate that all these assumptions are incorrect. While large seeds were present in 48% of moa gizzards analysed, analysis of 152 moa coprolites (subfossil faeces) revealed a very fine-grained consistency unparalleled in extant herbivores, with no intact seeds larger than 3.3 mm diameter. Secondly, prolonged experimental mechanical scarification of E. dentatus and P. ferruginea seeds did not reduce time to germination, providing no experimental support for the hypothesis that present-day slow germination results from the loss of scarification in moa guts. Paradoxically, although moa were New Zealand's largest native herbivores, the only seeds to survive moa gut passage intact were those of small-seeded herbs and shrubs.

    1. Introduction

    The last 50 000 years have seen the widespread extinction of large herbivores globally [1]. North and South America lost around 84 megafaunal genera at the end of the Pleistocene, while Australia lost 14 of its large herbivore genera around 30 000 years earlier [24]. Understanding the ecological consequences of these extinctions is important when evaluating current ecosystem functioning, but the ecosystem services provided by many of these lost species, and the legacy of their losses, are still unclear [1,57].

    Seed dispersal is a mutualism that has been studied in some depth for extinct fauna (e.g. [69]), but detailed data on the quality and quantity of the dispersal services provided by extinct species are rare. Consequently, the traits of dispersed fruits and the quality of dispersal service provided by extinct megafauna are usually inferred based on plausible assumptions. As a strong correlation exists between fruit size and disperser size [9,10], one of the most common assumptions is that extinct megafaunal frugivores consumed and dispersed the largest fruits. For example, Janzen & Martin [11] proposed that extinct Central American megafauna had a central role in the dispersal and evolution of several large angiosperm seeds. Their argument was supported by the existence of dispersal anachronisms—plant species with seed traits (such as size) that appear maladapted for dispersal by the contemporary fauna, haunted by the ‘ghosts of past mutualisms' [12]. As well as being large seeded, these putatively anachronistic plants had fruits that dropped to the ground when ripe (early abscission), where the terrestrial megafauna could access them; were unattractive to arboreal or volant frugivores; showed low contemporary dispersal rates; and had seeds that were protected by a thick, woody, endocarp [11]. Since Janzen and Martin first proposed their hypothesis for Central America, other putative anachronistic seeds with extinct dispersers have been recognized, involving the giant Malagasy elephant birds (Aepyornis) [13], giant tortoises in Mauritius [14] and Australian mihirungs (Dromornis) [15], among others [16]. The majority of these hypotheses are underpinned by the assumption that the largest extinct herbivores were dispersers of the largest seeds, but thus far fossil evidence has been insufficient to test whether this assumption actually holds true, although there have been some indications of its limitations (e.g. [17,18]). These limitations are of concern if they result in the erroneous identification of ‘gaps’ in seed dispersal networks. For example, the Amazonian motacú palm (Attalea princeps) conforms exactly to the classic megafaunal fruit syndrome outlined by Guimarães (fleshy fruits 4–10 cm diameter with up to five seeds), but is actually being effectively dispersed by several species of macaw (Ara spp.), despite the macaws' comparatively small size [17,18].

    New Zealand offers an ideal case study to test the validity of the assumption that large extinct frugivores were important dispersers of large-seeded, putatively anachronistic plants. The extinct moa of New Zealand (Aves: Dinornithiformes) were nine species [19] of flightless herbivorous ratites. With body masses ranging from 15 to greater than 200 kg [20], they were the largest terrestrial herbivores in New Zealand's prehistoric ecosystems before their extinction, driven by human hunting, ca 500 years ago [21]. They went extinct so recently that there is a rich subfossil record, including bones, skins, eggshell and coprolites (mummified faeces)—e.g. [2124], allowing an unusually comprehensive analysis of the ecosystem services these birds provided [25].

    Since the discovery of several subfossil moa gizzards containing large seeds [2629], moa have been widely proposed as dispersers of large, thick-endocarp-protected seeds. Clout & Hay [26] suggested that the giant New Zealand ratites might have performed a similar role to that of their extant relative the cassowary (Casuarius spp.), which consumes and disperses large amounts of large fallen fruit in tropical forests of Northern Australia, New Guinea and Yapen [30,31]. Although the New Zealand flora lacks species with fruits large enough to fit the classic megafaunal fruit syndrome (New Zealand mean fruit diameters ≤23 mm: [18,27,32]), five endemic plant species with early abscission and unusually thick-walled (2–3 mm) endocarps have been suggested as being adapted for dispersal by moa: Elaeocarpus dentatus, E. hookerianus (Elaeocarpaceae), Prumnopitys ferruginea, P. taxifolia (Podocarpaceae) and Vitex lucens (Labiatae) (figure 1) [2628]. All these species have moderate to large fruits by New Zealand standards (V. lucens 15 mm mean fruit diameter, P. ferruginea 13 mm, E. dentatus and P. taxifolia 9.4 mm, E. hookerianus 7.2 mm [27,32]; for further descriptions of these fruits, see electronic supplementary material, S1). The slow germination (2–7 years) currently experienced by four of these species (all except P. taxifolia) prompted speculations that these seeds were adapted to survive passage through a stone-filled moa gizzard, with the abrasion thinning the seedcoat and perhaps hastening germination times [26,27].

    Figure 1.

    Figure 1. New Zealand's five species of putative moa-dispersed fruit. (a) Elaeocarpus hookerianus; (b) Prumnopitys taxifolia; (c) P. ferruginea; (d) E. dentatus; (e) detail showing woody endocarp of P. ferruginea seed; (f) Vitex lucens. (Online version in colour.)

    Over the past decade, new insights into the role that these birds played in pre-human New Zealand have come from analyses of more than 150 subfossil moa coprolites [7,22,25,33,34]. This large sample size provides the basis for our quantitative examination of the role of moa as dispersers of large, endocarp-protected seeds. Coprolites are particularly useful for studying seed dispersal as they represent the post-digestion stage and the potential for dispersal, unlike gizzards, the contents of which can only demonstrate consumption. This distinction is especially important when studying birds with stone-filled gizzards, where the grinding action of the muscular gizzard may destroy many seeds [35,36].

    We analysed published macrofossil and ancient DNA (aDNA) data from 23 preserved gizzards and 152 coprolites, recovered from 13 sites spanning the South Island of New Zealand (figure 2), to determine whether moa ate and dispersed large seeds. We also tested whether prolonged mechanical scarification (used to simulate moa gut passage) of putatively moa-dispersed endocarps of E. dentatus and P. ferruginea decreased their time to germination. As we were interested in the protection conferred by the woody endocarps of the putative moa-dispersed plant species, we also tested the mechanical force needed to crush E. dentatus endocarps.

    Figure 2.

    Figure 2. The South Island of New Zealand, showing sites where analysed coprolites (diamonds) and gizzards (circles) were recovered. The * symbol denotes sites where Prumnopitys or Elaeocarpus were also indicated to be present, based on pollen records, aDNA and macrofossils. GIS layer from www.geographx.co.nz. (Online version in colour.)

    2. Material and methods

    (a) Dataset of seeds in gizzards and coprolites

    Gizzards were obtained from four sites in the South Island (Cheviot, Pyramid Valley, Scaifes Lagoon, Styx Mire; figure 2), and sampled four moa species (Dinornis robustus, Emeus crassus, Euryapteryx curtus, Pachyornis elephantopus).

    Coprolites were obtained from nine South Island sites (Takahe Valley, Sawers rockshelter, Roxburgh Gorge, Old Man Range, Mount Nicholas, Kawarau Gorge, Euphrates Cave, Earnscleugh Cave, Dart River; figure 2). The coprolites sampled three of the species represented by gizzard fossils (D. robustus, E. curtus, P. elephantopus) plus two others (Anomalopteryx didifornis, Megalapteryx didinus) for a total of five moa species.

    We compiled a dataset of all seeds found in these coprolites and gizzards from Falla [37], Burrows et al. [38], Horrocks et al. [39], Wood [40] and Wood et al. [22,25,33,34,41]. We also noted Prumnopitys and Elaeocarpus pollen and aDNA records for each sample (where tested) as evidence that these genera were growing locally or in the diet. There were no data on the presence of Prumnopitys aDNA in coprolites as the PCR primers used do not amplify conifer DNA. The presence of Prumnopitys or Elaeocarpus at the sites was indicated by the presence of the taxon's pollen, macrofossils or aDNA in the fossils, or from the presence of macrofossils recovered from the surrounding soil layer.

    (b) Seed size in coprolites and gizzards

    We used published records [42,43] to obtain mean seed lengths for each plant species with seeds detected in gizzards and coprolites. We used length rather than diameter as we were interested in the largest dimension of the seeds, but for New Zealand fruits (and seeds) under 10 mm diameter, length and diameter are similar [32]. Some seeds in the samples were only identified to genus level; these were given a mean length based on other identified species of that genus found in moa coprolites or gizzards. If no other species of that genus occurred within the fossils, the genus was generally given the mean of all native species within that genus. For Carex we used the mean of 20 randomly selected Carex species, due to the large number of native species in the genus. Seeds from a coprolite recovered from Earnscleugh Cave that were only identified as ‘Veronica’ were given the mean of 13 Veronica (previously Hebe) species that occurred in the region where the fossil was recovered. Species that were identified to above the genus level, or were identified to the genus level but only had one seed counted across all the fossils, were not included in the analysis (172 seeds). Analysis was conducted on a total of 9039 seeds. The number of seeds in a single E. curtus gizzard was not described so this sample was not included in this analysis.

    (c) Germination experiments on the putative moa-dispersed seeds

    Germination experiments commenced in August 2009. We subjected P. ferruginea and E. dentatus seeds collected in Blue Duck reserve, Kaikoura (42°14′ S, 173°47′ E) in August 2009 to five different scarification treatments to test whether treatments altered time to germination. The five scarification treatments were (1) 30 min mechanical scarification, (2) 4 h mechanical scarification, (3) 30 min mechanical scarification plus acid scarification, (4) 4 h mechanical scarification plus acid scarification and (5) acid scarification only (E. dentatus only). Mechanically scarified seeds were placed in a motorized portable concrete mixer with 2 l of water and 2 l of gravel chips (2–4 cm in length), where they were tumbled for either 30 min or 4 h. Acid-scarified seeds were placed on a tray (post-mechanical scarification, if relevant), covered with 0.5% hydrochloric acid, and held for 48 h at 38°C to crudely represent gut passage acidity and temperature. Sample sizes for each treatment ranged from 50–100 seeds for each species. The time to germination of treated seeds was compared to the time to germination for whole seeds (E. dentatus only) and seeds that had been processed in a way that mimics an extant frugivorous bird's gut passage (‘hand-cleaned’; [44]). Hand-cleaned seeds were rubbed by hand until the fleshy exocarp was removed. All whole seeds and 100 hand-cleaned seeds of E. dentatus were cold stratified at 1.6°C for 8 weeks prior to sowing, in case cold stratification was needed to break seed dormancy. The seeds from each treatment were planted in potting mix in one seed tray each and kept in an unheated shadehouse, where they were watered once daily. Trays were checked three times a year for seedling emergence for 5 years after sowing. Seven years after sowing trays were checked again and new seedlings were aged by size and bud scars to determine whether they germinated in years 6 or 7.

    (d) Endocarp strength tests

    We tested the force needed to rupture 14 fresh E. dentatus endocarps using a MTS Criterion Model 43 with a 2.5 kN cell and a crosshead speed of 5 mm min−1. Fresh E. dentatus fruits were collected from the ground beneath a tree at the University of Canterbury, Christchurch, and were tested without additional drying. Seeds were oriented with their sutures vertical to the loading direction and their shorter dimension parallel to the loading direction, following the method of Schüler et al. [45].

    (e) Statistical analysis

    All analyses were performed in R v. 3.3.0. We used Fisher's exact test to test whether there was a significant difference between the proportion of gizzards and coprolites that contained seeds of E. dentatus, E. hookerianus, P. ferruginea or P. taxifolia. Only subfossils that were recovered from sites where either Prumnopitys or Elaeocarpus were indicated to be present were included in this analysis. We did not test for the presence of V. lucens as it is native to only the North Island of New Zealand, whereas moa gizzard and coprolite samples have only been found in the South Island. The assumption of independence was potentially violated as some of the coprolites may have been produced by the same individual bird, but the large number of coprolites obtained from multiple different sites reduces this risk.

    We used the lme4 package [46] to perform a linear mixed effects analysis on the length of seeds found in the two subfossil types to test whether coprolites had significantly smaller seeds than gizzards. Seed lengths were log transformed to improve normality. We used subfossil type as a fixed effect, and sample nested within site as random effects. Moa species was not included as an effect because species identity was not known for some coprolites and gizzards. p-Values were obtained by likelihood ratio tests of the full model with a fixed effect for subfossil type against the model without the fixed effect for subfossil type.

    We used Cox proportional hazard (CPH) regression models to test whether the scarification treatments decreased the germination time for seeds of E. dentatus and P. ferruginea that germinated within 7.2 years [47]. We fitted CPH models to data consisting of the number of years between initiation of germination trials and seedling emergence, for each individual seed. Only data from seeds that germinated by the end of the study were included to separate the effects of treatment on the rate of germination from the effects on final germination percentages ([44]; see Figuerola et al. [48] for a similar approach). We used the ‘survival’ package in R to fit the models. Schoenfeld tests were conducted to confirm the assumption of proportionate hazards. We also tested for differences between the final percentages of germinated seeds after 7.2 years for the two plant species. We grouped treatments as mechanically scarified, hand-cleaned, or whole seeds (E. dentatus only) and tested for differences between them using a χ2 test. We used the ‘fifer’ package in R to run post hoc tests with a false discovery rate adjustment for multiple comparisons. All germination treatments which included acid scarification gave extremely low numbers of germinated seeds, so those treatments were excluded from all analyses.

    3. Results

    (a) Presence of putative moa-adapted seeds in coprolites and gizzards

    None of the gizzards or coprolites came from the North Island where Vitex occurred (figure 2), but moa ate fruits of P. taxifolia and E. hookerianus. Moa coprolites were found to have a very finely ground texture with no large seeds present, with a consistency more similar to the droppings of ruminant mammals than extant ratites. Although particle size was not measured, the fine-grained consistency is clear in comparative photos (figure 3). There was a significant difference between the proportion of gizzards and coprolites that contained the woody endocarp-protected seeds (hereafter just referred to as ‘seeds’) of E. dentatus, E. hookerianus, P. ferruginea or P. taxifolia. Almost half (48%) of the 23 gizzards contained identifiable remains of either P. taxifolia or E. hookerianus seeds, and 17% contained both. None of the 152 coprolites contained any intact Prumnopitys or Elaeocarpus seeds, or identifiable remains of these seeds, (p < 0.001; Fisher's exact χ2-test), despite the fact that 82.9% of the coprolites were recovered from sites where Prumnopitys or Elaeocarpus were present, based on pollen, macrofossil or aDNA evidence (S2). Elaeocarpus aDNA was present in 27% of coprolites that had been analysed for aDNA (n = 30), and a further 23% contained Oxalidales aDNA (the order that contains Elaeocarpus), although this aDNA could have come from foliage as well as pulverized seeds. The abundance of small seeds that ripen in summer or autumn (e.g. Fuchsia excorticata, Gaultheria spp.) in these coprolites indicated that many of the coprolites were deposited when fruits of Elaeocarpus and Prumnopitys would have been available [42]. Taken together, these data suggest that Elaeocarpus and Prumnopitys fruit were being eaten by moa, but did not survive gut passage intact. Wood et al. [22] recorded several gymnosperm and podocarp seed fragments in coprolites that could have come from Prumnopitys, although none of the fragments could be identified microscopically to the species level. Many small seeds discovered in coprolites were also broken.

    Figure 3.

    Figure 3. Photos showing the consistency of faecal deposits from (a) moa, (b) red deer Cervus elaphus, (c) cassowary Casuarius sp. and (d) emu Dromaius novaehollandiae. Emu and cassowary faeces have abundant coarse unground material, including seeds, whereas deer and moa faeces are finely ground. The small pale objects in the moa coprolite are Coprosma seeds (approx. 2 mm length). Scale bars are 1 cm. Cassowary photo provided by Barbara Maslen; emu photo provided by Jason Smith. (Online version in colour.)

    (b) Seed size in coprolites versus gizzards

    We analysed the sizes of 2253 seed remains identified from coprolites (S3) and 6786 seeds identified from gizzards (figure 4). The linear mixed effect model showed that seed length was significantly smaller in coprolites than in gizzards (LMM with a fixed effect for subfossil type and a random effect for sample nested within site: likelihood ratio test between full and null model: χ2(1) = 10.46, p = 0.0012). The weighted mean seed length in gizzards was 4.0 mm (±0.02 s.e.). The weighted mean seed size dispersed by moa (i.e. intact seeds in coprolites) was 1.64 mm (±0.02 s.e.), far smaller than the 4.6 mm (±0.03 s.e.) weighted mean seed size dispersed by New Zealand's smallest extant avian frugivore, the silvereye Zosterops lateralis (figure 5). The largest-seeded species found in gizzards was from E. hookerianus (10.25 mm average length), while the largest found in coprolites was Muehlenbeckia axillaris (3.3 mm average length).

    Figure 4.

    Figure 4. Frequency distributions of seed length for seeds found in moa gizzards (light grey, a) and moa coprolites (dark grey, b). Means are marked by the dashed lines. Note the different y-axis scales. (Online version in colour.)

    Figure 5.

    Figure 5. Mean seed size dispersed by (a) moa, weighing from 15–250 kg, and (b) silvereyes (Z. lateralis), weighing 0.01–0.013 kg. Error bars are standard error of the mean. Moa data for coprolites; silvereye data from [49]. (Online version in colour.)

    (c) Germination effects of simulated moa gut passage

    Germination began 2.5 years after sowing for P. ferruginea and was largely complete after 4 years, whereas for E. dentatus germination began after 1.6 years and continued for at least 7.2 years. There was germination from untreated and mechanically scarified seeds, but almost none from acid-scarified seeds for either E. dentatus or P. ferruginea (electronic supplementary material, table S4). Although mechanical scarification using stones in a concrete mixer (simulating moa gut passage) abraded and polished the endocarps, CPH models found that scarification did not decrease the time to germination for seeds that germinated within 7.2 years for either of the two species (χ2 = 1.491, d.f. = 5, p = 0.914 for E. dentatus, figure 6; χ2 = 3.1096, d.f. = 2, p = 0.211 for P. ferruginea). For E. dentatus, seeds that were mechanically scarified had higher final total percentages germinating than whole or hand-cleaned seeds (hand-cleaned seeds simulate the effect of fruit pulp removal by an extant frugivore's gut) (χ2 = 51.83, d.f. = 2, p ≤ 0.001). For P. ferruginea, there was no significant difference between the final percentages of germinated seeds for hand-cleaned and mechanically scarified treated seeds (χ2 = 0.015, d.f. = 1, p = 0.90). Overall, there was no evidence that the very slow germination of E. dentatus and P. ferruginea was an artefact of no longer passing through moa guts.

    Figure 6.

    Figure 6. Mean cumulative germination (%) against time for all E. dentatus seeds which germinated within 7.2 years, for mechanically scarified (black line) and non-scarified (grey line) treatments. Error bars are bootstrapped 95% confidence intervals.

    (d) Endocarp strength tests

    The mean force needed to rupture the endocarp of E. dentatus using a compression loaded MTS Criterion Model 43 was 927.4 (±68.9 s.e.) N, with a maximum force of 1226.4 N.

    4. Discussion

    Understanding whether extinctions of large herbivores have created gaps in mutualistic networks is important when assessing the wider ecological consequences of their loss [50]. Typically, large seeded plants (especially those with woody endocarps) are assumed to be most at risk of mutualistic disruption due to megafaunal extinctions [11,16,18]. However, our results show that while New Zealand's largest herbivore (the moa) consumed the large, thick-endocarp seeds of Prumnopitys and Elaeocarpus (as indicated by almost half of the gizzards containing them), they did not defaecate these seeds intact. None of the 152 coprolites contained whole seeds larger than 3.3 mm, strongly suggesting that large seeds consumed by moa were ground by the gizzard into finely crushed particles. Small seeds generally pass through the gut of gizzard-containing birds faster than large seeds [36], and were probably subjected to less grinding and, therefore, higher survival rates than large seeds. Similarly, there is actually a significant negative relationship between animal body mass and ingested seed size across all vertebrates, mainly driven by the tendency for large ungulates to ingest small, dry seeds [51]. Birds, however, generally show a positive relationship between animal body mass and ingested seed size [51], which further demonstrates how unusual moa are compared to their extant ratite relatives. Chen & Moles [51] suggested that while interactions between large vertebrates and small seeds are often overlooked, they are actually very frequent, and this appears to hold true for moa. Wood et al. [22] proposed that some of the small, dry seeds found in moa coprolites (e.g. Einadia, Poaceae) fit the ‘foliage is the fruit’ dispersal syndrome, where dry indehiscent fruits closely associated with nutritious foliage are adapted for dispersal by large herbivores [52]. Paradoxically, the loss of moa's interactions with these seeds may be far more concerning than those interactions involving large seeds, as New Zealand has no other large native herbivores and the role of exotic deer and ungulates as seed dispersers is poorly documented.

    (a) Moa gizzards have no extant analogues

    Our results demonstrate that the grinding efficiency of the moa gizzard has no analogue in extant ratites, and possibly not in any extant bird. Moa gizzards were filled with up to 5.6 kg of stones [38] measuring up to 110 mm in length [40]. By comparison, emu (Dromaius novaehollandiae) gizzards contain up to 0.75 kg stones, with the largest stones only 47 mm in length [53]; ostriches (Struthio spp., the world's largest extant birds) carry 1 kg of gizzard stones [54]; and cassowaries' gizzards are poorly developed and hold few stones [55]. We found that E. dentatus endocarps required forces of 930 N (equivalent to 95 kg of force) to rupture them and yet none survived moa gut passage, which further demonstrates the force moa gizzards must have exerted. This force is higher than those needed to rupture pecans (Carya illinoinensis), walnuts (Juglans spp.) and hazelnuts (Corylus spp.) [45], but less than the forces needed to rupture endocarp-protected palm species consumed by ungulates in the Peruvian Amazon [56]. The intense grinding efficiency of the moa gizzard explains why the consistency of moa coprolites appears very fine, especially when visually compared to representative faecal deposits of other large extant ratites such as emus and cassowaries (figure 3). The reason why moa required such powerful grinding gizzards is unclear, although for the larger moa species it could be an adaptation to maximize digestion of their woody, low-nutrient diets [25].

    The more time that food spends in the gizzard, the longer it is subjected to grinding. The time that food takes to pass through the entire gut (including the gizzard) scales allometrically with herbivore body mass [57], and is influenced by diet [58], suggesting that moa gut passage times were probably very long. For example, 40 kg emus have been recorded taking an average of 1-2 days to pass experimental pseudoseeds, with a number of pseudoseeds taking weeks and rarely even months to pass [59]. Fibrous foodstuffs likely increase gut passage time, so the wood and twig diet of certain moa species [40] would have increased gut passage times further. For example, the diet of South Island giant moa (D. robustus) was largely beech twigs [25]: low-quality food which would have required considerable amounts of time in the gizzard to grind into small enough particles to pass through to the gut. These long passage times would have given the moa gastric mill more time to grind large seeds into small particles. However, it is possible that juvenile moa may have had less developed gizzards, which could not destroy large seeds. Similarly, if adult moa regurgitated seeds to their chicks the seeds may not have had adequate time in the adult bird's gizzard to be destroyed, although it is unlikely such seeds would be dispersed to suitable microsites. Given the large sample size of coprolites analysed in this study it is possible that some of the coprolites came from juvenile moa (especially as evidence suggests that some coprolites may be associated with nesting sites[60]), yet all of the coprolites had the same finely ground consistency with no intact large seeds.

    Although our results convincingly demonstrate the destruction of large seed by moa, it is worth noting that we only analysed fossil evidence from six of the nine species of moa, and that some subfossils could not be identified to species level. Therefore, there may be some differences between moa species that remain to be uncovered. However, all six moa genera were sampled so differences are likely to be small.

    (b) Anachronistic seeds: the ‘slaying of a beautiful hypothesis by an ugly fact’

    While the seeds of P. ferruginea, P. taxifolia, E. dentatus, E. hookerianus and V. lucens appear ill-suited for dispersal by the contemporary fauna, we have shown that they almost certainly did not survive passage through the stone-filled gizzards of moa, and, therefore, are not anachronistic seeds adapted for dispersal by these giant ratites. Germination of most of these plant species is slow to extremely slow [27], and in fact we know of no other large-seeded tree where germination of unburied seeds continues for more than 7 years as in E. dentatus. While small, rounded seeds tend to get incorporated into the soil profile and if buried often delay germination for years [61], it is extremely unusual to have such slow germination in large-seeded tree species. While we aimed to test whether simulated moa gut passage increased germination speed in two of these species, this study has revealed that 4 h in a concrete mixer would have been a poor proxy for the powerful grinding action of a moa gizzard. But even gentle seed coat abrasion in the concrete mixer failed to hasten germination for P. ferruginea or E. dentatus, so these species' extremely slow germination cannot be explained by the extinction of moa. It is possible that the long germination times of seeds from these long-lived tree species are advantageous as they facilitate dispersal in time as well as space [62]. Additionally, early germination is not beneficial under all circumstances, as demonstrated by early germinating individuals of Linum catharticum and Gentianella amarella doing worse than those that germinated later [63]. Perhaps slow germination in canopy trees is favoured by New Zealand's high proportion of mast-seeding plants [64] and low pre-human rates of disturbance [65], meaning that trees had more need than in other parts of the world for delayed germination to exploit canopy gaps.

    While we have demonstrated that the large size of these seeds is not an adaptation for moa dispersal, their size probably confers other advantages, such as increased seedling survival. Seedlings that emerge from large seeds have been shown to be more tolerant of the many hazards encountered during establishment (e.g. drought, defoliation, shading) [66]. Large seeded tree species also typically have larger canopies and more reproductive years than small-seeded species, which provides a reproductive benefit [66]. It is also important to note that while these seeds are large by New Zealand standards, they are still within the gape size of New Zealand's largest extant frugivore, the kereru (Hemiphaga novaeseelandiae).

    Despite the unsolved puzzle of slow germination times, the remaining dispersal syndrome (early abscission, low contemporary dispersal rates, and visibility on the forest floor) of Elaeocarpus fruits suggest that they may be adapted for dispersal by flightless birds with less destructive gizzards [32,67]. For example, in Australia Elaeocarpus species are dispersed by cassowaries [68]. Flightless birds made up a significant proportion of the New Zealand avifauna [69], and the seed dispersal services these taxa provide are still largely unknown. Both North Island brown kiwi (Apteryx mantelli, an endemic ratite) and weka (Gallirallus australis, an endemic flightless rail) have been recorded consuming Elaeocarpus fruits [26]. We found that the final germination percentage of E. dentatus seeds increased with mechanical scarification, so it could be that the grit in kiwi and weka gizzards also abrades the seedcoat enough to increase the number of seeds that germinate. Additionally, volant frugivores such as the kereru have been recorded consuming E. dentatus fruits from the forest floor (J Carpenter 2017, unpublished data).

    5. Conclusion

    This study overturns widely held suppositions about moa as important dispersers of large seeds in pre-human New Zealand, and, therefore, contradicts assumptions that extinct herbivorous megafauna dispersed the largest seeds. While moa consumed seeds up to approximately 10 mm length, which are the largest seeds occurring in the areas yielding subfossil samples, the only seeds that survived their formidable grinding gizzards were less than 3.3 mm long. Therefore, New Zealand's largest herbivores only dispersed small seeds, such as those of dryland ‘spring annual’ herbs [22]. The loss of moa interactions with these small-seeded plants (some of which are now endangered; [70]) may paradoxically represent the actual gap in contemporary seed dispersal services. However, overall our results show that moa would have left little in the way of a seed dispersal gap, and focus should fall back to how New Zealand dispersal is faring with reduced populations and range sizes of extant seed dispersers [27].

    Our findings highlight the importance of using multiple lines of evidence when investigating the ecosystem roles of extinct fauna rather than simply relying on plausible assumptions. While the discovery of subfossil gizzard contents revealed important information on moa diet, it is only the subsequent recent discovery and analysis of large numbers of coprolites that have revealed that moa did not in fact disperse large or even medium-sized seeds. Similarly, although using the ‘ghosts of past mutualisms' to explain the existence of strange, apparently maladapted flora is appealing, further lines of evidence are required before gaps can be confirmed to be present in contemporary seed dispersal networks. This study also demonstrates the value of long-term palaeoecological data to help inform ecology and identify biological vulnerability and resilience [71]. With contemporary ecosystems undergoing significant environmental change, studies that use palaeoecological data to determine ecological baselines are becoming increasingly relevant [71,72]. While the majority of these studies have examined how taxa respond to environmental change, here we showcase how palaeoecological data (in particular, coprolite analysis) can also yield precious information on lost interaction networks (see also [73]).

    Data accessibility

    The datasets supporting this article have been uploaded as part of the electronic supplementary material, S5.

    Authors' contributions

    J.K.C. carried out all statistical analyses and drafted the initial manuscript. J.K.C. and J.M.W. conceived the initial study. D.K. conceived and designed the seed germination experiment. J.R.W. collated coprolite data. J.R.W., J.M.W. and D.K. edited the manuscript.

    Competing interests

    We declare we have no competing interests.

    Funding

    Funding was supplied by a University of Canterbury Roper Scholarship in Science, an Auckland Botanical Society Lucy Cranwell Scholarship, and a Graduate Women of New Zealand Fellowship to J.K.C. J.R.W. and J.M.W. were supported by funding from the Royal Society of New Zealand Marsden Fund.

    Acknowledgements

    We thank University of Canterbury staff Jenny Ladley for establishing and monitoring the germination experiments, Kim Roberts for compiling the list of seed sizes, Kevin Stobbs for assisting with seed rupture tests, and Elena Moltchanova for statistical advice. Juan Manuel Morales, Tomas A. Carlo, and an anonymous reviewer provided helpful suggestions on an earlier draft. Jim Watts created the base map used in figure 2. The quote—‘the slaying of a beautiful hypothesis by an ugly fact’—is from Thomas Huxley.

    Footnotes

    Electronic supplementary material is available online at https://dx.doi.org/10.6084/m9.figshare.c.4058807.

    Published by the Royal Society. All rights reserved.

    References