Study system

The sword-billed hummingbird,E.ensifera, is distributed in the high Andean forests (between 1,300 and 4,500 m a.s.l.) in Venezuela, Colombia, Ecuador, Peru and Bolivia. It exhibits the longest bill among hummingbirds, ranging between 81 and 120 mm in length (mean bill length of females and males: 103 and 96 mm, respectively), nearly as long as the body [9,21,27].

Flower traits, such as corolla length, have been used to determine the potential set of plant species pollinated byE.ensifera, since no other hummingbird pollinator in its Neotropical communities would be able to pollinate the deepest flowers (i.e. longer than 80 mm, [19]). We compiled published evidence ofE.ensifera interaction with long-flowered plant species. First, we explored available floras of the mentioned countries searching for flowers longer than 80 mm that had additional bird pollination traits (flower color and time of flowering) to differentiate them from similarly long hawkmoth-pollinated species. The threshold of 80 mm was chosen as a conservative limit of inclusion, since this is the minimum bill length ofE.ensifera. We also searched for long-flowered species among the specimens deposited in different herbaria (Universidad Nacional Mayor de San Marcos, New York Botanical Garden and Museo Botánico de Córdoba). These procedures resulted in a list of 24 species with mean corolla lengths from 95.59 to 146 mm potentially pollinated by the sword-billed hummingbird (Fig 1,Table 1,S1 Fig). Then we searched for Google Scholar articles containing the words “ensifera” and “hummingbird” or “sword-bill*”. We checked that these articles included field observations or anecdotal records of plants being visited by the sword-billed hummingbird (Table 1). Following this procedure, in addition to personal field observations, and study reports of the interaction [22–24,26,30–34], we compiled 24 plant species as the “plant guild” ofE.ensifera (Table 1).

Flower and bill lengths

We measured corolla tube length and flower operative length (i.e. distance from anthers and stigma to nectary) of the 24 species. Measurements were taken from 2 to 44 individuals per plant species, either in the field or from voucher specimens deposited in different herbaria (Table 1). Bill length data ofE.ensifera was obtained from 51 georeferenced museum records of male individuals published by Sánchez Osés [27]. Mean value of the corolla’ operative length of the guild of plants measured from herbarium vouchers (21 plant species,Table 1) was compared with bill length of 51E.ensifera individuals by fitting a generalized linear model (GLM) with gamma error distribution and log link function in R [42]. The model consisted of length as the response variable and organism trait length (hummingbird bill and plant tube) as the fixed factor. In addition a null model with 5000 simulations was performed to calculate a pseudo-F which was compared with the observed F.

Occurrence data

To reduce potential sampling bias, initial georeferenced points were spatially thinned; to this end, clusters were eliminated by keeping 4 km as the minimum distance between points [43]. Occurrence points forE.ensiferawere obtained from the Global Biodiversity Information Facility (http://www.gbif.org/). If information source was ambiguous, occurrence points were discarded for greater reliability.

Species distribution modeling and range polygons

Species distribution was modeled forE.ensifera using one topographic (altitude) and 19 bioclimatic variables obtained from WorldClim database (http://www.worldclim.org/bioclim) with a spatial resolution of 2.5 arc-min. SDM was performed from presence-only records via maximum entropy method using the program MAXENT v.3.3.3k [44]. Before deciding final MAXENT settings, we tested different preliminary models by changing default settings for each run [45]. Optimal prediction models were searched by modifying the extent of the prediction range [46]. Model performance was defined with the area under the curve (AUC) of the receiver operating characteristic (ROC) plot. The final resulting prediction area was located in the northeast region of South America (from Venezuela to Bolivia), from 12.347246 N to 23.014439 S and from 81.773802 W to 60.370377 E. Models were refined by alternatively changing the regularization multiplier to 0.5, 1, 1.5 or 2. Regularization penalizes each term included in the model, thus preventing overfitting [45]. In addition, autofeatures option was selected, which allows linear, quadratic, product, threshold, and hinge feature types to relate species records and environmental variables [47]. Background data was randomly selected by MAXENT from the prediction area using the default option which assumes that every pixel has the same probability of being selected as background [47]. The final settings used were: random test percentage = 25, regularization multiplier = 1, convergence threshold = 0.00001, and maximum iterations = 1,000. A total of 10 bootstrap replicates were generated for each model. A jackknife test was conducted to assess the relative importance of each one of the environmental variables in the MAXENT model. Only those variables that individually contributed more than 5% to the SDM were reported. Projections of the final SDM of MAXENT onto the geographical space were visualized and edited in QGIS v.2.14 [48]. The 10 percentile training presence threshold was used to visualize the predicted area onto the geographical space, thus excluding 10% of the presence points with the lowest predicted values.

Range polygons were obtained for 19 of the 24 plant species of the guild from the BIEN package of R [49]. Range polygons were not available forS.dombeyi,A.mutisii,P.bracteosa,P.leptomischa, andP.tenerifensis. Thus, these plant species were not included in the following analyses.

Mosaic of overlapping ranges

The SDM ofE.ensifera was converted into presence-absence maps considering the 10 percentile threshold. For the guild of plants, a map was generated by adding all the plant species range polygons (“richness map”). To quantify the areas of overlap betweenE.ensifera and its plant guild and reciprocally, i.e. between the plant guild andE.ensifera, the number of overlapped pixels on the resulting maps was counted and converted to km². A visual representation of the number of plant species overlapping withE.ensifera along its range was generated by trimming the richness map to fit the SDM ofE.ensifera.

To determine the relationship between bill length and plant richness we fitted a GLM with a gamma error distribution using the 51 bill length records of male hummingbird. We visualized the regressions using the ggplot2 package in R [50].

Flower and bill lengths

Mean operative length of the 24 species of the plant guild was between 95.59 ± 7.24 mm inP.parvifolia up to 146 ± 2.70 mm inP.loxensis (Table 1). Mean bill length ofE.ensifera was 99.39 ± 13.92 mm and varied geographically, without showing a clear clinal pattern (Fig 2A). The longest bills were recorded at mid-latitudes of its distribution range, around the Ecuadorian Andes (Fig 2A). Mean operative length of the guild of plants was significantly longer than bill length (F = 32.61, P < 0.001;Fig 3). The observed F value was significantly different from the pseudo-F generated by the null model (P < 0.001), i.e. the observed pattern is not a product of chance (Fig 3B).

Species distribution modeling

The SDM showed an AUC value of 0.97 ± 0.01. Altitude was the most important variable in contributing to the SDM, (Table 2). The boundary of the predicted distribution ofE.ensifera was similar to that in other maps of the range of this species [27,51]. Predicted ranges ofP.tripartita,P.mixta, andB.sanguinea were broader thanE.ensifera range, extending considerably further to the south than the area predicted for the pollinator (Fig 2A andS2 Fig).

Mosaic of overlapping range

The predicted range ofE.ensifera covered an area of 256,547 km², and was completely nested within that of the plant guild (1,729,836 km²;Fig 2A). In turn, the plant guild overlapped 21% of its distribution with that ofE.ensifera. The sword-billed hummingbird overlapped with eight of the 19 plant species across most of its entire range, and with 12 at the center of its distribution (Fig 2A). Plant richness was highest at mid-latitudes and lowest in the south ofE.ensifera range (Fig 2B).

The longest bills were detected near the Ecuadorian Andes, whereas the shortest bills were recorded in northern Peruvian Andes (Fig 2A).Ensifera ensifera bill length and guild of plant richness showed a positive trend (GLM: estimate = 0.01, t = 1.78, P = 0.08,Fig 4A). The longest bill lengths were detected in areas of high plant richness, i.e.: in the center ofE.ensifera distribution, between 0 and 5° S (Fig 4B and 4C).

Geographical ranges

The high geographical overlap betweenE.ensifera and its guild of long-flowered plants, together with the observed trait matching between hummingbird bill length and mean operative flower length, lends support to the existence of phenotypic specialization resulting from a coevolutionary process. This assumption is strengthened by the evidence of consistent phylogenetic divergence ages reported forE.ensifera (10.7 ma) and long-flowered species ofPassiflora supersection Tacsonia (11.6 ma) [23,52]. In addition, as a first geographical insight into this interaction system, our results reinforce the developing concept of interactions occurring in a geographically structured mosaic across the landscape. The geographical structuring of the interaction shows potential variation in the size and degree of asymmetry of the interaction networks, since the ranges of plant species did not completely overlap with each other and with the range ofE.ensifera. A similar pattern was also evidenced in the pollination mutualism between the extremely curved-billedEutoxeres hummingbirds and their matching plant species [53].

Trait matching

Previous studies suggest that reciprocal selection drives escalated lengthening of interacting traits of plant and insects [2]. Although this assumption needs to be confirmed for flowers and bills in hummingbird-plant interactions, there is evidence suggesting that this may indeed be the case. On the one hand, selection driven co-adaptation has been demonstrated for flower and bill morphology [53]. On the other hand, studies in plant-hummingbird networks have shown that long matching traits benefit both sides of the relationship [19,20]. Since hummingbirds with the longest bills will sip the most nectar from long flowers, and deepest flowers will be more often pollinated, then conditions are met for reciprocally driven positive directional selection and hence for an escalated lengthening of both bills and flowers [3,54,55]. For theE.ensifera guild, mean operative flower length was greater than bill length, as expected for morphological complementarity in coadapted plant-pollinator relationships (see [56]).

Ecological specialization

Extreme one-to-one specialization is rare in plant-pollinator systems, since the flower resources exploited by the nectar-feeding visitors are ephemeral, rendering the dependence on a single plant species unreliable. Even in the highly specialized interaction between the long-tongued hawkmothXantopham morganii praedicta and the long-spurred orchidA.sesquipedale, other long-spurred species have been proposed as possible food plants [3]. Plant-pollinator interactions most likely occur as a broad network of interactions [14], with long-billed hummingbirds acquiring specialization by more frequently interacting with flowers that match in length than with shorter ones [19]. Our results suggest that the sword-billed hummingbird locally relies on several long-flowered nectar sources, since at least three plant species simultaneously overlapped with the hummingbird distribution range.Ensifera ensifera bill length varied across the geographical range, possibly creating differences in the degree of asymmetry of interactions across the landscape, as expected under a geographical mosaic of coevolution [28].

Hotspots and coevolutionary vortices

We detected an area where species are predicted to participate in a “coevolutionary vortex” at mid-latitudes ofE.ensifera distribution. There,E.ensifera exhibited the longest bills, probably acquired by escalated reciprocal selection, which presumably drew several species to converge matching traits. We cannot ascertain whether escalation, if present, was driven by pairwise interaction and later increased in the number of interacting species, or if the joint action of the plant guild has driven coevolution in a diffuse manner. Both possibilities are suggested by the high richness of the plant community in these areas. Despite a non-significant relationship, probably due to the small sample size, we observed a positive trend between bill length and species plant richness. Accordingly, a study on the relationship between proboscides length of a long-tongued fly and the long-flowered guild it pollinates showed that morphological traits of pollinators were more exaggerated at sites where the local plant community consisted of several species with even longer tubes than the proboscides [56]. However, information about the degree of asymmetry of the interaction networks of these communities is lacking.

Coldspots

In a geographical context, the coevolutionary scenario in some communities may be intermingled with other communities in which selection is one-sided or absent [2]. For instance, at sites where the bills ofE.ensifera are much shorter than floral tubes, unilateral selection of bills is the most likely scenario, since the hummingbirds with the longest bills will be favored. We expect this to happen in the south of theE.ensifera range, where bills are short (mean = 96.91 mm) and plant richness is low. In these areas,P.tripartita,P.mixta andB.sanguinea overlapped withE.ensifera, and their flower tubes are much longer (mean = 113, 110, and 173 mm, respectively) than the bill length. Based on these facts, we postulate that these areas represent coevolutionary coldspots. When both hummingbird and flower match at low trait magnitudes, we expect an incipient hotspot since we assume that escalated reciprocal selection will likely shape long complementary traits over time. Other coldspots are represented by areas whereE.ensifera was absent within the range of the plant guild. Plants might colonize other areas outside the range of this hummingbird, where abiotic conditions are favorable, via alternative modes of reproduction (e.g. cultivation for ornamental purposes).

• 1.Anderson B, Johnson SD. Geographical covariation and local convergence of flower depth in a guild of fly-pollinated plants. New Phytol. 2009; 182: 533–40. 10.1111/j.1469-8137.2009.02764.x [DOI] [PubMed] [Google Scholar]
• 2.Pauw A, Stofberg J, Waterman RJ. Flies and flowers in Darwin’s race. Evolution. 2009; 63: 268–79. 10.1111/j.1558-5646.2008.00547.x [DOI] [PubMed] [Google Scholar]
• 3.Nilsson LA. The evolution of flowers with deep corolla tubes. Nature. 1988; 334:147–9. [Google Scholar]
• 4.Wasserthal LT. The pollinators of the Malagasy star orchidsAngraecum sesquipedale,A.sororium andA.compactum and the evolution of extremely long spurs by pollinator shift. Bot Acta. 1997; 110: 343–59. [Google Scholar]
• 5.Netz C, Renner SS. Long-spurredAngraecum orchids and long-tongued sphingid moths on madagascar: A time frame for Darwin’s predictedXanthopan/Angraecum coevolution. Biol J Linn Soc. 2017; 122: 469–78. [Google Scholar]
• 6.Vogel S, Michener CD. Long Bee Legs and Oil-Producing Floral Spurs, and a NewRediviva (Hymenoptera, Melittidae; Scrophulariaceae). J Kansas Entomol Soc. 1985; 58: 359–64. [Google Scholar]
• 7.Steiner KE, Whitehead VB. Oil Flowers and Oil Bees: Further Evidence for Pollinator Adaptation. Evolution. 1991; 45: 1493–501. 10.1111/j.1558-5646.1991.tb02651.x [DOI] [PubMed] [Google Scholar]
• 8.Muchhala N, Thomson JD. Going to great lengths: selection for long corolla tubes in an extremely specialized bat-flower mutualism. Proc Biol Sci. 2009;276: 2147–52. 10.1098/rspb.2009.0102 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 9.Snow DW, Snow B. Relationships between hummingbirds and flowers in the Andes of Colombia. Bull Br Museum Nat Hist. 1980; 38: 105–39. [Google Scholar]
• 10.Darwin C. On the various contrivances by which British and foreign orchids are fertilized by insects. London: John Murray: Oxford University Press; 1862. [Google Scholar]
• 11.Arditti J, Elliott J, Kitching IJ, Wasserthal LT. “Good Heavens what insect can suck it”- Charles Darwin,Angraecum sesquipedale andXanthopan morganiipraedicta. Bot J Linn Soc. 2012; 169: 403–32. [Google Scholar]
• 12.Bascompte J, Jordano P, Olesen JM. Asymmetric coevolutionary networks facilitate biodiversity maintenance. Science. 2006; 312: 431–3. 10.1126/science.1123412 [DOI] [PubMed] [Google Scholar]
• 13.Thompson JN. The coevolving web of life (American Society of Naturalists Presidential Address). Am Nat. 2009; 173: 125–40. 10.1086/595752 [DOI] [PubMed] [Google Scholar]
• 14.Sazatornil FD, Moré M, Benitez-Vieyra S, Cocucci AA, Kitching IJ, Schlumpberger BO, et al. Beyond neutral and forbidden links: morphological matches and the assembly of mutualistic hawkmoth–plant networks. J Anim Ecol. 2016; 85: 1586–94. 10.1111/1365-2656.12509 [DOI] [PubMed] [Google Scholar]
• 15.Benkman CW, Holimon WC, Smith JW. The influence of a competitor on the geographic mosaic of coevolution between crossbills and lodgepole pine. Evolution. 2001; 55: 282–94. [DOI] [PubMed] [Google Scholar]
• 16.Stinchcombe JR, Rausher MD. The evolution of tolerance to deer herbivory: modifications caused by the abundance of insect herbivores. Proc R Soc B Biol Sci. 2002; 269: 1241–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 17.Thompson JN. The geographic mosaic of coevolution. Chicago: The University of Chicago Press; 2005. [Google Scholar]
• 18.Thompson JN. Mutualistic webs of species. Science. 2006; 312: 372–3. 10.1126/science.1126904 [DOI] [PubMed] [Google Scholar]
• 19.Maglianesi MA, Blüthgen N, Böhning-Gaese K, Schleuning M. Morphological traits determine specialization and resource use in plant–hummingbird networks in the neotropics. Ecology. 2014; 95: 3325–34. [Google Scholar]
• 20.Weinstein BG, Graham CH. Persistent bill and corolla matching despite shifting temporal resources in tropical hummingbird-plant interactions. Ecol Lett. 2017; 20: 326–35. 10.1111/ele.12730 [DOI] [PubMed] [Google Scholar]
• 21.Pelayo RC, Rengifo C, Soriano PJ. Avian nectar robbers ofPassiflora mixta (Passifloraceae): do they have positive effect on the plant?Interciencia. 2011; 36: 587–92. [Google Scholar]
• 22.Ridgely R. Hummingbirds of Ecuador. Field guide. Ediecuatorial; 2011. [Google Scholar]
• 23.Abrahamczyk S, Souto-Vilaros D, Renner SS. Escape from extreme specialization: passionflowers, bats and the sword-billed hummingbird. Proc R Soc B Biol Sci. 2014; 281: 20140888. [DOI] [PMC free article] [PubMed] [Google Scholar]
• 24.Lindberg AB, Olesen JM. The fragility of extreme specialization:Passiflora mixta and its pollinating hummingbirdEnsifera ensifera. J Trop Ecol. 2001; 17: 323–9. [Google Scholar]
• 25.Gutierrez Z. A, Rojas-Nossa SV, Stiles F. G. Dinamica anual de la interaccion colibri-flor en ecosistemas altoandinos. Ornitol Neotrop. 2004; 15: 205–13. [Google Scholar]
• 26.Amico GC, Vidal-Russell R, Nickrent DL. Phylogenetic relationships and ecological speciation in the mistletoeTristerix (Loranthaceae): the influence of pollinators, dispersers, and hosts. Am J Bot. 2007; 94: 558–67. 10.3732/ajb.94.4.558 [DOI] [PubMed] [Google Scholar]
• 27.Sánchez Osés C. Taxonomy, phylogeny, and biogeography of the Andean hummingbird genera Coeligena Lesson, 1832; Pterophanes Gould, 1849; Ensifera Lesson 1843; and Patagona Gray, 1840 (Aves:Trochiliformes). [Bonn, Germany]: Dissertation, University of Bonn; 2003.
• 28.Thompson JN. The coevolutionary process. University of Chicago Press; 1994. [Google Scholar]
• 29.Elith J, Leathwick JR. Species Distribution Models: Ecological Explanation and Prediction Across Space and Time. Annu Rev Ecol Evol Syst. 2009; 40: 677–97. [Google Scholar]
• 30.Kuijt J. Revision ofTristerix (Loranthaceae). Syst Bot Monogr. 1988; 19: 1–61. [Google Scholar]
• 31.Segura S, Coppens d`Eeckenbrugge G, Lopez L, Grum M, Guarino L. Mapping the potential distribution of five species ofPassiflora in Andean countries. Genet Resour Crop Evol. 2003; 50: 555–66. [Google Scholar]
• 32.Wood JRI. 785.Salvia dombeyi. Curtis’s Bot Mag. 2014; 31: 143–53. [Google Scholar]
• 33.Tinoco BA, Graham CH, Aguilar JM, Schleuning M. Effects of hummingbird morphology on specialization in pollination networks vary with resource availability. Oikos. 2017;126: 52–60 126: 52–60 [Google Scholar]
• 34.Ocampo Pérez J, Coppens d’Eeckenbrugge G. Morphological characterization in the genusPassiflora L.: an approach to understanding its complex variability. Plant Syst Evol. 2017; 303: 531–58. [Google Scholar]
• 35.Holm-Nielsen LB, Jørgensen PM, Lawesson JE. 126. Passifloraceae In: Harling G, Andersson L, editors. Flora of Ecuador N° 31. Denmark: Department of Systematic Botany, University of Goteborg, and Pontificia Universidad Catolica del Ecuador, Quito; 1988. [Google Scholar]
• 36.Killip EP. The American species of Passifloraceae. Chicago: Field Museum of Natural History; 1938. [Google Scholar]
• 37.Escobar L. Flora de Colombia. Universidad Nacional de Colombia. Fondo Colombiano de Investigaciones Científicas y Proyectos Especiales “Francisco José de Caldas.” Bogotá, Colombia.: COLCIENCIAS; 1988.
• 38.Keel SHK. A revision of the genus Salpichroa (Solanaceae). City University of New York: University Microfilms International; 1984. 196 p. [Google Scholar]
• 39.Wester P, Claßen-Bockhoff R. Pollination syndromes of new worldSalvia species with special reference to bird pollination. Ann Missouri Bot Gard. 2011; 98: 101–55. [Google Scholar]
• 40.Wood JRI. The Salvias(Lamiaceae) of Bolivia. Kew Bull. 2007; 62: 177–221. [Google Scholar]
• 41.Dueñas-Gómez HC. Sinopsis de las Loranthaceae de Colombia. Caldasia. 2001; 23: 81–99. [Google Scholar]
• 42.R Development Core Team. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria: URLhttp://www.R-project.org. 2015. [Google Scholar]
• 43.Veloz SD. Spatially autocorrelated sampling falsely inflates measures of accuracy for presence-only niche models. J Biogeogr. 2009; 2290–2299. [Google Scholar]
• 44.Phillips SJ, Anderson RP, Schapire RE. Maximum entropy modeling of species geographic distributions. Ecol Modell. 2006; 190: 231–59. [Google Scholar]
• 45.Anderson RP, Gonzalez I. Species-specific tuning increases robustness to sampling bias in models of species distributions: An implementation with Maxent. Ecol Modell2011; 222: 2796–811. [Google Scholar]
• 46.Anderson RP, Raza A. The effect of the extent of the study region on GIS models of species geographic distributions and estimates of niche evolution: preliminary tests with montane rodents (genusNephelomys) in Venezuela. J Biogeogr. 2010; 37: 1378–93. [Google Scholar]
• 47.Merow C, Smith MJ, Silander JAJ. A practical guide to MaxEnt for modeling species’ distributions: what it does, and why inputs and settings matter. Ecography. 2013; 36: 1058–1069. [Google Scholar]
• 48.QGIS Development Team. QGIS geographic information system. Open Source Geospatial Foundation Project.http://www.qgis.org/. 2016. [Google Scholar]
• 49.Maitner B. BIEN: Tools for accessing the botanical information and ecology network database. Version 1.2.3.https://CRAN.R-project.org/package=BIEN. 2018.
• 50.Wickham H. ggplot2: elegant graphics for data analysisSpringer-Verlag; New York: 2016. [Google Scholar]
• 51.eBird. eBird: An online database of bird distribution and abundance [web application]. eBird, Ithaca, New York: 2016. [Google Scholar]
• 52.Mcguire JA, Witt CC, Remsen J V, Corl A, Rabosky DL, Altshuler DL, et al. Report molecular phylogenetics and the diversification of hummingbirds. CURBIO. 2014; 24: 910–6. [DOI] [PubMed] [Google Scholar]
• 53.Abrahamczyk S, Poretschkin C, Renner SS. Evolutionary flexibility in five hummingbird/plant mutualistic systems: testing temporal and geographic matching. J Biogeogr. 2017; 44: 1847–55. [Google Scholar]
• 54.Temeles EJ, Pan IL, Brennan JB, Horwitt JN. Evidence for ecological causation of sexual dimorphism in a hummingbird. Science. 2000; 289: 441–443. [DOI] [PubMed] [Google Scholar]
• 55.Alexandersson R, Johnson SD. Pollinator-mediated selection on flower-tube length in a hawkmoth-pollinatedGladiolus (Iridaceae). Proc Biol Sci. 200222; 269: 631–6. 10.1098/rspb.2001.1928 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 56.Anderson B, Terblanche JS, Ellis AG. Predictable patterns of trait mismatches between interacting plants and insects. BMC Evol Biol.2010; 10: 20410.1186/1471-2148-10-204 [DOI] [PMC free article] [PubMed] [Google Scholar]
• 57.Brown JH, Bowers MA. Community organization in hummingbirds: relationships between morphology and ecology. Auk. 1985; 102(2): 251–269 [Google Scholar]
• 58.Hillebrand H. On the generality of the latitudinal diversity gradient. The Am Nat. 2004; 163(2): 192–211. 10.1086/381004 [DOI] [PubMed] [Google Scholar]

Supplementary Materials

Data Availability Statement

All relevant data are within the paper and its Supporting Information files.