Tropical deforestation reduces plant mating quality by shifting the functional composition of pollinator communities

Most species of flowering plants rely on animal pollinators for effective mating (Ollerton et al., 2011), often interacting with diverse communities floral visitors 2007; Waser 1996). However, many commonly receive incompatible pollen, self-pollen or pollen from closely related individuals, thus influencing the quality that reaches target (Aizen & Harder, Ashman 2004). Such reductions in plant (i.e. received by a plant) can decrease likelihood beneficial genotype combinations among progeny (Karron 2012; Paschke 2002) and limit female choice more genetically compatible (Breed 2012, 2015; Skogsmyr Lankinen, 2002). This may result less vigorous offspring lower germination rates (Aguilar 2019; Marshall Ellstrand, 1986), competitive abilities (Gómez, 2000; Herrera, 2000) survival probabilities 2014; Karron Marshall, 1990). As such, characterizing mechanisms govern variation is essential to understand ecological evolutionary dynamics reproduction. Changes amount spatial configuration habitat (hereafter deforestation) negatively affect reduced reproductive success 2006; Eckert 2010; Leimu 2010). Deforestation substantially impact genetic diversity loads (Barrett 2017; Pannell Labouche, 2013), rate self-fertilization (Brunet Holmquist, 2009; Brunet Sweet, 2006) frequency individuals (Griffin Eckert, 2003; Herlihy 2004) altering three main components plant–pollinator interactions (Hadley Betts, Xiao 2016): demography (e.g. population size, phenology), pollinator availability behaviour. Accumulating evidence suggests deforestation directly influences reducing abundance size Galetto, 2004; Harder Aizen, Wagenius 2007). At local landscape scales, these number conspecific donors, limiting both quantity) potential options outcrossing Knight, Ward Johnson, 2005). Furthermore, also important microenvironmental conditions regulate phenological patterns (Xiao 2016). Asynchronous flower production shortened periods donor reduce cross-pollination Nason Hamrick, 1997). In addition direct, donor-mediated effects, influence indirectly, through cascading, pollinator-mediated effects driven shifts composition community (Kremen Owing morphological foraging behavioural differences, distinct functional groups have divergent particular (Castilla Rhodes Rodríguez-Rodríguez 2013; Valverde 2019). For example, forage across long distances acquire resources highly mobile pollinators) be efficient at transferring multiple sources (Krauss facilitating (Bezemer Byrne Llorens Ohashi Thomson, 2009) unrelated 2017). has been shown Hadley 2017) restricting their daily movement (Kormann 2016; Kremen Volpe 2014, 2016) access (Brosi Briggs, Fenster 2002), shifting which interact Feinsinger, Ashworth Brosi 2008). even small changes Fründ 2013). Studies assessing how are generally lacking, especially tropical plants. this paper, we use path analysis approach quantify direct donor-mediated) indirect pollinator-mediated) Heliconia tortuosa. forest herb pollinated two (sensu hummingbirds differ (Betts 2015). Territorial aggressively defend areas (<100 m diameter) contain high density Dobkin, 1984). ‘low-mobility’ strategy results restricted facilitate nearby (Linhart, 1973; Stiles, 1975). contrast, traplining typically long-distance routes (up 1 km per day) nectar (Gill, 1988; 2014), potentially enhancing transfer landscape. ‘high-mobility’ increase mate receipt non-self (Ohashi 2009). Our recent work showed hierarchical pool differentiation (quantified patches neighbouring H. tortuosa within patches) were consistent high-mobility hummingbirds, as found sample pools (Torres-Vanegas largely responsible next step quality. surrogates quality, estimated grains resulted successful pollination events individual sampled human-modified landscape, well corresponding degree biparental inbreeding extent individuals). Combined data communities, used here allowed us evaluate hypothesis deforestation, quantified cover patch affects donors indirectly community. Specifically, predict will haplotype) plants, inbreeding, abundance, availability) cross-pollination. Beyond indirect, cascading effects. expect induce filtering out hummingbird capacity distances. turn, lowering increasing inbreeding. The study was conducted an (approx. 310 km2) area surrounding Organization Tropical Las Cruces Biological Station southern Costa Rica (8°47?7?N, 82°57?32?W). included elevation gradient 850 1,500 a.s.l. originally covered Pacific premontane forest. Today, only about 30% remains (Zahawi existing 2,200) range <1 >1,300 ha span 1% 99% 1-km radius (Figure 1). non-forested matrix dominated pasture agriculture, some regrowth secondary Zahawi focal study, Griggs (Heliconiaceae), perennial, hermaphroditic exclusively understorey forests, where it occurs individually clonal clumps (Stiles, one most common hummingbird-pollinated 2014). Previous research hub network, interacts exhibit bill length curvature Borgella 2006). Although butterfly, Anartia fatima, experiments visitor fails deliver any kind service During peak season (February–May), produce inflorescences, although inflorescences observed. Each inflorescence holds up 12 bracts, each subtending 15 curved tubular flowers open sequentially fertile single day Across area, patchily distributed patches, mean 127 ha. reproduce clonally rhizomatous growth partially self-compatible (Kress, 1983). exclusion absence same (autogamy), required Self-fertilization possible between different (geitonogamy). Upon pollination, produces fleshy fruits seeds. Seed dispersal mediated several generalist frugivore birds, clay-coloured thrush, Turdus grayi (L.A. Arias-Medellin unpublished data). sampling design based previous area. al. (2014) stratified-random select 40 represented independent gradients size. We subset 30 selected long-term system. Within subset, ranged 0.6 than 1,300 7.8% 74.4% cover, measured proportion forested 1; Table S1). distance corresponds maximum Due design, close 100% not selected, confounded, because expected occur strongly amounts (Andrén, 1994; Betts 2013 season, 25 sampling. 2016, disturbance prevented 2013. Thus, during 2016 resampled 13 five additional compensate. Therefore, our dataset comprised (2013: n = 25; 2016: 18). patch, identified road point randomly location site) anywhere 500 confounding edge very large >500 From random site, marked nearest maternal plants). To avoid marking minimum (distances 162 m; 28.40 median 20.39 m). end leaf tissue fruit removal Once mature, bracts collected seeds all fruits. abortion collecting sufficient (?5) 2013, 87 while 71 new 2016. Combined, final materials 158 (Table haphazardly average 10 (range 5–21) DNA extraction genotyping, resulting total 1,584 770 plants; 814 These samples represented, average, 5.6 plant, 5.1 bract 1.8 fruit. Since confounds (Knight 2005; 2007), counted 20 calculated estimate site. Measures log-transformed normalize distribution. extracted genomic (leaf individuals) (1,584 embryos carefully dissected seeds). All extractions completed using QIAGEN DNeasy Plant Mini Kit following manufacturer's protocol (QIAGEN). amplified genotyped 11 microsatellite loci (Hac_C7, Hb_C115, Hac_D1, Hb_B9, Hac_B4, Hac_B6, Hac_C114, Hac_A103, Hc_C7, Hac_A116, Hc_C126) markers tested departures Hardy–Weinberg equilibrium, linkage disequilibrium null alleles (for details see Torres-Vanegas Genotyping errors scored locus mismatches observed plant-offspring arrays. account possibility seasons, probability observing identical multilocus genotypes (P(ID); Waits 2001) CERVUS 3.0.7. (Kalinowski haplotypes Pollen correspond set present paternal seed). they do represent complete array plant. obtained subtracting contribution seed minus.mom function r package gstudio (Dyer, resolves ambiguous cases (where mother–offspring pair heterozygous genotype) estimating posterior paternal–maternal gametic allele, given allele frequencies (Smouse 2001). allele.frequencies haplotype (h) measure (averaged loci) chosen (Nei, 1987) positively correlated (NEP; Smouse developed estimation (see associated digital repository). Corresponding standard performing 1,000 bootstrap replicates. analyses performed 3.5.0 (R Development Core Team, MLTR 3.4 (Ritland, (tm) (tm ? ts) outcrossed produced including individuals. single-locus (ts) differs considered event difference estimates (biparental inbreeding). above parameters Newton–Raphson method allowing equal those ovule. Standard bootstraps resampling family level. Hummingbird capture describe S2). mist nets separately placed front understorey, Note necessarily net captures intended provide forest-patch level, 05:30 12:30 hr 2010 2011. had effort patches: hours approx. 1,906 136 Captured level counting recaptures over observation methods counts walkabout surveys), rapid movements low lighting make identification challenging. Using variety radio-transmitter tracking, radio-frequency identification, field observations), categorized them either low- 0.5 classified (Campylopterus hemileucurus, Phaethornis guy longirostris). other (Amazilia decora, Amazilia edward, tzacatl, Heliodoxa jacula, Phaeochroa cuvierii striigularis) low-mobility pollinators, latter group contains short-distance (P. territorial species, whereas first Stiles Freeman, 1993). Aviary compared examined effect (proportion size) separate (n 30) 13). determine representative patches. subsequent linear mixed-effects models fitted later incorporated into confirmatory analysis. (log-transformed) significantly (Pearson's 0.304, p 0.102). variables 0.879, < 0.05) data. Further, (estimated site 0.086, 0.649), 0.253, 0.177), 0.313, 0.297). Also, 0.034, 0.663), 0.072, 0.367) 0.051, 0.519). confounded excluded further evaluated (forest-patch predictor variables) pools, (maternal plant-level response (LMM), variable (predictor) surrogate (response). multiple, non-independent ID effect. removed non-independence caused repeated years. treated fixed statistically controlled differences years (binary factor: 0, 1) model. scaled (mean SD allow comparison regression coefficients. variance explained ( ) r.squaredGLMM MuMIn (Barton, 2020) 3.5.0. explicitly autocorrelation, LMM re-fitted correlation structure. structures (corExp, corGaus, corSpher, corRatio, corLin) specifying argument lme nlme (Pinhero 2021). LMMs (REML) AICc score did autocorrelation then lowest best-supported detected variable. these, corExp structure corGaus S3). include structure, suggesting accounting needed ts), piecewise structural equation modelling (SEM) (Lefcheck, data, 761 73 338 37 423 36 described section (based analysis, knowledge system build model represents hypothetical causal links (Li, 1975; Shipley, act predictors responses, hypotheses simultaneously model, quantification (Shipley, Unlike traditional SEM, SEM relies tests hypothesized link (Grace 2000). advantage types defined appropriately decomposes another third referred ‘causal modeling’, causality merely a-priori cannot established observational Prior defining km, hummingbirds) collinearity inflation factors (VIF) vif car (Fox Weisberg, A conservative value VIF > 2.5 (Myers, 1990) detect predictors. (VIF 4.55, Pearson's investigated SEMs: km) representing initial 6: combination alternative surrogates) postulated influenced deforestation. would increase, decrease, addition, find enhance delivery SEMs pathway S1) combining (LM) (LMM). Linear variable). year diversit