We conducted fieldwork in the valley of San Gerardo de Dota (9°33´´N, 83°47´W, 2300 masl), located in the Pacific slope of the Talamanca Mountain Range, San José province, Costa Rica. The upper montane cloud forest of the Talamanca Mountain range in Costa Rica is an evergreen forest dominated by oaks (Quercus spp.). Epiphytes (including bromeliads, orchids, mosses and lichens), tree ferns, and bamboo form a characteristic component of the montane forest vegetation (Kappelle et al., 1992). Daytime average temperature at the study site is 17°C with a median annual precipitation of 2,500 mm. This region experiences two seasons: a dry season from December to March and a wet, rainy season from April to November.

The valley of San Gerardo is flanked by Los Quetzales National Park, Tapantí National Park, and Los Santos Forest Reserve. Most human activities in San Gerardo occur in the upper basin of the Savegre River ( Figure 1A illustrates the presence of montane forest on the hilltops of San Gerardo and fruit crops in the valley). Low-intensity fruit orchards (apple and peach) were planted to replace pastures in the 1980’s. Avocado plantations were established at the beginning of the 1990’s. Currently, avocado is the dominant fruit crop, followed by apple, blackberry, and remnant patches of plum trees. Ecotourism is the predominant economic activity in San Gerardo de Dota, which encourages farmers to minimize the use of agrochemicals and to attract avifauna and insects by means of ornamental gardens and native trees (e.g., Ocotea sp.). Farmers perform mowing every two months to manage weed growth.

We conducted monthly surveys to estimate the diversity of insect visitors to the flowers of local native plants, focusing on herbs and shrubs, including ruderals commonly found growing within the fruit crops. The four transects were about 150 m in length and 4km away from the fruit farms, respectively. We sampled insect visitors of avocado (2 farms), apple (2 farms), plum (1 farm), and blackberry (2 farms) ( Supplementary Table 1 ). Two farms (Lauraceas and Sueños del Bosque) rent Apis mellifera apiaries from October to February. The farms were roughly 4 km apart from each other. We visited farms during the peak blooming periods of each fruit crop. Crop and wild plant surveys consisted in collecting all insects in a focal flowering plant or tree during a 15 min period. We collected insects on flowers using handheld nets or by positioning a clean transparent plastic bag on the plant and directing the insects into the bag. Insects were then stored in collection tubes with 70% ethanol for later identification in the lab. We carried out surveys during the morning hours of clear, warm days, avoiding rainy and windy days. Surveys lasted between 2 and 8 hours, depending on weather conditions and flower availability.

Insects were identified by trained taxonomists and specialists (G. R-M., N. G-B, P. H., and M. A. Z.) using available keys (Brown et al., 2009; Brown et al., 2010; Michener et al., 1994) and, when needed, comparing specimens collected with those deposited in entomology collections, namely the Museo de Zoología, Universidad de Costa Rica (UCR), and the Museo Nacional de Costa Rica. Most tropical insects lack taxonomic keys at the species level, and a considerable number of these insect species are undescribed; thus, many of the species collected in this study were assigned to morphospecies based on the criteria of expert taxonomists. We deposited voucher specimens in the Museo de Zoología (UCR). A subset of 242 insects collected from the fruit crops were further processed to isolate DNA from the pollen grains attached to the insects. Collection tubes were stored at -20°C in the lab at UCR until further processing.

We used Hill-diversity estimates to describe the diversity patterns of flower visitors according to season and site (i.e. fruit crops and wild plant transects). A larger number of taxa (40%) were unique to fruit crops sampled during the rainy season compared to the dry season (3%) ( Figure 1B ). We found that the diversity of flower visitors was similar in fruit crops compared to wild plant transects, a result that was consistent across seasons and Hill-diversity estimates ( Figure 1C ; Supplementary Figures 2 , 3 ). Fruit crops flowered at different times of the year. Avocado trees had the most extended flowering periods, followed by blackberry, apple, and plum crops ( Supplementary Table 2 ). Apple and plum varieties planted in the region bloom twice a year (in July-August and January), with short (approx. 1 month) windows of flowering bouts. We observed that avocado plantations differ from the other fruit crops in that they tend to attract a higher diversity of insect visitors during the flowering period although diversity was not significantly different across fruit crops ( Figure 1D ; Supplementary Figures 4 , 5 ).

We found that there is little overlap in the composition of insect visitors between fruit crops and wild plants (C_{q =1N}: F = 7.18, p < 0.0001, R ² = 0.11) and a significant, albeit weaker effect of season (C_{q =1N}: F = 2.42, p = 0.004, R ² = 0.04) ( Figure 2A ). Results were consistent across estimates based on the three Hill q levels analyzed ( Supplementary Figure 6 ). Furthermore, we found a strong effect (R ² = 0.24) of type of fruit crop on community composition (F = 2.51, p < 0.0001), suggesting fruit crop-specific insect communities ( Figure 2B ; Supplementary Figure 7 ).

We amplified and sequenced the ITS2 region of 157 out of 242 pollen samples (65%) to characterize the plant sources found in flower visitors of fruit crops. Obtaining high quality DNA from pollen samples for sequencing was a major challenge in our study; 12% of samples yielded no DNA, 7% did not amplify, and 15% were not indexed for sequencing due to very low concentrations in the first step of PCR. Close to half of the pollen samples that failed to yield high-quality DNA were from insects in the order Diptera (49%, all collected in the avocado farms), followed by Apis (27%), and Coleoptera (10%). The low amount of pollen grains attached to the bodies of dipterans may have been the main limiting factor in obtaining sequencing data for this group. As stated in the previous section, we found a higher diversity of insect visitors in avocado trees compared to the other fruit crops. Hence, the number of insect taxa used to study the diversity of plant sources via pollen metabarcoding from fruit crop insect visitors was greater for avocado, particularly for Diptera, since taxa in this group were diverse and abundant ( Supplementary Table 9 ).

We detected 2,387 sequence variants in a total of 3,064,745 quality-filtered, denoised, and chimera filtered sequencing reads with an average sequencing depth of 18,508 reads per sample (range 880 - 40,627). In our study, a reference database that incorporated local diversity and built from sequences specific to the Neotropics (i.e., with a geographic restriction applied to sequence retrieval from NCBI, Supplementary Text S2 ) yielded the highest proportion of reads assigned to genus (95%) and to species (85%) ( Supplementary Figure 8A ). We did not observe an impact of dereplication (i.e., merging identical sequences with identical taxonomic annotations into a unique sequence, Supplementary Text S1 ) but extracting primer-specific amplicons can result in a low fraction of reads with taxonomic assignment ( Supplementary Figure 8 ).

Among the 2,387 pollen sequence variants, we identified a total of 153 species of plants, spanning 140 genera, 68 families, and 37 orders. Of these, 29 plant taxa (25 of which were identified to species and 4 to genus level) were relatively common (i.e. > 20% prevalence) ( Figures 3A–K ). A large proportion of plant taxa were rare (i.e., close to half (48.7%) of plant taxa were found in up to two samples). Moreover, we identified a mean of 18.5 (range 3 - 48) plant taxa per sample ( Supplementary Figure 9 ). We found that two endemic plant species, Meliosma irazuensis (Sabiaceae, 72.4%) and Viburnum costaricanum (Adoxaceae, 71.8%), and other herbaceous plants, such as flatweed (Hypochaeris radicata), glossy nightshade (Solanum americanum), and pink knotweed (Persicaria capitata), were among the most common plant species in pollen samples (> 65% prevalence) ( Figure 3A ). Other endemic plants were also commonly detected, including Hypericum irazuense (Hypericaceae, 26.3%) and Diplostephium costarricense (Asteraceae, 18.6%). The oak species Quercus salicifolia, native to central Mexico and Central America, featured among the most frequent plant species (52.56%) recovered from pollen samples ( Figure 3A ). Regarding pollen from the fruit crops, we found that plum (57.1%), followed by blackberry (37.8%) and apple (37.1%) were more commonly detected as plant sources, compared to an observed low prevalence of avocado (0.04%).

We used incidence data to test for differences in alpha diversity of plant species (or genera) identified in the pollen samples. Sampling completeness was relatively consistent across the four fruit crops (range: 94% - 99%, Supplementary Figure 10 ) and the groups of insect visitors (range: 89% - 97%, Supplementary Figure 12 ) studied in San Gerardo. We found that the diversity of plant sources according to richness was similar across fruit crops and across insect visitors ( Figures 4A, B , overlapping CI in q = 0 in Supplementary Figure 11 and Supplementary Figure 13 ). Richness error estimates were particularly high for pollen samples collected from syrphid flies, likely due to pooling individual flies for pollen analyses ( Figure 4B ). In contrast to richness estimates, the patterns of diversity, according to Hill-Shannon and Hill-Simpson, revealed consistent differences between crops. Estimates of evenness in apple and plum crops were higher compared to avocado and blackberry ( Figure 4A ; Supplementary Figure 11 ). These results suggest that in avocado and blackberry farms, insects visited fewer dominant plant taxa compared to the other fruit crops. We found a similar diversity of plant sources across insect groups according to Hill-Shannon estimates, but Hill-Simpson estimates revealed lower diversity in syrphid flies compared to all the other insect groups ( Figure 4B ; Supplementary Figure 13 ).

With respect to patterns of plant species composition, we found that fruit crops were a stronger predictor (C_{q =1N}: F = 7.75, p < 0.0001, R ² = 0.13) compared to insect group (C_{q =1N}: F = 2.39, p < 0.0001, R ² = 0.059). We observed that pollen samples from plum and apple farms cluster closer together, while pollen samples from blackberry appear to have a distinct composition of plant species. The community of plant taxa in avocado seems to be highly heterogeneous ( Figure 4C ). We found a large overlap in plant sources among the insect groups ( Figure 4D ; Supplementary Figure 14 ).

We found that fruit crops attract a large diversity of insect pollinators. These farming practices appear to have a positive impact on pollinator diversity by promoting the presence of native plants that provide additional food sources that attract a diverse community of pollinators. Indeed, we observed that close to half of the species of herbaceous and ruderal plants surveyed in the wild plant transects were also present in fruit crop farms.

An additional factor that might have a positive impact on the diversity of flower visitors is the proximity of protected forest to the fruit crops in San Gerardo, as has been previously shown for other tropical crops such as avocado, coffee, and mango (Carvalheiro et al., 2010; Celis-Diez et al., 2023; Hipólito et al., 2018; Villamil et al., 2018). Fruit crop farms in our study region are surrounded by ornamental gardens, stands of introduced cypress trees (Cupressus lusitanica), patches of secondary montane forest, and/or protected forest on the hilltops. The montane forests of San Gerardo are characterized by a species-rich plant community shaped in part by deforestation and other human activities since the mid 50’s. Clearance led to the emergence of open habitats such as pasture land and shrublands that enabled the migration and establishment of herbaceous plants from the higher altitude páramo and subalpine dwarf forests (Kappelle et al., 2000). It is possible that, besides providing additional feeding resources for pollinators, ruderal vegetation in and around fruit crops and undisturbed nearby montane forests provide microhabitats that sustain rich and abundant populations of insect pollinators. For instance, larvae of the dipteran family Sciaridae, feed on decaying plant material, an abundant substrate on the ground covered by the layer of ruderal plants and inside forest edges. Other larvae scavenge on dead bodies of wild (small and large) animals that die in the forest or feed on mammal dung (e.g., Sarcophagidae); yet the larvae of various fly families recorded in our study (e.g., Syrphidae) prey on the larvae of other insects (Rojo et al., 2003; Zumbado, 2006). The bumblebee Bombus ephippiatus, the most common native species in our study, builds nests underground, usually in drained terrain covered with a dense herbaceous layer (Laverty and Plowright, 1985); GB unpublished data).

Our study revealed that avocado farms tend to host a higher diversity of insect floral visitors compared to other fruit crops. The presence of a diverse floral visitor community is consistent with previous studies of avocado orchards (Okello et al., 2021; Perez-Balam et al., 2012; Villamil et al., 2018). We found that fruit crops and wild plants have a distinct community composition of flower visitors, and that insect composition was also specific to the type of fruit crop. Together, these results suggest that pronounced differences in microhabitat influence the presence or absence of insect taxa at different sampling sites. However, we are currently unable to distinguish the factors shaping these diversity patterns. Studies that incorporate surveys of wild plant diversity in parallel with insect surveys in farmlands might provide useful insights for future farming practices, such as evaluating the relationship between wild plant diversity and insect visitor diversity.

We found an effect of season on insect visitor community composition. Climate models predict that high-elevation tropical environments will experience a strong decrease in precipitation, higher temperatures, and increased variability of both precipitation and temperature (Karmalkar et al., 2008). Our results suggest that insect composition in San Gerardo are sensitive to seasonal effects. Given that high-elevation tropical ecosystems are regions of high endemism but also emerging hot-spots for climate change (Giorgi, 2006), long-term assessments of flower visitor community dynamics are crucial for understanding the potential effects of changing climatic conditions on plant-pollinator interactions.

Overall, the observed diversity of flower visitors in this study provides a baseline understanding of pollinator communities and their potential role as bioindicators of ecosystem health. The observed variability in diversity patterns of flower visitors across fruit crops and wild plant transects suggests the presence of fine-scale habitat differences that deserve further attention. For instance, the need to conduct studies that aim at identifying the main predictors of insect diversity and composition, such as the influence of distance to protected forest, the effect of farming practices, and the relationship between different wild plant assemblages and pollinator diversity. Currently, Latin America’s landscape conversion is largely driven by regional development (Keys, 2010; Klepeis and Turner, 2001; Newcomer et al., 2022). A main challenge to preserving biodiversity is therefore not only to effectively manage protected areas but also to ensure that rural areas develop in a more sustainable manner (Lambin et al., 2003). Costa Rica, despite its known efforts to protect biodiversity, represents a good example of this struggle (Broadbent et al., 2012; Kappelle and Juárez, 1995; Sanchez-Azofeifa et al., 2002). Our results suggest that tropical highland farmers have the potential to improve their crop yields and preserve ecosystem services by encouraging the growth of ruderals in and around their fields and by preserving tracts of undisturbed habitat.

A main outcome of this study was the relatively large number of plant species identified in pollen samples that were not necessarily observed in the fruit crops nor in the wild plant transects. We found that a high proportion of plant sources (close to 50%) were detected in up to two samples, illustrating the sensitivity of pollen metabarcoding in detecting rare taxa among flower visitors (Bell et al., 2017; Milla et al., 2021; Richardson et al., 2021). Insect visitors carried pollen from common ruderals and a variety of wild and endemic species. Some of these correspond to plant species with distributions restricted to the Talamanca mountain range (i.e., Vaccinium consanguineum (Ericaceae), Ageratina kupperi (Asteraceae), Viburnum costaricanum (Adoxaceae), Hypericum irazuense (Hypericaceae), Sciodaphyllum pittieri (Araliaceae), Quercus costaricensis (Fagaceae) and Diplostephium costaricense (Asteraceae)). These findings provide evidence that insect visitors benefit from the resources provided by vegetation co-occurring in fruit crop farms, plants from ornamental gardens, and nearby undisturbed habitat. This diverse community of flower visitors likely contributes to the pollination services of fruit crops and native plant species in San Gerardo, with nearby protected areas playing a crucial role in maintaining pollinator populations during periods when crop flowers are not in bloom.

Results of plant sources among pollen samples in fruit crops mirror the differences in composition observed in insect visitor data, i.e., the type of fruit crop was a strong predictor of similarity among pollen samples, suggesting that fruit crop farms have a relatively distinct community of herbaceous wild plants that attract different insect visitors and that these visitors exploit different foraging resources. Furthermore, we found that estimates of plant species diversity (Hill-evenness) in pollen samples were higher in apple and plum farms compared to avocado and blackberry crops. Differences in the distance between crops and the natural protected areas might contribute to the observed variation in diversity. The plum and one of the two apple farms were closer to the montane protected forest compared to the other study sites. Higher diversity in these fruit crops might reflect the greater availability and diversity of floral resources associated with nearby undisturbed habitats (Muñoz et al., 2021). The fact that the number of plant taxa detected per sample was highly variable suggests that a large proportion of plant sources identified through analyses of pollen samples correspond to the detection of heterospecific pollen picked up by flower visitors. High levels of heterospecific pollen can be an indication of high diversity of floral resources (Willmer et al., 2017) and phenological overlap (Aizen and Rovere, 2010).

We found differences in diversity with respect to groups of insect visitors; estimates of Hill-evenness were lower in syrphid flies compared to the other insect groups in our study. Syrphid flies appear to carry pollen of a few dominant plant species, including common ruderals planted by farmers to attract pollinators (e.g., pinkweed, Persicaria capitata). Lower diversity of plant sources in Syrphid flies might be explained by differences in morphological traits such as smaller body length and fewer setae compared to other flies like Tachinidae, which hinder pollen adherence (Cullen et al., 2021; Földesi et al., 2021).

Montane tropical forests are characterized by low temperatures and very humid conditions, which may have a negative effect on the activity of managed populations of honeybees (Corbet et al., 1993). In contrast, non-syrphid flies are more resilient to these unfavorable conditions (Levesque and Burger, 1982), which explains similar estimates of diversity of plant sources for non-syrphid flies compared to Apis and Bombus bees. Moreover, the high diversity of plant sources found in bumblebee samples highlights the importance of these insects as crop and wild plant pollinators. Bumblebees are the most prevalent wild pollinators in the understory of montane forests in the region (Cristóbal-Perez et al., 2024), remain active throughout the day, and have been shown to carry a high percentage of conspecific pollen grains (Wesselingh et al., 2000). Together, our findings echo previous work demonstrating the importance of studying wild pollinators in farmlands and their role in ecosystem services (Larson et al., 2001; Orford et al., 2015).

Our findings suggest that small-scale fruit crops may facilitate insect movement between forest patches, increasing gene flow and decreasing genetic structure among native and endemic plant species (Jha and Dick, 2010). Increased connectivity is crucial for maintaining the genetic diversity of native species, which is essential for their adaptability and population viability (Frankham, 2005; Newman and Tallmon, 2001; Richards, 2000). However, agricultural expansion may compromise landscape connectivity, resulting in habitat fragmentation that could disrupt insect movement between forest patches, reducing gene flow and thereby increasing the detrimental effects of genetic drift on native plant populations (Fuchs et al., 2003). Consequently, our results highlight the importance of preserving natural habitats and managing agricultural landscapes in a way that safeguards the genetic variability of native flora (Kamm et al., 2009; Van Geert et al., 2010).

Pollen metabarcoding is a valuable tool to assess the diversity of plant taxa available to insect visitors; however, this information is largely qualitative (Bell et al., 2019) and can be subject to amplification, sequencing and data processing biases (Arstingstall et al., 2021). While this method enabled us to characterize potential foraging resources associated with farmland near protected montane forests in Costa Rica, distinguishing pollination from other types of interactions (i.e., neutral or resource parasitism) remains a challenge. Future studies that combine direct measurements of pollinator effectiveness (e.g., via studies of single-visit deposition King et al., 2013) with pollen metabarcoding to identify plant sources in flower visitors are needed to evaluate the extent of pollination services provided by the diverse community of flower visitors observed in tropical montane agrosystems.