Cacao flower visitation: Low pollen deposition, low fruit set and dominance of herbivores

1. Pollination services of cacao are crucial for global chocolate production, yet remain critically understudied, particularly in regions of origin of the species. Notably, uncertainties remain concerning the identity of cacao pollinators, the influence of landscape (forest distance) and management (shade cover) on flower visitation and the role of pollen deposition in limiting fruit set. 2. Here, we aimed to improve understanding of cacao pollination by studying limiting factors of fruit set in Peru, part of

5. The diversity and high relative abundances of herbivore flower visitors limit our ability to draw conclusions on the functional role of different flower visitors. The remarkably low fruit set of naturally and even hand pollinated flowers indicates that other unaddressed factors limit cacao fruit production. Such factors could be, amongst others, a lack of effective pollinators, genetic incompatibility or resource limitation. Revealing efficient pollinator species and other causes of low fruit set rates is therefore key to establish location-specific management strategies and develop high yielding native cacao agroforestry systems in regions of origin of cacao.

K E Y W O R D S
agroforestry, cocoa, flower visitors, forest proximity, hand pollination, pollen, pollination services, shade cover INTRODUCTION Despite pollination services being central to successful fruit production of the cacao tree (Theobroma cacao L.), the underlying processes and limiting factors are still poorly understood (Klein et al., 2008;Toledo-Hernández et al., 2017). This is striking, considering that the tree is an important tropical cash crop used to manufacture chocolate and cacao cultivation sustains ∼6 million farmers globally, most of which are smallholders (Clay, 2004). While being an understorey tree native to the Amazon basin, cacao is nowadays mainly cultivated outside its native distribution range (Thomas et al., 2012). As a consequence, most research on cacao pollination services has been restricted to non-native countries (Toledo-Hernández et al., 2021). Yet, in recent years, cacao production in Amazonian countries has been on the rise (FAO, 2020), but yields of native cacao are often low (Romero & Vargas, 2016). Therefore, identifying limitations of pollination success ( Figure 1) and closing the multiple knowledge gaps concerning fruit set in the native range of cacao is crucial for improving livelihoods of rural smallholders.
Productivity of cacao is, amongst others, limited by the plants' reproductive biology, for example entomophily and low abundances of presumed cacao pollinators reported by older studies (reviewed by Toledo-Hernández et al., 2017). Half of all cacao flower-visiting species worldwide are midges from the Ceratopogonidae and Cecidomyiidae families, yet, relative abundances observed on cacao flowers in Latin America can be as low as 2%, while other visitors such as thrips and ants have been found to be more abundant (Chumacero de Schawe et al., 2016;Toledo-Hernández et al., 2021). For example, in a study in Indonesia not a single Ceratopogonid was trapped visiting flowers (Toledo-Hernández et al., 2021). Owing to the variation in observed visitation patterns across study locations, the taxonomic identity of the main pollinators remains debated; it is likely that several arthropod taxa beyond midges contribute to pollination in cacao. Studying patterns of flower visitors across different cacao geographies is thus crucial to clarify pollination potential of different insects, as to improve pollination services.
Landscape properties and management features are known to drive pollination services of tropical agroforestry crops, including cacao, but patterns are still not fully understood. In Asia, flower visitation by potential coffee pollinators increased with forest proximity (Klein et al., 2008), but thus far, no such association has been detected for cacao (Toledo-Hernández et al., 2021). The integration of shade trees in cacao agroforests can provide multiple economic and ecological benefits (Blaser et al., 2018;Jezeer et al., 2017), such as increased Dipteran visitation rates under higher canopy closure detected in Indonesia (Toledo-Hernández et al., 2021). However, forest distance and shade cover patterns remain to be studied in cacao agroforestry outside of Asia.
Cacao yields also depend on characteristics of pollen deposition: Only a small fraction of the thousands of flowers receives a sufficient quantity of pollen to result in fruit set (Groeneveld et al., 2010).
Because low pollen deposition can be linked to suboptimal cacao fruit set (Falque et al., 1996;Mena-Montoya et al., 2020), it is important to better understand the link between pollen deposition rates in the field and actual fruit setting rates. Limiting effects of pollen quantity and compatibility on yield can be alleviated by hand pollination (Toledo-Hernández et al., 2020), particularly so in self-incompatible cacao varieties . Manual pollen supplementation has been found to triple yields and increase cacao farmers' incomes by up to 69% (Toledo-Hernández et al., 2020). However, yield gains through hand pollination depend on environmental factors, cross-compatibility levels and timing (de Almeida & Valle, 2009;Forbes et al., 2019). Successes also might fluctuate locally, but no large-scale studies have addressed hand pollination gains in countries of origin of cacao. Resource availability F I G U R E 1 Conceptual overview of the cacao pollination process, depicting several steps preceding fruit set (ovals), including relevant drivers and limitations (rectangles). Variables addressed in this study are highlighted in grey. Insect visitation is necessary for pollen deposition and may depend on a plethora of factors, such as farm and landscape-level management such as canopy closure and forest distance (Toledo-Hernández et al., 2021). Pollen deposition can be influenced by visitation rates of insects and the amount and quality of pollen carried by different visitor species. When sufficient viable and compatible pollen is deposited on the style of a cacao flower, pollen tubes are formed, and the sperm nuclei migrate to the ovary for fertilization (Claus et al., 2018;Falque et al., 1995). Finally, pollen compatibility and resource availability can affect setting of fruits even until after fertilization (de Almeida & Valle, 2009;Ford & Wilkinson, 2012) gradients of forest distance and shade cover in biogeographically distinct regions; (Q2) how much pollen is deposited during natural pollination and how does this affect fruit set rates in the field; and (Q3) to what extent does hand pollination improve cacao fruit set rates. Drawing on our findings, we discuss next steps to improve knowledge on pollination services in smallholder agroforestry systems in cacao's native range.

Study regions
We conducted our research in two cacao-growing areas in Peru with a distinct climate, vegetation type and biogeography: the dry north-ern lowlands, west of the Andes, and the humid south-eastern Andean slopes. The study area in the north was located around the farmer community of La Quemazón, in the department of Piura, in the coastal northwest of Peru (S5.312249 • , W79.718996 • , 240 m.a.s.l.; Figure   S1a) where the local variety, Piura white cacao, is cultivated under irrigation. The area is characterized by the dominance of seasonally dry tropical forest vegetation and the climate is hot and semi-arid (SENAMHI, 2020a). Annual rainfall averages to 235 mm per year. Most of the annual rainfall (235 mm) occurs during the short, wet season from December until March. In the dry months, rainfall is close or equal to 0 mm.

Site selection and characterization
In the northern study region, 12 smallholder organic cacao agroforests were selected, between 0.2 and 2 ha in size, consisting of 5to 10-year-old trees mainly from the native Piura white cacao. During the dry season, these agroforests are irrigated every 15-20 days by means of gravity-fed flood canals. In the southern study region, we selected eight organic smallholder agroforests, smaller than 3 ha and ranging between 5 and 65 years old. Here, gravity-fed flood canals and aspersion were used for irrigation, mainly during the dry season.
We calculated forest proximity, that is the shortest distance from each study site to the nearest forest (km) using ArcMap 10.5.1. To this end, we used updated versions of land-use map of Piura in the north (Otivo Barreto, 2010) and the vegetation cover map of Cusco in the south (MINAM, 2015). Canopy closure, assessed with a spherical densitometer, was used as measure for shade cover. For the northern agroforests, we averaged canopy closure over 25 readings spread out over an area of about ∼0.2 ha, and in the southern agroforests, we averaged 20 readings over ∼0.15 ha, to account for slightly larger subplot sizes in the north. Cacao tree density and abundance were comparable throughout the study: in most of the agroforests, trees were planted following a 3 × 3 m grid, with few exceptions of 3.5 m grids.

Flower visitors
To trap arthropods visitors of cacao flowers, we applied non-drying, odourless and colourless insect adhesive (Schacht Raupenleim) on the reproductive parts of cacao flowers (mainly around the style), between 5:15 AM and 11:30 AM. We retrieved the flowers about 24 h later. In the north, we sampled flowers during the dry season (Oct-Dec), and in the south, during the rainy season (Jan-Feb) in 2018/2019.
All agroforests were sampled three times, with minimum 4 and maximum 40 days between sampling rounds. During each sampling round, we selected 50 flowers distributed among 10 trees and covered the reproductive parts with glue, totalling to 150 flowers per agroforest.
Upon flower retrieval, 24 h after glue application, most of the flowers had abscised, a process that is normal in cacao (24-36 h; Toledo-Hernández et al., 2017). Therefore, not all flowers could be recollected and numbers of retrieved flowers differed among trees and farms (Table S1). Arthropod specimens were retrieved from the flowers, and sorted into morphological and functional groups, based on general taxonomic keys (Gibb & Oseto, 2006) and keys to family level for Diptera (Brown et al., 2009). Cecidomyiidae and Ceratopogonidae were lumped, representing potential cacao-pollinating midges, hereafter referred to as midges. Other dipteran families were categorized as other Diptera; Hymenopterans were either classified as parasitoid wasps, ants or other Hymenoptera.

Pollen quantity
To study how pollen deposition affects fruiting success in northern Peru, we took ultra-macro photographs of flowers directly on the tree and estimated the amount of pollen grains deposited on the style, following Macinnis and Forrest (2017).
Pollen deposition is usually quantified destructively, that is by removing pollinated flowers or flower parts. Here, flowers were monitored whilst developing further on the tree and as such, we avoided the risk of interfering with pollination success. We used a DSLR camera with ultra-macro lens (LAOWA, five times magnification) and a LED lamp and ring to increase light intensity. Photographs were taken at ISO 400 with shutter speed 1:40 and aperture F8. Of each flower, two series of photographs with different focusing depth were used for capturing the two opposite sides of the style ( Figure S2).
We took 7704 macro photographs of 518 flowers, spread over five agroforests and different shooting days. Data of two consecutive years were included (Table S2)

Statistical analyses
All statistical analyses were performed with R (R Core Team, 2020); plots were built with the package ggplot2 (Wickham, 2016). Spatial analyses and maps were performed and created with ArcMap 10.5.1.

Flower visitors
We used generalized linear mixed effect models (GLMM) with the package lme4 (Bates et al., 2015) to investigate the effect of region, distance from forest (km) and canopy closure ( (Table S1). Because surveys were conducted during the dry season in the north, and during the wet season in the south, seasonality is implicitly included in region.
In all three models, identity of agroforest was included as random effect variable to account for multiple sampling in each agroforest.
Data from one southern agroforest were excluded from all models, because of incomplete canopy closure assessments (Q14; Table S1; Figure S1b). Aphid visits were modelled with a Poisson distribution.
Due to over-dispersion in the models constructed for thrips and other visitors, we used a negative binomial distribution. All model residuals were inspected with package "DHARMa" (Hartig, 2018); no significant deviations were detected.
In our models, we integrated the differences in retrieved flowers per agroforest by including this value as offset, which is a good way to standardize count data of visits per flower (Reitan & Nielsen, 2016).
For plotting, we used visitation rates (i.e. total visitors/retrieved flowers) instead of total visitors, and held the offset held constant at one to obtain predictions that are easy to compare.

Pollen quantity
We recorded extremely low fruit sets during the experiment: the proportion of successes and failures was unbalanced (1:128). Although unbalanced data is a common phenomenon in ecological data (Salas-Eljatib et al., 2018), the success events were too rare to perform any meaningful statistical analysis.

Hand pollination
To examine differences in fruit set rates (proportion ranging from 0 to 1) between naturally and hand pollinated flowers, we used a generalized linear mixed model (package "lme4"). Fruit set rates were pooled over seven counting rounds and compared between pollination treatment (fixed effect variable) using a binomial distribution, whereby the total number of open flowers was included as weights argument.
DHARMa residual plots signalled no model violations. Since counts of cherelles and flowers were performed on eight trees per farm (Table   S3), we included trees nested in farms as random effect variables. Trees with incomplete counts were excluded: only 93 were considered in this analysis (N Manual = 90, N Natural = 91; Table S3).
Overall, visitation rates of flower-visiting arthropods increased along higher canopy closure in the north and decreased in the south, whereas forest distance did not play an important role in flower visitation patterns (Table S4) (Table S4)
On these four flowers, an average of 111 ± 19.2 pollen grains were deposited, while an average of 30.7 ± 1.2 pollen grains were deposited on styles of flowers that did not set fruit (n = 513).

Hand pollination
Fruit set was remarkably low in both pollination treatments, but significantly higher for hand-pollinated flowers (GLMM: z = −6.76, P < 0.001; Figure 4;

DISCUSSION
In this study, we aimed to reveal key drivers of cacao pollination services ( Figure 1)  likely that their net effect on fruit set is neutral or adverse (Entwistle, 1972). Aphids are likely to negatively affect fruit set, because of their sap-sucking diets and association with honeydew-collecting ants (Maas et al., 2013). Thrips might contribute to pollination mainly through their high relative abundances which may compensate for the minimal amount of pollen they typically carry with their hairy-fringed wings, although a substantial part of pollen transported by thrips might be self-pollen (Entwistle, 1972;Mound, 2005). In our study, the functional role of midges, aphids and thrips remains unconfirmed. In the light of these uncertainties, methodologies that allow to demonstrate transport of outcross-pollen should be developed to confirm functional roles of flower visitors in future investigations.
The lack of a strong relationship between forest distance and visitation rates was contrary to our expectations of finding higher visitation rates in forest vicinity, as was the case in studies carried out in Asia (Klein et al., 2008;Toledo-Hernández et al., 2021). Possibly, other management variables, such as canopy closure and habitat management, play a bigger role in insect visitation to flowers of native cacao.
In the north, visitation rates tended to be associated with increasing canopy closure, while in the south, during the wet season, an opposite trend prevailed. Shade trees decrease transmitted radiation, lower air temperatures and increase relative humidity (Niether et al., 2018;Tscharntke et al., 2011). Especially under intensely dry circumstances as in the north, buffering of extreme environmental conditions in the agroforests could have benefited flower visitation. In the south, the high cloud cover during the wet season might have limited transmitted radiation. Under denser canopies, the radiation could have been below the threshold necessary for insects to visit flowers (Liporoni et al., 2020).
We were not able to relate fruit set with pollen quantities measured directly on cacao trees in the northern study region, because fruit set rates were extremely low (0.8%) compared to the 10% reported from Indonesia (Groeneveld et al., 2010). This could be problematic for final yields, because in cacao, the majority of pollinated flowers do not develop into harvestable fruits (Bos et al., 2007). Considering that we observed several cases of pollination failure in spite of high amounts of pollen deposited, other factors such as pollen viability, pollen compatibility and resource availability may be limiting fruit set even more than previously thought. Pollination failures are also commonly caused by low pollen viability (Wilcock & Neiland, 2002) and viability in turn can be affected by high temperatures and drought. Potentially, extraordinarily high temperatures in our northern study region have induced more pollination failures than expected. Alternatively, and more likely, the narrow genetic basis of the native variety used for our experiments (Thomas et al., 2012) resulted in limited compatibility (Rodger & Ellis, 2016), while climatic conditions could have aggravated fruit set failures. It is critical that future studies aim to understand the relative contributions of pollen quantity, resource availability and compatibility to pollination failure to allow designing locally adapted (hand-)pollination strategies that improve fruit set.
The average pollen deposition on freely pollinated flowers (30 grains) was much lower than the threshold for pollination success (115 grains) established from experimental evidence (Falque et al., 1995) (Bos et al., 2007). Properties of cacao varieties might influence contrasts between continents: outside of the Americas, plantations consist mainly of hybrid varieties bred in clonal design for steady production and auto-compatibility (Zhang & Motilal, 2016), whereas productivity of the native variety we studied is more variable, and potentially more reliant on cross-pollination than hybrid varieties. Conducting inter-and cross-compatibility trials with planted varieties to maximize gains is therefore strongly recommended. In the light of pollinator uncertainty, hand pollination could be applied to mitigate pollen limitations in the field and improve fruit set rates, though thorough assessments would be needed to calculate yield gains in the longer term.

CONCLUSION
Despite years of intensive research on the pollination services in cacao, multiple knowledge gaps remain, underpinning the difficulty of related research. Based on the dominance of herbivore visitors and the low pollen deposition and fruit set rates we found, we urge the confirmation of the main cacao pollinator in regions of origin of cacao, and the cause of low fruit set rates. Our results demonstrate that with hand pollination, it is possible to alleviate fruit set limitations, although only partly. The limited hand pollination gains in native cacao might be due to pollen incompatibility-and it will be crucial to determine the relative importance of limitations other than pollen quantity (i.e. pollen compatibility and resource availability) to increase fruit set rates. Confirming pollinator identity will also be key to make recommendations on farm and landscape management to maximize visitation rates. To this end, we recommend combining new and existing techniques to study pollen deposition quantities of different arthropod visitors, permitting the development of management interventions to maximize the visitation rates of the groups that deposit sufficient viable and compatible pollen.

INCLUSION STATEMENT
Several authors from different countries collaborated on the work presented in this study, conducted within a larger collaborative framework in Peru. Hypotheses and research questions were developed after stakeholder meetings with local universities, farmers' organizations and governmental institutions ensuring applied relevance of the research. Peruvian students and field assistants made a significant contribution to the implementation of field experiments and data collection. Further, preliminary results have been presented to regional farmer communities and printed results were distributed to farmers participating in the project. Local literature was consulted whenever relevant.

ACKNOWLEDGEMENTS
We thank all farmers from Norandino Ltda. for granting permission to conduct various experiments on their land, as well as Diego P.
Zavaleta and Fredy Yovera for their assistance with logistics and organization. We thank Mathil Vandromme for her help with insect identification. Lisanne Abts, Dror Noe and Steffen Dammeyer deserve our gratitude for assistance with collecting insects, photographing flowers and counting pollen and Cristina, Gemma, and Elena for proofreading the Spanish abstract. Thanks are also due to two anonymous reviewers and the editor, Dr. Marc Cadotte, for their insightful suggestions.
We are grateful for the financial support from the Federal Ministry for Economic Cooperation and Development of Germany (GIZ contract number 81219430) and from the CGIAR Fund Donors. The Open Access Publication Fund of the University of Wuerzburg supported this publication.
Open access funding enabled and organized by Projekt DEAL.

CONFLICT OF INTEREST
The authors declare no conflict of interest.

PEER REVIEW
The peer review history for this article is available at https://publons.