Zoophilous flowering plants have features to attract their pollinators, which may also be used by exploiters like nectar robbers. Nectar robbers access nectar by probing flower corollas (primary robbing) or using existing holes (secondary robbing). Nectar robbing can negatively impact a plant’s fitness directly by damaging the reproductive structures of the flower or indirectly by modifying the visitation patterns of pollinators. We tested the hypothesis that the robbed flowers are less visited by legitimate pollinators by comparing floral visitors and visitation frequency in primary-, secondary- and un-robbed flowers of Tecoma fulva spp. altoandina, a native shrub species that is an important nectar source for nectarivores in the dry valleys of the Bolivian Andes (Bignoniaceae). The Giant Hummingbird, Patagona gigas (Trochilidae), was the main visitor of T. fulva, and visited more frequently un-robbed flowers, followed by secondary robbed and primary robbed flowers. The variation in visitation frequency may result in less pollen transport to robbed flowers, which may have negative consequences for biological fitness, probably manifested in decreased seed production.

We conducted the study in the municipality of Mecapaca (16.639818º S and 68.015356º W; see Figure S1) near La Paz city, at ~2900 m asl in the Andean dry valley region of Bolivia (Beck et al., 2015). The average annual temperature is ~15 ºC and annual precipitation is ~400 mm (SENAMHI 2017). According to Schmida (1985) this region comprises a semiarid area. The dominant vegetation is scrub, with shrubs, herbaceous plants, and columnar cacti (López, 2003). Cacti such as Corryocactus melanotrichus and Trichocereus bridgessi, thorny shrubs such as Colletia spinossisima, micro-foliated and/or thorny trees, and shrubs such as Neltuma alba, Vachelia acuminate, and Schinus areira are common (Beck et al., 2015; names were actualized according to Catalogue of Bolivia from Tropicos: http://legacy.tropicos.org/Project/BC). Sunrise was between 6:30 - 6:45 h and sunset was between 18:00 - 18:15 h.

Tecoma fulva spp. altoandina (Bignoniaceae), T. fulva hereafter, has a disjunct distribution between 2200 and 3600 m asl, from southern Peru in the Huancavelica and Apurimac valleys, and northern Bolivia in the provinces of Camacho (Sorata locality) and Murillo (municipalities of Nuestra Señora de La Paz, Mecapaca and Luribay) (Wood, 2008; see Figure S1 its distribution in Bolivia). It is the only subspecies present at elevations of >2700 m. It is a shrubby plant but can become arborescent (Beck and Zenteno-Ruiz, 2015). Its funnel-shaped flowers have relatively long corollas (up to 8 cm long) and a red-orange color (Wood, 2008), typical of flowers pollinated by hummingbirds and also used by nectar robbers (Rojas-Nossa et al., 2016; Wilson et al., 2004). The stamens are slightly exserted, with the most exposed pair longer than the inner pair. Its flowers have styles longer than the stamen filaments and the corolla is longer than in the other subspecies (Wood, 2008), suggesting a low probability of autogamy (Biernaskie and Cartar, 2004). An individual plant can produce > 100 flowers in a year, depending on its size, which can reach 3 m in height (Beck et al., 2015). It presents flowers asynchronously throughout the year, with flowering peaks between March and May, as well as August to September in some years (Pacheco et al., 2015, and unpublished data).

Tecoma fulva in our study area show flowers with diurnal anthesis, moderate nectar volume (mean±SD: 8.16 ± 2.78 µl) and moderate nectar concentration (26.9 ± 3.07 ºBx), with peaks at the end of the afternoon (Figure S2 shows nectar volume and concentration during the day), typical nectar features in ornithophilous flowers (Faegri and Van der Pjil, 1979; Fenster et al., 2004; Wilson et al., 2004). These nectar features are also common in flowers subject to bird- and bee-mediated nectar robbing (Genini et al., 2010; Rojas-Nossa et al., 2016, Pelayo et al., 2017, Tie et al., 2023).

2.2. Sampling design

Sampling was conducted between May and August 2018, focusing on approximately 50 T. fulva individuals occupying an area of ~2 ha, although not all of them were in bloom. The site was at the edge of a rough road with almost no vehicular traffic.

In May, we selected 20 adult plants between 1.5 and 3 m tall, with at least 30 flowers each, for observations of floral visitors. Type of visit (legitimate - anther and stigma contact, primary robbing, secondary robbing, pollen collection, corolla contact, stigma contact, and antagonists visits; Figure 1) and the number of visits were quantified, both per plant (considering all available flowers per plant) and per marked flower of different types (unrobbed, primary robbed and secondary robbed). Assuming the possibility of nectar replenishment after nectar removal by pollinators or nectar robbers (Ordano and Ornelas 2004, Lou et al. 2014), we established marked flowers of the second day of anthesis with different types as starting flowers for our observations. Assignment of flowers to different types (unrobbed, primary robbed, and secondary robbed) was done before sunrise (5:30-6:30 am). Visits to flowers were recorded in 30-minute periods, every two hours, from 7:00 to 18:00 h, on five flowers per type of flower of 20 individual plants (300 flowers observed), making a total of 3.5 h of observation per day and an overall total of 70 hours in 10 days. The records were made by direct observation, at approximately 10 m from the observed plant, using binoculars.   

Marked flowers from different inflorescences were examined to determine whether they were unrobbed, primary robbed, or secondary robbed; with five flowers per type of flower. Flowers categorized as primary robbed already had a hole or cut at the base of the flower, either in the bract or corolla. Secondary robbed flowers also had a hole as a result of primary robbing, but secondary robbing was simulated by removing as much of the remaining nectar as possible with a medical syringe, avoiding damage to the flower structure. Unrobbed flowers did not show any of the features of primary or secondary robbed flowers and, to ensure that they were not robbed throughout the day, they were checked periodically for nectar robbing holes. Occasionally some flowers changed from unrobbed to robbed flowers or from primary to secondary theft, as some robbers visited them. These flowers with their new status were excluded from their respective flower type for analysis, but this only occurred in some plants. To evaluate the effects of robbing on nectar production, 60 flowers were marked on their first day of anthesis with (28 flowers) and without natural nectar robbing (30 flowers) at the end of the day (17:00-18:00 h). On the morning of the following day (7:00-9:00 h), nectar was extracted from these marked flowers to quantify the volume of nectar available under natural conditions (unbagged flowers).

2.3. Data analysis

The flower number of visits at the plant level was analyzed using Generalized Linear Models to evaluate differences in visitation rates to flowers between visiting species and time of the day. For this analysis, residuals were adjusted to a negative binomial distribution with a “log” link. We did not divide the number of visits by flower number and time of observation to obtain visit rates as response variable; instead, the number of observed flowers and time of observation were assigned as offset variables in the models, as the former varied between plants and observation time was restricted to 0.5 h every 2 h (Reitan and Nielsen, 2016).

The indirect effects of nectar robbing were evaluated in two ways. First, the difference in nectar volume between robbed and unrobbed flowers was analyzed with a linear model, considering flowers as replicates. Second, indirect effects of nectar robbing on the number of visits by different types of visitors (legitimate, primary robbers, secondary robbers, and pollen collectors) were evaluated with generalized linear mixed-models with Poisson error distribution and the alternative square root link (Buckley, 2001, Bolker, 2015). Individual plants were considered as random factors. The pooled number of visits per flower category (unrobbed, primary robbed, and secondary robbed) was used as the response variable, with 20 replicates per category (i.e., Kohl and Steffan-Dewenter, 2022). In addition, we also evaluated pooled visits to compare unrobbed vs robbed flowers (all primary and secondary robbed flowers). The pooled number of visits was not divided by flower number and time of observation to obtain a visit rate as a response variable; instead, the number of observed flowers and time of observation were assigned as offset variables in the models, as the former varied between flower types and observation time was restricted to 3.5 h (Reitan and Nielsen, 2016). As we found differences among legitimate visitors, we also conducted separate analyses for each legitimate visitor, using a similar approximation. In all cases, residuals were tested to verify normality. To test for overdispersion in the residuals, we calculated the dispersion parameter (residual deviance/degrees of freedom) and considered values less than 1.5 as indicating no overdispersion (Carruthers et al., 2008).

All models were selected by comparing the Akaike Information Criterion value for small samples between submodels (Delta AICc<2). When the difference between the AICc values of the best and null models was small, the Likelihood Ratio Test (LRT) was used to verify differences (Anderson et al., 2008). Models that accounted for random effects were also compared against null models with random effects. Paired comparisons between factor levels were done using Tukey post-hoc tests, based on the best models without random effects. Analyses were carried out in the R platform (version 4.2.2; R Core Team, 2022) using packages “lme4” to build GLMMs (Bates et al., 2015), “MuMIn” for model selection (Barton, 2018), “MASS” to build GLMs with a negative binomial distribution (Venables, 2002), and “emeans” for the post-hoc analysis (Lenth et al., 2019).