Niche theory predicts high specialization in species-rich environments such as neotropics, where plant-herbivore interactions are ubiquitous in all terrestrial systems. However, there is much to be explored on how different insect guilds respond to plant phenology in these specialized environments.

 Using a Fabaceae-herbivore community as an ecological model, we evaluated how plant phenology affects the abundance and shapes the temporal distribution of exophytic (folivore chewers) and endophytic (seed chewers) insects. We tested the specialization hypothesis, predicting insect species on only one or a few host plants, with a seasonal pattern of occurrence, many rare and few abundant species. Additionally, we analyzed the relationship between plant phenology and environmental factors (temperature and precipitation). We conducted fieldwork with five species and 97 individuals of Fabaceae in Brazilian Savannah and performed an insect-rearing study in the laboratory.

We found an immediately synchronized relationship between endophytic insects and plant phenology in the dry season, whereas exophytic were throughout the year. Only six species from the exophytic guild, out of a total of 87 species for both guilds, fed on more than one host plant, showing a specialized network. Endophytic herbivores are more abundant, ephemeral, and predictable than exophytic ones. We also found a negative relationship between fruiting and precipitation and a positive relationship between new leaves and temperature and precipitation.

Our study shows that the temporal distribution of insect-herbivore guilds responds to plant phenology. In a scenario of global warming with effects on the annual precipitation, endophytic insects are most vulnerable to local extinction due to temporal changes in plant phenology than the exophytic ones.

The study was performed in a Brazilian Cerrado area of the Clube Caça e Pesca Itororó (CCPIU) in Uberlândia city, southeastern Brazil (18°59ʹS, 48°17ʹW). There are two main phytophysiognomies of the savanna formation, the sensu stricto cerrado and the Vereda (Silva et al., 2022), a swampy area that crosses the reserve. Herbaceous plants dominate local vegetation with some trees ranging between 2 and 8 m tall. The climate in the region is markedly seasonal, characterized by a wet (October to April) and a dry season (May to September) (see Sousa-Lopes et al., 2016). The monthly mean temperature during the study period, from October 2016 to September 2017, ranged from 19 (July) to 24.9°C (October), and 80% of the precipitation was concentrated in the wet season (INMET 2019; Fig. 2). We used preexistent trails of CCPIU to locate all plants evaluated in this study.

Visual inspections were carried out monthly from October 2016 to September 2017 on 97 randomly tagged individuals of A. humilis (n= 20), B. rufa (n= 20), C. cathartica (n= 17), M. setosa var. paludosa (n= 20) and S. polyphyllum (n= 20). In the randomization process, we chose a pre-existing trail in the study area and marked the plants found of the previously selected species. From 08 am to 06 pm, we carefully inspected all plant structures, i.e. leaves, flower buds, flowers, and fruits, searching for insects for 30 minutes divided into two sessions per month (totaling 582 hours of observation in the year). As soon as we found insects, specimens of each species were carefully collected with a brush or voile net and then placed individually into transparent plastic pots (500 ml). Pots were then labeled with information about the host plant, plant structure, and date and taken to the Behavioral Ecology and Interactions Laboratory at the Federal University of Uberlândia (LECI-UFU), where insects were reared in laboratory conditions (12-h light and 20–30°C) and with the same plant structure where they were found. When available, we also collected leaves, flowers, and fruits to describe internal insect herbivores (n= 20 plant structures per plant, monthly). We consider true herbivores those insects that were found feeding on structures of plants, in the field or laboratory. The folivore endophytic insects were not sampled, possibly because when we took the leaves to the laboratory, they dried out and the plant tissues hardened and prevented the development of the immature ones, even when moistened cotton was added to the petiole of the leaves. On the other hand, the fruits removed from the plants contain water and nutrients stored in their seeds, which allows the insects to continue their development under laboratory conditions and be sampled. After rearing, insects were identified by specific taxonomic keys at the lowest taxonomic level possible (e.g. Rafael et al., 2012 for family and subfamilies) and with the help of specialists (e.g. for some Bruchinae, Dr. Cibele Stramare Ribeiro Costa). Specimens of insect herbivores were deposited at the LECI-UFU and Coleção Entomológica Padre Santiago Jesus Moure (DZUP), Departamento de Zoologia, Universidade Federal do Paraná.

The number of new (not fully expanded) and mature leaves (fully expanded), flower buds, flowers, and fruits were also monthly recorded for each of the 97 previously tagged individual plants after searching for insects.

We used the package bipartite from R 3.5.3 (R Development Core Team, 2019) to build the network through a matrix with values of the absolute frequencies (total abundance) for each herbivore on its host plant throughout the year. In that, the strength of interactions, calculated by frequencies of insect herbivores, indicates weak (the lines of the graph are narrower) or strong relationships (the lines of the graph are thicker) (Tylianakis et al., 2010). We calculated the degree of specialization of the network (H₂’), whose values range from 0 (no specialization) to 1 (perfect specialization; Blüthgen et al., 2006). Modularity (Q) was calculated through the QuanBiMo method (Dormann & Strauss, 2014). This metric is characterized by the presence of species subsets more densely connected among themselves than with the other species in the network, also named modules/compartments (Dormann et al., 2009). The number of steps taken for the modularity analysis was 10x10⁵. The significance of metrics (H₂’ and Q) was estimated through the comparison with 1000 null networks generated using the Patefield algorithm (Patefield, 1981).

The circular analysis from Oriana (Kovach Computing Services, Pentraeth, Isle of Anglesey, UK) was used to describe the phenology and calculate the mean month for new leaves and fruits in the plant community. We used only new leaves and fruits because these phenophases were with higher insect herbivores abundance. The months were converted into angles (30° intervals) and combined with the respective values of each phenophase (Sousa-Lopes et al., 2016, 2019b). The monthly percentage of each phenophase was calculated by dividing the value of the phenophase in the month of interest by the sum of the values of the same phenophase throughout the year. Pearson correlations with time lag from one to three months, from October 2016 to September 2017, were run to test the effect of temperature and precipitation on the number of new leaves and fruits and new leaves on leaf chewers and fruits on seed chewers. These guilds of herbivores were chosen as they were the most abundant in this study. We also used the package DataCombine from R to perform the correlations. Data were previously checked as to their normality and variance with the Shapiro-Wilk and Levene tests, respectively. OriginPro (OriginLab, Northampton, Massachusetts, USA) was also used to perform figures.

Field data were insert into thirteen tabs in the spreadsheet. The first shows the abundances of the herbivores (columns) and their host plants in the rows. N/A means not applicable (not found). The second shows the abundance of herbivores from October 2016 to September 2017 in the columns and the associated plants in the rows. The third elucidates the phenology of the host plants over time and the number of flower buds, flowers, fruits, young leaves and mature leaves for each of the study species. The fourth tab shows the phenology of all the plants, accumulated by phenophase on a monthly basis (columns). The fifth shows the temperature and rainfall data over the aforementioned period. The correponding units for the variables "average temperature" and "total monthly rainfall" are ºC and millimeters, respectively. The sixth to tenth tabs show the abundance of herbivores per host plant over time (columns), each tab containing an individual plant: 6 - Andira, 7 - Bauhinia, 8 - Chamaecrista, 9 - Mimosa and 10 - Stryphnodendron. The eleventh tab shows the abundance of all the insects found over time (columns). The twelfth tab shows the abundance of insects that were considered tourists, i.e. they did not feed on any of the phenophases of the plants studied. The thirteenth tab shows the young leaf and fruit phenophases and the abundance of herbivores associated with them over time. Fruit data and the abundance of herbivores associated with this phenophase are available with the Log 10 transformation.