1. Optimal Defense Theory (ODT) predicts that to maximize the benefits of defense against herbivores while minimizing its costs, plants will invest in defenses to structures according to their value and to the likelihood that they will be attacked. Constitutive defenses are expected in structures of high value, whereas induced defenses are expected in structures of low value. Regarding the biotic defense mediated by extrafloral nectaries (EFNs) and based on ODT, we predicted that under control conditions EFNs on higher-value structures would produce more nectar than would EFNs on lower-value structures, attracting more ants; however, when damaged, EFNs on higher-value structures would not increase the production of extrafloral nectar (since constitutive defenses should be employed in this region), whereas EFNs on lower-value structures would so (since induced defenses should be employed in this region), at a level commensurate with the extent of damage. 
2. Here we test these predictions in a Brazilian ant-plant mutualism. Qualea multiflora (Vochysiaceae), a savanna tree, presents EFNs on both lower-value structures (leaves) and higher-value structures (inflorescences). We simulated herbivory by cutting 10% or 40% of the leaves, or 10% of the flowers, then monitoring extrafloral nectar production and ant attendance. 
3. Extrafloral nectar volume and calorie content, as well as ant abundance, were higher in EFNs of inflorescences compared to EFNs of leaves both before and after simulated herbivory, consistent with one of our predictions. However, EFNs on both leaves and inflorescences, not leaves only, were induced by simulated herbivory, a pattern opposite to our prediction. Plants subjected to higher levels of leaf damage (i.e., more damage to lower-value tissues) produced more and higher-calorie extrafloral nectar, but showed similar ant abundance, partially consistent with our prediction. 
4. Our results show that extrafloral nectar production before and after simulated herbivory, as well as the ant recruitment, vary according to the plant structure on which EFNs are located. Our study is unique showing that ant recruitment via extrafloral nectar follows predictions from Optimal Defense Theory, and that the ant foraging patterns may be shaped by the level and region damaged in the plant.

We selected 45 plants with similar phenotypic characteristics (1.5-2 m in height, producing leaves with EFNs but not yet producing flowers) and at least 10 m apart. We randomly allocated the plants to one of three treatments (N = 15 plants per treatment; treatments are summarized in Table 1). In the first treatment (Foliar control), no manipulation was conducted. In the other two treatments, herbivory was simulated by cutting the apical part of leaves with scissors. In one treatment (Foliar 10%), 10 % of the leaf area was removed from all leaves (including young and mature leaves) of each plant, and in the other treatment (Foliar 40%), we removed 40 % of the leaf area from all leaves in each plant. 

In addition, we selected a different set of 30 similar plant individuals (1.5-2 m in height, developed leaves, 10-15 inflorescences), but now during flowering and without EFNs active on leaves and with EFNs active on inflorescences. We randomly allocated the plants to one of two treatments (N = 15 plants per treatment). In the first treatment (Floral control), no experimental manipulation was conducted; in the second treatment (Floral 10%), we cut 10 % of the apical part of all buds and flowers of each plant with the aid of a pair of scissors. We did not do a 40 % cutting treatment on buds and flowers as we had on the leaves, due to the small size of floral buds and the associated difficulty in their handling.

Leaves, buds and flowers were cut in the evening at 2100 h, during the period of highest productivity of extrafloral nectar in Q. multiflora (Lange, Calixto, & Del-Claro, 2017). Simulated herbivory has been used in many studies to test induced plant responses, including the response of extrafloral nectaries (Heil, Fiala, Baumann, & Linsenmair, 2000; Jones & Koptur, 2015; Wäckers & Wunderlin, 1999). In the case of Q. multiflora, natural foliar herbivory rates vary from 2.64 ± 1.9 % (mean ± SD) in ant-attended plants to 8.16 ± 4.08 % in plants without ants (Calixto et al., unpublished). Thus, our treatments mimicked natural herbivory rates. 

On each individual, we selected one EFN. If studying leaf EFNs, we selected an EFN on the adaxial surface of a young leaf near the apical meristem (Fig. 1b), and if studying inflorescence EFNs, we chose the most basal EFN of an inflorescence (Fig. 1c). The marked EFNs were isolated with a mesh bag and Tanglefoot resin strip (Tanglefoot Company®, Rapids, Michigan), decreasing dilution by rain and dew and reducing access to and removal of nectar by ants and other arthropods. Both factors (dilution by rain and dew, and removal by ants and other arthropods) might influence the amount of nectar present at the time of assessment. Foliar experiments and data collection occurred during October, while Floral experiments occurred in January.

Nectar produced in all selected EFNs on plants in all five treatments was collected 1, 6, 24, 48, 72, and 96 hours after cutting (method adapted from Heil, Fiala, Baumann, & Linsenmair, 2000). At each census, we measured the volume of nectar produced and the quantity of sugar (Brix % - mg sugar per ml solution) with the aid of 5μL graduated microcapillary tubes and manual refractometer (Eclipse® model, Bellingham & Stanley, Tunbridge Wells, UK). All evaluated EFNs were washed with distilled water and dried with filter paper immediately after simulated herbivory and after each evaluation. During evaluations, we recorded ant abundance and richness on plants at the time of nectar collection. An individual of each ant species was collected, fixed in 70% alcohol, and identified with confirmation by specialists from the Universidade Federal do Paraná, in Curitiba, Brazil. Data on ant identity are presented elsewhere (Supporting Information Table S1).