This study was conducted in an inundated forest dominated by 10-m-tall Metopium brownei (Anacardiaceae) situated in southern Quintana Roo, Mexico (18.4267° N; 88.8043° W; 120 meters a.s.l). The mean temperature varies from 25.5 to 26.5 °C. The dry season lasts from February to May and the rainy season from June to January.

Aechmea bracteata is a large tank bromeliad (leaves ca. 1-m-long; inflorescences up to 1.7-m-long) found from Mexico to northern Colombia. Each plant is constituted of a group of shoots at different stages of maturity that develop from a rhizome (Fig. 1). As the rhizome grows, each shoot begins as a bud, which gives rise to a seedling, and then develops into a shoot whose leaves collect water until the inflorescence is formed, which is followed by the progressive drying out of the shoot before it begins to rot [8]. Each well-developed shoot forms a rosette with numerous reservoirs, where an abundant aquatic fauna develops. In the heart of the rosette, an amphora-shaped leaf grows around the inflorescence, creating a central watertight cavity suitable for housing ants [21,22].

Like most myrmecophytes, A. bracteata can be associated with several ant species, Neoponera villosa and Dolichoderus bispinosus being the most frequent [22]. The selection of host plants by arboreal ants depends on two factors: a “genetically determined attraction” toward a given plant species rather than another, and a kind of imprinting – due to the contact by individuals with their host plant during their larval life and the first days of their adult life – that supplants the genetically determined attraction (the contrary is true for species-specific associations) [24–26]. This process is at the origin of “regional traditions” for nest-site selection in certain arboreal ants and social wasps (i.e. they nest on the same plant species over very wide areas) [19,22,27]. When a given myrmecophyte can be associated with various species of ants, the first formed domatia and those liberated after the death of an ant colony are colonized by winged queens swarming by chance at the right moment. One species of ant wins out through a “lottery competition”, the whole process resulting in a sequential partnering with different species of ants because the lifespan of myrmecophytes is longer than that of ant colonies [28,29].

Concerning the aquatic macroinvertebrate communities, much of the adult biomass of most taxa is obtained during the aquatic part of their life cycle as larvae, illustrating the importance of breeding- and oviposition-site selection. The principal parameters that favor oviposition by adult females followed by the adequate development of larvae are: (1) the size of the tank of the bromeliads [9,30–32], (2) their complexity (i.e. the number of wells corresponding to the number of leaves) [31–34], (3) their color [33,35], (4) the quantity of organic material [31–34], (5) the presence of competitors [9,33], and (6) for certain bromeliad specialists, an attractive odor emitted by the bromeliad or the water in their tanks [34,36].

We examined 92 A. bracteata individuals bearing inflorescences situated in the area studied. We determined the presence or absence of any ant colony in the central watertight cavity of shoots bearing an inflorescence and, when present, the species of ant.

We then examined 30 well-developed Ae. bracteata individuals including three devoid of any associated ant colony. They were selected based on the likelihood of obtaining an invertebrate sample without losing the contents of their tanks (details in Table 1 and Appendix S1). We detached from the rhizome a shoot bearing a mature inflorescence (i.e. with fruits and not yet dry) and noted which ant species it sheltered in the central cavity. We measured the height at which these shoots were located on their host tree and the percentage of incident radiation was quantified using hemispherical photography (see [37] for details). To characterize the habitat structure of the aquatic invertebrates, as per Céréghino et al. [5,6], we measured the water volume in the tanks and several plant parameters (see details in Appendix S1).

Each selected shoot was transported in a sealed plastic bag to the laboratory where it was dismantled in a large bucket and all the water, detritus and invertebrates were poured into wide-mouthed plastic bottles whose contents were preserved in 4% formalin (final concentration) until we were able to sort the macrofauna. Then, the leaves were placed in fine mesh sieves (100 μm) and rinsed to maximize the invertebrate and detritus sampling efficiency. After sorting the invertebrates (see paragraph below), the water was filtered through sieves of different mesh sizes (100 μm–1 mm; 1 mm–10 mm; and > 10 mm) to separate the detritus by size fractions (fine, mid-size, and coarse detritus, respectively). The detritus was then oven-dried (48 h at 60 °C) and weighed (grams; dry weight) (see details in Appendix S1).

The aquatic invertebrates were sorted in the laboratory and preserved in 70% ethanol. They were mostly identified to species or morphospecies and enumerated. Professional taxonomists provided assistance for the identification of the Oligochaeta and other invertebrates (Prof. N. Giani and Dr. A.G.B. Thomas; University Toulouse, France).

Species rarefaction and extrapolation curves were plotted on the abundance data matrices using EstimateS 9.1.0 software [38] with 100 randomizations of the sampling order without replacement. In order to standardize the comparisons between bromeliad categories (classified according to ant presence/absence and species of ant when present) and to estimate sampling completeness, these curves and the Chao1 non-parametric estimator of total species richness were calculated [39].

We calculated Hill numbers (i.e. “true diversities”) for three different orders (q) of diversity. True diversity indices obey the doubling property, preventing a mistaken interpretation of results [40]. Order q is related to the sensitivity of the index to the frequency of the species in the community: when q = 0, all species are given the same weight (rare species are thus favored); when q = 1, species are weighted for their frequency in the community (neither common nor rare species are favored); and when q = 2, more abundant species are favored. Accordingly, species richness is a measure of diversity of order zero (⁰D), the exponential of Shannon's entropy index is the measure of diversity of order one (¹D), and the inverse of Simpson's index is a measure of order two (²D) [40,41]. All three indices are in units of equivalent, equally abundant species and were calculated using Spade R online software [42]. Differences in species richness between habitats were assessed using the non-overlapping of 95% confidence intervals as a conservative criterion of statistical difference in species richness between habitats [38].

The global turnover between the different aquatic macroinvertebrate communities that depends on the species of associated ant was analyzed using the second version of Harrison's indices (β_H2 = [(S/α_max)–1]/(N–1)), [43] as a beta-diversity index obtained using PAST 3 software. β_H2 is an improvement over Whittaker's index (β_W = (S/α)–1), which was modified to be effective in analyzing pairwise differentiation between sites; it is insensitive to species richness trends [44]. We also conducted pairwise comparisons of the diversity profiles of the four situations using “diversity t-tests” based on the Shannon index (PAST 3 software); multiple comparisons were adjusted using the false discovery rate (FDR) (see Table 1) [45].

The environmental parameters recorded allowed us to build a “17 (characteristics of the A. bracteata monitored plus abiotic parameters) × 30 shoots” matrix.

A first Principal Component Analysis (PCA) was conducted to ordinate the A. bracteata shoots based on the environmental variables (e.g., their structure, their height from the ground and the incident radiation; see Appendix S1) and the species of ant they sheltered (or absence of any associated ant colony). Another PCA was conducted to ordinate the A. bracteata shoots based on the aquatic macroinvertebrates sheltering in their tanks according to the species of associated ant (or absence of any associated ant colony). Also, three generalized linear model (GLM) analyses were conducted on the data corresponding to species richness, evenness, and the Shannon index. Another GLM analysis was conducted to determine the effects of ants (presence and species) on predators in the aquatic macroinvertebrate communities. In both cases, the fixed factors were the A. bracteata coordinates on Axes 1, 2 and 3 and the factor “Ant” (species of the associated ant colony or the absence of any associated ant colony), the A. bracteata shoots representing the random factor. These analyses were conducted with the ADE4, nlme and dplyr packages in R software [46].

Of the 92 A. bracteata studied, three individuals (3.26%) were devoid of associated ants, whereas all others sheltered an ant colony in the central, watertight amphora-shaped domatia of the mature shoots. Among these, 13 (14.13%) were associated with Azteca serica, 43 (46.74%) with Dolichoderus bispinosus (both Dolichoderinae) and 33 (35.87%) with Neoponera villosa (Ponerinae). These percentages are intermediary between those for the associations found in another low inundated forest and a semi-evergreen forest in northern Quintana Roo [22].

The low percentage of individuals devoid of any associated ant colony likely results from the much longer lifespan of the myrmecophytes compared to that of their associated ant colonies, so that myrmecophytes shelter several ant colonies successively throughout their life (see cases in [28,29]).

The global beta diversity (Harrison β_H2 = 0.018; a very low value as this index varies between 0 and 1) denoted a large global species overlap (i.e. a small turnover) between the macrofauna in the tanks of the A. bracteata sheltering different species of ants or devoid of any associated ant colony; the same was true for pairwise comparisons between ant species (Table 1). Nevertheless, the multiple comparison of “diversity t-tests” based on the Shannon index showed significant differences for all comparisons involving A. bracteata devoid of any associated ant colony, whereas no significant differences were noted due to ant species (Table 1).

The same was true for the diversity profiles of the aquatic macroinvertebrate communities, those of bromeliads devoid of ants being significantly lower (Fig. 2a).

Because the rarefied species richness of macroinvertebrate communities was not affected by the species of ant (Table 1), we pooled the corresponding data showing that the rarefied species richness tended to an asymptote. Here, too, the diversity profile of bromeliads devoid of ants was significantly lower than that of the individuals sheltering an ant colony (Fig. 2b). Note that the Chao1 estimators of species richness indicated a sampling completeness of 97% (19/19.3) and 94% (8/8.5) for bromeliads with or without ants, respectively (Table 1).

Therefore, these results, where the ant–plant relationships are narrow, complement those of the facultative relationships between ants and the tank bromeliad Aechmea aquilega [20], as, in both cases, the favorable impact of ant presence on the diversity of the aquatic macroinvertebrate communities developing in the tanks of their host plants was noted.

We previously noted that associated ants provide their tank bromeliads with discarded nitrogen-rich debris and feces [19,21]. Because we recorded numerous cocoons in the tanks of A. bracteata individuals sheltering a N. villosa colony, we deduced that the workers not only discarded these cocoons after the emergence of imagoes from the pupae (i.e. new workers), but, likely, also other waste (the two other associated ant species, of the Dolichoderinae subfamily, have naked pupae).

In the PCA ordination of the A. bracteata shoots based on the species of ant they sheltered (or the absence of associated ants) according to the environmental variables selected, the first three components account for 60% of the variance (eigenvalues: Axes 1, 2 and 3 account for 30%, 17% and 13% of the variance, respectively). The parameters corresponding to the structure of the A. bracteata shoots (grouped along the right part of Axis 1) are separated from the parameters “height above ground” and “incident radiation”, which are related (grouped at the bottom of Axis 3) (Fig. 3a–b). Thus, these environmental variables did not influence the distribution of the ant species sheltered by these bromeliads (Fig. 3c–d). This result reflects the fact that the longevity of A. bracteata individuals is longer than that of the ant colonies they shelter (see above) and founding queens are guided to an available plant through a kind of imprinting to chemicals released by the host plant on which they developed [24,25]. Note that these environmental variables can influence the aquatic macrofauna developing in the tanks of bromeliads ([6,7,31,47]; this study, see F1:F3, Table 2A).

In the PCA ordination of the A. bracteata shoots based on the distribution of the aquatic fauna found in their tanks according to the species of ant they sheltered (or absence of associated ants), the first three components account for 43.4% of the variance (eigenvalues: Axes 1, 2, and 3 account for 17.8%, 15.0% and 10.6% of the variance, respectively) (Fig. 3e–f). The A. bracteata shoots sheltering an A. serica colony are separated from the three other cases that overlap (i.e. shoots from individuals without associated ants and individuals sheltering a colony of N. villosa or of D. bispinosus) (Fig. 3e–f).

The GLM analysis conducted on the data corresponding to evenness and the Shannon index resulted in non-significant differences (results not shown), whereas species richness was explained by the species of associated ant or the absence of associated ants (i.e. factor “Ant”; Table 2A). The same was true for the GLM analysis conducted on the data corresponding to the numbers of aquatic predators (Table 2B). The ant effect is at least partially due to the higher presence of predators (mostly Bezzia) in the A. bracteata sheltering an A. serica colony (see Appendix S1).

Therefore, in addition to the impact of ant presence compared to ant absence, the species of ant also plays a role in the composition of the macroinvertebrate communities developing in the tanks of A. bracteata shoots. Indeed, the tanks of A. bracteata associated with A. serica sheltered a community whose richness was different from those associated with D. bispinosus, N. villosa or in the absence of associated ants; the three latter categories did not differ from each other (see Fig. 3d, e). The aforementioned differences can be due to the various food regimes of these ants that, by discarding their feces and waste, fuel the aquatic food webs in the tanks, or may occur if workers of only certain species capture emerging adult insects and/or deter ovipositing females [37,48,49].

Nevertheless, an indirect effect of the ants on their host bromeliad morphology (e.g., size and shape) via the intermediary of the aquatic macroinvertebrate community was not noted in this snapshot study. Indeed, only “height above ground” and “incident radiation”, which are related, can have a slight, indirect influence on the parameters corresponding to the structure of the A. bracteata individuals (Fig. 3a–b). Although this is in line with the results noted for the ant-garden bromeliad A. mertensii, the latter depends on ants for germination resulting in an indirect effect of ants on this plant's size and shape [37]. This is not the case for A. bracteata whose seeds are disseminated by birds [8], so that further studies are needed to verify if there is an indirect ant effect on the fitness of the plant (e.g., the number and/or size of the seeds; see [50]).

There is thus a kind of continuum in the relationships between ants and tank bromeliads as ant presence impacts the aquatic macroinvertebrate diversity developing in the tanks of their host plants in both facultative relationships, the case of A. aquilega [20], and in narrow relationships with the myrmecophytic A. bracteata (this study; see Table 1; Fig. 1a, b). In the latter case, the species of ant also has an impact on the composition of the aquatic macroinvertebrate communities, in line with results noted for the ant-garden tank bromeliad A. mertensii [5,6].

In conclusion, both ant presence (versus absence) and the species of ant mediate the composition of the aquatic macroinvertebrate communities in the tanks of the myrmecophytic bromeliad A. bracteata through direct interactions (i.e. the ants discard their waste and feces into the tanks), whereas we did not note a perceptible indirect effect on the host bromeliad morphology.