Leaf-cutter ants engineer large nitrous oxide hot spots in tropical forests

Abstract

Though tropical forest ecosystems are among the largest natural sources of the potent greenhouse gas nitrous oxide (N₂O), the spatial distribution of emissions across landscapes is often poorly resolved. Leaf cutter ants (LCA; Atta and Acromyrmex, Myrmicinae) are dominant herbivores throughout Central and South America, and influence multiple aspects of forest structure and function. In particular, their foraging creates spatial heterogeneity by concentrating large quantities of organic matter (including nitrogen, N) from the surrounding canopy into their colonies, and ultimately into colony refuse dumps. Here, we demonstrate that refuse piles created by LCA species Atta colombica in tropical rainforests of Costa Rica provide ideal conditions for extremely high rates of N₂O production (high microbial biomass, potential denitrification enzyme activity, N content and anoxia) and may represent an unappreciated source of heterogeneity in tropical forest N₂O emissions. Average instantaneous refuse pile N₂O fluxes surpassed background emissions by more than three orders of magnitude (in some cases exceeding 80 000 µg N₂O-N m⁻² h⁻¹) and generating fluxes comparable to or greater than those produced by engineered systems such as wastewater treatment tanks. Refuse-concentrating Atta species are ubiquitous in tropical forests, pastures and production ecosystems, and increase density strongly in response to disturbance. As such, LCA colonies may represent an unrecognized greenhouse gas point source throughout the Neotropics.

1. Introduction

Leaf cutter ants (LCA; genera Atta and Acromyrmex; figure 1a) are prominent ecosystem engineers throughout Neotropical forests [1–3]. Their foraging and colony construction behaviours relocate plant biomass and excavate soil, altering nutrient distribution and patchiness [3,4], and influencing forest structure, understorey microclimate and regeneration patterns (figure 1b) [5,6]. As dominant forest herbivores [4,7], LCA harvest a wide range of material including leaves, twigs, flowers, fruits, seeds [8] and scientific equipment, and can remove up to 8% of standing live forest foliage annually (equivalent to 100–500 kg ha⁻¹ y⁻¹ [4,9]). The majority of this foraged plant material is relocated to subterranean gardens where it is used as a substrate to cultivate the mutualistic fungus that feeds the colony [10]. The resulting partially degraded plant material is enriched in nitrogen (N), partly as a result of symbiotic N fixation that also occurs in the fungal gardens [11]. This material, along with other colony detritus such as ant necromass, is then transported to designated colony refuse dumps [4,12]. Depending on the LCA species, these refuse dumps can be located in subterranean chambers (many Atta genera, including the widespread A. cephalotes), or as a single aboveground pile in the open or at the base of a tree at the periphery of the colony (e.g. A. colombica [13]; figure 1c). These refuse piles play host to diverse microbial and litter arthropod communities, including species that specialize on Atta waste [4,14,15]. The piles also serve to concentrate carbon (C) and N, with a typical pile (area approx. 1–5 m²) receiving inputs of approximately 70 kg of C and 4.5 kg of N per year [4]. Nitrogen accrued in refuse piles may then be taken up by tree roots [16], or leach and persist in the soil below the piles for months after abandonment [17].

Figure 1.

Atta colombica: typical colony and refuse pile. (a) Atta colombica forager with harvested leaf fragment. (b) Colony area characterized by disturbed soil and defoliated understorey. (c) Aboveground refuse pile consisting of degraded leaf material, fungal biomass, deceased ants and other colony waste. Pile width 1.7 m. Photo credit (a): Manuel Sanchez Mendoza. (Online version in colour.)

Here, we demonstrate an important additional pathway of N processing from LCA refuse piles: extremely high rates of loss to the atmosphere as the greenhouse gas nitrous oxide (N₂O). We characterized N₂O fluxes from Atta colombica refuse piles across primary and secondary tropical rainforest of southwest Costa Rica, analaysed the chemical and microbial properties of refuse to determine the mechanisms supporting such high rates of gas loss, and surveyed pile distribution across the forest landscape to approximate the potential contributions of LCA-generated N₂O to local emissions.

2. Material and methods

(a). Study site

Sampling was conducted in lowland tropical rainforest in southwest Costa Rica, adjacent to the Osa Conservation Piro Biological Station (8°24′42″ N, 83°20′00″ W). The site is a matrix of mature primary tropical rainforest with no recorded history of human disturbance and adjacent regenerating secondary rainforest, regrown from pasture for more than 50 years. Site elevation ranges from 0 to 100 m. a.s.l. and is characterized by Ultisols (Chromic Cambisols) derived from basaltic and andesitic volcanic debris. Mean annual temperature is approximately 26°C and mean annual rainfall approximately 3400 mm with a dry season typically extending January–May, and frequent heavy rains for the rest of the year (peaking in September and October).

(b). Sampling design

Twenty-two Atta colombica Guerin colonies were sampled within an area of approximately 4 km², across primary (n = 9) and secondary forest (n = 13). GPS location, the areal extent of understorey defoliation/soil disturbance around the active colony and areal extent of the single colony refuse pile were recorded. Active colony area was measured along two perpendicular axes and calculated as oval in shape. Refuse pile area was measured as for colonies (if free-standing) or broken up into multiple polygons as necessary to approximate dimensions (if irregular in shape as a result of occurring tree in root wells, adjacent to logs, etc.).

N₂O fluxes were measured using a static chamber technique in June 2017 (n = 10 refuse piles, plus adjacent defoliated colony area and adjacent undisturbed forest for each), October 2017 (n = 18 refuse piles only) and July 2018 (n = 10 refuse piles only). To sample refuse piles, three 22 cm diameter PVC collars per pile were placed approximately 3 cm into the substrate 10 min prior to sampling and topped with chamber lids (vented PVC sewer end caps with butyl rubber gaskets). Headspace samples were withdrawn every 10 min for 0.5 h and placed into pre-evacuated 10 ml exetainers (LabCo, Lampeter, UK). Collar depth was measured at four points and averaged. In June 2017, the defoliated colony area adjacent to each refuse pile (3–10 m away, figure 1b) was sampled by deploying four chambers per colony plus an additional three chambers in undisturbed forest greater than 5 m beyond the colony edge. Over the same period, N₂O fluxes were sampled monthly from 30 collars placed across the surrounding undisturbed primary forest as part of a separate study [18]. Samples were analysed for N₂O, CO₂ and CH₄ concentrations using a gas chromatograph fitted with a methaniser, and flame ionization and electron capture detectors (ECD; Shimadzu GC-2014, Shimadzu, Kyoto, Japan) at the University of Nevada, Reno. The ECD was calibrated with multiple standards spanning 0.1–440 ppm N₂O, bracketing all sample concentrations measured in this study. Fluxes were calculated based on the linear rate of increase in concentration over the four time points.

Refuse material from 10 piles (0–5 cm) and 10 adjacent forest soils (0–10 cm) were analysed to compare chemical properties. Unless indicated, one sample was analysed per pile/site, composed of five homogenized subsamples. In October 2017, triplicate refuse and soil samples were extracted within 2 h of collection by shaking for 1 min h⁻¹ for 4 h in 1:3.75 w/v 2 M KCl and filtered through pre-leached Whatman 1 filter paper (GE Life Sciences, Piscataway, NJ, USA). Samples were analysed for NH₄⁺ and NO₃⁻ colorimetrically [19,20] using a Synergy 2 Microplate Reader (Biotek, Winooski, VT, USA). In June and October 2017, additional replicate samples were dried for 3 days at 65°C to determine moisture content. Dried samples were homogenized in a ball mill and analysed for C and N content using a continuous flow isotope ratio mass spectrometer (Model Delta V Advantage; Thermo-Scientific, Bremen, Germany) at the Cornell University Stable Isotope Laboratory. Sample pH was analysed in a 1 : 2 deionized water solution using an Accumet AP60 electrode (Fischer Scientific, Waltham, MA, USA).

Refuse and soil was kept at 4°C for approximately 3 days and assayed to determine potential denitrification enzyme activity independent of substrate limitation [21]. Samples from 10 piles/sites were combined, aliquoted and incubated with or without the addition of 10% acetylene headspace for 90 min (n = 15 each). Headspace samples were analysed as described above. To determine microbial biomass, replicate samples (8 g) were extracted in 1 : 5 w/v 0.5 M K₂SO₄ for 1 h before and after 3 days of chloroform fumigation and filtered as described above [22]. Extracts were analysed for total organic C and total N using a Shimadzu TOC-V CPN/TNM-1 analyser (Shimadzu Inc., Kyoto, Japan). Microbial biomass C was calculated as the difference in extractable organic C before and after fumigation using a correction factor of 0.45 for the chloroform-labile C pool [23].

(c). Molecular analysis of denitrifying communities

From each of 10 refuse piles in June 2017, a composite of five subsamples was combined and stored at 4°C for 48 h. Genomic DNA was extracted from approximately 0.25 g material using a MoBio DNeasy Powersoil Kit (MoBio, Carlsbad, CA, USA) and stored in elution buffer at −80°C. DNA sample concentrations were quantified using Quant-iT DNA Assay Kit (Invitrogen, Thermo-Scientific, Villebon/Yvette, France) and diluted to a target concentration of 1 ng µl⁻¹ with UP water. Real-time PCR assays were used to measure the abundance of markers for bacterial 16S rDNA and internal transcribed spacer regions of fungi [24–26].

Reactions were carried out using a ViiA7 (Life Technologies, Villebon/Yvette, France). Quantification was based on the increasing fluorescence intensity of the SYBR Green dye during amplification. The real-time PCR assay was carried out in a 15 µl reaction volume containing Takyon low ROX SYBR 2X MasterMix blue dTTP (Eurogentec, Angers, France), 1–2 µM of each primer, and 3 ng of DNA. Two independent quantitative PCR assays were performed for each gene. Standard curves were obtained using serial dilutions of linearized plasmids containing the studied genes. PCR efficiency for the different assays ranged between 77–99%. For each qPCR assay, two to three no-template controls were run. The presence of PCR inhibitors in DNA extracted from soil was estimated by mixing a known amount of plasmid DNA with soil DNA extract or water prior to qPCR. The measured cycle threshold (Ct) values obtained when quantifying the added plasmid DNA were not significantly different between the different soil DNA extracts and the controls with water, which indicates that no inhibition occurred.

(d). Colony density surveys

Field transects were performed to estimate colony density at the study site. For 1.5–5 ha plots (n = 4 each in primary and secondary forest), each encountered LCA colony (of either A. colombica or other locally present species, A. cephalotes) was geo-located. Density at disturbed forest edges was calculated by recording all colonies less than 20 m from the edge of 1.8 km of well-trafficked unpaved road. A literature search was conducted to estimate colony density of Atta colombica and other Atta leaf cutter species throughout Neotropical forests or production systems (see electronic supplementary material, table S1).

(e). Extrapolation

Per-pile fluxes were calculated by multiplying refuse pile areal footprint (electronic supplementary material, figure S3) by average flux rate (mean of n = 3 collars) for that pile. The interquartile range (25th–75th percentile) of pile fluxes was multiplied by A. colombica colony density for three cover types (primary forest, secondary forest and road edges) surveyed in this study (electronic supplementary material, table S1).

(f). Statistical analyses

Statistical analyses were conducted in Graphpad Prism 7.0 (GraphPad Software, La Jolla, USA). Mixed-effects models were performed to test for relationships between N₂O fluxes and refuse pile area, N concentrations, sampling date, CO₂ fluxes and CH₄ fluxes, where the plot was included as a random effect. Where necessary, Johnson SI transformations were applied to normalize gas flux data prior to model building [27]. Welch two-sample t-tests (p < 0.05) were used to assess differences between refuse and soil for chemical and microbial properties.

3. Results and discussion

(a). Flux rates and mechanisms

N₂O fluxes from A. colombica refuse piles ranged widely, from 180 µg N₂O-N m⁻² h⁻¹ up to 89 000 µg N₂O-N m⁻² h⁻¹ (figure 2a; note log scale), and thus represent some of the largest point fluxes of N₂O measured in a natural ecosystem. With a mean flux of 11 200 µg N₂O-N m⁻² h⁻¹ (geometric mean 5830 µg N₂O-N m⁻² h⁻¹, measured from 22 refuse piles in 2017 and 2018; figure 2a), LCA refuse fluxes are often comparable to or greater than point source emissions measured from engineered or intensively managed anthropogenic systems such as anaerobic dairy manure lagoons or wastewater management facilities (figure 2b).

Figure 2.

N₂O flux on and around Atta colombica colonies, and flux comparisons with anthropogenic systems. Note log scale. (a) N₂O (µg N₂O-N m⁻² h⁻¹) fluxes from 22 discrete colonies, with samples taken from adjacent tropical forest (n = 30, green), understorey-defoliated disturbed colony area (n = 40, brown) in June 2017, or refuse piles (n = 104, orange) in June 2017, October 2017 or July 2018. Points are individual measurements and overlay indicates range, median and interquartile range. (b) Range and mean (black bar) for N₂O fluxes from leaf cutter ant refuse piles and a variety of engineered or intensively managed anthropogenic point sources [28–33]. (Online version in colour.)

These fluxes contrasted strikingly with the surrounding mature or secondary tropical rainforest, as well as with the adjacent defoliated colony area where ants create a notable disturbance (figures 1b and 2a). Even when refuse pile emissions peaked in June 2017 at the onset of the wet season, N₂O emissions from the adjacent colony area were not notably different from surrounding forest, and did not exceed 67 µg N₂O-N m⁻² h⁻¹ (mean 14 µg N₂O-N m⁻² h⁻¹; figures 1b and 2a). Instantaneous N₂O fluxes from refuse piles also greatly exceeded those measured simultaneously from the undisturbed forest, by up to four orders of magnitude. Rainforest adjacent to colonies produced emissions of only 11 µg N₂O-N m⁻² h⁻¹ (figure 2a) and annual forest emissions across this forest site average 2 µg N₂O-N m⁻² h⁻¹ year-round [23].

Nitrous oxide is produced predominantly by nitrification and denitrification, microbial processes that are highly responsive to substrate supply and oxygen partial pressure [34]. There are several unique features of refuse piles that may have promoted the extremely high rates of N₂O loss we observed. Firstly, refuse entering the piles was comparatively enriched in N, with a lower C : N ratio (13.5 ± 1; figure 3e,f) than green leaf tissue (21 ± 5 [36]) or leaf litter (32–42 [37]) at this site. While soil typically experiences relatively even inputs from litterfall, ants actively concentrate large inputs of organic matter and nutrients into their refuse piles. Refuse was characterized by five-fold higher total N than surrounding forest soil (2.5% by mass; figure 3f) and 500 times higher concentrations of NH₄⁺ (877 µg g⁻¹ dry material), though not NO₃⁻ (figure 3g,h). Low NO₃⁻ concentrations (with correspondingly high concentrations of NH₄⁺) and evidence of anoxia were consistent with high rates of denitrification, rather than nitrification, as the primary pathway of N₂O loss, but could also be indicative of contributions from dissimilatory nitrate reduction to ammonium [34]. Refuse material was approximately one-third C by mass (figure 3e), and respiration rates (CO₂ flux) from piles exceeded soil by a factor of greater than 20 (mean 2910, maximum 5800 mg CO₂-C m⁻¹ h⁻¹, electronic supplementary material, figure S1a). This suggests that abundant microbially accessible labile C is available to support denitrification [34] and is consistent with observations that decomposition (as measured by mass loss) is greater for refuse material than for original leaf fragments entering colonies [38].

Figure 3.

Properties of forest soil and Atta colombica refuse material. (a) Microbial biomass carbon (μg C g⁻¹ dry material). (b) Bacterial 16S rRNA (copies g⁻¹ dry material). (c) Fungal internal transcriber spacer regions (copies g⁻¹ dry material). (d) Potential denitrification enzyme activity (μg N g⁻¹ h⁻¹). (e) Bulk carbon (mg g⁻¹ dry material). (f) Bulk nitrogen (mg g⁻¹ dry material). (g) Ammonium concentration (μg N g⁻¹ dry material note log scale). (h) Nitrate concentration (μg N g⁻¹ dry material). Plots indicate range, median and interquartile range (n = 7–12). Soil C and N values from [35]. (Online version in colour.)

Consistently high moisture content (75 ± 10% H₂O by mass), combined with high respiration rates (electronic supplementary material, figure S1a), also probably promoted anoxic microsites that favoured denitrification [34]. Previous sequencing of DNA collected from refuse material has shown that microbial communities are enriched in anaerobic bacteria [14]. In this study, evidence of anoxia was also supported by the positive (and often very high) methane (CH₄) fluxes that we observed from the piles, which ranged from 0.01 up to 250 mg CH₄-C m⁻¹ h⁻¹ (electronic supplementary material, figure S1b). Methanogenesis is far less energetically favourable than denitrification at pH 5.5 (refuse pH: 5.5 ± 0.1), indicating that redox potentials in the refuse can get very low [39]. The surrounding forest soil was a consistent sink for CH₄ as a result of oxidation, demonstrating that characteristics unique to refuse piles—rather than the soil or climate—drive these anaerobic conditions (electronic supplementary material, figure S1b). Refuse material was also highly enriched in microbial biomass C, copies of 16S rRNA (a proxy for bacterial community size) and fungal ITS sequences (a proxy for fungal community size, some of which probably originates from discarded fungal garden material; figure 3a–c). Finally, comparison of potential denitrification enzyme activity, though not an accurate method for estimating in situ total denitrification, showed markedly (25-fold) higher maximum rates for ant refuse than for adjacent soil (figure 3d).

Together, this evidence suggests that the refuse pile environment displays a combination of high substrate availability, ideal abiotic conditions and high levels of microbial biomass and activity that drive elevated greenhouse gas fluxes of CO₂, CH₄ and especially N₂O. Though always markedly greater than soil emissions, N₂O fluxes between individual refuse piles varied by orders of magnitude (figure 2a) for reasons that are not readily apparent; we found no relationship between N₂O fluxes and pile footprint, colony area, refuse N concentrations or forest type (primary or secondary), for example, (electronic supplementary material, figure S2). Instead, it seems likely that inter-pile differences may be driven by a combination of factors such as microclimate, disturbance by other arthropods [15], local forage quality [4] or observed inter-colony variation in microbial community composition [14].

(b). Local N₂O emissions and potential for broader regional importance

To approximate the magnitude of LCA N₂O emissions, we combined measurements of refuse pile flux rates over three time periods with measured pile areas and overall pile density across the primary forest, secondary forest and edge habitats (figure 4). Atta colony density was estimated both for our sites in Costa Rica specifically and from the literature for Neotropical primary and secondary wet/moist tropical forests and their disturbed edges (roads, paths, forest boundaries; electronic supplementary material, table S1). Because N₂O fluxes from refuse piles were non-normally distributed, we present landscape emissions generated using the 25th–75th percentile of flux values. For an average A. colombica density of 1.2 colonies ha⁻¹ in primary rainforest, we estimate that refuse piles may generate 0.01–0.35 kg N₂O-N ha⁻¹ y⁻¹, sufficient to increase N₂O emissions by up to 200% compared with the 12-month primary forest average for this site [18] (figure 4). Baseline N₂O emissions from this lowland rainforest are low in the context of the wet tropics generally [18], and so the relative contribution of LCA-generated emissions at other sites may vary.

Figure 4.

Landscape-scale N₂O emissions for primary tropical forest or Atta colombica refuse piles at densities corresponding to primary forest, secondary forest or edge habitats (kg N₂O-N ha⁻¹ y⁻¹). Bars indicate 25th–75th percentile range. Mean annual value for primary tropical forest at this site (green) is 0.17 kg N₂O-N ha⁻¹ y⁻¹ [21]. Colony densities are presented in electronic supplementary material, table S1. (Online version in colour.)

In most ecosystems, elevated N₂O fluxes are typically episodic and short-lived, characterized as ‘hot moments’ occurring, for example, during freeze–thaw cycles or re-wetting events [40]. In LCA refuse piles, however, annual temperatures are stable, rainfall persists throughout most of the year, and continuing addition of moist N-enriched organic material is maintained throughout the diel and annual cycle, and even enhanced (by approximately 30%) during periods of reduced rainfall [4,41]. In addition, up to one-quarter of A. colombica colonies relocate annually [42], and refuse pile sites within colonies are abandoned and relocated nearby at a similar rate (F.M.S. 2017, personal observation). This leads to a high number of abandoned piles on the landscape. Although these piles typically decompose within approximately 1 year, elevated inorganic N concentrations persist in the soil beyond this period [17], suggesting that abandoned piles could also contribute to additional N₂O production, albeit at unknown rates.

The significance of LCA-driven emissions could increase notably when considering either a broader range of habitat types or the potential for similar N₂O production across a greater range of refuse-concentrating LCA species. Other Acromyrmex and Atta species form refuse piles that can be located both aboveground (e.g. A. mexicana, many Acromyrmex) or belowground (common species A. cephalotes and A. sexdens) and are widely distributed within the latitudinal extents of Neotropical forests, from the southern USA to Argentina [13]. Overall colony density is also 50–150% greater when considering multiple co-occurring Atta species compared with A. colombica alone (electronic supplementary material, table S1). We propose that the same conditions that promote high denitrification (e.g. high rates of N and labile C inputs, high respiration rates and moisture, anoxic microsites) in aboveground refuse piles may also occur in subterranean refuse chambers. Though it is possible that the longer path length between subterranean piles and the atmosphere may allow for more complete reduction of N₂O to N₂ [34], Atta colonies have been shown to remain well ventilated to reduce CO₂ build-up [43], which would reduce this effect.

Leaf cutter ants are also highly responsive to habitat fragmentation, disturbance and land-cover change, which are increasingly common throughout the Neotropics. Colony densities have been shown to increase substantially in secondary forests [5] and at forest edges [9,44] (electronic supplementary material, table S1). As such, extrapolated LCA N₂O emissions from road edges at this site are twice as great as for primary forest (0.1–0.7 kg N₂O-N ha⁻¹ y⁻¹; figure 4). Most species of Atta (and the second LCA genus, Acromyrmex) also increase dramatically in abundance in response to agricultural conversion, to the point that they are considered a major economic pest across cropping, pasture and agroforestry ecosystems [45,46]. In forest production ecosystems such as coffee or Eucalyptus plantations, densities of up to 100 Atta colonies ha⁻¹ have been reported (electronic supplementary material, table S1) [47]. This suggests that production ecosystems, in particular, could represent a significant source of unrecognized, LCA-engineered N₂O. Though N₂O makes up less than 10% of anthropogenic greenhouse gas emissions, both tropical forests and production systems contribute disproportionality to emissions, making parsing patterns in these ecosystems particularly important.

4. Conclusion

Together, our findings suggest that LCA activities can engineer substantial N₂O hot spots in tropical forest ecosystems, actively creating sites with unique physical and biological conditions to support N₂O production rates many orders of magnitude greater than those typically observed in natural systems. Because their distribution does not mimic that of other variables thought to influence N₂O production (such as topography or soil moisture), LCA activity may also represent an unappreciated source of heterogeneity in N₂O emissions that would not be captured by conventional ground-based sampling or modelling approaches. This finding expands our understanding of the mechanisms by which LCA can move nutrients through the environment, adding to a growing list of examples of macrofauna that may play a key role in shaping local biogeochemical cycling [48,49].

Ethics

Site access permission was granted by the Sistema Nacional de Areas de Conservación (Ministerio de Ambiente y Energía de Costa Rica) and Osa Conservation.

Data accessibility

Data collected for this study are available through the Dryad Data Repository at: http://dx.doi.org/10.5061/dryad.27t0v5m [50].

Authors' contributions

F.M.S., C.C.C., B.W.S. and L.P. conceived of and designed the study. F.M.S., B.B.O., A.N.S., B.W.S. and L.P. performed the research and analysed samples. F.M.S. and A.N.S. analysed data. All authors interpreted results and contributed to writing the manuscript, led by F.M.S.

Competing interests

The authors declare no competing interests.

Funding

This work was supported by National Science Foundation DEB grant #1264031 to C.C.C.