Invasive Andropogon gayanus (Gamba grass) alters litter decomposition and nitrogen fluxes in an Australian tropical savanna

The African grass Andropogon gayanus Kunth. is invading Australian savannas, altering their ecological and biogeochemical function. To assess impacts on nitrogen (N) cycling, we quantified litter decomposition and N dynamics of grass litter in native grass and A. gayanus invaded savanna using destructive in situ grass litter harvests and litterbag incubations (soil surface and aerial position). Only 30% of the A. gayanus in situ litter decomposed, compared to 61% of the native grass litter, due to the former being largely comprised of highly resistant A. gayanus stem. In contrast to the stem, A. gayanus leaf decomposition was approximately 3- and 2-times higher than the dominant native grass, Alloteropsis semilata at the surface and aerial position, respectively. Lower initial lignin concentrations, and higher consumption by termites, accounted for the greater surface decomposition rate of A. gayanus. N flux estimates suggest the N release of A. gayanus litter is insufficient to compensate for increased N uptake and N loss via fire in invaded plots. Annually burnt invaded savanna may lose up to 8.2% of the upper soil N pool over a decade. Without additional inputs via biological N fixation, A. gayanus invasion is likely to diminish the N capital of Australia’s frequently burnt savannas.

senesces and the litter remains as part of the standing tussock, unlike native grasses that have a higher proportion of litter in contact with the soil surface ( Fig. 1) 29,30 . There are two main fates for A. gayanus biomass; being burnt or decomposed 31 . Whatever biomass is not burnt or decomposed is carried over to the next year, and can accumulate to up to 30 t ha −1 26,27 . We have previously documented that A. gayanus increases the magnitude of fire-mediated N fluxes 11,13 . N fluxes during a high intensity fire were ∼22 and 16.6 kg N ha −1 yr −1 in A. gayanus invaded and native grass savanna, respectively 13 . However, the fate of litter in years without fire has not been quantified. Understanding Ninputs to soil N pool via litter decomposition will advance understanding of the impact of A. gayanus invasion on the overall ecosystem soil N pool, particularly important considering this is an N-depauperate system 11,32 .
A. gayanus invasion increases the total amount of grass available for decomposition. Here we tested the hypothesis that A. gayanus invasion increases grass litter decomposition (mass loss) and therefore the total quantity of N released from the litter N pool, when compared to native savanna, and that this impacts on the soil N pool. We quantified in situ litter decomposition rates and litter N release using published data of harvested litter 11 . We conducted litterbag experiments, at soil surface and aerial positions, to examine the potential mechanisms driving differences between in situ litter decomposition of the standing litter, specifically differences in litter quality and habitat. We then used these rates to examine the impact of invasion on the soil total N pool over a 10-year period for three fire frequency scenarios.

Results
Production and in situ decomposition of grass litter. The in situ decomposition of the standing crop of grass litter was compared in native and invaded plots by repeat measurement in a 120-day period over the wet season and calculating the litter loss m −2 . As reported in Rossiter-Rachor et al. 11 , the native and invaded plots had 41.5 ± 5.6 and 357.6 ± 43.9 g m −2 of litter respectively at the commencement of the wet season. By the end of the wet season, 61% of the in situ native grass litter had decomposed compared to only 30% in the invaded plots.
The daily litter mass loss (decomposition), and litter N loss (N release) was calculated for both native and invaded plots. Despite the lower proportion of A. gayanus in situ litter that decomposed, the larger quantity of litter meant that overall there was a significantly higher daily litter mass loss in invaded plots compared to native plots (mean 0.90 ± 0.41 versus 0.21 ± 0.04 g −1 m −2 day −1 , respectively; Supplementary Table S1). The daily litter N loss (N release) was also significantly higher in invaded plots compared to native plots (mean 3.29 ± 1.49 versus 0.39 ± 0.23 mg g −1 N m −2 day −1 ), respectively; See Supplementary Table S1 for ANOVAs). The mechanisms explaining these differences in in situ litter decomposition are explained by differences in litter position (surface versus aerial, see below). Scientific RepoRTS | 7: 11705 | DOI:10.1038/s41598-017-08893-z Surface litter decomposition. A surface litterbag experiment was used, whereby litter from the native grass A. semialata and the invasive grass A. gayanus was incubated at the soil surface in: (1) their habitat of origin (native or invaded), and (2) their reciprocal habitats. At the beginning of the experiment the litter N and lignin concentrations of A. gayanus were significantly lower than those of the dominant native grass A. semialata (Table 1; Initial N F [1,18] = 19.32, P = <0.001; Initial Lignin F [1,18] = 15.33, P = 0.001), and had a significantly higher C:N ratio (Table 1; F [1,18] = 17.35, P = 0.001). Litter mass declined rapidly for the first 60 days (from December to February) (Fig. 2a). After 150 days in the field, 21.8 ± 3.8 and 22.6 ± 4.8% of the initial A. gayanus litter remained in the invaded and native habitats, respectively (Fig. 2a). By comparison, 42.8 ± 4.7 and 45.1 ± 3.1% of initial A. semialata litter remained in the invaded and native habitats, respectively (Fig. 2a). The decomposition rate constant (k s ) for A. gayanus litter was approximately double that of A. semialata (Table 2; F [1,16] = 13.21, P = <0.01) with no significant effect of grass habitat type on k s . The N concentration of the surface litter increased over the duration of the study in both A. gayanus and A. semialata (Fig. 2b). While both species displayed the same general trend, the N concentration of A. gayanus litter was significantly higher than that of A. semialata litter by the end of the study, indicating a greater immobilisation   Table S2). The magnitude of difference in litter N concentration between litter types was higher at some plot-pairs, resulting in a significant plot-pair × litter type interaction (Supplementary  Table S2). Litter N loss (expressed as the percentage of initial litter N pool remaining) was significantly greater in A. gayanus compared to A. semialata ( Fig. 2c and Table S2). By the end of the reciprocal litterbag transplant experiment (May, beginning of dry season) the mean litter N pool for A. gayanus had decreased by 38.2 ± 9.6 and 39.1 ± 7.1% (invaded and native habitats, respectively) and 47.4 ± 4.8 and 54.9 ± 5.5% of the initial N pool for A. semialata litter (in invaded and native habitats respectively, Fig. 2c). There was no significant effect of grass habitat type on litter N loss, but there was a significant interaction between habitat and litter type at some sampling times (Supplementary Table S1). Termite activity was more commonly associated with A. gayanus litter (Fig. 2d). By the end of the study 72 ± 10 and 80 ± 6% of A. gayanus litterbags had visible termite activity (in native and invaded habitats), compared to only 28 ± 12 and 40 ± 9% of A. semialata litterbags (in native and invaded habitats) (Fig. 2d). The mass loss of litter was highly correlated with termite activity in both litter type and habitat combinations (A. semialata litter in native grass habitat R 2 = 0.91; A. semialata litter in A. gayanus habitat R 2 = 0.94; A. gayanus litter in native grass habitat R 2 = 0.90; A. gayanus litter in A. gayanus habitat R 2 = 0.87). Only one termite species, Nasutitermes eucalypti (Mjoberg), was found feeding on litter in the litterbags.
Aerial litter decomposition. An aerial litterbag experiment was used, whereby litter from A. semialata and A. gayanus was incubated at 1 m high in their habitat of origin (native or invaded). The results of the aerial litterbag decomposition experiment showed that litter mass declined more slowly in the aerial position and after 189 days in the field (November to May), with 80.7% of original A. semialata leaf, 70.3% of the A. gayanus leaf and 92.2% A. gayanus stem remaining by the end of the study period (Fig. 3). The magnitude of difference in litter mass loss between litter types was higher at some plot-pairs, resulting in a significant plot-pair × litter type interaction (Supplementary Table S3). The decomposition rate constant for aerial incubations (k a ) was significantly greater for A. gayanus leaf, compared to A. semialata leaf, while decomposition of A. gayanus stems was significantly lower than the leaf of either species (F [2,6] = 83.3, P = <0.001, Table 3). Table 4 we present the dominant inputs and output fluxes to the soil N pool to predict decadal consequences of invasion on N cycling. This table consists of fluxes quantified during this and previous studies at this location. We have compiled soil N fluxes for both a burnt and an unburnt state, and for both grass types. We include N input via rainfall, which for this region, is low at approximately 2 kg N ha −1 33 and  Scientific RepoRTS | 7: 11705 | DOI:10.1038/s41598-017-08893-z N input via fixation, which is largely unknown in these savannas, but is likely to be small 10 . The other significant soil N input is from grass decomposition and subsequent N release as quantified by this study (Table 4). Outputs from the soil N pool include uptake via plant growth and incorporation into biomass, and for the burnt condition, loss of N via biomass combustion and volatilisation and ash loss 13 . These annual input and output fluxes were summed over a 10 year period to examine the impact on soil N pool as a function of grass type and three fire regimes: (1) annually burnt (2) burnt 2 in 3 years, and (3) 1 in 5 years. The estimated annual N pool is plotted in Fig. 4 for each grass type and fire regime. At the native savanna plots, the mean total soil N pool was 1692 ± 10 kg N ha −1 (0-30 cm 13 ) a value we assume represents the native savanna soil N pool. Satellite monitoring of fires occurrence indicate that this location experiences a regime of one fire in five years (NAFI, http://www.firenorth.org.au, accessed 8 December, 2016). Using this fire regime and soil N inputs (rainfall, litter N release estimates) and outputs (plant N uptake, N loss from fire, Table 4), the total soil N pool in native savanna would be largely stable after a decade (0.3% loss) while invaded savannas would experience a 3.1% loss in soil N (Fig. 4). This decline occurs despite N inputs over 10 years from rainfall and 4.6 kg N ha −1 and 31.4 kg N ha −1 due to litter decomposition in the native and invaded savanna respectively. However, as fire frequency increases, the soil N input via litter decomposition declines and loss via fire increases. If the savanna was burnt annually, or burnt two in every three years (a fire regime more typical of high rainfall Australian savannas 31 , the loss from the N pool is more substantial in invaded areas compared to native areas (Fig. 4). This is because N loss through volatilisation is twice as great following invasion (Table 4). An annual fire regime in invaded areas results in an 8.2% loss in the total soil N pool after a decade (Fig. 4).

Discussion
We confirm that Andropogon gayanus invasion results in marked changes to litter decomposition and concomitant N fluxes. Compared to native grasses, A. gayanus produced more litter. However, the majority (>80%) of A. gayanus litter was distributed in the aerial position (between 30 cm and 4 m off the ground) whereas 100% of native grass litter was distributed within 30 cm of the soil surface (Fig. 1). Only 30% of the A. gayanus in situ litter decomposed over the wet season compared to 61% of the native grass litter, resulting in a considerable amount of   Table 4. Annual N fluxes (kg N ha −1 year −1 ) to the soil N pool in burnt and unburnt native and invaded savanna at Mary River National Park, Northern Territory Australia. These data were used to estimate the net effect on the soil N pool (kg N ha −1 ) integrating the fluxes over a 10 year period, for the three contrasting fire regimes: (1) annually burnt (2) burnt 2 in 3 years, and (3) 1 in 5 years (see Fig. 4). *Decomposition estimated from this study (from ratio of production to in situ decomposition).
aerial A. gayanus litter being carried over into the dry (and fire) season. Standing aerial litter decomposes slower than litter close to the ground due to lower moisture availability and reduced presence of decomposers communities. A. gayanus stems were particularly resistant to decay, with only ∼8% of stem material decomposing over the wet season. The accumulation of standing litter in African C 4 pasture grass swards is well documented in South American pasture systems 34 . Similarly, in Australian savannas, this resistance to aerial litter decomposition results in the accumulation of large quantities of litter in invaded plots, with up to 30 t ha −1 of grass litter in long unburnt A. gayanus invaded systems 26 , compared to up to 5.8 t ha −1 in long unburnt native grass savanna 10 . Litter quality also impacted on decomposition rates. In tropical wet seasons, litter quality, not climate, is the key determinant of litter decomposition rates, as temperature and moisture are generally non-limiting in the wet season 9,35-37 . Litter quality parameters include initial litter N concentrations, lignin concentrations, C:N and lignin:N ratios 19 . Soil surface decomposition rates (k s ) of A. gayanus were approximately double that of A. semialata litter. A. gayanus litter had significantly lower N and lower lignin concentration than native grass litter (Table 1). While live A. gayanus leaf tissue has a higher N concentration than native grasses (Rossiter-Rachor et al. 11 ), it is highly effective at translocating this N from senescing tissues, leaving A. gayanus litter that has a low N, and a low lignin concentration. Lignin concentration has been found to be negatively correlated with the rates of litter decomposition in several tropical grasses, including A. gayanus 37 . Our lignin results for A. gayanus (10 ± 0.4%) were within the range reported by Thomas & Asakawa 37 for A. gayanus (9.2-11.7%). The relatively low litter quality of A. gayanus reported in our study is common among C 4 grasses 38 , and may result in litter in the early stages of decomposition taking up and storing N 39-41 . Bacteria and/or fungi take up N to supplement the N in the litter being decomposed 42 . N immobilisation is particularly common in C 4 grasses 21,37 . Importantly, our results highlight that changes in litter decomposition can occur as soon as the invasion commences, as the altered litter decomposition rates were driven by the altered litter quality of A. gayanus, not the habitat in which decomposition occurred.
Litter quality is also likely to have influenced termite feeding preferences, which preferentially consumed A. gayanus litter, regardless of the habitat the litterbags were incubated in (native or invaded savanna), a finding that is consistent with studies of A. gayanus in its native range in Nigeria 43,44 . In Africa, the termite Trinervitermes geminatus (Wasmann) had a stronger preference for A. gayanus litter than five other African grasses tested, even though the density of A. gayanus stands at their study sites was significantly lower than the other grasses.  10 year period, as a function of fire regime. Net soil N pool calculated by summing N inputs (fixation, wet deposition, N release from decomposition) and N outputs (fire losses, grass N uptake) in burnt and unburnt years for each fire regime. Three fire regimes were used; annual burning; fire 2 in 3 years (typical fire regime) and 1 in 5 years. Numbers indicate %N loss of gain relative to the initial total soil N pool. Soil N, grass N uptake under burnt and unburnt conditions, and fire N loss data were taken from Rossiter-Rachor et al. 11,13 . See Table 4 for further information on N inputs and N outputs in burnt/unburnt years.
It remains unknown which properties of A. gayanus litter are responsible for such preferential behaviour by termites.
The high decomposition rates found in this study are comparable to those reported in the literature for C 4 African grasses; with k values ranging between 0.0020 g g −1 d −1 for Brachiaria dictyoneura 37 , 0.0025 g g −1 d −1 for A. gayanus 37 and 0.0174 g g −1 d −1 for Panicum maximum 45 . The A. gayanus k s values from this study were (0.0089-0.0113 g g −1 d −1 ) were at the higher end of this range.

Implications for soil N fluxes. This study demonstrates that replacing native savanna grasses with A. gayanus,
and altering the ecosystem process of litter decomposition, has important implications for soil N cycling. Overall, the N flux data shows that the increased litter N release from A. gayanus (in situ decomposition data) does not compensate for the increased outputs from the soil N due to fire mediated N losses 13 or increased N uptake by A. gayanus 11 given the ambient fire regime. Longer-term (decadal) implications of changed N relations are a reduction in soil total N which will increase with fire frequency. In high rainfall Australian savannas impacted by invasive grasses, fire frequency is typically annual or biennial burning 2 . In the absence of additional inputs via N fixation, this fire regime would reduce total soil N by up to 8.2% in invaded savanna within a decade (Fig. 4). This is likely to have substantial implications given that the savannas studied here are among the most N-depauperate systems globally (Table 4). However, to date, we have not detected a consistent change in soil N over time following invasion 30 . This may be due to the death and decomposition of woody vegetation in the invaded areas 24 that is co-occurring as a consequence of invasion and increased fire intensity, and this process would be releasing substantial amounts of N. The N losses may be compensated for by N fixation, with speculation 46 but no detailed investigation that A. gayanus encourages diazotrophic soil microbes. Clearly, longer-term studies of extremely impacted A. gayanus sites (i.e. where fires have significantly reduced the presence of live trees 25 ) are needed to quantify the consequences of A. gayanus invasion on whole of ecosystem N balance in invaded savanna ecosystems.

Site description and experimental design. Our research was conducted in Mary River National Park
(Northern Territory, Australia (12°64′S, 131°75′E). Temperatures are high throughout the year (27 °C, annual average), while rainfall is highly seasonal (1591 mm, annual average, Weather station number 014263, Australian Bureau of Meteorology) and concentrated in the wet season between October and April. The savanna overstorey is dominated by Eucalyptus miniata (Cunn. Ex Schauer) and Eucalyptus tetrodonta (F. Muell), with an understory dominated by native perennial grasses Alloteropsis semialata (R. Br.) Hitchc. and Eriachne triseta Nees ex Steud., with patches of annual grasses, including Pseudopogonatherum irritans (Br.) A. Camus. Andropogon gayanus (gamba grass) has invaded extensive areas of the park, replacing the short savanna grass communities (~0.5 m) with dense, tall, almost monospecific swards up to 4 m high 47 .
We compared litter decomposition of A. semialata and A. gayanus using a randomised block design. Paired-plots (blocks) were established, with each plot-pair consisting of an area dominated by native grass (hereafter referred to as 'native' plots), and a nearby (approximately 50 m distance) A. gayanus dominated area (hereafter referred to as 'invaded' plots). Plot-pairs were located up to 600 m apart, and each plot was 50 m × 50 m in size. In native plots, the grass component included several native grass species, while A. gayanus was the only grass species present in invaded plots.
Litter decomposition. We used two complementary methods based on the loss rate of grass litter over time to quantify litter decomposition. We followed an in situ cohort of native grass and A. gayanus litter through a wet season to quantity the decomposition rate based on litter loss over time. In addition, we used litterbags located on the soil surface and also suspended aerially at 1 m in height to quantify decomposition rate constants (k) for litter exposed to microbial, fungal and invertebrate decomposition (surface) versus exposure to microbial, fungal and photodegradation (aerial). The wet season is the peak period of decomposition in these savannas due to the high rainfall and humidity 48 and very little new grass litter is produced during this time.
Quantification of in situ litter decomposition and litter N loss. The in situ decomposition of the standing crop of grass litter was compared in native grass and invaded plots by following the cohort of grass litter over a wet season. Plots were not burnt during the study period and any reduction in litter mass between sampling events was assumed to be due to decomposition of the previous year's grass production 11,19 , because very little new grass litter is produced during the wet season 11 . We used litter data published in  to calculate the litter loss over time. As described in Rossiter-Rachor et al. 11 , grass phytomass (biomass + necromass) was destructively harvested at the beginning (November 2002) and then end (March 2003) of the wet season, some 120 days later. Samples were collected from three random replicate 2 × 2 m quadrats, within each of the five plot pairs (n = 15 quadrats per grass type, per sampling time). Samples were returned to the lab, and sorted into green leaves and stems (live biomass, data not presented) and dead standing grass (necromass; hereafter referred to as grass litter). All litter was dried for 48 h at 60 °C and weighed. Subsamples were ground in a Wiley mill, analysed for %N and %C using a Carlo Erba analyser (Thermo Electron, MA, USA), and the initial litter N pool calculated. In this study we calculated the in situ grass litter decomposition as the decrease in grass litter mass between November 2002 and March 2003. Similarly, litter N release (product of litter mass x litter tissue N concentration; g N m −2 ; following) 49 was calculated as the change in the grass litter N pool during this period. This litter N release was also expressed as a release rate per day (calculated as the decrease in the litter N pool, divided by the number of days between the November and March harvests).
Scientific RepoRTS | 7: 11705 | DOI:10.1038/s41598-017-08893-z Quantification of surface litter decomposition rate constant. We quantified surface litter decomposition using a reciprocal transplant experiment following Rothstein et al. 50 in which leaf litterbags were incubated in the field in plot pairs. Senesced A. semialata and A. gayanus leaf litter was collected from the site in November 2003, dried at 60 °C to a constant mass and 5 g sub-samples were sewn into 15 × 15 cm mesh bags (∼1.5 mm mesh size). Fifteen sub-samples were ground in a Wiley mill, analysed for %N and %C using a Carlo Erba analyser (Thermo Electron, MA, USA), and the initial litter N pool calculated. Samples were also analysed for % Acid Detergent Lignin using an Ankom A200 fibre analyser (Ankom Technology, NY, USA). Litterbags were collected every thirty days up to 150 days (7 th January, 6 th February, 7 th March, 6 th April and 6 th of May 2004, respectively). A total of 1509.1 mm rainfall was recorded during this period 51 . At each harvest date, 50 litterbags of both grasses were retrieved (5 plot-pairs × 2 grass habitats × 5 replicates). Litter was carefully washed (without pressure) with de-ionised water to remove soil, roots or live material (following) 50 , before being dried at 60 °C and weighed for mass loss. Litter decomposition was expressed as percent of the initial mass remaining. We used time series of litter mass and N concentrations to calculate k s for A. gayanus and A. semialata surface litter using a simple negative exponential decay model 52 ; where M t is the litter mass at time t (years), M i the initial litter mass, k the decomposition rate constant for surface litter and m is a regression constant. We randomly chose three bags of each litter type, from each plot-pair, at each sampling time. Litter was ground, analysed for %N and the litter N pool (litter mass × N concentration) was calculated for each harvest date. Litter N release was expressed as percent of the initial litter N pool remaining, showing net losses (mineralisation) or gains (immobilisation). During collection of the litterbags, visible termite activity was noted via evidence of termites in the litterbags, soil sheeting attached to the litterbags and damage to the side of the litterbag in contact with the soil. Termites present in the litterbags were collected, placed in 95% ethanol, and sent to the CSIRO, Darwin laboratories for identification.
Quantification of aerial litter decomposition rate constant. We quantified aerial decomposition rates of leaf and stem litter using an aerial litterbag experiment following Liao et al. 19 . A. semialata and A. gayanus leaf litter, as well as A. gayanus stem litter, was collected in November 2014, and processed into litterbags as previously described. We placed litterbags in 3 plot-pairs in early December 2014, to coincide with the first wet-season rains. Nine litterbags of A. semialata leaf litter were placed in each native plot. In invaded plots, nine A. gayanus litterbags of leaf litter, and nine of A. gayanus stem were incubated. We attached litterbags to aluminium stakes 1 m in height. Three randomly selected bags were collected from each plot after 47, 83, 113, 155 and 189 days in the field respectively (3 rd February, 11 th March, 10 th April, 22 nd May, 25 th June). A total of 1282.2 mm of rain was recorded during this period 51 . Litterbags were processed as previously described and the decomposition rate constant for aerial litters (k a ) was calculated as per Equation 1.
Impacts of invasion and fire on soil N cycling. The effect of A. gayanus invasion and fire on the soil N pool was investigated by collating measured N fluxes for native grass and invaded savanna. The initial N pool was assumed to be 1692 kg ha −1 (0-30 cm) as measured by Rossiter-Rachor et al. 13 . The soil N inputs included wet deposition, litter N release from in situ decomposition, with N outputs including fire losses and grass N uptake from grass growth. These N inputs and outputs for burnt and unburnt years., and these data were used to investigate the net effect of different fire resgimes on the soil N pool. Three fire regimes used; annual burning; fire 2 in 3 years (as occurs at the study location) and 1 in 5 years. Changes to the soil N pool were estimated over a 10-year period by summing net N inputs and outputs as described above for burnt and unburnt years for each fire regime.
Data describing the initial soil N pool and grass N uptake rate under burnt and unburnt conditions, and fire loss data were calculated using the data presented in Rossiter-Rachor et al. 11,13 . Grass N uptake was estimated by summing annual increments of biomass N (see ref. 11 for details of biomass N determination). N input via rainfall was based on data of Noller et al. 33 in a study conducted in the nearby (~100 km east) Alligator River Region of the Northern Territory; a region very comparable in climate, vegetation and fire regime to that of the current study.

Statistical analysis.
In situ litter decomposition rates were compared using a two-factor analysis of variance (ANOVA), with factors litter type (A. semialata or A. gayanus, fixed) and plot-pair (5 locations, random). Differences in surface litter decomposition rate, litter N concentration, and litter N release were analysed using a four-way mixed model ANOVA, with factors time (5 levels, fixed), plot-pair (3 locations, random), grass habitat type (native or invaded, fixed), and litter type (A. semialata or A. gayanus, fixed). Differences in aerial litter decomposition rates were analysed using a three-way ANOVA, with the factors time (5 levels, fixed), plot-pair (5 locations, random), and litter type (A. semialata or A. gayanus, fixed). Before statistical analyses were undertaken, assumptions of ANOVA were checked using Cochran's test, and where necessary data were transformed prior to analyses to improve normality and homogeneity. Tukey's pairwise contrasts were used to explore the main effects. Analyses were carried out using Systat Version 10 (SPSS, Chicago, IL, USA).