Plant Soil (2020) 450:93–110 https://doi.org/10.1007/s11104-019-04012-1 REGULAR ARTICLE Regulation of nitrogen fixation from free-living organisms in soil and leaf litter of two tropical forests of the Guiana shield Leandro Van Langenhove & Thomas Depaepe & Sara Vicca & Joke van den Berge & Clement Stahl & Elodie Courtois & James Weedon & Ifigenia Urbina & Oriol Grau & Dolores Asensio & Josep Peñuelas & Pascal Boeckx & Andreas Richter & Dominique Van Der Straeten & Ivan A. Janssens Received: 29 June 2018 / Accepted: 25 February 2019 / Published online: 1April 2019 # The Author(s) 2019 Abstract understood. The aim of this study was to quantify FLNF Background and aims Biological fixation of atmospher- rates and determine its drivers in two tropical pristine ic nitrogen (N2) is the main pathway for introducing N forests of French Guiana. into unmanaged ecosystems. While recent estimates Methods We used the acetylene reduction assay to mea- suggest that free-living N fixation (FLNF) accounts for sure FLNF rates at two sites, in two seasons and along the majority of N fixed in mature tropical forests, the three topographical positions, and used regression anal- controls governing this process are not completely yses to identify which edaphic explanatory variables, Responsible Editor: Bernardo M Flores. Electronic supplementary material The online version of this article (https://doi.org/10.1007/s11104-019-04012-1) contains supplementary material, which is available to authorized users. L. Van Langenhove (*) : S. Vicca : J. van den Berge : J. Weedon E. Courtois : I. A. Janssens Department of Ecological Science, Vrije Universiteit Amsterdam, Centre of Excellence PLECO (Plants and Ecosystems), De Boelelaan 1085, 1081 HVAmsterdam, The Netherlands Department of Biology, University of Antwerp, Wilrijk, Belgium e-mail: leandro.vanlangenhove@uantwerpen.be I. Urbina : O. Grau : D. Asensio : J. Peñuelas CSIC, Global Ecology Unit CREAF-CSIC-UAB, T. Depaepe : D. Van Der Straeten 08193 Bellaterra, Catalonia, Spain Laboratory of Functional Plant Biology, Department of Biology, Ghent University, Ledeganckstraat 35, B-9000 Ghent, KL, I. Urbina : O. Grau : D. Asensio : J. Peñuelas Belgium CREAF, 08193 Cerdanyola del Vallès, Catalonia, Spain C. Stahl P. Boeckx INRA, UMR Ecology of Guiana Forests (Ecofog), Department of Applied Analytical and Physical Chemistry, AgroParisTech, Cirad, CNRS, Université des Antilles, Université Faculty of Bioscience Engineering, Isotope Bioscience Laboratory de Guyane, 97387 Kourou, French Guiana – ISOFYS, Ghent University, Coupure Links 653, 9000 Ghent, Belgium E. Courtois Laboratoire Ecologie, évolution, interactions des systèmes A. Richter amazoniens (LEEISA), Université de Guyane, CNRS, IFREMER, Department of Microbiology and Ecosystem Science, University French Guiana, 97300 Cayenne, France of Vienna, Althanstr. 14, 1090 Vienna, Austria 94 Plant Soil (2020) 450:93–110 including carbon (C), nitrogen (N), phosphorus (P) and N fixation was attributed to symbiotic organisms due to molybdenum (Mo) content, pH, water and available N the high abundance of leguminous trees (Losos and and P, explained most of the variation in FLNF rates. Leigh 2004), typically associated with symbiotic N Results Overall, FLNF rates were lower than measured fixers. The relative importance of symbiotic N fixation, in tropical systems elsewhere. In soils seasonal variabil- however, has been questioned because it is facultative ity was small and FLNF rates differed among topogra- (Menge et al. 2009) and may decline to near zero in phies at only one site. Water, P and pH explained 24% of mature tropical rainforests (Barron et al. 2011; the variation. In leaf litter, FLNF rates differed season- Batterman et al. 2013; Sullivan et al. 2014). In this ally, without site or topographical differences. Water, C, context, N fixation by free-living organisms is thus N and P explained 46% of the observed variation. We increasingly considered to be more important, with re- found no regulatory role of Mo at our sites. cent estimates from multiple rain forests suggesting that Conclusions Rates of FLNF were low in primary a substantial amount of N is indeed fixed by free-living rainforest on poor soils on the Guiana shield. Water was diazotrophs (e. g. Reed et al. 2007, Cusack et al. 2009, the most important rate-regulating factor and FLNF in- Černá et al. 2009, Barron et al. 2009, Wurzburger et al. creased with increasing P, but decreased with increasing N. 2012, Matson et al. 2014). Our results support the general assumption that N fixation Because of its relevance for ecosystem functioning, N in tropical lowland forests is limited by P availability. fixation and its rate-controlling factors have been the focus of previous research. Nitrogen fixation is energetically Keywords Free-living nitrogen fixation . Tropical expensive, requiring much ATP for both the reaction itself forest . French Guiana . Nutrients . Phosphorus . (Gutschick 1981) and for maintaining an oxygen-poor Molybdenum intracellular environment for nitrogenase, the enzyme re- sponsible for N fixation (Robson and Postgate 1980). The oxygen-poor environment can be created, either by an Introduction upregulated respiration rate, or developing cellular struc- tures that can limit the entry of oxygen (Robson and Nitrogen (N) availability is a limiting factor for plant Postgate 1980). This regulation is needed because oxygen growth in a wide range of terrestrial ecosystems world- binds to nitrogenase and inactivates the enzyme. Free- wide (LeBauer and Treseder 2008) and restricts the living N fixation (FLNF) is thought to decrease as N amount of carbon (C) that can be assimilated and stored availability increases in the environment (Hartwig 1998), in the terrestrial biosphere (Hungate et al. 2003; because uptake of inorganic N is less costly than N fixation Penuelas et al. 2013; Zaehle et al. 2015). Globally, (Gutschick 1981; Menge et al. 2009)). biological N fixation is the most important natural path- Low phosphorus (P) availability has been reported to way for introducing previously inert N - namely N2 gas limit FLNF in several tropical environments (e.g., -, into unmanaged ecosystems (Galloway et al. 1995; Pearson and Vitousek 2002; Reed et al. 2007), an ob- Vitousek et al. 2013). Nitrogen is fixed by microorgan- servation often attributed to the diazotrophs’s high P isms, known as diazotrophs, that can reduce gaseous N requirement. Nonetheless, studies of P addition to trop- (N2) into ammonia. Diazotrophs are divided in two ical forests have reported contradictory responses. Some groups: symbiotic and free-living. Symbiotic N fixers reported that FLNF in soils and plant litter increased in are generally found in root nodules and live in a mutu- response to P additions (Benner et al. 2007; Reed et al. alistic relationship with higher plants that allocate C to 2007), while others found no effect (Pérez et al. 2008; the N fixers in exchange for N. Free-living N fixers are Barron et al. 2009). Recent studies suggest that the hetero- or autotrophic bacteria or archaea inhabiting stimulation of N fixation by P addition may be due to water, soil, rocks and leaf litter (Dynarski and Houlton molybdenum (Mo) contamination of the commonly 2018). Global terrestrial N inputs from biological N used superphosphate fertiliser (Barron et al. 2009; fixation have been estimated at 60 Tg y−1 (Vitousek Wurzburger et al. 2012). Molybdenum is a rock- et al. 2013), and biome-level comparisons suggest that derived trace element required to produce the FeMo tropical rain forests may fix more N than any other cofactor necessary for the functioning of the most com- unmanaged ecosystem (Galloway et al. 2004; Reed mon group of nitrogenases (Igarashi and Seefeldt 2003). et al. 2011). Until recently, a large proportion of tropical Molybdenum is thus associated more with fundamental Plant Soil (2020) 450:93–110 95 enzymatic requirements than with the high energy con- Here we present results from a study carried out in sumption of N fixation, but limitation of both Mo and P mature tropical rainforests of French Guiana, where we has been documented in some forests (Barron et al. measured FLNF in both soil and leaf litter. The rolling 2009; Wurzburger et al. 2012; Reed et al. 2013). hills of French Guianese tropical forests, part of the Energy in the form of external organic carbon (C) is Guiana Shield and perched upon parent material that is another likely factor regulating the activity of free-living amongst the oldest and most weathered in the world N fixers. Like other heterotrophic microbes, diazotrophs (Hammond 2005), are characterised as very poor in rely on extracellular organic C for respiration, and C mineral nutrients (van Kekem et al. 1996), with topog- supply may be more limiting than nutrient availability in raphy inducing spatially heterogeneous nutrient avail- free-living fixers (Hofmockel and Schlesinger 2007), abilities. We studied three distinct topographical posi- even though the C cost of FLNF is not well-quantified tions at two different forest sites in the wet and dry (Dynarski and Houlton 2018). Lastly, seasonal variation season, with the aim of maximizing the range in soil in soil moisture can play a large role in the regulation of nutrients and moisture. Our goals were (I) to determine N fixation rates, because the rates are positively corre- and compare rates of FLNF in soils and leaf litter to lated with moisture content (Roskoski 1980). Seasonal other tropical forest studies, (II) to evaluate whether or differences in N fixation have indeed been observed, but not N fixation rates differed between sites or between the direction of these seasonal changes differed among seasons, and (III) to identify which environmental fac- studies. Studies conducted at different tropical sites have tors (or combination thereof) best explained the spatial reported both higher (cf Reed et al. 2007) and lower (cf and seasonal variation in N fixation rates. Matson et al. 2014) rates in the wet season compared to the dry season with likely factors besides moisture, such as changes in oxygen supply, causing the discrepancy in Materials and methods results. Lowland tropical forests found on highly weathered Study sites old soils are typically considered to be N-rich, because they can accumulate, recycle and export large quantities The study was conducted at two primary forest sites in of N (Vitousek and Sanford 1986; Hedin et al. 2009; French Guiana, in the research stations of Paracou and Brookshire et al. 2012). This open N cycle is also Nouragues. Paracou is situated 15 km from the coast corroborated by enriched δ15N values of soils and plant (5°15’N, 52°55’W), while Nouragues is located inland tissues, due to high gaseous and leaching losses of (4°05’N, 52°41’W). Annual rainfall quantities at both depleted 15N sources (Amundson et al. 2003). Simulta- sites were similar, with Paracou receiving an average of neously, these lowland tropical forests have usually 3100 mm year−1 for the period 2004–2015 (Aguilos been described as P limited because of biogeochemical et al. 2018) and Nouragues receiving an average of theory predicting that P limitation should be prevalent in 2990 mm year−1 (Bongers et al. 2001). Mean annual old, strongly weathered soils (Walker and Syers 1976; air temperature is near 26 °C for both sites (Gourlet- Wardle et al. 2004). There is also a wealth of indirect Fleury et al. 2004; Bongers et al. 2001). The French evidence including high N availability (Brookshire et al. Guianese climate is characterized by a wet and a dry 2012), high N:P ratios in leaves (Vitousek 1984) and season due to the north/south movement of the Inter- correlations between forest properties and soil fertility at Tropical Convergence Zone. The region receives heavy continental scale (Quesada et al. 2012; ter Steege et al. rains from December to July and a dry period, typically 2006. Locally, however, nutrient availability in tropical characterized by less than 100 mm rainfall month−1, landscapes can vary with topography, but the magnitude from August to November (Aguilos et al. 2018). and direction of this variation is variable and influenced Topography at both sites is undulating with maxi- by, e.g., terrain steepness and rainfall, leading to differ- mum slopes of approximately 15°. The elevational dif- ences in physical denudation rates and solute transpor- ference between hill summits and valleys is 20–50 m tation (Porder et al. 2005; Weintraub et al. 2015). Dif- over horizontal distances of 200–400 m. Soils at the ferences in soil type and redox status along topograph- Paracou site are schist soils with veins of pegmatite ical gradients may also affect nutrient availability along the Bonidoro series, a Precambrian metamorphic (Tiessen et al. 1994). formation (Epron et al. 2006). Soils at the Nouragues 96 Plant Soil (2020) 450:93–110 site are also derived from the same Bonidoro series, and grade, Air Liquide) to create a 10% headspace concen- consist of mainly Caraib gneiss (Bongers et al. 2001). tration by volume. The samples were then incubated in This Precambrian geological substrate is particularly situ at ambient forest light (no direct sunlight) and low in P content compared to the generally younger, temperature for 18 h. Sample moisture was not manip- nutrient-richer soils of western Amazonia (Hammond ulated in any way, but was determined after the incuba- 2005; Grau et al. 2017) and therefore soils at both sites tion as the weight loss after oven drying at 70 °C during are classified as nutrient-poor Acrisols (FAO-ISRIC- 48 h. ISSS 1998). Soils at Paracou range from loamy sand After incubation, a subsample from the sample head- to sandy loam, while soils at Nouragues contain much space was injected into a pre-evacuated 12 ml borosili- more clay and span the range of sandy loam to silty clay cate vial (Labco Limited, Ceredigion, UK) and shipped according to the USDA texture classification chart (Fig. to Ghent, Belgium for analysis. Ethylene concentrations S1). were measured using laser-based photo-acoustic spec- troscopy (ETD-300, Sensor Sense, The Netherlands). Experimental design Parallel acetylene blanks (no leaf litter or soil) were created to assess background levels of ethylene in the To maximize the natural variation in soil nutrient avail- acetylene gas (1.5 + − 0.4 nl ethylene ml−1 acetylene ability the experimental plots encompassed a topograph- gas), which were subtracted from the sample ethylene ical gradient. At each forest site twelve plots were concentrations. Controls for ethylene production in the installed, distributed over four hillslopes, with each soil or litter in the absence of acetylene gas were also hillslope having three plots located at three topograph- assayed and were consistently below the detection limit ical levels: (1) bottom, i.e. just above the creek running of 0.01 nl ethylene ml−1 air. Soil and leaf litter samples through the valley, (2) slope, i.e. the intermediate section that over the incubation time produced ethylene concen- of the elevation, and (3) top, i.e., where the slope evens trations below the detection limit (0.01 nl ethylene ml−1 out and becomes the hilltop. Each plot was of 20 × 20 m air) were recorded as half of this value. in size. In total, 24 plots spread over two sites and three We converted the rate of ethylene production, topographical categories per site were studied. Distances expressed as nmol ethylene g−1 sample h−1, into N between the plots were 10–100 m and in each plot five fixation rates, expressed as kg N fixed ha−1 y−1 using soil and litter samples were collected. Four samples the densities of the soil (to a depth of 5 cm) or leaf litter, were collected at 2 m distance from each corner and a and the theoretical ratio of 3 mol ethylene produced per fifth sample was taken in the middle of the plot. A total mole N fixed (Benner et al. 2007; Cusack et al. 2009; of 120 sampling points (2 study sites, 3 topographical Matson et al. 2014; Pearson and Vitousek 2001; Reed categories per site, 4 plots per category and 5 samples et al. 2007). The latter being based upon the conclusion per plot) were sampled for both soil and leaf litter in that reducing 3 mol of acetylene to ethylene is equiva- both the wet and dry season (240 samples in total). lent to the 6 electron transfer involved in the reduction of Courtois et al. (2018) have provided more detailed one mole of N2 to ammonium (Seitzinger and Garber information on the experimental design. 1987). We attempted to measure uptake of 15N through the pool dilution assay (Furnkranz et al. 2008) in a N fixation subset of soil samples to gauge the actual ratio of moles ethylene produced per mole N fixed, but we failed to do Nitrogen fixation rates were determined using the acet- so due to a combination of low soil FLNF rates and high ylene reduction assay (Hardy et al. 1968). Samples were background levels of N in these environmental samples. collected in May and September 2016, in the wet and Other authors have reported encountering similar issues dry season, respectively. Samples of leaf litter were and could not measure 15N uptake in soil (Matson et al. collected manually from the soil surface and soil sam- 2014) or leaf litter (Menge et al. 2009) samples. Soil ples were collected with a 2-cm diameter corer to a samples for determining bulk density to a depth of 5 cm depth of 5 cm after removing all litter from the surface. were taken with Kopeck rings. The samples were oven All samples were placed in clear 100 ml borosilicate dried at 105 °C for 48 h and bulk density was calculated jars. The jars were sealed with rubber septa and 10 ml of by dividing soil weight by Kopeck ring volume. Leaf air was replaced with 10 ml of acetylene gas (welding litter was collected in a 0.5 m2 wooden frame at five Plant Soil (2020) 450:93–110 97 locations per plot and once per season for determining and five 2 cm2 strips of anion-exchange membrane litter density. The litter was dried at 105 °C for 48 h and (VWR Chemicals, USA) were added. This mixture the density was calculated by dividing weight by area was stirred for 24 h and the strips were then rinsed and (kg ha−1). eluted with 10% HNO3. Available P and Mo contents were determined with a iCAP 6300 Duo ICP optical Chemical analyses emission spectrometer (Thermo Fisher Scientific, Germany). Total C, N, P and Mo contents in the soil and litter were determined on the same samples that were used for Literature comparison acetylene reduction. Ethylene production was first mea- sured (see N fixation), and afterwards samples were To compare the FLNF rate results we found to those of oven dried at 70 °C for two days and then ground in a other authors we performed a database search similar to ball mill (Retsch, Germany). Total C and N contents the search carried out by Dynarski and Houlton (2018). were determined by dry combustion with an elemental We searched Web of Science using the keywords nitro- analyser (Flash 2000, Thermo Fisher Scientific, Germa- gen, free-living, asymbiotic, fixation, soil, leaf litter and ny). Total P (mg kg−1) and Mo (mg kg−1) contents were tropical forest. From the resulting studies we selected determined by the sequential digestion of the ground those that were performed in natural terrestrial tropical soil and litter samples in heated strong acid (69% HNO3 ecosystems. For studies that presented results from mul- and 30% H2O2), followed by analysis on an iCAP 6300 tiple time points or seasons, we averaged the data over Duo ICP optical emission spectrometer (Thermo Fisher the course of the study period. Studies that did not report Scientific, Germany). any measure of variance were assigned a standard error Soil texture and nutrient availability were determined of ¼ of the mean. Reported rates of FLNF were con- on a composite sample made of three soil cores per verted to nmol ethylene produced g−1 dry substrate h−1 sampling point, each core to a depth of 15 cm. Texture in order to compare N fixation rates between studies. was determined on fresh soil using the hydrometric When N fixation rates were presented on a per area basis method (Gee and Bauder 1986). Soil particles were we used the bulk density of the N fixation substrate (soil dispersed with sodium metaphosphate and the amounts or leaf litter) and the ethylene: N2 conversion factor of sand, silt and clay were determined using a hydrom- indicated in the study to convert to nmol ethylene pro- eter. Soil samples for measuring nutrient availability duced g−1 dry substrate h−1. When no conversion factor were collected in May and September 2015, sieved was indicated, we assumed the standard 3: 1 conversion (2 mm) and split into three subsamples. The first sub- factor (Hardy et al. 1968). Results of this database sample was extracted with 1 M KCl in a 1:2.5 w:v ratio search are summarised in supplementary information for pH measurement and determination of available N. Table S4. On this extract pHKCl was measured with a pH meter (HI 2210–01, Hanna Instruments, USA), after which the Data analysis extract was filtered through a 42 μm filter and the filtrate’s concentration of NH4+ and NO3− was deter- To assess the differences in N fixation rates and soil and mined colorimetrically (SAN++ continuous flow leaf litter variables between sites and seasons we used analyser, Skalar Inc., The Netherlands). Available N linear mixed effects regression models (LMER), (mg kg−1) was defined as the sum of the NH4+ and analysing soil and leaf litter data separately. We used NO3− concentrations. The second subsample was used site (Paracou or Nouragues), season (Wet or Dry) and to determine the gravimetric water content by measuring topographical position (Bottom, Slope or Top) as fixed weight loss after drying at 105 °C during 48 h. The third factors and plot identity as a random factor. The validity subsample was used to determine extractable P and Mo. of the linear models’ assumptions (linearity, normality Soils were oven dried at 60 °C for 48 h after which of residuals, no influential outliers, homoscedasticity) extractable P was determined with Bray-P acid fluoride were evaluated with standard functions of R (R core extraction (Bray and Kurtz 1945). Available Mo was team 2017, version 3.4.3), including diagnostic plots. determined through resin extraction (Wurzburger et al. Prior to analysis, data were log transformed if their 2012). The samples were mixed with water in a 1:6 ratio distribution was right-skewed to improve normality of 98 Plant Soil (2020) 450:93–110 model residuals. Multiple comparisons within a factor Anderson 2002). We did this for all abovementioned were analysed using Tukey post hoc tests. We per- predictors. This approach yielded very similar results to formed principal component analyses to visualize the the forward stepwise model selection, confirming the correlations of previously standardised soil and leaf robustness of our analysis, but for readability reasons litter variables according to site, season and, if neces- are not presented nor discussed in this paper. These sary, topography. We observed that a large proportion of results are, however, shown in supplements (Figs. S2 our measured samples yielded ethylene production rates and S3). All analyses were carried out with the software that fell below the detection limit (25% of samples in package R (version 3.4.3) using packages lme4 (Bates soil). To investigate if there were site-specific or season- et al. 2015), MASS (Venables and Ripley 2002), al patterns in the occurrence of values below detection ggfortify (Tang et al. 2016) and AICcmodavg limit, we transformed our ethylene production rates into (Mazerolle 2017). binomial data (1 for measured rate and 0 for below detection limit rate) and analysed the resulting data with binomial generalized linear model (GLM) with season Results and site as factors. To identify which set of physico-chemical variables Rates of FLNF were on average 0.015 ± 0.003 (standard significantly contributed to the observed variation in N error) nmol ethylene g−1 h−1 or 0.57 ± 0.10 kg N ha−1 fixation rates we performed a forward stepwise model y−1 in soil and 0.25 ± 0.04 nmol ethylene g−1 h−1 or selection, i.e. starting from a null model and retaining 0.09 ± 0.01 kg N ha−1 y−1 in leaf litter. Per unit mass, the predictor variable that led to the largest decrease in FLNF rates were thus, on average, tenfold higher in leaf the Akaike information criterion (AIC), corrected for litter than in the top 5 cm of the soil (Table 1). However, sample size (AICc). This process was iterated until no when reported per unit area, FLNF rates were lower in additional predictor reduced the model AICc by at least litter than soil, due to the huge difference in density two units. This procedure was performed for the dataset between the sample types (Table 1). as a whole, as well as for each combination of site and season that was shown to be significantly different in the Soil – N fixation Overall, soil FLNF rates did not differ mixed-effects models (see above) in soil and in leaf between seasons (P = 0.27). The effect of topography on litter. For these analyses the measurements of FLNF that soil FLNF rates differed between Paracou and were below the detection limit were excluded to assess Nouragues (Site x Topography interactive effect; P = which predictor variables participate in regulating the 0.021) (Fig. 1a, b). In Paracou, soil FLNF rates were 20 FLNF that were detectable and thus participate to the times higher (+/-SE = 6 to 65 times higher) in the bottom ecosystem scale N fixation. The potential predictor var- and slope plots than in the top plots (P = 0.047 and P = iables for both leaf litter and soil were standardised to a 0.009, respectively), but did not differ significantly be- mean = 0 and standard deviation = 1 prior to the model tween the bottom and slope plots. In Nouragues no fitting procedure to avoid potential issues in interpreta- differences were observed in the FLNF rates of the tion and numerical stability due to differences in mag- bottom, slope or top plots. A considerable proportion nitude between variables. Potential predictor variables of the soil FLNF rates were below the detection limit were gravimetric water content, C:N ratio, N:P ratio, (25%, out of 230 samples). Although the LMER did not total carbon, total nitrogen, total phosphorus and total reveal a significant effect of season on the soil FLNF molybdenum. For soil we additionally included avail- rates, season did affect the number of samples whose able nitrogen, phosphorus and molybdenum, soil pH FLNF rates were below the detection limit (Fig. S4 A & and soil texture. We present the best-fit model for each B), albeit differently in Paracou than in Nouragues (Site data subset, based on this forward stepwise procedure. x Season interactive effect P = 0.004). Whereas in As an additional check of robustness we used an Paracou the number of soil FLNF rates that were below Akaike weight approach to assess the importance of the detection limit did not differ between wet and dry predictor variables. We summed Akaike weights com- season, in Nouragues values below the detection limit puted for all possible first-order models containing a occurred primarily in the dry season (18 out of 52 given predictor, to obtain a measure of the relative samples versus 2 out of 56 samples, in the dry and wet variable importance for this predictor (Burnham and season, respectively, P < 0.001). When conducting the Plant Soil (2020) 450:93–110 99 Table 1 Range, median, mean and standard error (SE) of FLNF rates at forest sites Paracou and Nouragues Paracou Nouragues Wet Dry Wet Dry Bottom Slope Top Bottom Slope Top Bottom Slope Top Bottom Slope Top Soil Ethylene Minimum BDL* BDL BDL BDL BDL BDL BDL BDL BDL BDL BDL BDL production Median 0.006 0.005 BDL 0.015 0.008 0.001 0.002 0.003 0.004 0.014 0.006 0.012 (nmol g−1 h−1) Mean 0.012 0.017 0.002 0.014 0.013 0.003 0.003 0.005 0.028 0.055 0.010 0.037 Maximum 0.044 0.191 0.022 0.029 0.056 0.018 0.006 0.034 0.404 0.298 0.086 0.460 SE 0.003 0.009 0.001 0.002 0.002 0.001 0.000 0.002 0.020 0.030 0.004 0.021 N fixation rate Minimum BDL BDL BDL BDL 0.001 BDL 0.020 BDL BDL 0.001 BDL BDL (kg N ha−1 y−1) Median 0.296 0.183 0.008 0.738 0.350 0.060 0.084 0.119 0.129 0.537 0.228 0.395 Mean 0.565 0.724 0.091 0.663 0.528 0.149 0.100 0.176 0.742 1.301 0.405 1.279 Maximum 2.082 7.836 1.215 1.449 2.168 0.858 0.181 1.157 8.944 10.513 3.635 15.850 SE 0.168 0.379 0.060 0.077 0.099 0.054 0.010 0.055 0.460 1.073 0.171 0.741 Sign. a a b a a b a a a a a a Litter Ethylene Minimum 0.026 0.025 BDL BDL BDL BDL 0.014 0.010 0.003 BDL BDL BDL production Median 0.186 0.143 0.112 0.111 0.053 0.000 0.109 0.059 0.030 0.087 0.045 0.043 (nmol g−1 h−1) Mean 0.524 0.433 0.403 0.376 0.290 0.001 0.346 0.300 0.150 0.207 0.046 0.139 Maximum 3.018 2.500 3.976 2.981 4.554 0.007 2.898 2.407 1.252 0.930 0.120 2.112 SE 0.188 0.127 0.186 0.156 0.219 0.001 0.145 0.119 0.063 0.056 0.008 0.103 N fixation rate Minimum 0.011 0.005 BDL BDL BDL BDL 0.005 0.003 0.001 BDL BDL BDL (kg N ha−1 y−1) Median 0.052 0.045 0.044 0.028 0.015 0.000 0.042 0.017 0.008 0.043 0.015 0.019 Mean 0.166 0.127 0.142 0.109 0.102 0.000 0.126 0.112 0.045 0.111 0.017 0.065 Maximum 0.889 0.580 1.332 0.898 1.707 0.002 0.615 0.979 0.418 0.800 0.053 0.867 SE 0.054 0.034 0.062 0.046 0.083 0.000 0.044 0.048 0.021 0.041 0.002 0.042 Sign. a a ab b b c a abd ac bcd ce be Rates measured in sample types soil and leaf litter are given for the wet and dry season separately and split up by landscape unit. Rates are expressed both on a mass basis as ethylene production rates and on an area basis as N fixation rates. Significant differences (P < 0.05) within a single site and sample type are indicated by differing letters. Calculation of median, mean and standard error and statistical analyses were performed from data including BDL* measurements. *BDL, below detection limit of 0.01 nl ethylene produced per ml air. This equates to a detection limit of 1*10−4 nmol g−1 h−1 and 7*10−4 kg N ha− LMER analysis on a subset of the Nouragues soil FLNF correlated with total P and available N and was highest rates, excluding samples where FLNF rates were below on the slopes. PC2, explaining 27% of the variation, the detection limit, we identified an effect of season contained the seasonally-induced variation (Fig. 2a). Total (P = 0.008), with the highest rates occurring in the dry C correlated with total N and was higher in the dry season, season. (Fig. S5 B). while total Mo correlated well with pH and was higher in the wet season. Available Mo correlated well with mois- Soil – Environmental variables As each site had different ture and was also higher in the wet season. Available P was soil FLNF rates relating to their topography, we performed not heavily affected by season, but was higher in the a principal component analysis (PCA) for each site sepa- bottom plots than on either the slopes or tops. In rately, allowing us to visualize site specific relationships Nouragues PC1 and PC2 together explained 68.6% of among soil parameters. In Paracou, PC1 and PC2 ex- the variation (Fig. 2b). The first principal component plained 63.3% of the variation, with PC1 explaining (PC1) explained 55.2% of the variation and, just as in 36.3% of the variation and containing most of the topo- Paracou, mostly contained topographically-induced varia- graphically induced variation (Fig. 2a). Clay content tion. In Nouragues soil properties for top plots were clearly 100 Plant Soil (2020) 450:93–110 Fig. 1 Area-based N fixation rates for soil at Paracou (a) and differences amongst different seasons and landscape units within Nouragues (b) and for leaf litter at Paracou (c) and Nouragues (d). a single site for soil or leaf litter at the P < 0.05 level. Significance Rates represent the means (±1 SE) of the samples collected from testing was performed by mixed effects regression models using the bottom, slope and top landscape positions in the wet (May) and the log transformed values as measured values were non-normal. dry (September) seasons (N = 20). Letters denote significant Data include measurements that were below the detection limit different from bottom and slope plots. Clay content corre- difference between both seasons was further empha- lated with total P and available N and all were highest in sized by the distinct seasonal difference in the propor- the top plots. In contrast to Paracou, however, total Mo tion of FLNF measurements that were below the detec- correlated with available P and both were higher on the tion limit (Fig. S4 C & D): the vast majority (97%) of the bottom and slope plots than on the top plots. Moisture, 16% (33 out of 235) of litter FLNF measurements that total C and total N correlated well with each other and were below the detection limit were from the dry season clustered together at a ± 45° angle from PC1. Available (binomial GLM, P < 0.001). Nonetheless, the impact of Mo and C:N were correlated along PC2, which explained season interacted with that of topographical position 13.4% of the variation. Lastly, pH also varied topograph- (Season x Topography interactive effect P < 0.001) ically and was highest on the bottom and slope plots, yet (Fig. 1c, d). In the wet season FLNF rates were similar the variation was small. For all Pearson correlations, see among topographic positions, but in the dry season, Table S3 A and B. rates on the top plots were 2 to 7.5 times (± SE) lower (P < 0.05) than rates on the bottom or slope plots, which Leaf litter – N fixation Leaf litter N fixation rates were did not differ from each other. not different between sites. On average, litter FLNF rates in the wet season were 7.5 to 13.5 times (± SE) Leaf litter – Environmental variables Because the higher (P < 0.001) than in the dry season. This LMER stated that FLNF rates were similar between sites Plant Soil (2020) 450:93–110 101 Fig. 2 Principal Component Analysis of soil variables in Paracou in = available P, N in = available N, Mo in = available Mo, Mois- (a) and Nouragues (b). Points on the graph are plot averages. C = ture = water content, pH = pH, Clay = % clay, Sand = % sand and total C, N = total N, P = total P, Mo = total Mo, C:N = C:N ratio, P BD = bulk density we analysed the leaf litter stoichiometry of both sites related to FLNF, and together with P, explained 36% of together. Together, the first two principal components of the variation. For Nouragues a model containing only total the PCA accounted for 73.3% of the variation (Fig. 3). N, total C or water content explained 40% of the variation PC1 explained 49.8% of the variation and distinguished in FLNF rates. These three parameters were strongly between the wet and dry season. Along this component, correlated and the regression analysis deemed the models moisture, N and P content all correlated positively with containing either one of them singularly equally valid each other and negatively with C:N and N:P ratio, (Table 2a). indicating that moisture, N and P content was higher In litter, across both sites and seasons, about 46% of in the wet season, while C:N and N:P ratios were higher the variation in FLNF rates was explained by water and in the dry season. This was confirmed by linear effects N:P ratio (Table 2b). While water had a positive effect regression analysis (Table S2). PC2, accounting for on FLNF rates, N:P ratio showed a negative effect. As 23.5% of the variation, correlated positively with C FLNF rates differed between seasons we ran the step- content. The absence of a site effect on PC1 and PC2 wise analysis again for each season separately. In the indicates that leaf litter chemistry was similar for both wet season 33% of the variation could be explained by sites. For all Pearson correlations, see Table S3 C. water, C content, P content and C:N ratio. The model for the dry season explained about 68% of the variation and Drivers of FLNF in soils and litter In soils, stepwise was dependent on the positive influence of water and the regression analyses indicated that soil water content, avail- negative influence of N content. able P and pH were the primary drivers of N fixation rates (Table 2a), explaining 24% of the variation in soil FLNF rates. Because soil FLNF rates differed significantly be- tween both sites, we ran the stepwise regression analysis Discussion again for each site separately. In Paracou, P exerted a strong effect, as both total P and available P had a positive Overall, the sum of FLNF rates of soils and leaf litter influence on FLNF rates. Available N was negatively measured in this study fall into the lower end of the 0.1– 102 Plant Soil (2020) 450:93–110 Fig. 3 Principal Component Analysis of leaf litter variables in P = total P, Mo = total Mo, C:N = C:N ratio, N:P = N:P ratio and both Paracou and Nouragues and for both wet and dry season. Moisture = water content Dots on the graph are averaged for plot. C = total C, N = total N, Table 2 Results of the forward stepwise selection analyses for N fixation rates in soil (A) and leaf litter (B). For these analyses we excluded the measurements that were below the detection limit A Site Season Parameters R2 Par & Nou Wet & dry +0.57 Water* +0.20 Available P* +0.17 pH* 0.240 F1,169 = 15.6 F1,179 = 4.8 F1,180 = 4.5 Par Wet & dry +1.1 Total P*** +0.34 Available P** −0.51 Available N** 0.362 F1,88 = 44.7 F1,91 = 7.2 F1,68 = 9.3 +0.25 Total N* 0.398 F1,88 = 5.0 Nou Wet & dry +0.25 Total C* 0.396 F1,88 = 4.7 +0.25 Water* 0.401 F1,88 = 5.0 B Site Season Parameters Par & Nou Wet & Dry +0.76 Water*** −0.32 N:P** 0.457 F1,40 = 24.67 F1,195 = 8.34 Par & Nou Wet +1.71 Water** +0.65 Total P** +1.38 Total C** + 0.81 CN* 0.328 F1,65 = 9.10 F1,117 = 10.69 F1,117 = 8.64 F1,115 = 5.87 Par & Nou Dry +2.63 Water*** −0.79 Total N* 0.676 F1,82 = 26.26 F1,73 = 6.70 Each row lists the parameters that were included into the model that provided the best fit, based on AICc values. Columns one and two give information regarding the dataset that was used. For soils (A) a model including data from both sites (Par, Paracou; Nou, Nouragues) and seasons was made along with models for each site separately because N fixation rates were different between sites. For leaf litter (B) a model including data from both sites and seasons was made along with models for each season separately because seasonality affected N fixation. Parameters that had a significant effect (* = p < 0.05, ** = p < 0.01 and *** = p < 0.001) on N fixation rates are given with their F values. Plus or minus numbers represent beta values. R2 was calculated for the fit of the modelled and measured fixation rates 1 The stepwise selection identified total N, total C and water content as equally important for determining FLNF rates in Nouragues soils. This is because these three variables are highly correlated and behave similarly in the model. Explanations as to why this occurs are offered in the text Plant Soil (2020) 450:93–110 103 60 kg N ha−1 y−1 range reported for FLNF on tropical metabolism require sufficient amounts of water, but forest floors worldwide (Reed et al. 2011), and much nitrogenase activity is hindered in the presence of oxy- below the more recent estimate of 15–36 kg N ha−1 y−1 gen (O2) (Hill 1988). Oxygen binds to nitrogenase and fixed in tropical forests (Pajares and Bohannan 2016) inhibits its function (Hartmann and Burris 1987) and and the average ethylene production rate of 5.32 nmol because increased soil moisture decreases soil O2 con- ethylene g−1 h−1 reported in a recent meta-analysis on N centrations, water content can have a large impact on fixation rates in tropical forest ecosystems (Dynarski soil FLNF rates. As diazotrophs are mainly heterotro- and Houlton 2018). The FLNF rates found in the present phic the thickness of the soil water film, which is im- study are much lower than those found in Costa Rica portant for the diffusion rates of extracellular enzymes (Reed et al. 2007; Reed et al. 2010; Reed et al. 2013), and soluble organic-C substrates and is directly affected Panama (Barron et al. 2009; Wurzburger et al. 2012), by soil water content (Davidson and Janssens 2006), Puerto Rico (Cusack et al. 2009) or in the younger sites will also play a role. along a Hawaiian chronosequence (Crews et al. 2000). In Paracou, we observed no seasonal effect on soil An overview of FLNF rates reported by these authors FLNF rates and although rates were typically reported to can be found in Table S4. Sullivan et al. (2014) mea- be higher in the wet season than in the dry season (e. g. sured N fixation rates in Costa Rica, about 35 km away Hofmockel and Schlesinger 2007; Reed et al. 2007), from where Reed et al. (2007) conducted their study. Matson et al. (2014) have shown that the opposite can These authors reported soil and leaf litter FLNF rates also occur. They postulated that it is likely that moisture that were lower than those previously published by fluctuations were not directly responsible for their ob- Reed et al. (2007), but while their reported soil FLNF served seasonal changes in FLNF rates, just as in our study rates were similar to those found in our study, their it might not have contributed to unchanging rates. Their reported leaf litter FLNF rates were still twice as high reasoning was based on the fact that cyanobacteria can fix as the rates we found. Other studies of FLNF rates N at moisture levels as low as 6% and in one study reached conducted in less fertile tropical forests, e.g. on the maximum N fixation rates at soil moistures between 22 higher altitudes of an altitudinal transect in Ecuador and 42% (Jones 1977), though it did not specify whether (Matson et al. 2014) or in the older sites of a Hawaiian this was in sandy or clayey soil. As soil moisture in chronosequence (Crews et al. 2000), reported FLNF Paracou always remained between 15 and 44% rates comparable to those found in the present study. (Table S1) it is likely that diazotrophs had sufficient mois- In Hawaii, Crews et al. (2000) measured decreasing ture to keep oxygen levels low and continue N fixation. rates of leaf litter FLNF on increasingly older soils and At Nouragues, the effect of soil moisture on FLNF attributed this decrease to a combination of lower con- rates was high (40% of variability explained by soil centrations of geologically cycled nutrients, such as P, moisture alone, Table 2a) and seasonality caused an and high N pools at their oldest sites. This, coupled with interesting pattern in the distribution of very low FLNF increases in FLNF rates after P fertilization, led them to rate measurements (below detection limit). The vast one of their main conclusions; that low P availability majority of below detection limit measurements oc- limited FLNF rates at their older sites. It would, how- curred during the dry season (Fig. S4 B) and when ever, be inaccurate to conclude that soil P alone deter- analysing only data for which FLNF rates were above mines N fixation since, for example, in Panama the detection limit, we found a significant season effect (Wurzburger et al. 2012) two of the studied sites along in Nouragues with quantifiable rates (i.e. in samples a total soil P gradient containing high (AVA) and medi- above the detection limit) actually being higher in the um (Gigante) levels of total soil P displayed the lowest dry season than in the wet season (Fig. S5 B). This rates of FLNF in their study. meant that the range of FLNF rates was broader during the dry season than during the wet season (Fig. S5 A), Soil free-living N fixation indicating that N fixation hotspots were of increasing importance. These are typically found in tropical eco- At both sites moisture was important in the regulation of systems (Pajares and Bohannan 2016) and reflect the FLNF (Table 2a). Water is essential for all microbes, but very small-scale spatial heterogeneity of abiotic factors for diazotrophs in particular water plays an important affecting the dynamics of the diazotroph community role in regulating activity. Not only does their (Reed et al. 2010). As the soil dried out at the onset of 104 Plant Soil (2020) 450:93–110 the dry season the heterogeneity of water-filled pore FLNF rates. This is in line with what was observed in space increased, leading to the creation of aerobic and previous tropical forest studies, where additions of N led anaerobic, as well as dry and mesic microsites that co- to the decrease of N fixation rates (e. g. Barron et al. occured on a small spatial scale (Sexstone et al. 1985). 2009; Crews et al. 2000; Cusack et al. 2009). This Besides water content, available P and pH were the observation is generally attributed to the theory that most important predictors for soil FLNF at our sites. The many heterotrophic N2 fixers are facultative fixers and pH in our soils is very low, ranging between 3.8 and 4.2 able to down-regulate their fixation pathway when other (Table S1). This is a relatively small range, yet pH has sources of N are available and more cost-efficient to been identified to affect bacterial community composi- acquire (Menge et al. 2009). The topographic effect on tion and diversity (Rousk et al. 2010; Tripathi et al. FLNF rates in Paracou, effectively resulting in lower 2014) or soil CO2 effluxes (Courtois et al. 2018) at the rates on the top than on bottom and slope plots is likely local scale even when changes are small. We were caused by the interplay of P and N in the Paracou soils as unable to find studies assessing the effect of pH on they also varied with topographic position. Total P was FLNF in tropical ecosystems, but nitrogenase activity highest on the slope and available P highest on the of soil diazotrophs has been shown to increase with bottom, while both were lowest on the top. Because of increased pH in a German pine forest situated in north- the higher clay content of the Paracou slope soils there east Bavaria (Limmer and Drake 1996). Roper and are likely more aluminium and iron oxides (Tiessen Smith (1991) found that nitrogenase activity of bacteria et al. 1994) that are able to occlude more P, resulting extracted from clayey Australian soils reached its peak in higher soil total P contents as seen on the slope. On around pH 7.5, matching the nitrogenase pH optimum the more sandy bottom landscape position metal oxides found by Pham and Burgess (1993), and decreased in that can occlude P are likely scarcer and the higher water more acidic soils. Their study, however, was carried out content, especially during rain events when runoff on soils that at the start of the experiment were only causes disproportionate changes in water status, leads slightly acidic after which the pH was decreased during to reduction of iron oxides (Colombo et al. 2014), the course of the study. Their result is therefore not liberating occluded P and provoking higher concentra- necessarily representative for N fixing microbial com- tions of available P. munities that developed in acidic soils, such as in our In Nouragues, we found that water content, C and N study. However, taking into account an enzymatic pH performed equally well, each individually explaining optimum of 7.5 (Pham and Burgess 1993) and decreased 40% of the observed variation in FLNF rates, suggesting nitrogenase activity in more acidic temperate soils (Zahran that the absolute concentrations of C and N in the soil 1999), a positive relationship between soil pH and FLNF explain a substantial part of the variation in FLNF rates. rates in our acidic tropical forest sites is plausible. It is surprising that the model identified a positive effect Phosphorus can play a pivotal role in regulating N of N content on FLNF rates, given that most studies fixation rates and is often limiting the rate of this process associate increased N with decreased or unchanged in highly weathered lowland tropical soils (e. g. fixation rates (Camenzind et al. 2018). Additionally, Camenzind et al. 2018; Reed et al. 2011). In Paracou, we would expect that available N, which was also both P predictors were positively correlated with the included as potential variable but was not selected by FLNF rates, suggesting a higher activity of N fixing the model, would be identified as a variable affecting microbes when more P is present in the soil. Additions FLNF rates instead of N content because no decompo- of P to the forest floor have shown to increase both sition is needed before assimilation. Likely, the identifi- diversity and abundance of diazotrophs in tropical soils cation of N by the model is purely mathematical and (Reed et al. 2010), as well as increases in FLNF rates (e. caused by its near perfect correlation with C content g. Benner et al. 2007; Reed et al. 2013). The positive (Table S3 B). The positive relationship of C with FLNF correlation between available P and total P on the one rates, which is predominantly carried out by heterotro- hand, and FLNF rates on the other, is in line with the phic diazotrophs (Sprent and Sprent 1990), is the result longstanding idea that increased P benefits FLNF rates of decomposition of organic material and subsequent in tropical lowland forests. In addition to P, the model assimilation of additional C providing the energy nec- for Paracou also identified available N as a predictor, essary to carry out fixation. In Nouragues, the high indicating that higher available N is linked to decreased correlation between C content and moisture (Table S3 Plant Soil (2020) 450:93–110 105 B) partly explains their equal importance in the model. model explaining 46% of the variation (Table 2b). The Their correlation might be due to the influence of soil negative influence of N:P ratio is in line with what was moisture on decomposition rates, especially when it is postulated by Reed et al. (2007) and likely illustrates a very wet (negative relation) or very dry (positive rela- stoichiometric and energetic balance shift. In a tion). At high moisture levels, soil organic carbon and decomposing leaf a decreasing N:P ratio leads to P be- nitrogen stocks increase because of the slower decom- coming comparatively more abundant, shifting the stoi- position in water-saturated soil (Van Sundert et al. chiometric balance to move away from P limitation to N 2018). Moreover, soils in Nouragues are clay-rich, ex- limitation, favouring diazotrophs. Additionally, increased acerbating this effect as soils containing more clay sta- P in the environment may alleviate energetic constraints bilize and store more soil organic matter than sandy soils (see above) and enable diazotrophs to invest the required (Reis et al. 2014), such as those in Paracou. Both in- energy into fixing N. As the general model included data creased water and C can be beneficial for free-living from both seasons, a large influence of water content, as diazotrophs as many species possess heterotrophic an- evidenced by the largest beta value (Table 2b), was to be aerobic metabolisms (Dixon and Kahn 2004) and pro- expected because litter FLNF rates were significantly af- liferate under oxygen poor and carbon rich conditions. fected by seasonality; wet season FLNF rates were up to In Nouragues the topographical patterns of N and P nearly two times higher than dry season FLNF rates were different from Paracou, with much higher total P (Table 1) and the number of samples for which FLNF concentrations that occurred on the top landscape position rates were below the detection limit was much smaller in instead of on the slopes, and much smaller differences in the wet than in the dry season (Fig. S4 C & D). As available P concentrations among the topographic posi- expected, litter dried out severely during the dry season tions. This likely played a role in the regulation of soil and water content decreased from an average ~64 to ~40% FLNF rate, causing them to remain high on the top land- across both sites. Given that diazotrophs grow and prolif- scape position. Because of this strong difference in topog- erate in aqueous environments (Scott 1957), it is likely that raphy effects between Nouragues and Paracou, we cannot leaf litter water content in the dry season dropped below a draw general conclusions about landscape-scale variations threshold of minimum water required for diazotroph pro- in soil FLNF rates across topographies in French Guianese liferation, resulting in a net drop of FLNF rate. This tropical forests. Instead, our results support the idea that response of diazotroph activity to leaf litter moisture was soil FLNF rates at our sites varied in function of water and already observed in a sub-tropical karst forest, where nutrient availability, similar to what was reviewed by researchers found that decreases in leaf litter moisture Dynarski et al. (2018). resulted in decreased FLNF rates (Li et al. 2018). The Lastly, it is interesting to note that total or available large beta value (2.63, Table 2b) associated with water Mo was never selected as explanatory variable for soil content in the dry season model hints towards the disrup- FLNF rates, in spite of playing a regulatory role in tive effect of water shortage. This model explained 67% of FLNF at other sites (Barron et al. 2009; Reed et al. the variation in litter FLNF rates and included only water 2013; Wurzburger et al. 2012). Molybdenum concen- content and litter N, which, just as the N:P ratio in the trations in our study sites were slightly higher than those general model, had a negative effect on the FLNF rates. As found elsewhere (Gupta and Lipsett 1981), and avail- mentioned earlier, N assimilation is cheaper than N fixa- able Mo was ten times higher than in Panama tion from an energetic point of view and when more N is (Wurzburger et al. 2012), which was the only study present fixation will likely be down regulated (Menge reporting the effect of available molybdenum on FLNF. et al. 2009). On the top landscape position in Paracou, Likely, the concentration of available Mo at our sites is also during the dry season, leaf litter water content was high enough to preclude any regulatory role of Mo in lowest while N:P ratio was highest. This likely led to a this P-limited environment. complete collapse of N fixation (Fig. 1c), because N was in ample supply compared to P and water was scarce. Leaf litter free-living N2 fixation During the wet season, litter stoichiometry explained 33% of the variation in FLNF rates which likely reflects, Overall, leaf litter FLNF rates did not vary with topogra- at least partially, the interactions between seasonal changes phy and were instead driven mainly by water content and in labile C availability, P content and N demand (Reed the N:P ratio of the leaf litter, as shown by the overall et al. 2007). Litterfall peaks at the onset of the dry season 106 Plant Soil (2020) 450:93–110 (Chave et al. 2010; Wagner et al. 2013) and once the wet 2004) and that N fixing Archaea within the community season starts the daily rainfall provides a vehicle for the may possess different nutritional requirements than bac- movement of readily decomposable, dissolved organic C terial diazotrophs (Leigh 2000) are additional reasons to (DOC) within the litter layer (Courtois et al. 2018). The study the diazotroph community composition. Also, we N2-fixing microbial community in the litter layer is dom- did not assess the iron (Fe) or the vanadium (V) avail- inated by heterotrophic microorganisms (Sprent and abilities at our sites while they have the potential of Sprent 1990) and the influx of litter DOC provides an participating in FLNF rate regulation (Zhang et al. easily accessible energy source for these diazotrophs. Ad- 2016). While Mo is a necessary co-factor of most nitro- ditionally, in our study, we found that the concentration of genase enzymes (Igarashi and Seefeldt 2003), Fe is both P and N in leaf litter was higher in the wet season than found in all known nitrogenases and the occurrence of in the dry season, similar to what was found in another ‘iron-only’ (Fe-Fe) nitrogenases has been widely docu- study in Costa Rica (Wood et al. 2005). However, relative mented (e. g. Yang et al. 2014; Zheng et al. 2018). The to N content, P content increased more towards the wet role of V in regulation of FLNF is understudied, but it is season, resulting in a lower N:P ratio during the wet certain that the occurrence of an alternative enzymatic season. In combination with more labile C input, this could co-factor, namely the vanadium-iron (V-Fe) cofactor, is stimulate N fixation through the relief of energetic con- widespread and V availability may therefore also play a straints and the added advantage of being able to fix N regulatory role in nitrogenase biosynthesis (Hu et al. compared to assimilating N from an environment where it 2012). A follow-up study investigating the community is, comparatively, less abundant than in the dry season. composition of the diazotrophs, specifically looking into This process was already observed in tallgrass prairie soils the prevalence of aerobic and anaerobic organisms, (Eisele et al. 1989). combined with research into the occurrence of V-Fe In contrast to litter P content, which changed signif- nitrogenases and V and Fe availabilities in the soil and icantly across seasons yet showed no significant change leaf litter could enable us to explain more of the ob- along the topographical gradient, litter Mo content served variation than we presently could. changed significantly between the two seasons and along the topographical gradient. Just as in soils, how- ever, and in spite of its importance for N fixation (Kaiser Conclusion et al. 2005; Seefeldt et al. 2009), we found no evidence to support a regulatory role for Mo content. Similar as in Our study has shed light on the drivers behind soils, litter Mo concentrations were ten-fold higher than FLNF in tropical soil and leaf litter on the Guiana those reported from other tropical sites (Barron et al. Shield and has shown that the rates of FLNF are 2009; Bowell and Ansah 1993; Reed et al. 2013; much lower than those estimated for most tropical Wurzburger et al. 2012) and likely too high to render a forests elsewhere. Water, N and P played the main regulatory role to Mo in litter FLNF rates. roles in determining FLNF rates in both sample Lastly, in both soils and leaf litter our models were types, while pH only regulated in the soil. The unable to explain more than 67% of the observed vari- effect of seasonality differed between sample type ation, and in most cases only around 40%. This means and differences in FLNF regulation between sites that often the majority of variation in FLNF rates could could be observed in soils, but not in litter. De- not be explained by the variables we selected and thus it spite having been shown to influence N fixation is very likely that other factors not measured in this rates in other mature tropical forests, the micro study contribute to the regulation of FLNF in our trop- nutrient Mo played no role in the regulation of ical lowland forests. Knowing the diazotroph commu- FLNF rates at our sites in French Guiana. nity composition, in both soil and leaf litter, could In the sandy site, Paracou, the stimulating effect enhance our understanding of their nutritional and envi- of P and the inhibiting effect of N were the main ronmental needs and help us estimate at what point drivers behind soil FLNF, but in the clayey site, parameters such as moisture, pH and P or N availability Nouragues, soil FLNF was mainly stimulated by are beneficial or, inversely, detrimental. The likelihood water or C content. Our models for soil FLNF that within a single diazotroph community both aerobic could not explain more than 40% of the observed and anaerobic lifestyles can occur (Dixon and Kahn variation, illustrating the complexity of predicting Plant Soil (2020) 450:93–110 107 fixation rates upon measured variables in an envi- References ronment that is highly heterogeneous on a regional, local and even micro scale. In leaf litter we also Aguilos M, Hérault B, Burban B, Wagner F, Bonal D (2018) What identified water, N and P as main drivers, but the drives long-term variations in carbon flux and balance in a underlying mechanisms that caused variation may tropical rainforest in French Guiana? Agric For Meteorol 253 -254 :114– 123 . https://doi.org/10.1016/j . have been different compared to the soil. In the leaf agrformet.2018.02.009 litter we observed no differences in FLNF rates Amundson R, Austin AT, Schuur EAG, Yoo K, Matzek V, Kendall between sites, but during the dry season litter rates C, Uebersax A, Brenner D, Baisden WT (2003) Global exhibited a drastic decline that was mainly related patterns of the isotopic composition of soil and plant nitro- gen. Global Biogeochem Cy 17. https://doi.org/10.1029 to water insufficiency and the inhibiting effect of /2002gb001903 N. During the wet season water was still of impor- Barron AR, Wurzburger N, Bellenger JP, Wright SJ, Kraepiel tance, but now stimulating effects of C and P also AML, Hedin LO (2009) Molybdenum limitation of came into play. It is important to note that our litter asymbiotic nitrogen fixation in tropical forest soils. Nat Geosci 2:42–45. https://doi.org/10.1038/Ngeo366 wet season model explained only about 30% of the Barron AR, Purves DW, Hedin LO (2011) Facultative nitrogen observed variation. fixation by canopy legumes in a lowland tropical forest. It is likely that in both soil and leaf litter Oecologia 165:511–520. https://doi.org/10.1007/s00442- diazotroph community composition and iron or 010-1838-3 vanadium availability had an influence on FLNF Bates D, Mächler M, Bolker B, Walker S (2015) Fitting linear mixed-effects models using lme4. 67:48. https://doi. rates making them interesting to measure in future org/10.18637/jss.v067.i01 studies. In the larger framework of global change, Batterman SA, Hedin LO, van Breugel M, Ransijn J, Craven DJ, where N deposition is expected to increase Hall JS (2013) Key role of symbiotic dinitrogen fixation in (Penuelas et al. 2013), P deposition to tropical tropical forest secondary succession. Nature 502:224–227. https://doi.org/10.1038/nature12525 forests may change (Gross et al. 2016) and the Benner JW, Conroy S, Lunch CK, Toyoda N, Vitousek PM (2007) possibility of a drier Amazon basin (IPCC 2013) Phosphorus fertilization increases the abundance and may cause disruptive changes to the FLNF rates in Nitrogenase activity of the Cyanolichen Pseudocyphellaria soil and litter. Nutrient addition studies may offer crocata in Hawaiian montane forests. Biotropica 39:400–405. https://doi.org/10.1111/j.1744-7429.2007.00267.x clues as to the response of FLNF to changes in N Bongers F, Charles-Dominique P, Forget P, Thèry M (2001) or P supply, but testing the influence of climatic Nouragues dynamics and plant-animal interactions in a changes in situ calls for a very specific type of Neotropical rainforest. Kluwer Academic Publishers, studies. Dordrecht, The Netherlands https://doi.org/10.1007/978-94- 015-9821-7 Bowell RJ, Ansah RK (1993) Trace element budget in an African Acknowledgements This research was supported by the Euro- savannah ecosystem. Biogeochemistry 20:103–126. pean Research Council Synergy grant ERC-2013-SyG. https://doi.org/10.1007/bf00004137 610028-IMBALANCE-P. We thank the staff of the Nouragues Bray RH, Kurtz LT (1945) Determination of Total, organic, and station, managed by USR mixte LEEISA. available forms of phosphorus in soils. Soil Sci 59:39–46 (CNRS; Cayenne), and the Paracou station, managed by UMR Brookshire ENJ, Gerber S, Menge DNL, Hedin LO (2012) Large Ecofog (CIRAD, INRA; Kourou). losses of inorganic nitrogen from tropical rainforests suggest Both research stations received support from BInvestissement a lack of nitrogen limitation. Ecology letters 15:9–16. d’Avenir^ grants managed by Agence Nationale de la Recherche https://doi.org/10.1111/j.1461-0248.2011.01701.x (CEBA: ANR-10-LABX-25-01,ANAEE France: ANR-11-INBS- Burnham K, Anderson D (2002) Model selection and multimodel 0001). We thank Tom Van der Spiet, Anne Cools and Miguel inference: a practical information-theoretic approach. Portillo Estrada for their help with analyses and Philippe Gaucher, Springer-Verlag, New York Nadine Calluy, Lore Verryckt and Kevin Van Sundert for their help Černá B, Rejmánková E, Snyder JM, Šantrůčková H (2009) with logistics and statistics. Heterotrophic nitrogen fixation in oligotrophic tropical marshes: changes after phosphorus addition. Hydrobiologia 627:55–65. https://doi.org/10.1007/s10750-009-9715-y Open Access This article is distributed under the terms of the Chave J, Navarrete D, Almeida S, Álvarez E, Aragão LEOC, Creative Commons Attribution 4.0 International License (http:// Bonal D, Châtelet P, Silva-Espejo JE, Goret JY, von creativecommons.org/licenses/by/4.0/), which permits unrestrict- Hildebrand P, Jiménez E, Patiño S, Peñuela MC, Phillips ed use, distribution, and reproduction in any medium, provided OL, Stevenson P, Malhi Y (2010) Regional and seasonal you give appropriate credit to the original author(s) and the source, patterns of litterfall in tropical South America. provide a link to the Creative Commons license, and indicate if Biogeosciences 7:43–55. https://doi.org/10.5194/bg-7-43- changes were made. 2010 108 Plant Soil (2020) 450:93–110 Colombo C, Palumbo G, He J-Z, Pinton R, Cesco S (2014) Grau O, Peñuelas J, Ferry B, Freycon V, Blanc L, Desprez M, Review on iron availability in soil: interaction of Fe minerals, Baraloto C, Chave J, Descroix L, Dourdain A, Guitet S, plants, and microbes. J Soils Sediments 14:538–548. Janssens IA, Sardans J, Hérault B (2017) Nutrient-cycling https://doi.org/10.1007/s11368-013-0814-z mechanisms other than the direct absorption from soil may Courtois EA, Stahl C, van den Berge J, Bréchet L, van control forest structure and dynamics in poor Amazonian Langenhove L, Richter A, Urbina I, Soong JL, Peñuelas J, soils. Sci Rep 7:45017. https://doi.org/10.1038/srep45017 Janssens IA (2018) Spatial variation of soil CO2, CH4 and Gross A, Turner BL, Goren T, Berry A, Angert A (2016) Tracing N2O fluxes across topographical positions in tropical forests the sources of atmospheric phosphorus deposition to a trop- of the Guiana shield. Ecosystems. 21:1445–1458. https://doi. ical rain Forest in Panama using stable oxygen isotopes. org/10.1007/s10021-018-0232-6 Environ Sci Technol 50:1147–1156. https://doi.org/10.1021 Crews TE, Farrington H, Vitousek PM (2000) Changes in /acs.est.5b04936 Asymbiotic, heterotrophic nitrogen fixation on leaf litter of Gupta UC, Lipsett J (1981) Molybdenum in soils, plants, and Metrosideros polymorpha with long-term ecosystem devel- animals. In: Brady NC (ed) advances in agronomy, vol 34. opment in Hawaii. Ecosystems 3:386–395. https://doi. Academic press, pp 73-115. https://doi.org/10.1016/S0065- org/10.1007/s100210000034 2113(08)60885-8 Cusack DF, Silver W, McDowell WH (2009) Biological nitrogen Gutschick VP (1981) Evolved strategies in nitrogen acquisition by fixation in two tropical forests: ecosystem-level patterns and plants. American naturalist:607–637 effects of nitrogen fertilization. Ecosystems 12:1299–1315. Hammond D (2005) Tropical forests of the Guiana shield: ancient https://doi.org/10.1007/s10021-009-9290-0 forests in a modern world. CABI Publishing, Cambridge Davidson EA, Janssens IA (2006) Temperature sensitivity of soil Hardy RW, Holsten RD, Jackson EK, Burns RC (1968) The carbon decomposition and feedbacks to climate change. acetylene-ethylene assay for N(2) fixation: laboratory and Nature 440:165–173. https://doi.org/10.1038/nature04514 field evaluation. Plant Physiol 43:1185–1207 Dixon R, Kahn D (2004) Genetic regulation of biological nitrogen Hartmann A, Burris RH (1987) Regulation of nitrogenase activity fixation. Nat Rev Microbiol 2:621–631. https://doi. by oxygen in Azospirillum brasilense and Azospirillum org/10.1038/nrmicro954 lipoferum. J Bacteriol 169:944–948 Dynarski KA, Houlton BZ (2018) Nutrient limitation of terrestrial Hartwig UA (1998) The regulation of symbiotic N2 fixation: a free-living nitrogen fixation. The New phytologist 217: conceptual model of N feedback from the ecosystem to the 1050–1061. https://doi.org/10.1111/nph.14905 gene expression level. Perspectives in Plant Ecology, Eisele KA, Schimel DS, Kapustka LA, Parton WJ (1989) Effects Evolution and Systematics 1:92–120. https://doi. of available P and N:P ratios on non-symbiotic dinitrogen org/10.1078/1433-8319-00054 fixation in tallgrass prairie soils. Oecologia 79:471–474. Hedin LO, Brookshire ENJ, Menge DNL, Barron AR (2009) The https://doi.org/10.1007/bf00378663 nitrogen paradox in tropical Forest ecosystems. Annu Rev Epron D, Bosc A, Bonal D, Freycon V (2006) Spatial variation of Ecol Evol Syst 40:613–635. https://doi.org/10.1146/annurev. soil respiration across a topographic gradient in a tropical rain ecolsys.37.091305.110246 forest in French Guiana. J Trop Ecol 22:565–574. https://doi. org/10.1017/s0266467406003415 Hill S (1988) How is nitrogenase regulated by oxygen? FEMS FAO (1998) World Reference Base for Soil Resources, by ISSS– Microbiol Rev 4:111–129. https://doi.org/10.1111/j.1574- ISRIC–FAO. World Soil Resources Report No. 84., Rome, p 6968.1988.tb02738.x 88 Hofmockel KS, Schlesinger WH (2007) Carbon dioxide effects on Furnkranz M, Wanek W, Richter A, Abell G, Rasche F, Sessitsch heterotrophic dinitrogen fixation in a temperate pine Forest A (2008) Nitrogen fixation by phyllosphere bacteria associ- abbreviations: FACE, free air carbon dioxide enrichment; ated with higher plants and their colonizing epiphytes of a NPP, net primary production. Soil Sci Soc Am J 71:140– tropical lowland rainforest of Costa Rica. The ISME journal 144. https://doi.org/10.2136/sssaj2006.110 2:561–570. https://doi.org/10.1038/ismej.2008.14 Hu Y, Lee CC, Ribbe MW (2012) Vanadium Nitrogenase: a two- Galloway JN, Schlesinger WH, Levy H, Michaels A, Schnoor JL hit wonder? Dalton transactions (Cambridge, England: 2003) (1995) Nitrogen fixation: Anthropogenic enhancement- 41:10 https://doi.org/10.1039/c1dt11535a environmental response. Global Biogeochem Cy 9:235– Hungate BA, Dukes JS, Shaw MR, Luo Y, Field CB (2003) 252. https://doi.org/10.1029/95GB00158 Atmospheric science. Nitrogen and climate change. Science Galloway JN, Dentener FJ, Capone DG, Boyer EW, Howarth RW, (New York, NY) 302:1512‑1513. https://doi.org/10.1126 Seitzinger SP, Asner GP, Cleveland CC, Green PA, Holland /science.1091390 EA, Karl DM, Michaels AF, Porter JH, Townsend AR, Igarashi RY, Seefeldt LC (2003) Nitrogen fixation: the mechanism Vörösmarty CJ (2004) Nitrogen cycles:past, present, and of the Mo-dependent nitrogenase. Critical reviews in bio- future. Biogeochemistry 70:153–226 chemistry and molecular biology 38:351–384. https://doi. Gee GW, Bauder JW (1986) Particle-size analysis. In: methods of org/10.1080/10409230391036766 soil analysis part 1: physical and mineralogical methods 2nd IPCC (2013) Climate change 2013: the physical science basis. edition edn. American Society of Agronomy & soil science Contribution of working group I to the fifth assessment report Society of America, Madison, Wisconsin, USA of the intergovernmental panel on climate change [stocker, Gourlet-Fleury S, Guehl JM, Laroussinie O (2004) Ecology and T.F., D. Qin, G.-K. Plattner, M. Tignor, S.K. Allen, J. management of a neotropical rainforest. Lessons drawn from Boschung, A. Nauels, Y. Xia, V. Bex and P.M. Midgley Paracou, a long-term experimental research site in French (eds.)]. Cambridge, United Kingdom and New York, NY, Guiana. Elsevier, Meppel, the Netherlands USA. https://doi.org/10.1017/CBO9781107415324 Plant Soil (2020) 450:93–110 109 Jones K (1977) The effects of moisture on acetylene reduction by landscape. Proc Natl Acad Sci U S A 102:10909–10912. Mats of blue-green algae in sub-tropical grassland. Ann Bot https://doi.org/10.1073/pnas.0504929102 41:801–806. https://doi.org/10.1093/oxfordjournals.aob. Quesada CA, Phillips OL, Schwarz M, Czimczik CI, Baker TR, a085355 Patiño S, Fyllas NM, Hodnett MG, Herrera R, Almeida S, Kaiser BN, Gridley KL, Brady JN, Phillips T, Tyerman SD (2005) Alvarez Dávila E, Arneth A, Arroyo L, Chao KJ, Dezzeo N, The role of molybdenum in agricultural plant production. Erwin T, di Fiore A, Higuchi N, Honorio Coronado E, Ann Bot 96:745–754. https://doi.org/10.1093/aob/mci226 Jimenez EM, Killeen T, Lezama AT, Lloyd G, López- van Kekem AJ, Pulles JHM, Khan Z (1996) Soils of the rainforest González G, Luizão FJ, Malhi Y, Monteagudo A, Neill DA, in Central Guyana. Series 2, Georgetown, Guyana Núñez Vargas P, Paiva R, Peacock J, Peñuela MC, Peña Cruz LeBauer DS, Treseder KK (2008) Nitrogen limitation of net A, Pitman N, Priante Filho N, Prieto A, Ramírez H, Rudas A, primary productivity in terrestrial ecosystems is globally Salomão R, Santos AJB, Schmerler J, Silva N, Silveira M, distributed. Ecology 89:371–379. https://doi.org/10.1890 Vásquez R, Vieira I, Terborgh J, Lloyd J (2012) Basin-wide /06-2057.1 variations in Amazon forest structure and function are medi- Leigh JA (2000) Nitrogen fixation in methanogens: the archaeal ated by both soils and climate. Biogeosciences 9:2203–2246. perspective. Current issues in molecular biology 2:125–131 https://doi.org/10.5194/bg-9-2203-2012 Limmer C, Drake HL (1996) Non-symbiotic N2-fixation in acidic R CT (2017) R: A language and environment for statistical com- and pH-neutral forest soils: aerobic and anaerobic differen- puting., 3.4.3 edn., Vienna, Austria tials. Soil Biol Biochem 28:177–183. https://doi.org/10.1016 ter Steege H, Pitman NCA, Phillips OL, Chave J, Sabatier D, /0038-0717(95)00118-2 Duque A, Molino JF, Prévost MF, Spichiger R, Castellanos Losos E, Leigh EG (2004) Tropical Forest diversity and dyna- H, von Hildebrand P, Vásquez R (2006) Continental-scale mism: findings from a large-scale plot network. Univ, patterns of canopy tree composition and function across Chicago Press, Chicago Amazonia. Nature 443:444–447. https://doi.org/10.1038 Matson AL, Corre MD, Burneo JI, Veldkamp E (2014) Free-living /nature05134 nitrogen fixation responds to elevated nutrient inputs in trop- Reed SC, Cleveland CC, Townsend AR (2007) Controls over leaf ical montane forest floor and canopy soils of southern litter and soil nitrogen fixation in two lowland tropical rain Ecuador. Biogeochemistry 122:281–294. https://doi. forests. Biotropica 39:585–592. https://doi.org/10.1111 org/10.1007/s10533-014-0041-8 /j.1744-7429.2007.00310.x Mazerolle MJ (2017) AICcmodavg: model selection and Reed SC, Townsend AR, Cleveland CC, Nemergut DR (2010) multimodel inference based on (Q)AIC(c) Microbial community shifts influence patterns in tropical Menge DN, Levin SA, Hedin LO (2009) Facultative versus obli- forest nitrogen fixation. Oecologia 164:521–531. https://doi. gate nitrogen fixation strategies and their ecosystem conse- org/10.1007/s00442-010-1649-6 quences. Am Nat 174:465–477. https://doi.org/10.1086 Reed SC, Cleveland CC, Townsend AR (2011) Functional ecolo- /605377 gy of free-living nitrogen fixation: a contemporary perspec- Pajares S, Bohannan BJ (2016) Ecology of nitrogen fixing, nitri- tive. Annu Rev Ecol Evol S 42:489–512. https://doi. fying, and denitrifying microorganisms in tropical Forest org/10.1146/annurev-ecolsys-102710-145034 soils. Front Microbiol 7:1045. https://doi.org/10.3389 Reed SC, Cleveland CC, Townsend AR (2013) Relationships /fmicb.2016.01045 among phosphorus, molybdenum and free-living nitrogen Pearson HL, Vitousek PM (2001) Stand dynamics, nitrogen accu- fixation in tropical rain forests: results from observational mulation, and symbiotic nitrogen fixation in regenerating and experimental analyses. Biogeochemistry 114:135–147. stands of Acacia Koa. Ecol Appl 11:1381–1394. https:// https://doi.org/10.1007/s10533-013-9835-3 doi.org/10.1890/1051-0761(2001)011[1381: Reis CES, Dick DP, Caldas JS, Bayer C (2014) Carbon seques- SDNAAS]2.0.CO;2 tration in clay and silt fractions of Brazilian soils under Pearson HL, Vitousek PM (2002) Soil phosphorus fractions and conventional and no-tillage systems. Sci Agric 71:292–301. symbiotic nitrogen fixation across a substrate-age gradient in https://doi.org/10.1590/0103-9016-2013-0234 Hawaii. Ecosystems 5:587–596. https://doi.org/10.1007 Robson RL, Postgate JR (1980) Oxygen and hydrogen in biolog- /s10021-002-0172-y ical nitrogen fixation. Annu Rev Microbiol 34:183–207. Penuelas J et al (2013) Human-induced nitrogen-phosphorus im- https://doi.org/10.1146/annurev.mi.34.100180.001151 balances alter natural and managed ecosystems across the Roper MM, Smith NA (1991) Straw decomposition and nitroge- globe. Nat Commun 4:2934. https://doi.org/10.1038 nase activity (C2H2 reduction) by free-living microorgan- /ncomms3934 isms from soil: effects of pH and clay content. Soil Biol Pérez S, Pérez C, Carmona M, Farina J, Armesto J (2008) Effects Biochem 23:275–283. https://doi.org/10.1016/0038-0717 of labile phosphorus and carbon on non-symbiotic N2 fixa- (91)90064-Q tion in logged and unlogged evergreen forests in Chiloé Roskoski JP (1980) Nitrogen fixation in hardwood forests of the Island, Chile. Rev Chil Hist Nat 81:12 northeastern United States. Plant Soil 54:33–44. https://doi. Pham DN, Burgess BK (1993) Nitrogenase reactivity: effects of org/10.1007/bf02181997 pH on substrate reduction and carbon monoxide inhibition. Rousk J, Bååth E, Brookes PC, Lauber CL, Lozupone C, Caporaso Biochemistry 32:13725–13731. https://doi.org/10.1021 JG, Knight R, Fierer N (2010) Soil bacterial and fungal /bi00212a043 communities across a pH gradient in an arable soil. The Porder S, Asner GP, Vitousek PM (2005) Ground-based and ISME journal 4:1340–1351. https://doi.org/10.1038 remotely sensed nutrient availability across a tropical /ismej.2010.58 110 Plant Soil (2020) 450:93–110 Scott WJ (1957) Water relations of food spoilage microorganisms. Wagner F, Rossi V, Stahl C, Bonal D, Hérault B (2013) In: Mrak EM, Stewart GF (eds) advances in food research, Asynchronism in leaf and wood production in tropical for- vol 7. Academic press, pp 83-127. https://doi.org/10.1016 ests: a study combining satellite and ground-based measure- /S0065-2628(08)60247-5 ments. Biogeosciences 10:7307–7321. https://doi. Seefeldt LC, Hoffman BM, Dean DR (2009) Mechanism of Mo- org/10.5194/bg-10-7307-2013 dependent Nitrogenase. Annu Rev Biochem 78:701–722. Walker TW, Syers JK (1976) The fate of phosphorus during https://doi.org/10.1146/annurev.biochem.78.070907.103812 pedogenesis. Geoderma 15:1–19. https://doi.org/10.1016 Seitzinger SP, Garber JH (1987) Nitrogen-fixation and N-15(2) /0016-7061(76)90066-5 calibration of the acetylene-reduction assay in coastal ma- Wardle DA, Walker LR, Bardgett RD (2004) Ecosystem proper- rine-sediments. Mar Ecol Progr Ser 37:8–73. https://doi. ties and Forest decline in contrasting long-term org/10.3354/Meps037065 Chronosequences. Science (New York, NY) 305:509-513. Sexstone AJ, Revsbech NP, Parkin TB, Tiedje JM (1985) Direct https://doi.org/10.1126/science.1098778 measurement of oxygen profiles and denitrification rates in Weintraub SR, Taylor PG, Porder S, Cleveland CC, Asner GP, soil Aggregates1. Soil Sci Soc Am J 49:645–651. https://doi. Townsend AR (2015) Topographic controls on soil nitrogen org/10.2136/sssaj1985.03615995004900030024x availability in a lowland tropical forest. Ecology 96:1561– Sprent JI, Sprent P (1990) Nitrogen fixing organisms. Chapman 1574. https://doi.org/10.1890/14-0834.1 and Hall, London, UK Wood TE, Lawrence D, Clark DA (2005) Variation in leaf litter Sullivan BW, Smith WK, Townsend AR, Nasto MK, Reed SC, nutrients of a Costa Rican rain forest is related to precipita- Chazdon RL, Cleveland CC (2014) Spatially robust estimates tion. Biogeochemistry 73:417–437. https://doi.org/10.1007 of biological nitrogen (N) fixation imply substantial human /s10533-004-0563-6 alteration of the tropical N cycle. Proc Natl Acad Sci U S A Wurzburger N, Bellenger JP, Kraepiel AM, Hedin LO (2012) 111:8101–8106. https://doi.org/10.1073/pnas.1320646111 Molybdenum and phosphorus interact to constrain Tang Y, Horikoshi M, Li WtUItVSRoPRP (2016) ggfortify: asymbiotic nitrogen fixation in tropical forests. PLoS One Unified Interface to Visualize Statistical Result fo Popular 7:e33710. https://doi.org/10.1371/journal.pone.0033710 R Packages. The R Journal 8:12 Yang J, Xie X, Wang X, Dixon R, Wang Y-P (2014) Tiessen H, Chacon P, Cuevas E (1994) Phosphorus and nitrogen Reconstruction and minimal gene requirements for the alter- status in soils and vegetation along a toposequence of dys- native iron-only nitrogenase in Escherichia coli. Proc Natl trophic rainforests on the upper Rio Negro. Oecologia 99: Acad Sci 111:E3718–E3725. https://doi.org/10.1073 145–150. https://doi.org/10.1007/bf00317095 /pnas.1411185111 Tripathi BM, Lee-Cruz L, Kim M, Singh D, Go R, Shukor NAA, Zaehle S, Jones CD, Houlton B, Lamarque J-F, Robertson E Husni MHA, Chun J, Adams JM (2014) Spatial scaling (2015) Nitrogen availability reduces CMIP5 projections of effects on soil bacterial communities in Malaysian tropical twenty-first-century land carbon uptake*. J Clim 28:2494– forests. Microb Ecol 68:247–258. https://doi.org/10.1007 2511. https://doi.org/10.1175/jcli-d-13-00776.1 /s00248-014-0404-7 Zahran HH (1999) Rhizobium-legume Symbiosis and nitrogen Van Sundert K, Horemans JA, Stendahl J, Vicca S (2018) The fixation under severe conditions and in an arid climate. influence of soil properties and nutrients on conifer forest Microbiol Mol Biol Rev 63:968–989 growth in Sweden, and the first steps in developing a nutrient Zhang X, McRose DL, Darnajoux R, Bellenger JP, Morel FMM, availability metric. Biogeosciences 15:3475–3496. Kraepiel AML (2016) Alternative nitrogenase activity in the https://doi.org/10.5194/bg-15-3475-2018 environment and nitrogen cycle implications. Venables WN, Ripley BD (2002) Modern applied statistics with S, Biogeochemistry 127:189–198. https://doi.org/10.1007 Fourth edn. Springer, New York /s10533-016-0188-6 Vitousek PM (1984) Litterfall, nutrient cycling, and nutrient lim- Zheng Y, Harris DF, Yu Z, Fu Y, Poudel S, Ledbetter RN, Fixen itation in tropical forests. Ecology 65:285–298. https://doi. KR, Yang ZY, Boyd ES, Lidstrom ME, Seefeldt LC, org/10.2307/1939481 Harwood CS (2018) A pathway for biological methane pro- Vitousek PM, Sanford RL (1986) Nutrient cycling in moist trop- duction using bacterial iron-only nitrogenase. Nat Microbiol ical Forest. Annu Rev Ecol Syst 17:137–167. https://doi. 3:281–286. https://doi.org/10.1038/s41564-017-0091-5 org/10.1146/annurev.es.17.110186.001033 Vitousek PM, Menge DN, Reed SC, Cleveland CC (2013) Biological nitrogen fixation: rates, patterns and ecological controls in terrestrial ecosystems. Philos Trans R Soc Lond Publisher’s note Springer Nature remains neutral with regard to Ser B Biol Sci 368:20130119. https://doi.org/10.1098 jurisdictional claims in published maps and institutional /rstb.2013.0119 affiliations.