This is the accepted version of the article: Van Langenhove, Leandro; Depaepe, Thomas; Vicca, Sara; [et al.]. «Regulation of Nitrogen Fixation from Free-Living Organisms in Soil and Leaf Litter of two tropical forests of the Guiana shield». Plant and soil, (April 2019). DOI 10.1007/s11104-019-04012-1 This version is avaible at https://ddd.uab.cat/record/203986 under the terms of the license 1 Regulation of Nitrogen Fixation from Free-Living Organisms in Soil and Leaf 2 Litter of two tropical forests of the Guiana shield 3 Leandro Van Langenhove1, Thomas Depaepe2, Sara Vicca1, Joke Van den Berge1, Clement Stahl3, Elodie 4 A. Courtois1,4, James Weedon5, Ifigenia Urbina6,7, Oriol Grau6,7, Dolores Asensio6,7, Josep Peñuelas6,7, 5 Pascal Boeckx8, Andreas Richter9, Dominique Van Der Straeten2, Ivan A. Janssens1 1 6 Centre of Excellence PLECO (Plants and Ecosystems), Department of Biology, University of Antwerp, 7 Wilrijk, Belgium 2 8 Laboratory of Functional Plant Biology, Department of Biology, Ghent University, KL Ledeganckstraat 9 35, B-9000, Gent, Belgium 3 10 INRA, UMR Ecology of Guiana Forests (Ecofog), AgroParisTech, Cirad, CNRS, Université de Guyane, 19 11 Université des Antilles, 97387 Kourou, France 4 12 Laboratoire Ecologie, évolution, interactions des systèmes amazoniens (LEEISA), Université de 13 Guyane, CNRS, IFREMER, 97300 Cayenne, French Guiana 5 14 Department of Ecological Science, Vrije Universiteit Amsterdam, De Boelelaan 1085, 1081 HV 15 Amsterdam, The Netherlands 6 16 CSIC, Global Ecology Unit CREAF-CSIC-UAB, 08193 Bellaterra, Catalonia, Spain 7 17 CREAF, Cerdanyola del Vallès, 08193 Catalonia, Spain 8 18 Department of Applied Analytical and Physical Chemistry, Faculty of Bioscience Engineering, Isotope 19 Bioscience Laboratory – ISOFYS, Ghent University, Coupure Links 653, 9000 Gent, Belgium 9 20 University of Vienna, Department of Microbiology and Ecosystem Science, Althanstr. 14, 1090 Vienna, 21 Austria 22 23 Keywords: free-living nitrogen fixation, tropical forest, French Guiana, nutrients, phosphorus, 24 molybdenum 25 26 Abstract 27 Background and aims Biological fixation of atmospheric nitrogen (N2) is the main pathway for 28 introducing N into unmanaged ecosystems. While recent estimates suggest that free-living N fixation 29 (FLNF) accounts for the majority of N fixed in mature tropical forests, the controls governing this 30 process are not completely understood. The aim of this study was to quantify FLNF rates and 31 determine its drivers in the tropical pristine forest of French Guiana. 32 Methods We used the acetylene reduction assay to measure FLNF rates at two sites, in two seasons 33 and along three topographical positions, and used regression analyses to identify which edaphic 34 explanatory variables, including carbon (C), nitrogen (N), phosphorus (P) and molybdenum (Mo) 35 content, pH, water and available N and P, explained most of the variation in FLNF rates. 36 Results Overall, FLNF rates were lower than measured in tropical systems elsewhere. In soils seasonal 37 variability was small and FLNF rates differed among topographies at only one site. Water, P and pH 38 explained 24 % of the variation. In leaf litter, FLNF rates differed seasonally, without site or 39 topographical differences. Water, C, N and P explained 46 % of the observed variation. We found no 40 regulatory role of Mo at our sites. 41 Conclusions Rates of FLNF were low in primary rainforest on poor soils on the Guiana shield. Water 42 was the most important rate-regulating factor and FLNF increased with increasing P, but decreased 43 with increasing N. Our results support the general assumption that N fixation in tropical lowland forests 44 is limited by P availability. 45 46 Introduction 47 Nitrogen (N) availability is a limiting factor for plant growth in a wide range of terrestrial ecosystems 48 worldwide (LeBauer and Treseder 2008) and restricts the amount of carbon (C) that can be assimilated 49 and stored in the terrestrial biosphere (Hungate et al. 2003; Penuelas et al. 2013; Zaehle et al. 2015). 50 Globally, biological N fixation is the most important natural pathway for introducing previously inert N 51 - namely N2 gas - , into unmanaged ecosystems (Galloway et al. 1995; Vitousek et al. 2013). Nitrogen 52 is fixed by microorganisms, known as diazotrophs, that can reduce gaseous N (N2) into ammonia. 53 Diazotrophs are divided in two groups: symbiotic and free-living. Symbiotic N fixers are generally found 54 in root nodules and live in a mutualistic relationship with higher plants that allocate C to the N fixers 55 in exchange for N. Free-living N fixers are hetero- or autotrophic bacteria or archaea inhabiting water, 56 soil, rocks and leaf litter (Dynarski and Houlton 2018). Global terrestrial N inputs from biological N 57 fixation have been estimated at 60 Tg y-1 (Vitousek et al. 2013), and biome-level comparisons suggest 58 that tropical rain forests may fix more N than any other unmanaged ecosystem (Galloway et al. 2004; 59 Reed et al. 2011). Until recently, a large proportion of tropical N fixation was attributed to symbiotic 60 organisms due to the high abundance of leguminous trees (Losos and Leigh 2004), typically associated 61 with symbiotic N fixers. The relative importance of symbiotic N fixation, however, has been questioned 62 because it is facultative (Menge et al. 2009) and may decline to near zero in mature tropical rainforests 63 (Barron et al. 2011, Batterman et al. 2013, Sullivan et al. 2014). In this context, N fixation by free-living 64 organisms is thus increasingly considered to be more important, with recent estimates from multiple 65 rain forests suggesting that a substantial amount of N is indeed fixed by free-living diazotrophs (e. g. 66 Reed et al. 2007, Cusack at al., 2009, Černá et al. 2009, Barron et al. 2009, Wurzburger et al. 2012, 67 Matson et al. 2014). 68 Because of its relevance for ecosystem functioning, N fixation and its rate-controlling factors have been 69 the focus of previous research. Nitrogen fixation is energetically expensive, requiring much ATP for 70 both the reaction itself (Gutschick 1981) and for maintaining an oxygen-poor intracellular environment 71 for nitrogenase, the enzyme responsible for N fixation (Robson and Postgate 1980). The oxygen-poor 72 environment can be created, either by an upregulated respiration rate, or developing cellular 73 structures that can limit the entry of oxygen (Robson and Postgate 1980). This regulation is needed 74 because oxygen binds to nitrogenase and inactivates the enzyme. Free-living N fixation (FLNF) is 75 thought to decrease as N availability increases in the environment (Hartwig 1998), because uptake of 76 inorganic N is less costly than N fixation (Gutschick 1981; Menge et al. 2009)). 77 Low phosphorus (P) availability has been reported to limit FLNF in several tropical environments (e.g., 78 Pearson and Vitousek 2002; Reed et al. 2007), an observation often attributed to the diazotrophs’s 79 high P requirement. Nonetheless, studies of P addition to tropical forests have reported contradictory 80 responses. Some reported that FLNF in soils and plant litter increased in response to P additions 81 (Benner et al. 2007; Reed et al 2007), while others found no effect (Pérez et al 2008; Barron et al. 82 2009). Recent studies suggest that the stimulation of N fixation by P addition may be due to 83 molybdenum (Mo) contamination of the commonly used superphosphate fertiliser (Barron et al. 2009, 84 Wurzburger et al. 2012). Molybdenum is a rock-derived trace element required to produce the FeMo 85 cofactor necessary for the functioning of the most common group of nitrogenases (Igarashi and 86 Seefeldt 2003). Molybdenum is thus associated more with fundamental enzymatic requirements than 87 with the high energy consumption of N fixation, but limitation of both Mo and P has been documented 88 in some forests (Barron et al. 2009; Wurzburger et al. 2012; Reed et al. 2013). 89 Energy in the form of external organic carbon (C) is another likely factor regulating the activity of free- 90 living N fixers. Like other heterotrophic microbes, diazotrophs rely on extracellular organic C for 91 respiration, and C supply may be more limiting than nutrient availability in free-living fixers (Hofmockel 92 and Schlesinger 2007), even though the C cost of FLNF is not well-quantified (Dynarski and Houlton 93 2018). Lastly, seasonal variation in soil moisture can play a large role in the regulation of N fixation 94 rates, because the rates are positively correlated with moisture content (Roskoski 1980). Seasonal 95 differences in N fixation have indeed been observed, but the direction of these seasonal changes 96 differed among studies. Studies conducted at different tropical sites have reported both higher (cf 97 Reed et al. 2007) and lower (cf Matson et al. 2014) rates in the wet season compared to the dry season 98 with likely factors besides moisture, such as changes in oxygen supply, causing the discrepancy in 99 results. 100 Lowland tropical forests found on highly weathered old soils are typically considered to be N-rich, 101 because they can accumulate, recycle and export large quantities of N (Vitousek and Sanford 1986; 102 Hedin et al. 2009; Brookshire et al. 2012). This open N cycle is also corroborated by enriched δ15N 103 values of soils and plant tissues, due to high gaseous and leaching losses of depleted 15N sources 104 (Amundson et al. 2003). Simultaneously, these lowland tropical forests have usually been described as 105 P limited because of biogeochemical theory predicting that P limitation should be prevalent in old, 106 strongly weathered soils (Walker and Syers 1976; Wardle et al 2004). There is also a wealth of indirect 107 evidence including high N availability (Brookshire et al. 2012), high N:P ratios in leaves (Vitousek 1984) 108 and correlations between forest properties and soil fertility at continental scale (Quesada et al. 2012; 109 ter Steege et al. 2006. Locally, however, nutrient availability in tropical landscapes can vary with 110 topography, but the magnitude and direction of this variation is variable and influenced by, e.g., terrain 111 steepness and rainfall, leading to differences in physical denudation rates and solute transportation 112 (Porder et al. 2005; Weintraub et al. 2015). Differences in soil type and redox status along 113 topographical gradients may also affect nutrient availability (Tiessen et al. 1994). 114 Here we present results from a study carried out in mature tropical rainforests of French Guiana, where 115 we measured FLNF in both soil and leaf litter. The rolling hills of French Guianese tropical forests, part 116 of the Guiana Shield and perched upon parent material that is amongst the oldest and most weathered 117 in the world (Hammond et al 2005), are characterised as very poor in mineral nutrients (van Kekem et 118 al 1996), with topography inducing spatially heterogeneous nutrient availabilities. We studied three 119 distinct topographical positions at two different forest sites in the wet and dry season, with the aim of 120 maximizing the range in soil nutrients and moisture. Our goals were (I) to determine and compare rates 121 of FLNF in soils and leaf litter to other tropical forest studies, (II) to evaluate whether or not N fixation 122 rates differed between sites or between seasons, and (III) to identify which environmental factors (or 123 combination thereof) best explained the spatial and seasonal variation in N fixation rates. 124 Materials and Methods 125 Study sites 126 The study was conducted at two primary forest sites in French Guiana, in the research stations of 127 Paracou and Nouragues. Paracou is situated 15 km from the coast (5°15’N, 52°55’W), while Nouragues 128 is located inland (4°05’N, 52°41’W). Annual rainfall quantities at both sites were similar, with Paracou 129 receiving an average of 3100 mm year-1 for the period 2004 -2015 (Aguilos et al. 2018) and Nouragues 130 receiving an average of 2990 mm year-1 (Bongers et al. 2001). Mean annual air temperature is near 26 131 °C for both sites (Gourlet-Fleury et al. 2004; Bongers et al. 2001). The French Guianese climate is 132 characterized by a wet and a dry season due to the north/south movement of the Inter-Tropical 133 Convergence Zone. The region receives heavy rains from December to July and a dry period, typically 134 characterized by less than 100 mm rainfall month-1, from August to November (Aguilos et al. 2018). 135 Topography at both sites is undulating with maximum slopes of approximately 15°. The elevational 136 difference between hill summits and valleys is 20-50 m over horizontal distances of 200-400 m. Soils 137 at the Paracou site are schist soils with veins of pegmatite along the Bonidoro series, a Precambrian 138 metamorphic formation (Epron et al. 2006). Soils at the Nouragues site are also derived from the same 139 Bonidoro series, and consist of mainly Caraib gneiss (Bongers et al. 2001). This Precambrian geological 140 substrate is particularly low in P content compared to the generally younger, nutrient-richer soils of 141 western Amazonia (Hammond 2005; Grau et al. 2017) and therefore soils at both sites are classified as 142 nutrient-poor Acrisols (FAO-ISRIC-ISSS 1998). Soils at Paracou range from loamy sand to sandy loam, 143 while soils at Nouragues contain much more clay and span the range of sandy loam to silty clay 144 according to the USDA texture classification chart (Fig S1). 145 Experimental design 146 To maximize the natural variation in soil nutrient availability the experimental plots encompassed a 147 topographical gradient. At each forest site twelve plots were installed, distributed over four hillslopes, 148 with each hillslope having three plots located at three topographical levels: (1) bottom, i.e. just above 149 the creek running through the valley, (2) slope, i.e. the intermediate section of the elevation, and (3) 150 top, i.e., where the slope evens out and becomes the hilltop. Each plot was of 20 x 20 m in size. In total, 151 24 plots spread over two sites and three topographical categories per site were studied. Distances 152 between the plots were 10 – 100 m and in each plot five soil and litter samples were collected. Four 153 samples were collected at 2 m distance from each corner and a fifth sample was taken in the middle 154 of the plot. A total of 120 sampling points (2 study sites, 3 topographical categories per site, 4 plots 155 per category and 5 samples per plot) were sampled for both soil and leaf litter in both the wet and dry 156 season (240 samples in total). Courtois et al (2018) have provided more detailed information on the 157 experimental design. 158 N fixation 159 Nitrogen fixation rates were determined using the acetylene reduction assay (Hardy et al. 1968). 160 Samples were collected in May and September 2016, in the wet and dry season, respectively. Samples 161 of leaf litter were collected manually from the soil surface and soil samples were collected with a 2-cm 162 diameter corer to a depth of 5 cm after removing all litter from the surface 163 All samples were placed in clear 100 ml borosilicate jars. The jars were sealed with rubber septa and 164 10 ml of air was replaced with 10 ml of acetylene gas (welding grade, Air Liquide) to create a 10 % 165 headspace concentration by volume. The samples were then incubated in situ at ambient forest light 166 (no direct sunlight) and temperature for 18 hours. Sample moisture was not manipulated in any way, 167 but was determined after the incubation as the weight loss after oven drying at 70°C during 48h. 168 After incubation, a subsample from the sample headspace was injected into a pre-evacuated 12 ml 169 borosilicate vial (Labco Limited, Ceredigion, UK) and shipped to Ghent, Belgium for analysis. Ethylene 170 concentrations were measured using laser-based photo-acoustic spectroscopy (ETD-300, Sensor 171 Sense, The Netherlands). Parallel acetylene blanks (no leaf litter or soil) were created to assess 172 background levels of ethylene in the acetylene gas (1.5 +- 0.4 nl ethylene ml-1 acetylene gas), which 173 were subtracted from the sample ethylene concentrations. Controls for ethylene production in the soil 174 or litter in the absence of acetylene gas were also assayed and were consistently below the detection 175 limit of 0.01 nl ethylene ml-1 air. Soil and leaf litter samples that over the incubation time produced 176 ethylene concentrations below the detection limit (0.01 nl ethylene ml-1 air) were recorded as half of 177 this value. 178 We converted the rate of ethylene production, expressed as nmol ethylene g-1 sample h-1, into N 179 fixation rates, expressed as kg N fixed ha-1 y-1 using the densities of the soil (to a depth of 5cm) or leaf 180 litter, and the theoretical ratio of 3 moles ethylene produced per mole N fixed (Benner et al. 2007; 181 Cusack et al. 2009; Matson et al. 2014; Pearson and Vitousek 2001; Reed et al. 2007). The latter being 182 based upon the conclusion that reducing 3 moles of acetylene to ethylene is equivalent to the 6 183 electron transfer involved in the reduction of one mole of N2 to ammonium (Seitzinger and Garber 184 1987). We attempted to measure uptake of 15N through the pool dilution assay (Furnkranz et al. 2008) 185 in a subset of soil samples to gauge the actual ratio of moles ethylene produced per mole N fixed, but 186 we failed to do so due to a combination of low soil FLNF rates and high background levels of N in these 187 environmental samples. Other authors have reported encountering similar issues and could not 188 measure 15N uptake in soil (Matson et al. 2014) or leaf litter (Menge et al. 2009) samples. Soil samples 189 for determining bulk density to a depth of 5 cm were taken with Kopeck rings. The samples were oven 190 dried at 105°C for 48 h and bulk density was calculated by dividing soil weight by Kopeck ring volume. 191 Leaf litter was collected in a 0.5 m² wooden frame at five locations per plot and once per season for 192 determining litter density. The litter was dried at 105°C for 48 h and the density was calculated by 193 dividing weight by area (kg ha-1). 194 Chemical analyses 195 Total C, N, P and Mo contents in the soil and litter were determined on the same samples that were 196 used for acetylene reduction. Ethylene production was first measured (see N fixation), and afterwards 197 samples were oven dried at 70°C for two days and then ground in a ball mill (Retsch, Germany). Total 198 C and N contents were determined by dry combustion with an elemental analyser (Flash 2000, Thermo 199 Fisher Scientific, Germany). Total P (mg kg-1) and Mo (mg kg-1) contents were determined by the 200 sequential digestion of the ground soil and litter samples in heated strong acid (69 % HNO3 and 30 % 201 H2O2), followed by analysis on an iCAP 6300 Duo ICP optical emission spectrometer (Thermo Fisher 202 Scientific, Germany). 203 Soil texture and nutrient availability were determined on a composite sample made of three soil cores 204 per sampling point, each core to a depth of 15 cm. Texture was determined on fresh soil using the 205 hydrometric method (Gee and Bauder 1986). Soil particles were dispersed with sodium 206 metaphosphate and the amounts of sand, silt and clay were determined using a hydrometer. Soil 207 samples for measuring nutrient availability were collected in May and September 2015, sieved (2 mm) 208 and split into three subsamples. The first subsample was extracted with 1M KCl in a 1:2.5 w:v ratio for 209 pH measurement and determination of available N. On this extract pHKCl was measured with a pH 210 meter (HI 2210-01, Hanna Instruments, USA), after which the extract was filtered through a 42µm filter 211 and the filtrate’s concentration of NH4+ and NO3- was determined colorimetrically (SAN++ continuous 212 flow analyser, Skalar Inc, The Netherlands). Available N (mg kg-1) was defined as the sum of the NH4+ 213 and NO3- concentrations. The second subsample was used to determine the gravimetric water content 214 by measuring weight loss after drying at 105 °C during 48 h. The third subsample was used to 215 determine extractable P and Mo. Soils were oven dried at 60 °C for 48 h after which extractable P was 216 determined with Bray-P acid fluoride extraction (Bray and Kurtz 1945). Available Mo was determined 217 through resin extraction (Wurzburger et al. 2012). The samples were mixed with water in a 1:6 ratio 218 and five 2 cm² strips of anion-exchange membrane (VWR Chemicals, USA) were added. This mixture 219 was stirred for 24h and the strips were then rinsed and eluted with 10 % HNO3. Available P and Mo 220 contents were determined with a iCAP 6300 Duo ICP optical emission spectrometer (Thermo Fisher 221 Scientific, Germany). 222 Literature comparison 223 To compare the FLNF rate results we found to those of other authors we performed a database search 224 similar to the search carried out by Dynarski and Houlton (2018). We searched Web of Science using 225 the keywords nitrogen, free-living, asymbiotic, fixation, soil, leaf litter and tropical forest. From the 226 resulting studies we selected those that were performed in natural terrestrial tropical ecosystems. For 227 studies that presented results from multiple time points or seasons, we averaged the data over the 228 course of the study period. Studies that did not report any measure of variance were assigned a 229 standard error of ¼ of the mean. Reported rates of FLNF were converted to nmol ethylene produced 230 g-1 dry substrate h-1 in order to compare N fixation rates between studies. When N fixation rates were 231 presented on a per area basis we used the bulk density of the N fixation substrate (soil or leaf litter) 232 and the ethylene: N2 conversion factor indicated in the study to convert to nmol ethylene produced g- 1 233 dry substrate h-1. When no conversion factor was indicated, we assumed the standard 3 : 1 conversion 234 factor (Hardy et al. 1968). Results of this database search are summarised in supplementary 235 information table S4 236 Data analysis 237 To assess the differences in N fixation rates and soil and leaf litter variables between sites and seasons 238 we used linear mixed effects regression models (LMER), analysing soil and leaf litter data separately. 239 We used site (Paracou or Nouragues), season (Wet or Dry) and topographical position (Bottom, Slope 240 or Top) as fixed factors and plot identity as a random factor. The validity of the linear models’ 241 assumptions (linearity, normality of residuals, no influential outliers, homoscedasticity) were 242 evaluated with standard functions of R (R core team 2017, version 3.4.3), including diagnostic plots. 243 Prior to analysis, data were log transformed if their distribution was right-skewed to improve normality 244 of model residuals. Multiple comparisons within a factor were analysed using Tukey post hoc tests. We 245 performed principal component analyses to visualize the correlations of previously standardised soil 246 and leaf litter variables according to site, season and, if necessary, topography. We observed that a 247 large proportion of our measured samples yielded ethylene production rates that fell below the 248 detection limit (25 % of samples in soil). To investigate if there were site-specific or seasonal patterns 249 in the occurrence of values below detection limit, we transformed our ethylene production rates into 250 binomial data (1 for measured rate and 0 for below detection limit rate) and analysed the resulting 251 data with binomial generalized linear model (GLM) with season and site as factors. 252 To identify which set of physico-chemical variables significantly contributed to the observed variation 253 in N fixation rates we performed a forward stepwise model selection, i.e. starting from a null model 254 and retaining the predictor variable that led to the largest decrease in the Akaike information criterion 255 (AIC), corrected for sample size (AICc). This process was iterated until no additional predictor reduced 256 the model AICc by at least two units. This procedure was performed for the dataset as a whole, as well 257 as for each combination of site and season that was shown to be significantly different in the mixed- 258 effects models (see above) in soil and in leaf litter. For these analyses the measurements of FLNF that 259 were below the detection limit were excluded to assess which predictor variables participate in 260 regulating the FLNF that were detectable and thus participate to the ecosystem scale N fixation. The 261 potential predictor variables for both leaf litter and soil were standardised to a mean = 0 and standard 262 deviation = 1 prior to the model fitting procedure to avoid potential issues in interpretation and 263 numerical stability due to differences in magnitude between variables. Potential predictor variables 264 were gravimetric water content, C:N ratio, N:P ratio, total carbon, total nitrogen, total phosphorus and 265 total molybdenum. For soil we additionally included available nitrogen, phosphorus and molybdenum, 266 soil pH and soil texture. We present the best-fit model for each data subset, based on this forward 267 stepwise procedure. 268 As an additional check of robustness we used an Akaike weight approach to assess the importance of 269 predictor variables. We summed Akaike weights computed for all possible first-order models 270 containing a given predictor, to obtain a measure of the relative variable importance for this predictor 271 (Burnham and Anderson 2002). We did this for all abovementioned predictors. This approach yielded 272 very similar results to the forward stepwise model selection, confirming the robustness of our analysis, 273 but for readability reasons are not presented nor discussed in this paper. These results are, however, 274 shown in supplements (Figs S2 and S3). All analyses were carried out with the software package R 275 (version 3.4.3) using packages lme4 (Bates et al. 2015), MASS (Venables and Ripley 2002), ggfortify 276 (Tang et al. 2016) and AICcmodavg (Mazerolle 2017) 277 Results 278 Rates of FLNF were on average 0.015 ±0.003 (standard error) nmol ethylene g-1 h-1 or 0.57 ±0.10 kg N 279 ha-1 y-1 in soil and 0.25 ±0.04 nmol ethylene g-1 h-1 or 0.09 ±0.01 kg N ha-1 y-1 in leaf litter. Per unit mass, 280 FLNF rates were thus, on average, tenfold higher in leaf litter than in the top 5 cm of the soil (Table 1). 281 However, when reported per unit area, FLNF rates were lower in litter than soil, due to the huge 282 difference in density between the sample types (Table 1). 283 SOIL – N FIXATION Overall, soil FLNF rates did not differ between seasons (P=0.27). The effect of 284 topography on soil FLNF rates differed between Paracou and Nouragues (Site x Topography interactive 285 effect; P=0.021) (Fig 1 A & B). In Paracou, soil FLNF rates were 20 times higher (+/-SE = 6 to 65 times 286 higher) in the bottom and slope plots than in the top plots (P=0.047 and P=0.009, respectively), but 287 did not differ significantly between the bottom and slope plots. In Nouragues no differences were 288 observed in the FLNF rates of the bottom, slope or top plots. A considerable proportion of the soil FLNF 289 rates were below the detection limit (25 %, out of 230 samples). Although the LMER did not reveal a 290 significant effect of season on the soil FLNF rates, season did affect the number of samples whose FLNF 291 rates were below the detection limit (Fig S4 A & B), albeit differently in Paracou than in Nouragues 292 (Site x Season interactive effect P=0.004). Whereas in Paracou the number of soil FLNF rates that were 293 below the detection limit did not differ between wet and dry season, in Nouragues values below the 294 detection limit occurred primarily in the dry season (18 out of 52 samples versus 2 out of 56 samples, 295 in the dry and wet season, respectively, P<0.001). When conducting the LMER analysis on a subset of 296 the Nouragues soil FLNF rates, excluding samples where FLNF rates were below the detection limit, we 297 identified an effect of season (P=0.008), with the highest rates occurring in the dry season. (Fig S5 B). 298 SOIL – ENVIRONMENTAL VARIABLES As each site had different soil FLNF rates relating to their 299 topography, we performed a principal component analysis (PCA) for each site separately, allowing us 300 to visualize site specific relationships among soil parameters. In Paracou, PC1 and PC2 explained 63.3 301 % of the variation, with PC1 explaining 36.3 % of the variation and containing most of the 302 topographically induced variation (Fig 2 A). Clay content correlated with total P and available N and 303 was highest on the slopes. PC2, explaining 27 % of the variation, contained the seasonally-induced 304 variation (Fig 2 A). Total C correlated with total N and was higher in the dry season, while total Mo 305 correlated well with pH and was higher in the wet season. Available Mo correlated well with moisture 306 and was also higher in the wet season. Available P was not heavily affected by season, but was higher 307 in the bottom plots than on either the slopes or tops. In Nouragues PC1 and PC2 together explained 308 68.6 % of the variation (Fig 2 B). The first principal component (PC1) explained 55.2 % of the variation 309 and, just as in Paracou, mostly contained topographically-induced variation. In Nouragues soil 310 properties for top plots were clearly different from bottom and slope plots. Clay content correlated 311 with total P and available N and all were highest in the top plots. In contrast to Paracou, however, total 312 Mo correlated with available P and both were higher on the bottom and slope plots than on the top 313 plots. Moisture, total C and total N correlated well with each other and clustered together at a ±45° 314 angle from PC1. Available Mo and C:N were correlated along PC2, which explained 13.4 % of the 315 variation. Lastly, pH also varied topographically and was highest on the bottom and slope plots, yet 316 the variation was small. For all Pearson correlations, see table S3 A and B. 317 LEAF LITTER – N FIXATION Leaf litter N fixation rates were not different between sites. On average, 318 litter FLNF rates in the wet season were 7.5 to 13.5 times (+/- SE) higher (P < 0.001) than in the dry 319 season. This difference between both seasons was further emphasized by the distinct seasonal 320 difference in the proportion of FLNF measurements that were below the detection limit (Fig S4 C & D): 321 the vast majority (97 %) of the 16 % (33 out of 235) of litter FLNF measurements that were below the 322 detection limit were from the dry season (binomial GLM, P<0.001). Nonetheless, the impact of season 323 interacted with that of topographical position (Season x Topography interactive effect P < 0.001) (Fig 324 1 C & D). In the wet season FLNF rates were similar among topographic positions, but in the dry season, 325 rates on the top plots were 2 to 7.5 times (+/- SE) lower (P < 0.05) than rates on the bottom or slope 326 plots, which did not differ from each other. 327 LEAF LITTER – ENVIRONMENTAL VARIABLES Because the LMER stated that FLNF rates were similar 328 between sites we analysed the leaf litter stoichiometry of both sites together. Together, the first two 329 principal components of the PCA accounted for 73.3% of the variation (Fig 3). PC1 explained 49.8 % of 330 the variation and distinguished between the wet and dry season. Along this component, moisture, N 331 and P content all correlated positively with each other and negatively with C:N and N:P ratio, indicating 332 that moisture, N and P content was higher in the wet season, while C:N and N:P ratios were higher in 333 the dry season. This was confirmed by linear effects regression analysis (Table S2). PC2, accounting for 334 23.5 % of the variation, correlated positively with C content. The absence of a site effect on PC1 and 335 PC2 indicates that leaf litter chemistry was similar for both sites. For all Pearson correlations, see table 336 S3 C. 337 DRIVERS OF FLNF IN SOILS AND LITTER- In soils, stepwise regression analyses indicated that soil water 338 content, available P and pH were the primary drivers of N fixation rates (Table 2 A), explaining 24 % of 339 the variation in soil FLNF rates. Because soil FLNF rates differed significantly between both sites, we 340 ran the stepwise regression analysis again for each site separately. In Paracou, P exerted a strong 341 effect, as both total P and available P had a positive influence on FLNF rates. Available N was negatively 342 related to FLNF, and together with P, explained 36 % of the variation. For Nouragues a model 343 containing only total N, total C or water content explained 40 % of the variation in FLNF rates. These 344 three parameters were strongly correlated and the regression analysis deemed the models containing 345 either one of them singularly equally valid (Table 2 A). 346 In litter, across both sites and seasons, about 46 % of the variation in FLNF rates was explained by 347 water and N:P ratio (Table 2B). While water had a positive effect on FLNF rates, N:P ratio showed a 348 negative effect. As FLNF rates differed between seasons we ran the stepwise analysis again for each 349 season separately. In the wet season 33 % of the variation could be explained by water, C content, P 350 content and C:N ratio. The model for the dry season explained about 68 % of the variation and was 351 dependent on the positive influence of water and the negative influence of N content. 352 Discussion 353 Overall, the sum of FLNF rates of soils and leaf litter measured in this study fall into the lower end 354 of the 0.1-60 kg N ha-1 y-1 range reported for FLNF on tropical forest floors worldwide (Reed et al. 2011), 355 and much below the more recent estimate of 15-36 kg N ha-1 y-1 fixed in tropical forests (Pajares and 356 Bohannan 2016) and the average ethylene production rate of 5.32 nmol ethylene g-1 h-1 reported in a 357 recent meta-analysis on N fixation rates in tropical forest ecosystems (Dynarski and Houlton 2018). 358 The FLNF rates found in the present study are much lower than those found in Costa Rica (Reed et al. 359 2007; Reed et al. 2010; Reed et al 2013), Panama (Barron et al. 2009; Wurzburger et al. 2012), Puerto 360 Rico (Cusack et al. 2009) or in the younger sites along a Hawaiian chronosequence (Crews et al. 2000). 361 An overview of FLNF rates reported by these authors can be found in Table S4. Sullivan et al (2014) 362 measured N fixation rates in Costa Rica, about 35 km away from where Reed et al. (2007) conducted 363 their study. These authors reported soil and leaf litter FLNF rates that were lower than those previously 364 published by Reed et al. (2007), but while their reported soil FLNF rates were similar to those found in 365 our study, their reported leaf litter FLNF rates were still twice as high as the rates we found. Other 366 studies of FLNF rates conducted in less fertile tropical forests, e.g. on the higher altitudes of an 367 altitudinal transect in Ecuador (Matson et al. 2014) or in the older sites of a Hawaiian chronosequence 368 (Crews et al. 2000), reported FLNF rates comparable to those found in the present study. In Hawaii, 369 Crews et al (2000) measured decreasing rates of leaf litter FLNF on increasingly older soils and 370 attributed this decrease to a combination of lower concentrations of geologically cycled nutrients, such 371 as P, and high N pools at their oldest sites. This, coupled with increases in FLNF rates after P 372 fertilization, led them to one of their main conclusions; that low P availability limited FLNF rates at 373 their older sites. It would, however, be inaccurate to conclude that soil P alone determines N fixation 374 since, for example, in Panama (Wurzburger et al 2012) two of the studied sites along a total soil P 375 gradient containing high (AVA) and medium (Gigante) levels of total soil P displayed the lowest rates 376 of FLNF in their study. 377 Soil free-living N fixation 378 At both sites moisture was important in the regulation of FLNF (Table 2 A). Water is essential for 379 all microbes, but for diazotrophs in particular water plays an important role in regulating activity. Not 380 only does their metabolism require sufficient amounts of water, but nitrogenase activity is hindered in 381 the presence of oxygen (O2) (Hill 1988). Oxygen binds to nitrogenase and inhibits its function 382 (Hartmann and Burris 1987) and because increased soil moisture decreases soil O2 concentrations, 383 water content can have a large impact on soil FLNF rates. As diazotrophs are mainly heterotrophic the 384 thickness of the soil water film, which is important for the diffusion rates of extracellular enzymes and 385 soluble organic-C substrates and is directly affected by soil water content (Davidson and Janssens 386 2006), will also play a role. 387 In Paracou, we observed no seasonal effect on soil FLNF rates and although rates were typically 388 reported to be higher in the wet season than in the dry season (e. g. Hofmockel and Schlesinger 2007; 389 Reed et al. 2007), Matson et al. (2014) have shown that the opposite can also occur. They postulated 390 that it is likely that moisture fluctuations were not directly responsible for their observed seasonal 391 changes in FLNF rates, just as in our study it might not have contributed to unchanging rates. Their 392 reasoning was based on the fact that cyanobacteria can fix N at moisture levels as low as 6 % and in 393 one study reached maximum N fixation rates at soil moistures between 22 and 42 % (Jones 1977), 394 though it did not specify whether this was in sandy or clayey soil. As soil moisture in Paracou always 395 remained between 15 and 44 % (Table S1) it is likely that diazotrophs had sufficient moisture to keep 396 oxygen levels low and continue N fixation. 397 At Nouragues, the effect of soil moisture on FLNF rates was high (40 % of variability explained by 398 soil moisture alone, Table 2 A) and seasonality caused an interesting pattern in the distribution of very 399 low FLNF rate measurements (below detection limit). The vast majority of below detection limit 400 measurements occurred during the dry season (Fig S4 B) and when analysing only data for which FLNF 401 rates were above the detection limit, we found a significant season effect in Nouragues with 402 quantifiable rates (i.e. in samples above the detection limit) actually being higher in the dry season 403 than in the wet season (Fig S5 B). This meant that the range of FLNF rates was broader during the dry 404 season than during the wet season (Fig S5 A), indicating that N fixation hotspots were of increasing 405 importance. These are typically found in tropical ecosystems (Pajares and Bohannan 2016) and reflect 406 the very small-scale spatial heterogeneity of abiotic factors affecting the dynamics of the diazotroph 407 community (Reed et al. 2010). As the soil dried out at the onset of the dry season the heterogeneity of 408 water-filled pore space increased, leading to the creation of aerobic and anaerobic, as well as dry and 409 mesic microsites that co-occured on a small spatial scale (Sexstone et al. 1985). 410 Besides water content, available P and pH were the most important predictors for soil FLNF at our 411 sites. The pH in our soils is very low, ranging between 3.8 and 4.2 (Table S1). This is a relatively small 412 range, yet pH has been identified to affect bacterial community composition and diversity (Rousk et 413 al. 2010; Tripathi et al. 2014) or soil CO2 effluxes (Courtois et al, 2018) at the local scale even when 414 changes are small. We were unable to find studies assessing the effect of pH on FLNF in tropical 415 ecosystems, but nitrogenase activity of soil diazotrophs has been shown to increase with increased pH 416 in a German pine forest situated in north-east Bavaria (Limmer and Drake 1996). Roper and Smith 417 (1991) found that nitrogenase activity of bacteria extracted from clayey Australian soils reached its 418 peak around pH 7.5, matching the nitrogenase pH optimum found by Pham and Burgess (1993), and 419 decreased in more acidic soils. Their study, however, was carried out on soils that at the start of the 420 experiment were only slightly acidic after which the pH was decreased during the course of the study. 421 Their result is therefore not necessarily representative for N fixing microbial communities that 422 developed in acidic soils, such as in our study. However, taking into account an enzymatic pH optimum 423 of 7.5 (Pham and Burgess 1993) and decreased nitrogenase activity in more acidic temperate soils, a 424 positive relationship between soil pH and FLNF rates in our acidic tropical forest sites is plausible. 425 Phosphorus can play a pivotal role in regulating N fixation rates and is often limiting the rate of 426 this process in highly weathered lowland tropical soils (e. g. Camenzind et al. 2018; Reed et al. 2011). 427 In Paracou, both P predictors were positively correlated with the FLNF rates, suggesting a higher 428 activity of N fixing microbes when more P is present in the soil. Additions of P to the forest floor have 429 shown to increase both diversity and abundance of diazotrophs in tropical soils (Reed et al. 2010), as 430 well as increases in FLNF rates (e. g. Benner et al. 2007; Reed et al. 2013). The positive correlation 431 between available P and total P on the one hand, and FLNF rates on the other, is in line with the 432 longstanding idea that increased P benefits FLNF rates in tropical lowland forests. In addition to P, the 433 model for Paracou also identified available N as a predictor, indicating that higher available N is linked 434 to decreased FLNF rates. This is in line with what was observed in previous tropical forest studies, 435 where additions of N led to the decrease of N fixation rates (e. g. Barron et al. 2009; Crews et al. 2000; 436 Cusack et al. 2009). This observation is generally attributed to the theory that many heterotrophic N2 437 fixers are facultative fixers and able to down-regulate their fixation pathway when other sources of N 438 are available and more cost-efficient to acquire (Menge et al. 2009). The topographic effect on FLNF 439 rates in Paracou, effectively resulting in lower rates on the top than on bottom and slope plots is likely 440 caused by the interplay of P and N in the Paracou soils as they also varied with topographic position. 441 Total P was highest on the slope and available P highest on the bottom, while both were lowest on the 442 top. Because of the higher clay content of the Paracou slope soils there are likely more aluminium and 443 iron oxides (Tiessen et al. 1994) that are able to occlude more P, resulting in higher soil total P contents 444 as seen on the slope. On the more sandy bottom landscape position metal oxides that can occlude P 445 are likely scarcer and the higher water content, especially during rain events when runoff causes 446 disproportionate changes in water status, leads to reduction of iron oxides (Colombo et al. 2014), 447 liberating occluded P and provoking higher concentrations of available P. 448 In Nouragues, we found that water content, C and N performed equally well, each individually 449 explaining 40 % of the observed variation in FLNF rates, suggesting that the absolute concentrations 450 of C and N in the soil explain a substantial part of the variation in FLNF rates. It is surprising that the 451 model identified a positive effect of N content on FLNF rates, given that most studies associate 452 increased N with decreased or unchanged fixation rates (Camenzind et al. 2018). Additionally, we 453 would expect that available N, which was also included as potential variable but was not selected by 454 the model, would be identified as a variable affecting FLNF rates instead of N content because no 455 decomposition is needed before assimilation. Likely, the identification of N by the model is purely 456 mathematical and caused by its near perfect correlation with C content (Table S3 B). The positive 457 relationship of C with FLNF rates, which is predominantly carried out by heterotrophic diazotrophs 458 (Sprent and Sprent 1990), is the result of decomposition of organic material and subsequent 459 assimilation of additional C providing the energy necessary to carry out fixation. In Nouragues, the high 460 correlation between C content and moisture (Table S3 B) partly explains their equal importance in the 461 model. Their correlation might be due to the influence of soil moisture on decomposition rates, 462 especially when it is very wet (negative relation) or very dry (positive relation). At high moisture levels, 463 soil organic carbon and nitrogen stocks increase because of the slower decomposition in water- 464 saturated soil (Van Sundert et al. 2018). Moreover, soils in Nouragues are clay-rich, exacerbating this 465 effect as soils containing more clay stabilize and store more soil organic matter than sandy soils (Reis 466 et al. 2014), such as those in Paracou. Both increased water and C can be beneficial for free-living 467 diazotrophs as many species possess heterotrophic anaerobic metabolisms (Dixon and Kahn 2004) and 468 proliferate under oxygen poor and carbon rich conditions. 469 In Nouragues the topographical patterns of N and P were different from Paracou, with much 470 higher total P concentrations that occurred on the top landscape position instead of on the slopes, and 471 much smaller differences in available P concentrations among the topographic positions. This likely 472 played a role in the regulation of soil FLNF rate, causing them to remain high on the top landscape 473 position. Because of this strong difference in topography effects between Nouragues and Paracou, we 474 cannot draw general conclusions about landscape-scale variations in soil FLNF rates across 475 topographies in French Guianese tropical forests. Instead, our results support the idea that soil FLNF 476 rates at our sites varied in function of water and nutrient availability, similar to what was reviewed by 477 Dynarski et al. (2018). 478 Lastly, it is interesting to note that total or available Mo was never selected as explanatory 479 variable for soil FLNF rates, in spite of playing a regulatory role in FLNF at other sites (Barron et al. 480 2009; Reed et al. 2013; Wurzburger et al. 2012). Molybdenum concentrations in our study sites were 481 slightly higher than those found elsewhere (Gupta and Lipsett 1981), and available Mo was ten times 482 higher than in Panama (Wurzburger et al. 2012), which was the only study reporting the effect of 483 available molybdenum on FLNF. Likely, the concentration of available Mo at our sites is high enough 484 to preclude any regulatory role of Mo in this P-limited environment. 485 Leaf litter free-living N2 fixation 486 Overall, leaf litter FLNF rates did not vary with topography and were instead driven mainly by 487 water content and the N:P ratio of the leaf litter, as shown by the overall model explaining 46 % of the 488 variation (Table 2 B). The negative influence of N:P ratio is in line with what was postulated by Reed et 489 al. (2007) and likely illustrates a stoichiometric and energetic balance shift. In a decomposing leaf a 490 decreasing N:P ratio leads to P becoming comparatively more abundant, shifting the stoichiometric 491 balance to move away from P limitation to N limitation, favouring diazotrophs. Additionally, increased 492 P in the environment may alleviate energetic constraints (see above) and enable diazotrophs to invest 493 the required energy into fixing N. As the general model included data from both seasons, a large 494 influence of water content, as evidenced by the largest beta value (Table 2 B), was to be expected 495 because litter FLNF rates were significantly affected by seasonality; wet season FLNF rates were up to 496 nearly two times higher than dry season FLNF rates (Table 1) and the number of samples for which 497 FLNF rates were below the detection limit was much smaller in the wet than in the dry season (Fig S4 498 C & D). As expected, litter dried out severely during the dry season and water content decreased from 499 an average ~64 to ~40 % across both sites. Given that diazotrophs grow and proliferate in aqueous 500 environments (Scott 1957), it is likely that leaf litter water content in the dry season dropped below a 501 threshold of minimum water required for diazotroph proliferation, resulting in a net drop of FLNF rate. 502 This response of diazotroph activity to leaf litter moisture was already observed in a sub-tropical karst 503 forest, where researchers found that decreases in leaf litter moisture resulted in decreased FLNF rates 504 (Li et al. 2018). The large beta value (2.63, Table 2 B) associated with water content in the dry season 505 model hints towards the disruptive effect of water shortage. This model explained 67 % of the variation 506 in litter FLNF rates and included only water content and litter N, which, just as the N:P ratio in the 507 general model, had a negative effect on the FLNF rates. As mentioned earlier, N assimilation is cheaper 508 than N fixation from an energetic point of view and when more N is present fixation will likely be down 509 regulated (Menge et al. 2009). On the top landscape position in Paracou, also during the dry season, 510 leaf litter water content was lowest while N:P ratio was highest. This likely led to a complete collapse 511 of N fixation (Fig 1 C), because N was in ample supply compared to P and water was scarce. During the 512 wet season, litter stoichiometry explained 33 % of the variation in FLNF rates which likely reflects, at 513 least partially, the interactions between seasonal changes in labile C availability, P content and N 514 demand (Reed et al. 2007). Litterfall peaks at the onset of the dry season (Chave et al. 2010; Wagner 515 et al. 2013) and once the wet season starts the daily rainfall provides a vehicle for the movement of 516 readily decomposable, dissolved organic C (DOC) within the litter layer (Courtois et al. 2018). The N2- 517 fixing microbial community in the litter layer is dominated by heterotrophic microorganisms (Sprent 518 and Sprent 1990) and the influx of litter DOC provides an easily accessible energy source for these 519 diazotrophs. Additionally, in our study, we found that the concentration of both P and N in leaf litter 520 was higher in the wet season than in the dry season, similar to what was found in another study in 521 Costa Rica (Wood et al. 2005). However, relative to N content, P content increased more towards the 522 wet season, resulting in a lower N:P ratio during the wet season. In combination with more labile C 523 input, this could stimulate N fixation through the relief of energetic constraints and the added 524 advantage of being able to fix N compared to assimilating N from an environment where it is, 525 comparatively, less abundant than in the dry season. This process was already observed in tallgrass 526 prairie soils (Eisele et al. 1989). 527 In contrast to litter P content, which changed significantly across seasons yet showed no 528 significant change along the topographical gradient, litter Mo content changed significantly between 529 the two seasons and along the topographical gradient. Just as in soils, however, and in spite of its 530 importance for N fixation (Kaiser et al. 2005; Seefeldt et al. 2009), we found no evidence to support a 531 regulatory role for Mo content. Similar as in soils, litter Mo concentrations were ten-fold higher than 532 those reported from other tropical sites (Barron et al. 2009; Bowell and Ansah 1993; Reed et al. 2013; 533 Wurzburger et al. 2012) and likely too high to render a regulatory role to Mo in litter FLNF rates. 534 Lastly, in both soils and leaf litter our models were unable to explain more than 67 % of the 535 observed variation, and in most cases only around 40 %. This means that often the majority of variation 536 in FLNF rates could not be explained by the variables we selected and thus it is very likely that other 537 factors not measured in this study contribute to the regulation of FLNF in our tropical lowland forests. 538 Knowing the diazotroph community composition, in both soil and leaf litter, could enhance our 539 understanding of their nutritional and environmental needs and help us estimate at what point 540 parameters such as moisture, pH and P or N availability are beneficial or, inversely, detrimental. The 541 likelihood that within a single diazotroph community both aerobic and anaerobic lifestyles can occur 542 (Dixon and Kahn 2004) and that N fixing Archaea within the community may possess different 543 nutritional requirements than bacterial diazotrophs (Leigh 2000) are additional reasons to study the 544 diazotroph community composition. Also, we did not assess the iron (Fe) or the vanadium (V) 545 availabilities at our sites while they have the potential of participating in FLNF rate regulation (Zhang 546 et al. 2016). While Mo is a necessary co-factor of most nitrogenase enzymes (Igarashi and Seefeldt 547 2003), Fe is found in all known nitrogenases and the occurrence of ‘iron-only’ (Fe-Fe) nitrogenases has 548 been widely documented (e. g. Yang et al. 2014; Zheng et al. 2018). The role of V in regulation of FLNF 549 is understudied, but it is certain that the occurrence of an alternative enzymatic co-factor, namely the 550 vanadium-iron (V-Fe) cofactor, is widespread and V availability may therefore also play a regulatory 551 role in nitrogenase biosynthesis (Hu et al. 2012). A follow-up study investigating the community 552 composition of the diazotrophs, specifically looking into the prevalence of aerobic and anaerobic 553 organisms, combined with research into the occurrence of V-Fe nitrogenases and V and Fe availabilities 554 in the soil and leaf litter could enable us to explain more of the observed variation than we presently 555 could. 556 Conclusion 557 Our study has shed light on the drivers behind FLNF in tropical soil and leaf litter on the Guiana Shield 558 and has shown that the rates of FLNF are much lower than those estimated for most tropical forests 559 elsewhere. Water, N and P played the main roles in determining FLNF rates in both sample types, while 560 pH only regulated in the soil. The effect of seasonality differed between sample type and differences 561 in FLNF regulation between sites could be observed in soils, but not in litter. Despite having been 562 shown to influence N fixation rates in other mature tropical forests, the micro nutrient Mo played no 563 role in the regulation of FLNF rates at our sites in French Guiana. 564 In the sandy site, Paracou, the stimulating effect of P and the inhibiting effect of N were the main 565 drivers behind soil FLNF, but in the clayey site, Nouragues, soil FLNF was mainly stimulated by water 566 or C content. Our models for soil FLNF could not explain more than 40 % of the observed variation, 567 illustrating the complexity of predicting fixation rates upon measured variables in an environment that 568 is highly heterogeneous on a regional, local and even micro scale. In leaf litter we also identified water, 569 N and P as main drivers, but the underlying mechanisms that caused variation may have been different 570 compared to the soil. In the leaf litter we observed no differences in FLNF rates between sites, but 571 during the dry season litter rates exhibited a drastic decline that was mainly related to water 572 insufficiency and the inhibiting effect of N. During the wet season water was still of importance, but 573 now stimulating effects of C and P also came into play. It is important to note that our litter wet season 574 model explained only about 30 % of the observed variation. 575 It is likely that in both soil and leaf litter diazotroph community composition and iron or vanadium 576 availability had an influence on FLNF rates making them interesting to measure in future studies. In 577 the larger framework of global change, where N deposition is expected to increase (Penuelas et al. 578 2013), P deposition to tropical forests may change (Gross et al. 2016) and the possibility of a drier 579 Amazon basin (IPCC 2013) may cause disruptive changes to the FLNF rates in soil and litter. Nutrient 580 addition studies may offer clues as to the response of FLNF to changes in N or P supply, but testing the 581 influence of climatic changes in situ calls for a very specific type of studies. 582 Acknowledgements 583 This research was supported by the European Research Council Synergy grant ERC-2013-SyG 584 610028-IMBALANCE-P. We thank the staff of the Nouragues station, managed by USR mixte LEEISA 585 (CNRS; Cayenne), and the Paracou station, managed by UMR Ecofog (CIRAD, INRA; Kourou). 586 Both research stations received support from “Investissement d’Avenir” grants managed by Agence 587 Nationale de la Recherche (CEBA: ANR-10-LABX-25-01,ANAEE France: ANR-11-INBS-0001). We thank 588 Tom Van der Spiet, Anne Cools and Miguel Portillo Estrada for their help with analyses and Philippe 589 Gaucher, Nadine Calluy, Lore Verryckt and Kevin Van Sundert for their help with logistics and statistics. 590 Tables and figure 591 Table 1 Range, median, mean and standard error (SE) of FLNF rates at forest sites Paracou and Nouragues. Rates measured in sample types soil and leaf litter are given for the wet and dry 592 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 < 593 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* 594 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- Paracou Nouragues Wet Dry Wet Dry Bottom Slope Top Bottom Slope Top Bottom Slope Top Bottom Slope Top Minimum BDL* BDL BDL BDL BDL BDL BDL BDL BDL BDL BDL BDL Median 0.006 0.005 BDL 0.015 0.008 0.001 0.002 0.003 0.004 0.014 0.006 0.012 Ethylene production 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 (nmol g-1 h-1) 0.044 0.191 0.022 0.029 0.056 0.018 0.006 0.034 0.404 0.298 0.086 0.460 Maximum 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 Soil Minimum BDL BDL BDL BDL 0.001 BDL 0.020 BDL BDL 0.001 BDL BDL 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 N fixation rate 0.565 0.724 0.091 0.663 0.528 0.149 0.100 0.176 0.742 1.301 0.405 1.279 Mean (kg N ha-1 y-1) 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 Minimum 0.026 0.025 BDL BDL BDL BDL 0.014 0.010 0.003 BDL BDL BDL 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 Ethylene production 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 (nmol g-1 h-1) 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 Litter Minimum 0.011 0.005 BDL BDL BDL BDL 0.005 0.003 0.001 BDL BDL BDL 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 N fixation rate 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 (kg N ha-1 y-1) 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 595 596 597 Table 2. Results of the forward stepwise selection analyses for N fixation rates in soil (A) and leaf litter (B). For these analyses 598 we excluded the measurements that were below the detection limit. A Site Season Parameters R² Par & Wet & dry +0.57 Water* +0.20 Available P* +0.17 pH* 0.240 Nou 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 & Wet & Dry +0.76 Water*** -0.32 N:P** 0.457 Nou F1,40= 24.67 F1,195= 8.34 Par & Wet +1.71 Water** +0.65 Total P** +1.38 Total C** + 0.81 CN* 0.328 Nou F1,65= 9.10 F1,117= 10.69 F1,117= 8.64 F1,115= 5.87 Par & Dry +2.63 Water*** -0.79 Total N* 0.676 Nou F1,82= 26.26 F1,73= 6.70 599 Each row lists the parameters that were included into the model that provided the best fit, based on AICc values. Columns 600 one and two give information regarding the dataset that was used. For soils (A) a model including data from both sites (Par, 601 Paracou; Nou, Nouragues) and seasons was made along with models for each site separately because N fixation rates were 602 different between sites. For leaf litter (B) a model including data from both sites and seasons was made along with models 603 for each season separately because seasonality affected N fixation. Parameters that had a significant effect (*= p<0.05, **= 604 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 605 was calculated for the fit of the modelled and measured fixation rates. 606 1The stepwise selection identified total N, total C and water content as equally important for determining FLNF rates in 607 Nouragues soils. This is because these three variables are highly correlated and behave similarly in the model. Explanations 608 as to why this occurs are offered in the text. 609 610 Figures 611 612 613 Figure 1. Area-based N fixation rates for soil at Paracou (A) and Nouragues (B) and for leaf litter at Paracou (C) and Nouragues 614 (D). Rates represent the means (±1 SE) of the samples collected from the bottom, slope and top landscape positions in the 615 wet (May) and dry (September) seasons (N = 20). Letters denote significant differences amongst different seasons and 616 landscape units within a single site for soil or leaf litter at the P < 0.05 level. Significance testing was performed by mixed 617 effects regression models using the log transformed values as measured values were non-normal. Data include 618 measurements that were below the detection limit. 619 620 621 622 Figure 2. Principal Component Analysis of soil variables in Paracou (A) and Nouragues (B). Points on the graph are plot 623 averages. C = total C, N = total N, P = total P, Mo = total Mo, C:N = C:N ratio, P in = available P, N in = available N, Mo in = 624 available Mo, Moisture = water content, pH = pH, Clay = % clay, Sand = % sand and BD = bulk density. 625 626 Figure 3. Principal Component Analysis of leaf litter variables in both Paracou and Nouragues and for both wet and dry 627 season. Dots on the graph are averaged for plot. C= total C, N = total N, P = total P, Mo = total Mo, C:N = C:N ratio, N:P = N:P 628 ratio and Moisture = water content. 629 630 Supplementary information 631 Table S1. Soil water content, stoichiometry and nutrient availabilities for both field sites and seasons on the three topographies Paracou Nouragues Wet season Dry season Wet season Dry season Bottom Slope Top Bottom Slope Top Bottom Slope Top Bottom Slope Top Gravimetric 37 (2)ab 44 (2)a 36 (1)b 19 (1)cd 22 (1)c 15 (1)d 48 (5)b 55 (3)b 66 (3)a 40 (5)c 30 (1)c 52 (3)b water (%) 17.8 C:N ratio 20.1 (1.1)ab 13 (0.8)c 18.7 (1.2)ab 18.8 (0.7)ab 21.0 (0.8)a 15.8 (0.3)a 17.5 (0.3)a 16.0 (0.5)a 16.3 (0.6)a 16.0 (0.3)a 16.3 (0.4)a (0.7)b Total C (%) 3.1 (0.3)b 4.9 (0.5)ab 3.3 (0.3)b 3.6 (0.7)b 4.8 (0.8)ab 5.9 (0.8)a 5.0 (0.7)a 5.8 (0.4)a 6.6 (0.4)a 4.6 (0.8)a 4.1 (0.2)a 8.2 (0.9)a 0.17 Total N (%) 0.24 (0.01)a 0.26 (0.01)a 0.18 (0.03)b 0.24 (0.02)a 0.27 (0.03)a 0.32 (0.04)b 0.33 (0.02)b 0.42 (0.02)a 0.27 (0.04)b 0.25 (0.01)b 0.49 (0.04)a (0.01)b Total P 78 (3)bc 112 (3)a 65 (3)ce 63 (3)de 96 (4)b 58 (4)d 85 (8)cd 93 (4)c 272 (5)a 69 (8)cd 75 (3)d 267 (6)b (mg kg-1) Total Mo 10.9 6.6 (0.7)b 10.3 (1)a 2.0 (0.1)c 1.4 (0.1)c 1.8 (0.1)c 6.3 (1.1)a 5.6 (0.6)a 1.5 (0.2)b 5.8 (0.7)a 4.7 (0.6)a 0.1 (0)b (mg kg-1) (1.1)a Available N 10.4 24.7 (1.9)a 15.3 (1.4)b 4.6 (0.3)c 11.4 (0.7)b 5.9 (0.3)c 9.4 (0.9)bc 7.7 (0.3)c 16.4 (0.9)a 9.0 (0.6)bc 11.0 (1.0)b 15.0 (1.0)a (mg kg-1) (0.8)b Available P 3.3 (0.3)a 1.3 (0.1)c 1.3 (0.1)cd 2.7 (0.2)b 0.9 (0.1)d 1.2 (0.1)cd 1.8 (0.2)b 0.9 (0)df 1.0 (0)ef 2.0 (0.1)a 1.2 (0.1)ce 1.3 (0.1)cd (mg kg-1) Available Mo 6.42 8.53 (1.61)a 3.96 (0.65)bc 3.97 (0.76)c 5.97 (0.83)b 3.52 (0.31)c 3.47 (0.34)a 1.49 (0.42)bc 0.36 (0.17)c 0.61 (0.26)c 2.48 (0.34)ab 3.76 (1.18)a (µg kg-1) (0.9)b 4.21 pH 4.04 (0.02)b 4.06 (0.01)b 3.99 (0.02)b 4.00 (0.02)b 3.88 (0.02)c 3.95 (0.02)a 3.81 (0.01)bc 3.78 (0.06)c 3.82 (0.03)bc 3.80 (0.02)c 3.90 (0.02)ab (0.03)a Clay (%) 8.5 (0.5)b 17.5 (0.9)a 9.3 (0.1)b 8.7 (0.5)b 17.5 (0.9)a 9.3 (0.1)b 18.2 (0.5)c 26.0 (1.4)b 42.8 (0.9)a 18.5 (0.6)c 25.7 (1.4)b 42.4 (0.8)a 77.3 Sand (%) 63.5 (1.4)b 76.2 (0.2)a 76.6 (0.9)a 63.5 (1.4)b 76.2 (0.2)a 63.8 (0.9)a 53.2 (2.6)b 22.8 (2.0)c 63.4 (1.0)a 53.5 (2.6)b 23.2 (2.0)c (0.9)a Bulk density 58.0 51.9 (1.4)b 56.4 (2.5)a 58.1 (1.8)a 51.9 (1.4)b 56.4 (2.5)a 47.1 (2)a 45.9 (1.2)a 39.1 (1.3)b 46.4 (2.2)a 45.9 (1.2)a 40.0 (1.4)b (kg m-2) (1.9)a 632 Values are means with standard errors in parentheses. Letters denote significant differences (linear mixed effects model with Season and Topography as factors, followed by post hoc 633 tests and with p<0.05 as significance level) within a site. 634 635 Table S2. Leaf litter water content and stoichiometry at Paracou and Nouragues for both seasons and on the three topographies. Paracou Nouragues Wet season Dry season Wet season Dry season Bottom Slope Top Bottom Slope Top Bottom Slope Top Bottom Slope Top Gravimetric 0.66 0.64 (0.01)a 0.67 (0.01)a 0.58 (0.02)b 0.42 (0.02)c 0.31 (0.01)d 0.22 (0.01)e 0.67 (0.01)a 0.62 (0.01)b 0.60 (0.01)b 0.44 (0.01)c 0.42 (0.03)c water (%) (0.01)ab C:N ratio 36.1 (0.9)b 34.9 (0.7)b 37.5 (1)b 42 (1.5)a 39.5 (0.9)a 42.1 (1.2)a 31.5 (1.8)b 34.2 (0.8)b 34.7 (0.9)b 33.7 (1.2)b 40.8 (1.6)a 38.9 (1.4)a N:P ratio 59.8 (2.2)c 60.2 (3.3)c 74.6 (3.1)b 63.0 (3.7)b 74.9 (3.7)ab 83.6 (5.5)a 60.4 (3.1)b 64.6 (4.0)b 58.7 (3.1)b 82.3 (5.2)a 79.6 (3.9)a 70.8 (2.9)a Total C (%) 40.5 (1.1)b 43.9 (0.7)b 44.6 (1.3)b 45.4 (0.6)a 46.0 (0.3)a 45.1 (0.6)a 41.6 (1.1)b 45 (0.6)a 45.6 (0.4)a 45.1 (0.4)a 47.0 (0.7)a 45.9 (0.5)a Total N (%) 1.14 (0.04)a 1.28 (0.03)a 1.21 (0.05)a 1.11 (0.05)b 1.19 (0.03)b 1.09 (0.03)b 1.39 (0.08)a 1.33 (0.03)a 1.34 (0.04)a 1.37 (0.06)b 1.19 (0.04)b 1.2 (0.03)b Total P (mg kg-1) 200 (14)a 227 (13)a 171 (14)a 190 (24)b 171 (10)b 140 (11)b 251 (26)a 220 (16)a 240 (14)a 180 (12)b 157 (9)b 176 (9)b Total Mo (mg kg- 1 2.8 (0.3)b 1.5 (0.2)c 1.8 (0.4)c 3.4 (0.7)a 1.9 (0.2)ab 3.8 (0.8)ab 1.7 (0.3)a 0.9 (0.2)b 0.5 (0.1)b 1.8 (0.4)a 0.7 (0.1)b 0.7 (0.1)b ) Bulk density (g 370.4 469.1 367.7 340.2 347.6 441.0 380.7 336.7 571.8 547.5 682.3 468.2 (62.5)a m-2) (23.2)b (18.9)a (23.0)b (20.7)c (12.8)bc (29.7)b (20.7)b (20.5)b (94.2)a (58.4)a (72.6)a 636 Values are means with standard errors in parentheses. Letters denote significant differences (linear mixed effects model with Season and Topography as factors, followed by post hoc 637 tests and with p<0.05 as significance level) within a site. 638 Table S3 Correlation matrix showing Pearson’s r for the variable used in the stepwise regression analysis. Data from A 639 Paracou soil, B Nouragues soil and C leaf litter from both sites. Data was averaged per plot prior to calculation. 640 Abbreviations: C = total C, N = total N, P = total P, Mo = total Mo, C:N = C:N ratio, N:P = N:P ratio, P in = available P, Nin = 641 available N, Moin = available Mo, Moisture = water content, pH = pH, Clay = % clay, Sand = % sand and BD = bulk density A Soil variables C N C:N P Mo Nin Pin Moin pH Clay Sand BD Moisture -0,16 0,06 -0,24 0,45 0,64 0,68 0,05 0,55 0,55 0,31 -0,33 -0,23 C 0,78 0,63 0,03 -0,48 0,01 -0,37 -0,06 -0,43 0,20 -0,20 -0,09 N 0,04 0,04 -0,18 0,12 -0,51 -0,14 -0,39 0,17 -0,18 -0,23 C:N 0,02 -0,52 -0,07 -0,01 0,12 -0,19 0,11 -0,12 0,04 P -0,06 0,69 -0,26 0,49 0,00 0,85 -0,83 -0,63 Mo 0,26 0,34 0,30 0,69 -0,31 0,31 0,21 Nin -0,25 0,33 0,09 0,64 -0,63 -0,45 Pin 0,02 0,36 -0,51 0,53 0,54 Moin 0,52 0,45 -0,44 -0,22 pH -0,14 0,09 0,18 Clay -0,99 -0,68 Sand 0,73 642 B Soil variables C N C:N P Mo Nin Pin Moin pH Clay Sand BD Moisture 0,75 0,77 0,14 0,65 -0,45 0,48 -0,36 -0,21 -0,20 0,59 -0,62 -0,64 C 0,97 0,25 0,67 -0,52 0,50 -0,18 0,22 -0,12 0,61 -0,63 -0,63 N 0,05 0,75 -0,63 0,62 -0,23 0,15 -0,11 0,70 -0,72 -0,74 C:N -0,11 0,37 -0,36 -0,02 0,13 -0,05 -0,18 0,20 0,34 P -0,70 0,80 -0,48 -0,10 0,03 0,90 -0,93 -0,79 Mo -0,70 0,40 0,06 0,19 -0,86 0,87 0,87 Nin -0,28 -0,05 -0,25 0,81 -0,82 -0,71 Pin 0,23 0,11 -0,63 0,60 0,54 Moin 0,36 -0,12 0,14 0,11 pH -0,19 0,15 0,09 Clay -0,99 -0,87 Sand 0,89 643 C Litter C N C:N P N:P Mo variables Moisture -0,29 0,55 -0,67 0,65 -0,50 -0,31 C 0,13 0,37 -0,11 0,25 -0,49 N -0,85 0,61 -0,13 -0,46 C:N -0,61 0,24 0,22 P -0,83 -0,26 N:P 0,13 644 645 646 647 Table S4 Comparison of FLNF rates measured in different studies carried out in primary tropical rainforests. Rates for soil 648 and leaf litter are given and expressed as nmol of ethylene produced per gram of substrate per hour (nmol g-1 h-1). Substrate Country Location FLNF Rate Reference (nmol g-1 h-1) Litter Hawai Pahoehoe 2.5 (0.4) Vitousek 1999 Hawai A’a 4.0 (1.4) Vitousek 1999 Hawai Thurston 3.15 (0.86) Crews 2000 Hawai Laupahoehoe 1.25 (0.31) Crews 2000 Hawai Kokee 1.08 (0.27) Crews 2000 Hawai Pahoehoe 7.42 (1.85) Vitousek 2000 Hawai Thurston 8.38 (2.10) Vitousek 2000 Hawai Laupahoehoe 1.93 (0.48) Vitousek 2000 Hawai Kokee 3.22 (0.81) Vitousek 2000 Costa Rica Osa Peninsula, Ultisol 8.82 (5.50) Reed et al 2007 Costa Rica Osa Peninsula, Mollisol 5.89 (4.75) Reed et al 2007 Panama 0.53 (0.17) Barron et al. 2009 Puerto Rico Wet tropical rainforest 2.0 (0.5) Cusack 2009 Puerto Rico Lower montane rainforest 1.2 (0.5) Cusack 2009 Costa Rica Osa Peninsula 11.39 (2.75) Reed et al 2010 Panama Fairchild 6.52 (1.00) Wurzburger et al. 2012 Panama AVA 0.34 (0.08) Wurzburger et al. 2012 Panama Gigante 0.38 (0.06) Wurzburger et al. 2012 Panama Barro Verde 1.84 (0.42) Wurzburger et al. 2012 Panama Zetek 0.48 (0.21) Wurzburger et al. 2012 Panama Rio Paja 1.58 (0.23) Wurzburger et al. 2012 Costa Rica Osa Peninsula 3.77 (0.46) Reed et al. 2013 Costa Rica Osa Peninsula 0.60 (0.15) Sullivan et al. 2014* French Guiana Paracou 0.32 (0.10) This study French Guiana Nouragues 0.18 (0.06) This study Soil Costa Rica Osa Peninsula, Ultisol 0.080 (0.013) Reed et al 2007 Costa Rica Osa Peninsula, Mollisol 0.042 (0.009) Reed et al 2007 Puerto Rico Wet tropical rainforest 0.11 (0.03) Cusack 2009 Puerto Rico Lower montane rainforest 0.06 (0.02) Cusack 2009 Ecuador 1000 m 0.179 (0.112) Matson et al. 2014 Ecuador 2000 m 0.313 (0.156) Matson et al. 2014 Ecuador 3000 m 0.223 (0.134) Matson et al. 2014 Costa Rica Osa Peninsula 0.017 (0.004) Sullivan et al. 2014* French Guiana Paracou 0.011 (0.005) This study French Guiana Nouragues 0.021 (0.011) This study 649 * For this study we found no bulk density reported for soil and litter. To calculate the amount of ethylene produced from the 650 kg N ha-1 y-1 reported in the study we used the bulk density values we measured in French Guiana. 651 652 653 Figure S1. Soil classification based on texture for each of the twelve plots in Paracou and Nouragues. Dots are plot averages. 654 655 656 Figure S2 Relative importance of physico-chemical variables in the overall soil dataset (A), in Paracou (B) and in Nouragues 657 (C). Higher relative importance means the predictor value is more likely to play a significant role in explaining the observed 658 variation in FLNF rate (Burnham and Anderson 2002). Relative importance was calculated by summing the Akaike weights of 659 each model, from all possible first order models, in which the variable participated. Moisture = water content, Ctot = total 660 C, Ntot = total N, Ptot = total P, MoTot = total Mo, CN = C:N ratio, Navail = available N, Pbray = available P, Moavail = 661 available Mo, pH = pH, Clay = percentage clay content and BD = bulk density. 662 663 Figure S3 Relative importance of physico-chemical variables in the overall litter dataset (A), in the wet season (B) and in the 664 dry season (C). Higher relative importance means the predictor value is more likely to play a significant role in explaining 665 the observed variation in FLNF rate (Burnham and Anderson 2002). Relative importance was calculated by summing the 666 Akaike weights of each model, from all possible first order models, in which the variable participated. Moisture = water 667 content, Ctot = total C, Ntot = total N, Ptot = total P, MoTot = total Mo, CN = C:N ratio and NP = N:P ratio. 668 669 Figure S4. Percentage of N fixation rates below (black) and above (grey) the detection limit as a function of season and 670 topography for (A) Paracou soil, (B) Nouragues soil, (C) Paracou leaf litter and (D) Nouragues leaf litter. 671 672 673 Figure S5. Boxplots comparing the effect of season on N fixation in Nouragues soils using the data set containing all datapoints 674 (A) and the data set excluding datapoints below the detection limit (B). 675 676 References 677 Aguilos M, Hérault B, Burban B, Wagner F, Bonal D (2018) What drives long-term variations in carbon flux and 678 balance in a tropical rainforest in French Guiana? Agricultural and Forest Meteorology 253-254:114-123 679 doi:10.1016/j.agrformet.2018.02.009 680 Amundson R et al. (2003) Global patterns of the isotopic composition of soil and plant nitrogen Global 681 Biogeochem Cy 17 doi:10.1029/2002gb001903 682 Barron AR, Purves DW, Hedin LO (2011) Facultative nitrogen fixation by canopy legumes in a lowland tropical 683 forest Oecologia 165:511-520 doi:10.1007/s00442-010-1838-3 684 Barron AR, Wurzburger N, Bellenger JP, Wright SJ, Kraepiel AML, Hedin LO (2009) Molybdenum limitation of 685 asymbiotic nitrogen fixation in tropical forest soils Nat Geosci 2:42-45 doi:Doi 10.1038/Ngeo366 686 Bates D, Mächler M, Bolker B, Walker S (2015) Fitting Linear Mixed-Effects Models Using lme4 2015 67:48 687 doi:10.18637/jss.v067.i01 688 Batterman SA, Hedin LO, van Breugel M, Ransijn J, Craven DJ, Hall JS (2013) Key role of symbiotic dinitrogen 689 fixation in tropical forest secondary succession Nature 502:224-227 doi:10.1038/nature12525 690 Benner JW, Conroy S, Lunch CK, Toyoda N, Vitousek PM (2007) Phosphorus Fertilization Increases the 691 Abundance and Nitrogenase Activity of the Cyanolichen Pseudocyphellaria crocata in Hawaiian Montane 692 Forests Biotropica 39:400-405 doi:doi:10.1111/j.1744-7429.2007.00267.x 693 Berg JM, Tymoczko JL, Stryer L (2002) Biochemistry, Fifth Edition. 5, illustrated edn. W. H. Freeman, 694 Bongers F, Charles-Dominique P, Forget P, Thèry M (2001) Nouragues Dynamics and Plant-Animal Interactions 695 in a Neotropical Rainforest. Kluwer Academic Publishers, Dordrecht, The Netherlands. doi:10.1007/978-94- 696 015-9821-7 697 Bowell RJ, Ansah RK (1993) Trace element budget in an African savannah ecosystem Biogeochemistry 20:103- 698 126 doi:10.1007/bf00004137 699 Bray RH, Kurtz LT (1945) Determination of Total, Organic, and Available Forms of Phosphorus in Soils Soil 700 Science 59:39-46 701 Brookshire ENJ, Gerber S, Menge DNL, Hedin LO (2012) Large losses of inorganic nitrogen from tropical 702 rainforests suggest a lack of nitrogen limitation Ecology letters 15:9-16 doi:doi:10.1111/j.1461- 703 0248.2011.01701.x 704 Burnham K, Anderson D (2002) Model selection and multimodel inference: a practical information-theoretic 705 approach. Springer-Verlag, New York 706 Černá B, Rejmánková E, Snyder JM, Šantrůčková H (2009) Heterotrophic nitrogen fixation in oligotrophic 707 tropical marshes: changes after phosphorus addition Hydrobiologia 627:55-65 doi:10.1007/s10750-009- 708 9715-y 709 Chave J et al. (2010) Regional and seasonal patterns of litterfall in tropical South America Biogeosciences 7:43- 710 55 doi:10.5194/bg-7-43-2010 711 Cleveland CC, Townsend AR (2006) Nutrient additions to a tropical rain forest drive substantial soil carbon 712 dioxide losses to the atmosphere Proceedings of the National Academy of Sciences 103:10316-10321 713 doi:10.1073/pnas.0600989103 714 Colombo C, Palumbo G, He J-Z, Pinton R, Cesco S (2014) Review on iron availability in soil: interaction of Fe 715 minerals, plants, and microbes Journal of Soils and Sediments 14:538-548 doi:10.1007/s11368-013-0814-z 716 717 Courtois EA et al. (2018) Spatial Variation of Soil CO2, CH4 and N2O Fluxes Across Topographical Positions in 718 Tropical Forests of the Guiana Shield Ecosystems doi:10.1007/s10021-018-0232-6 719 Crews TE, Farrington H, Vitousek PM (2000) Changes in Asymbiotic, Heterotrophic Nitrogen Fixation on Leaf 720 Litter of Metrosideros polymorpha with Long-Term Ecosystem Development in Hawaii Ecosystems 3:386- 721 395 doi:10.1007/s100210000034 722 Cusack DF, Silver W, McDowell WH (2009) Biological Nitrogen Fixation in Two Tropical Forests: Ecosystem-Level 723 Patterns and Effects of Nitrogen Fertilization Ecosystems 12:1299-1315 doi:10.1007/s10021-009-9290-0 724 Davidson EA, Verchot LV, Cattânio JH, Ackerman IL, Carvalho JEM (2000) Effects of soil water content on soil 725 respiration in forests and cattle pastures of eastern Amazonia Biogeochemistry 48:53-69 726 doi:10.1023/a:1006204113917 727 Davidson EA, Janssens IA (2006) Temperature sensitivity of soil carbon decomposition and feedbacks to climate 728 change Nature 440:165-173 doi:10.1038/nature04514 729 Dixon R, Kahn D (2004) Genetic regulation of biological nitrogen fixation Nat Rev Microbiol 2:621-631 730 doi:10.1038/nrmicro954 731 Dynarski KA, Houlton BZ (2018) Nutrient limitation of terrestrial free-living nitrogen fixation The New 732 phytologist 217:1050-1061 doi:10.1111/nph.14905 733 Eisele KA, Schimel DS, Kapustka LA, Parton WJ (1989) Effects of available P and N:P ratios on non-symbiotic 734 dinitrogen fixation in tallgrass prairie soils Oecologia 79:471-474 doi:10.1007/bf00378663 735 Epron D, Bosc A, Bonal D, Freycon V (2006) Spatial variation of soil respiration across a topographic gradient in 736 a tropical rain forest in French Guiana J Trop Ecol 22:565-574 doi:10.1017/s0266467406003415 737 Furnkranz M, Wanek W, Richter A, Abell G, Rasche F, Sessitsch A (2008) Nitrogen fixation by phyllosphere 738 bacteria associated with higher plants and their colonizing epiphytes of a tropical lowland rainforest of 739 Costa Rica The ISME journal 2:561-570 doi:10.1038/ismej.2008.14 740 Galloway JN et al. (2004) Nitrogen Cycles:Past, Present, and Future Biogeochemistry 70 741 Galloway JN, Schlesinger WH, Levy H, Michaels A, Schnoor JL (1995) Nitrogen fixation: Anthropogenic 742 enhancement‐environmental response Global Biogeochem Cy 9:235-252 doi:doi:10.1029/95GB00158 743 Gee GW, Bauder JW (1986) Particle-size analysis. In: Methods of Soil Analysis Part 1: Physical and Mineralogical 744 Methods 2nd edition edn. American Society of Agronomy & Soil Science Society of America, Madison, 745 Wisconsin, USA 746 Gourlet-Fleury S, Guehl JM, Laroussinie O (2004) Ecology and management of a neotropical rainforest. Lessons 747 drawn from Paracou, a long-term experimental research site in French Guiana. Elsevier, Meppel, The 748 Netherlands 749 Grau O et al. (2017) Nutrient-cycling mechanisms other than the direct absorption from soil may control forest 750 structure and dynamics in poor Amazonian soils Sci Rep 7:45017 doi:10.1038/srep45017 751 Gross A, Turner BL, Goren T, Berry A, Angert A (2016) Tracing the Sources of Atmospheric Phosphorus 752 Deposition to a Tropical Rain Forest in Panama Using Stable Oxygen Isotopes Environ Sci Technol 50:1147- 753 1156 doi:10.1021/acs.est.5b04936 754 Gupta UC, Lipsett J (1981) Molybdenum in Soils, Plants, and Animals. In: Brady NC (ed) Advances in Agronomy, 755 vol 34. Academic Press, pp 73-115. doi:https://doi.org/10.1016/S0065-2113(08)60885-8 756 Gutschick VP (1981) Evolved strategies in nitrogen acquisition by plants American Naturalist:607-637 757 Hammond D (2005) Tropical Forests of the Guiana Shield: Ancient Forests in a Modern World. CABI Publishing, 758 Cambridge 759 Hardy RW, Holsten RD, Jackson EK, Burns RC (1968) The acetylene-ethylene assay for N(2) fixation: laboratory 760 and field evaluation Plant Physiol 43:1185-1207 761 Hartmann A, Burris RH (1987) Regulation of nitrogenase activity by oxygen in Azospirillum brasilense and 762 Azospirillum lipoferum Journal of Bacteriology 169:944-948 763 Hartwig UA (1998) The regulation of symbiotic N2 fixation: a conceptual model of N feedback from the 764 ecosystem to the gene expression level Perspectives in Plant Ecology, Evolution and Systematics 1:92-120 765 doi:https://doi.org/10.1078/1433-8319-00054 766 Hayatsu M, Tago K, Saito M (2008) Various players in the nitrogen cycle: Diversity and functions of the 767 microorganisms involved in nitrification and denitrification Soil Science and Plant Nutrition 54:33-45 768 doi:10.1111/j.1747-0765.2007.00195.x 769 Hedin LO, Brookshire ENJ, Menge DNL, Barron AR (2009) The Nitrogen Paradox in Tropical Forest Ecosystems 770 Annual Review of Ecology, Evolution, and Systematics 40:613-635 771 doi:10.1146/annurev.ecolsys.37.091305.110246 772 Hill S (1988) How is nitrogenase regulated by oxygen? FEMS Microbiology Reviews 4:111-129 773 doi:10.1111/j.1574-6968.1988.tb02738.x 774 Hofmockel KS, Schlesinger WH (2007) Carbon Dioxide Effects on Heterotrophic Dinitrogen Fixation in a 775 Temperate Pine Forest Abbreviations: FACE, free air carbon dioxide enrichment; NPP, net primary 776 production Soil Science Society of America Journal 71:140-144 doi:10.2136/sssaj2006.110 777 Hu Y, Lee CC, Ribbe MW (2012) Vanadium Nitrogenase: A Two-Hit Wonder? Dalton transactions (Cambridge, 778 England : 2003) 41:10.1039/c1031dt11535a doi:10.1039/c1dt11535a 779 Hungate BA, Dukes JS, Shaw MR, Luo Y, Field CB (2003) Atmospheric science. Nitrogen and climate change 780 Science (New York, NY) 302:1512-1513 doi:10.1126/science.1091390 781 Igarashi RY, Seefeldt LC (2003) Nitrogen fixation: the mechanism of the Mo-dependent nitrogenase Critical 782 reviews in biochemistry and molecular biology 38:351-384 doi:10.1080/10409230391036766 783 IPCC (2013) Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth 784 Assessment Report of the Intergovernmental Panel on Climate Change [Stocker, T.F., D. Qin, G.-K. Plattner, 785 M. Tignor, S.K. Allen, J. Boschung, A. Nauels, Y. Xia, V. Bex and P.M. Midgley (eds.)]. Cambridge, United 786 Kingdom and New York, NY, USA. doi:10.1017/CBO9781107415324 787 Jones K (1977) The Effects of Moisture on Acetylene Reduction by Mats of Blue-green Algae in Sub-tropical 788 Grassland Annals of botany 41:801-806 doi:10.1093/oxfordjournals.aob.a085355 789 Kaiser BN, Gridley KL, Brady JN, Phillips T, Tyerman SD (2005) The role of molybdenum in agricultural plant 790 production Annals of botany 96:745-754 doi:10.1093/aob/mci226 791 LeBauer DS, Treseder KK (2008) NITROGEN LIMITATION OF NET PRIMARY PRODUCTIVITY IN TERRESTRIAL 792 ECOSYSTEMS IS GLOBALLY DISTRIBUTED Ecology 89:371-379 doi:doi:10.1890/06-2057.1 793 Leigh JA (2000) Nitrogen fixation in methanogens: the archaeal perspective Current issues in molecular biology 794 2:125-131 795 Limmer C, Drake HL (1996) Non-symbiotic N2-fixation in acidic and pH-neutral forest soils: Aerobic and 796 anaerobic differentials Soil Biology and Biochemistry 28:177-183 doi:https://doi.org/10.1016/0038- 797 0717(95)00118-2 798 Losos E, Leigh EG (2004) Tropical Forest Diversity and Dynamism: Findings from a Large-Scale Plot Network. 799 Chicago Press, Chicago: Univ. 800 Matson AL, Corre MD, Burneo JI, Veldkamp E (2014) Free-living nitrogen fixation responds to elevated nutrient 801 inputs in tropical montane forest floor and canopy soils of southern Ecuador Biogeochemistry 122:281-294 802 doi:10.1007/s10533-014-0041-8 803 Mazerolle MJ (2017) AICcmodavg: Model selection and multimodel inference based on (Q)AIC(c) 804 Menge DN, Levin SA, Hedin LO (2008) Evolutionary tradeoffs can select against nitrogen fixation and thereby 805 maintain nitrogen limitation Proceedings of the National Academy of Sciences of the United States of 806 America 105:1573-1578 doi:10.1073/pnas.0711411105 807 Menge DN, Levin SA, Hedin LO (2009) Facultative versus obligate nitrogen fixation strategies and their 808 ecosystem consequences Am Nat 174:465-477 doi:10.1086/605377 809 Pajares S, Bohannan BJ (2016) Ecology of Nitrogen Fixing, Nitrifying, and Denitrifying Microorganisms in 810 Tropical Forest Soils Front Microbiol 7:1045 doi:10.3389/fmicb.2016.01045 811 Pearson HL, Vitousek PM (2001) STAND DYNAMICS, NITROGEN ACCUMULATION, AND SYMBIOTIC NITROGEN 812 FIXATION IN REGENERATING STANDS OF ACACIA KOA Ecol Appl 11:1381-1394 doi:doi:10.1890/1051- 813 0761(2001)011[1381:SDNAAS]2.0.CO;2 814 Pearson HL, Vitousek PM (2002) Soil Phosphorus Fractions and Symbiotic Nitrogen Fixation across a Substrate- 815 Age Gradient in Hawaii Ecosystems 5:587-596 doi:10.1007/s10021-002-0172-y 816 Penuelas J et al. (2013) Human-induced nitrogen-phosphorus imbalances alter natural and managed 817 ecosystems across the globe Nature communications 4:2934 doi:10.1038/ncomms3934 818 Pérez S, Pérez C, Carmona M, Farina J, Armesto J (2008) Effects of labile phosphorus and carbon on non- 819 symbiotic N2 fixation in logged and unlogged evergreen forests in Chiloé Island, Chile. Rev CHil Hist Nat 820 81:12 821 Pham DN, Burgess BK (1993) Nitrogenase reactivity: Effects of pH on substrate reduction and carbon monoxide 822 inhibition Biochemistry 32:13725-13731 doi:10.1021/bi00212a043 823 Philippot L, Hallin S, Schloter M (2007) Ecology of Denitrifying Prokaryotes in Agricultural Soil. In: Advances in 824 Agronomy, vol 96. Academic Press, pp 249-305. doi:https://doi.org/10.1016/S0065-2113(07)96003-4 825 Porder S, Asner GP, Vitousek PM (2005) Ground-based and remotely sensed nutrient availability across a 826 tropical landscape Proceedings of the National Academy of Sciences of the United States of America 827 102:10909-10912 doi:10.1073/pnas.0504929102 828 Quesada CA et al. (2012) Basin-wide variations in Amazon forest structure and function are mediated by both 829 soils and climate Biogeosciences 9:2203-2246 doi:10.5194/bg-9-2203-2012 830 R CT (2017) R: A language and environment for statistical computing., 3.4.3 edn., Vienna, Austria 831 Reed SC, Cleveland CC, Townsend AR (2007) Controls over leaf litter and soil nitrogen fixation in two lowland 832 tropical rain forests Biotropica 39:585-592 doi:DOI 10.1111/j.1744-7429.2007.00310.x 833 Reed SC, Cleveland CC, Townsend AR (2011) Functional Ecology of Free-Living Nitrogen Fixation: A 834 Contemporary Perspective Annu Rev Ecol Evol S 42:489-512 doi:DOI 10.1146/annurev-ecolsys-102710- 835 145034 836 Reed SC, Cleveland CC, Townsend AR (2013) Relationships among phosphorus, molybdenum and free-living 837 nitrogen fixation in tropical rain forests: results from observational and experimental analyses 838 Biogeochemistry 114:135-147 doi:10.1007/s10533-013-9835-3 839 Reed SC, Townsend AR, Cleveland CC, Nemergut DR (2010) Microbial community shifts influence patterns in 840 tropical forest nitrogen fixation Oecologia 164:521-531 doi:10.1007/s00442-010-1649-6 841 Reis CESd, Dick DP, Caldas JdS, Bayer C (2014) Carbon sequestration in clay and silt fractions of Brazilian soils 842 under conventional and no-tillage systems Scientia Agricola 71:292-301 doi:10.1590/0103-9016-2013-0234 843 Robson RL, Postgate JR (1980) Oxygen and hydrogen in biological nitrogen fixation Annual review of 844 microbiology 34:183-207 doi:10.1146/annurev.mi.34.100180.001151 845 Roper MM, Smith NA (1991) Straw decomposition and nitrogenase activity (C2H2 reduction) by free-living 846 microorganisms from soil: Effects of pH and clay content Soil Biology and Biochemistry 23:275-283 847 doi:https://doi.org/10.1016/0038-0717(91)90064-Q 848 Roskoski JP (1980) Nitrogen fixation in hardwood forests of the northeastern United States Plant and Soil 849 54:33-44 doi:10.1007/bf02181997 850 Rousk J et al. (2010) Soil bacterial and fungal communities across a pH gradient in an arable soil The ISME 851 journal 4:1340-1351 doi:10.1038/ismej.2010.58 852 Scott WJ (1957) Water Relations of Food Spoilage Microorganisms. In: Mrak EM, Stewart GF (eds) Advances in 853 Food Research, vol 7. Academic Press, pp 83-127. doi:https://doi.org/10.1016/S0065-2628(08)60247-5 854 Seefeldt LC, Hoffman BM, Dean DR (2009) Mechanism of Mo-Dependent Nitrogenase Annual review of 855 biochemistry 78:701 doi:10.1146/annurev.biochem.78.070907.103812 856 Seitzinger SP, Garber JH (1987) Nitrogen-fixation and N-15(2) calibration of the acetylene-reduction assay in 857 coastal marine-sediments Mar Ecol Progr Ser 37:8 doi:10.3354/Meps037065 858 Sexstone AJ, Revsbech NP, Parkin TB, Tiedje JM (1985) Direct Measurement of Oxygen Profiles and 859 Denitrification Rates in Soil Aggregates1 Soil Science Society of America Journal 49:645-651 860 doi:10.2136/sssaj1985.03615995004900030024x 861 Sprent JI, Sprent P (1990) Nitrogen fixing organisms. Chapman and Hall, London, UK 862 Sullivan BW, Smith WK, Townsend AR, Nasto MK, Reed SC, Chazdon RL, Cleveland CC (2014) Spatially robust 863 estimates of biological nitrogen (N) fixation imply substantial human alteration of the tropical N cycle 864 Proceedings of the National Academy of Sciences of the United States of America 111:8101-8106 865 doi:10.1073/pnas.1320646111 866 Tang Y, Horikoshi M, Li WtUItVSRoPRP (2016) ggfortify: Unified Interface to Visualize Statistical Result fo 867 Popular R Packages The R Journal 8:12 868 ter Steege H et al. (2006) Continental-scale patterns of canopy tree composition and function across Amazonia 869 Nature 443:444-447 doi:10.1038/nature05134 870 Tiessen H, Chacon P, Cuevas E (1994) Phosphorus and nitrogen status in soils and vegetation along a 871 toposequence of dystrophic rainforests on the upper Rio Negro Oecologia 99:145-150 872 doi:10.1007/bf00317095 873 Tripathi BM et al. (2014) Spatial Scaling Effects on Soil Bacterial Communities in Malaysian Tropical Forests 874 Microbial Ecology 68:247-258 doi:10.1007/s00248-014-0404-7 875 van Kekem AJ, Pulles JHM, Khan Z (1996) Soils of the Rainforest in Central Guyana. Series 2, Georgetown, 876 Guyana 877 Van Sundert K, Horemans JA, Stendahl J, Vicca S (2018) The influence of soil properties and nutrients on conifer 878 forest growth in Sweden, and the first steps in developing a nutrient availability metric Biogeosciences 879 15:3475-3496 doi:10.5194/bg-15-3475-2018 880 Venables WN, Ripley BD (2002) Modern Applied Statistics with S. Fourth edn. Springer, New York 881 Vitousek PM (1984) Litterfall, Nutrient Cycling, and Nutrient Limitation in Tropical Forests Ecology 65:285-298 882 doi:10.2307/1939481 883 Vitousek PM, Menge DN, Reed SC, Cleveland CC (2013) Biological nitrogen fixation: rates, patterns and 884 ecological controls in terrestrial ecosystems Philosophical transactions of the Royal Society of London Series 885 B, Biological sciences 368:20130119 doi:10.1098/rstb.2013.0119 886 Vitousek PM, Sanford RL (1986) Nutrient Cycling in Moist Tropical Forest Annu Rev Ecol Syst 17:137-167 887 doi:DOI 10.1146/annurev.es.17.110186.001033 888 Wagner F, Rossi V, Stahl C, Bonal D, Hérault B (2013) Asynchronism in leaf and wood production in tropical 889 forests: a study combining satellite and ground-based measurements Biogeosciences 10:7307-7321 890 doi:10.5194/bg-10-7307-2013 891 Walker TW, Syers JK (1976) The fate of phosphorus during pedogenesis Geoderma 15:1-19 892 doi:https://doi.org/10.1016/0016-7061(76)90066-5 893 Wardle DA, Walker LR, Bardgett RD (2004) Ecosystem Properties and Forest Decline in Contrasting Long-Term 894 Chronosequences Science (New York, NY) 305:509-513 doi:10.1126/science.1098778 895 Weintraub SR, Taylor PG, Porder S, Cleveland CC, Asner GP, Townsend AR (2015) Topographic controls on soil 896 nitrogen availability in a lowland tropical forest Ecology 96:1561-1574 doi:doi:10.1890/14-0834.1 897 Wood TE, Lawrence D, Clark DA (2005) Variation in leaf litter nutrients of a Costa Rican rain forest is related to 898 precipitation Biogeochemistry 73:417-437 doi:10.1007/s10533-004-0563-6 899 Wurzburger N, Bellenger JP, Kraepiel AM, Hedin LO (2012) Molybdenum and phosphorus interact to constrain 900 asymbiotic nitrogen fixation in tropical forests PloS one 7:e33710 doi:10.1371/journal.pone.0033710 901 Yang J, Xie X, Wang X, Dixon R, Wang Y-P (2014) Reconstruction and minimal gene requirements for the 902 alternative iron-only nitrogenase in Escherichia coli Proceedings of the National Academy of 903 Sciences 111:E3718-E3725 doi:10.1073/pnas.1411185111 904 Zaehle S, Jones CD, Houlton B, Lamarque J-F, Robertson E (2015) Nitrogen Availability Reduces CMIP5 905 Projections of Twenty-First-Century Land Carbon Uptake* Journal of Climate 28:2494-2511 doi:10.1175/jcli- 906 d-13-00776.1 907 Zahran HH (1999) Rhizobium-Legume Symbiosis and Nitrogen Fixation under Severe Conditions and in an Arid 908 Climate Microbiology and Molecular Biology Reviews 63:968-989 909 Zhang X, McRose DL, Darnajoux R, Bellenger JP, Morel FMM, Kraepiel AML (2016) Alternative nitrogenase 910 activity in the environment and nitrogen cycle implications Biogeochemistry 127:189-198 911 doi:10.1007/s10533-016-0188-6 912 Zheng Y et al. (2018) A pathway for biological methane production using bacterial iron-only nitrogenase Nat 913 Microbiol 3:281-286 doi:10.1038/s41564-017-0091-5