(DCrossMark^ dick for updates

Spatially robust estimates of biological nitrogen (N) fixation ' , \ substantial human alteration of the tropical N cycleBenjamin W . Sullivan®' and Cory C. Cleveland®

^ W. Kolby Smith®, Alan R. Townsend*’, Megan K. Nasto®, Sasha C. Reed^ Robin L. Chazdon**,

‘‘D epa rtm en t o f Ecosystem and Conservation Sciences, University o f M ontana , Missoula, MT 59812; ‘‘D epa rtm en t o f Ecology and E vo lu tionary B iology, Environm ental Studies Program and Ins titu te o f A rc tic and A lp in e Research, U niversity o f Colorado, Boulder, CO 80309; 'US Geological Survey, Southwest Biological Science Center, M oab, UT 84532; and ‘‘D epa rtm en t o f Ecology and E vo lu tionary B iology, University o f Connecticut, Storrs, CT 06269-3043

Edited by Pamela A . M atson, S tan ford University, S tan ford , CA, and approved A p ril 14, 2014 (received fo r rev iew Novem ber 1, 2013)

Biological nitrogen fixatio n (BNF) is th e largest natural source o f exogenous nitrogen (N) to unm anaged ecosystems and also th e prim ary baseline against w hich anthropogenic changes to th e N cycle are m easured. Rates o f BNF in tropical ra in forest are th o u g h t to be am on g th e h ig h est on E arth , b u t th e y are n o to rio u s ly d ifficu lt to qu an tify and are based on little em pirical da ta . W e ad apted a sam pling strategy from com m unity ecology to generate spatial estim ates o f sym biotic and free-liv ing BNF in secondary and prim ary fo rest sites th a t span a typical range o f tropical forest legum e abundance. A lth ou gh to ta l BNF w as higher in secondary th an prim ary forest, overall rates w e re roughly five tim es low er th an previous estim ates fo r th e tropical fo rest biom e. W e found strong correlations b e tw een sym biotic BNF and legum e ab un­dance, bu t w e also show th a t spatia lly free-liv ing BNF o ften exceeds sym biotic inputs. Our results suggest th a t BNF in tropical fo rest has been overestim ated, and our d a ta are consistent w ith a recent to p -d o w n estim ate o f global BNF th a t im plied bu t did no t m easure lo w tropical BNF rates. Finally, com paring tropical BNF w ith in th e h istorical a rea o f tro p ica l ra in fo re s t w ith cu rren t an th ro p o g en ic N inputs ind icates th a t hum ans have a lre a d y a t least do u b led reactive N inputs to th e tro p ica l fo re s t b iom e, a fa r g re a te r change th a n prev iou s ly th o u g h t. Because N inputs are increasing fa s te r in th e trop ics th a n a n y w h e re on Earth, bo th th e p ro p o rtio n and th e e ffec ts o f hum an N en ric h m en t are lik e ly to g ro w in th e fu tu re .

adap tive c luster sam pling | free -liv ing n itrog e n fix a tio n n itrog e n depos ition | sym biotic n itrog e n fix a tio n

Over the last few decades, humans have dram atically altered the global nitrogen (N ) cycle (1-3). Three main processes—

Haber-Bosch fixation o f atmospheric N 2 , widespread cultivation o f leguminous N-frxing crops, and incidental N fixation during fossil fuel combustion— collectively add more reactive N to the biosphere each year than all natural processes combined (2). A lthough human perturbation o f the N cycle has brought sub­stantial benefits to society (most notably, an increase in crop production) (4), i t has also had a number o f negative effects on both ecosystems (5, 6) and people (7).

A lthough hum anity’s large im p rin t on the global N cycle is clear, quantifying the extent o f anthropogenic changes depends, in large part, on establishing baseline estimates o f nonanthropogenic N inputs (1, 8, 9). Before recent hum an activities, b io log ica l N fixa tion (B N F ) was the largest source o f new N to the b io ­sphere (9). Terrestria l B N F has been pa rticu la rly challenging to quantify, because i t displays high spatial and tem poral het­erogeneity at local scales, it arises from both symbiotic associations between bacteria and plants as well as free-living microorganisms (e.g., in leaf litte r and soil) (10), and high atmospheric concen­trations o f N 2 make direct flux measurements unfeasible. Conse­quently, spatial estimates o f BN F have always been highly uncertain (11), and global rate estimates have fallen precipitously in the last

15 y (fro m 100-290 to ~44 Tg N y *) (9). This decline in B N F im plies an increase in the re la tive m agnitude o f anthropogenic N inputs fro m 100-150% to 190-470% o f B N F (9).

H istorically, the largest anthropogenic changes to the N cycle have occurred in the northern temperate zone: firs t throughout the U n ited States and western Europe and more recently, in China (12,13). Large-scale estimates o f BN F in natural ecosystems in these regions are consistently low (11), leading some to conclude that anthropogenic N inputs in the northern temperate zone ex­ceed naturally occurring BN F and preindustrial atmospheric N deposition by an order o f magnitude o r more (1, 14). By contrast, the highest rates o f naturally occurring B N F have been thought to occur in the evergreen lowland tropical rainforest biome (11), impljdng that, on a regional basis, human alteration o f the tropical N cycle has been comparatively modest. However, in recent years, the tropics have seen some o f the most dramatic increases in an­thropogenic N inputs o f any region on Earth— a trend that is likely to continue (2, 6, 13). Anthropogenic N inputs are increasing in tropical regions, prim arily because o f increasing fossil fuel com­bustion (13) and expanding high-N-input agriculture fo r both food and biofuels (6). These anthropogenic N inputs are having a mea­surable effect on tropical ecosystems (15). However, understanding and forecasting the effects o f anthropogenic N depend, in part, on accurate estimates o f BN F in lowland tropical rainforest.

Significance

Biological n itro g e n fix a tio n (BNF) is th e largest na tura l source o f n e w n itro g en (N ) to te rres tria l ecosystem s. Tropical fo re s t ecosystem s are a p u ta tiv e g loba l h o tsp o t o f BNF, b u t d irec t, sp atia lly exp lic it m easurem ents in th e b iom e are v ir ­tu a lly n o nex is ten t. N onetheless, robu st es tim ates o f trop ical fo re s t BNF are critical fo r un derstand ing h o w these im p o r­ta n t ecosystem s m ay respond to g loba l change and assessing hum an p e rtu rb a tio n s to th e N cycle. H ere, w e in troduce a spatia l sam pling m eth o d to assess BNF and present e v i­dence th a t trop ica l fo re s t BNF is much lo w e r th a n previously assum ed. O ur results im p ly th a t hum ans have roug h ly d o u ­bled N inputs to th e trop ica l fo re s t b iom e re la tive to N inputs th ro u g h BNF.

A u thor contributions: B.W.S., M.K.N., R.L.C., and C.C.C. designed research; B.W.S., W.K.S., M.K.N., S.C.R., and C.C.C. perform ed research; B.W.S., W.K.S., and C.C.C. analyzed data; and B.W.S., W.K.S., A.R.T., M.K.N., S.C.R., and C.C.C. w ro te the paper.

The authors declare no conflict o f interest.

This article is a PNAS Direct Submission.

Present address: Department o f Natural Resources and Environmental Sciences, Univer­sity o f Nevada, Reno, NV 89557.

^To whom correspondence should be addressed. E-mail: bsully613@gmail.com.

This article contains supporting in form ation online at www.pnas.org/lookup/suppl/doi: 10. 1073/pnas. 1320646111/-/DCSupplemental.

W W W .pnas.org/cg i/do i/10.1073/pnas. 1320646111 PNAS Early E d ition | 1 o f 6

4

Unfortunate ly, the paradigm that the tropics have high rates o f B N F is based on a paucity o f evidence and several tenuous assumptions. For example, an early global synthesis o f terrestria l B N F (11)— which included contributions from both symbiotic and free-liv ing sources— included only one measured estimate o f symbiotic B N F from tropical forest (16 kg N ha“ ̂y “ ^) (16). That single estimate, scaled over thousands o f square kilometers,

^ 0 represented the on ly d irect evidence o f high trop ica l B N F rates available at that tim e (Fig. 1). Subsequent modeled estimates (17) tha t ind irectly estimated B N F have reinforced the notion that tropical B N F rates are high and dom inated by the symbiotic fo rm o f fixation (Fig. 1). Such high estimates o f symbiotic B N F are consistent w ith the large num ber o f legum inous trees in tropical forest (18-20). However, many legume species do not fo rm N-frxing nodules (21), and o f those species that do, nodu- la tion in individuals varies w ith soil nu trien t status, N demand, and tree age (22). Several recent analyses (10, 22-24) indicate lower tropical forest B N F and suggest that symbiotic B N F may not be as im portan t to to ta l B N F as previously thought (Fig. 1), although few studies have simultaneously measured symbiotic and free-living BNF.

There is also a sound theore tica l basis fo r questioning high estimates o f B N F in trop ica l forest. Namely, high concen­tra tions o f soil N in the legum e-rich tropics create som ething

P rior estim ates This study50

83%TO■= 40

76%

O) 76%30

LLZ 51%CQ 43%

20 55%

O28% 29%

34%-(0Oa.o 22%

H< GQ a>o o

Fig.1. Previous estim ates o f BNF in trop ica l ra in fo res t and BNF measured in th is study. Percentages ind ica te th e p ro p o rtio n o f to ta l BNF fro m sym biotic BNF. C leveland e t al. 1999 A (11) is a lite ra tu re database-derived estim ate o f trop ica l fo res t BNF; Cleveland e t al. 1999 B (11) is a m odeled estim ate o f BNF based on th e co rre la tion be tw een net p rim ary p ro d u c tiv ity (NPP) and BNF derived w ith rem ote ly sensed NPP and evergreen broadleaved fo res t (EBP) land cover classification. Central estim ates and variance fo r Cleveland e t al., 1999 A (11) and Reed e t al. 2011 (10) represent th e low , centra l, and h igh data-based estim ates o f BNF assuming 5% , 15%, and 15% legum e cover, respectively. Central estim ates and variance fo r W ang and F loulton 2009 (17) represent th e m ode led mean and SD o f BNF pred icted fo r th e EBF b iom e. Central estim ates and variance fo r Cleveland e t al. 2010 (23) represent th e low , centra l, and h igh estim ates o f sym bio tic BNF plus free -liv ing BNF o r m odeled BNF plus free -liv ing BNF. Central estim ates and variance fo r BNF in th e fo u r fo res t ages measured here (p rim ary, 5-15 y, 15-30 y, and 30-50 y) represent means ± 1 SD (n = 3). O ur estim ate o f BNF in a dynam ic p rim ary fo res t (gap dynamics) lacks SD, because it consisted o f on ly tw o measure­m ents: lo w and h ig h estim ates o f fo re s t tu rn o v e r tim es equa l to 150 and 75 y, respective ly .

o f a paradox. A ltho ug h B N F could create N -rich conditions, the substantial energetic cost o f B N F means— and some data show— tha t B N F should be suppressed under high N ava il­a b ility in p rim ary forests (25). Because o f high rates o f net p rim ary p ro du c tiv ity and high N demand in secondary forests (26, 27), regenerating canopy gaps o r abandoned ag ricu ltu ra l land may have higher rates o f B N F than late-successional forest ecosystems (26).

Resolving the uncertainty in the tropical (and global) N cycle requires that we overcome the enduring challenge o f quantifying B N F in any ecosystem. H ow do we estimate large-scale rates o f a process that displays extreme spatial heterogeneity at local scales? W hether using acetylene reduction assays, ^®N tracer incubations, o r the ^®N natural abundance method, most past approaches to em pirica lly estimate symbiotic B N F have relied on spatial extrapolations o f B N F rates measured at the level o f individual trees. Typically, such extrapolations are based on legume abundance (e.g., percent cover) and make species- or genera-Ievel assumptions about nodulation status o f putative N fixers. Here, we applied a m ethod commonly used by community ecologists to measure rare species abundances— stratified adap­tive cluster sampling (SACS) (28)— to measure symbiotic BNF. This approach could be used in any ecosystem, and in contrast to other methods, SACS generates unbiased estimates o f mean symbiotic B N F (independent o f legume abundance) and can more robustly capture the irregular d istribu tion o f nodules on the landscape. W e simultaneously measured symbiotic and free-living B N F m ultip le times over the course o f 1 y to generate spatially explicit rates o f B N F inputs in prim ary and secondary (5-50 y old) lowland tropical forest in Costa Rica and then used the un­derstanding gained from those estimates to revisit estimates o f B N F and anthropogenic N inputs in the tropical forest biome.

Results and DiscussionTaken together, our data suggest far lower rates o f to ta l B N F in a region o f mixed prim ary and secondary trop ica l forest than have been previously reported (Fig. 1). The mean rate o f to ta l BN F that we measured in primary forest was only 1.2 kg N ha“ ̂y“ ,̂ 10-20 times lower than previously published empirical (11.7 kg N ha“ ̂y “ ^) (10) or modeled rates (25.4-31.9 kg N ha“ ̂y“ ^) (Fig. 1) (11,17). Secondary forest had higher total BN F than primary forest (6.2-14.4 kg N ha“ ̂y“ ^), and rates increased w ith age in the three successional forest age classes (although w ith substantial intraage variability) (Fig. 1). However, the primary forest BN F estimate does not explicitly account for the important role o f frequent small-scale disturbances that create canopy gaps in primary forest. Gaps pro­mote species turnover and thereby contribute to small-scale BNF variability in primary forest. A lthough the primary forest sites that we studied showed no signs o f recent disturbance and lacked large canopy gaps, data from other Costa Rican rainforests suggest tree turnover times that average between 75 and 150 y (29). Assuming that secondary forest BN F approximately represents BN F in dis­turbed forest patches (a reasonable assumption based on a recent analysis in nearby Panama) (22), we suggest a time-integrated mean estimate fo r BN F o f 5.7 kg N ha“ ̂ y~ fo r primary forest in this region (Fig. 1, gap dynamics). A lthough substantially higher than our estimate o f BN F in undisturbed primary forest alone, this es­timate is still low relative to previous estimates (Fig. 1).

Symbiotic BNF, which is typically assumed to be the dominant source o f B N F in tropical forests, accounted fo r only 20-50% o f to ta l B N F in our study depending on forest age (Fig. 1). There­fore, free-living B N F may represent an equal, i f not greater, source o f new N to both prim ary and secondary tropical forests than symbiotic BNF. O n a per-mass basis, symbiotic nodules have the highest rates o f B N F measured in nature (30), whereas mass- based rates o f free-living B N F tend to be much lower (10). However, free-living B N F is much more consistent across the landscape than sym biotic B N F. Thus, N inputs through

2 of 6 I www.pnas.org/cgi/doi/10.1073/pnas.1320646111 Sullivan et al.

Table 1. Legume (Fabaceae) species abundance in five prim ary fo rest sites [Korup, Luqillo, Yasuni, Pasoh, and Barro Colorado Island (BCI)] th a t contribute to th e Center fo r Tropical Forest Science p lo t n e tw o rk (34) as w e ll as th e prim ary fo rest (1°) and secondary fo rest (2°) m easured in th is study near th e Piro Biological Station in southw est Costa Rica

Site

Fabaceae a b u n d a n ce a n d basal a rea

K o rup ,C am eroon

L u q illo , P u e rto Rico

Y asun i,E cuador

Pasoh,M alays ia

BCI,Panam a

P iro 1°, Costa R ica*

P iro 2°, Costa R ica*

Fam ily ra n k 3 6 1 2 2 8 ± 2 5 ± 2Basal a rea (m^ h a “ ') 2.9 0.8 2.2 2.6 3.2 1.0 ± 0.5 9.1 ± 3.3

Basal a rea (% ) 9.0 6.5 14.9 8.5 9.9 0.7 ± 0.39 10.3 ± 3.0Trees (ha “ ') 387.6 36.0 377.2 — 343.7 67.3 ± 20.7 371.1 ± 95.6

Trees (% ) 5.9 2.7 13.0 3.3 7.5 8.2 ± 3.9 39.9 ± 9.9T o ta l t re e species (ha “ '; 38 6 108 — 37 90 ± 6 58 ± 6

a ll fa m ilie s )

‘ Means ± 1 SE.

widespread (but low) rates o f free-liv ing B N F may exceed N inputs through isolated (but high) rates o f symbiotic BNF. Recent syntheses have made this argument (Fig. 1) (10, 23), but un til now, d irect evidence has been lacking. W e note that, although soil and litte r are like ly the dom inant sources o f free-living BN F in tropical forest (10), i f we included N inputs from unmeasured sources such as decajdng wood (31), canopy epiphjTes (32), or termites (33), the relative contribution o f free-living B N F would almost certainly increase. Surprisingly, the proportion o f the three sources o f B N F that we measured (soil, litte r, and symbiotic) was sim ilar between prim ary and secondary forests (Fig. S I), raising questions about the biophysical and/or biogeochemical factors tha t regulate sym biotic and free-liv ing B N F across forest age classes (10). However, d iffe rent B N F rates between prim ary and secondary forest occurred, despite sim ilar soil N and P availability among the sites (Table S I).

W e measured low rates o f symbiotic B N F in prim ary tropical forest, despite the presence o f legumes in all ou r sites and an abundance o f legumes in secondary forest sites. In fact, both legume abundance and legume basal area in o u r sites were s im ilar to those in m ultip le well-studied tropical forests around the w orld (Table 1 and Fig. S2) (34). Therefore, we suggest that differences between our direct spatial measurement o f symbiotic B N F and previous estimates re fle c t the a b ility o f the SACS m ethod to provide a spatially unbiased estimate o f nodule b io ­mass w ith in a site and to generate more robust spatial estimates o f symbiotic BNF.

Accurately scaling symbiotic B N F from nodules to ecosystems has been a m ajor im pedim ent to generating spatial estimates o f symbiotic B N F in tropical forest. Scaling approaches that rely on legume abundance assume that legumes are actively fixing and/or tha t sym biotic B N F is co rre la ted w ith tree size. However, studies using N isotopes have shown tha t on ly 36 -53% o f

1 ^is1*^

^ 100

80■>

' r e

■c 60

O)

i T 40

moo!nE

20

>(0

-20

1 1 1 1 w

y = 1.07x + 9.40

-r^= 1.0

-2 r • Predicted prim ary

sg> o Predicted secondary■ Measured p rim ary

/ □ Measured secondary ~

□ -------------------------- m t - m — y = 0 .0 7 x -0 .0 6

1

r ̂= 0.57

1 1 1

B

20 40 60 80

Relative legume abundance (%)

100

E2wwreEo!nre3

■co

1.5 -

1 -

0.5 -

□ Piro■ Barron et al. 2011

40%

Prim ary Secondary

Forest succession status

Fig. 2. Comparisons betw een th e SACS approach used in th is study and existing m ethods o f spatia lly e x tra p o la ting BNF and nodu le biomass. (A) Based on legum e abundance in th e p lo ts th a t w e used, th e regression equa tion used by Cleveland e t al. (11) predicts much h ig h e r BNF in bo th prim ary and secondary forests than w e actua lly measured. In p rim a ry forests, th e regression approach predicts a sym biotic BNF flu x o f 14-23 kg N ha“ ̂ y “ ;̂ w e measured a flu x o f 0.1-0.5 kg N ha“ ̂ y“ \ (6) Using m axim um like lihood estim ates o f nodu le biomass and legum e abundance, Barron e t al. (22) estim ated nodu le biomass, on a spatia l basis, in bo th p rim ary and secondary forests in Panama. Using th e SACS design, w e measured 45% less nodu le biomass in p rim a ry forests near the Piro B iolog ica l S ta tion and 40% less nodu le biomass in secondary forests near th e Piro Biological S tation th a n estim ated by Barron e t al. (22) in Panama.

Sullivan e t al. PNAS Early E d ition | 3 o f 6

legumes are actively fix ing in the Am azon (35, 36), and legume tree size and nodule biomass were only weakly correlated in a Panamanian rainforest (22). By comparing our estimates o f B N F using the SACS method w ith two other methods that have scaled B N F based on legume abundance, we show tha t the SACS m ethod was the proxim ate cause o f the d ifference between the existing B N F estimates and our lower measurement o f BNF.

^ 0 W e used legume re la tive abundance in o u r p lo ts (Table 1) toestimate B N F w ith the same regression equa tion tha t was used by C leveland et al. (11). Com pared w ith o u r measured BNF,

^ 2 the legume abundance-based regression pred ic ted muchh igher rates o f B N F in ou r p rim a ry fo rest sites (14 -22 kg N ha“ ̂ y “ ^) (F ig . 2)— B N F rates were s im ila r to the rates tha t generated the biom e-Ievel estim ate in the w o rk by C leveland et al. (11). Because legumes were m ore abundant in secondary forest sites, the regression approach pred ic ted secondary forest B N F rates as h igh as 100 kg N ha“ ̂ y “ ,̂ a value s tr ik ­ing ly h igher than the measured values (F ig. 2). Furthermore, our estimate o f nodule biomass was 40% lower in prim ary forest and 45% lower in secondary forest than a legume-based maxi­mum like lihood scaling approach used in Panama (22) that pre­dicted much lower rates o f symbiotic B N F than had been reported in the literature (16) (Fig. 2). Because the SACS method incorporated both isolated areas, where nodules are abundant, and extensive areas, where nodules are absent, we argue that it represents a significant advance in the measurement o f symbiotic B N F in tropical forest.

The SACS approach estimated nodule biomass independently o f legume abundance, thereby provid ing the opportun ity to correlate nodule biomass and legume abundance at the ecosys­tem scale. There has been a surprising lack o f em pirical evidence to validate the in tu itive positive relationships among legume abundance, nodule biomass, and symbiotic BN F rates. F o r in ­stance, ter Steege et al. (18) suggested that, across the Amazon basin, legume abundance and nodule biomass were negatively correlated, im pijdng tha t legume abundance may no t predict symbiotic B N F rates. Here, we observed strong positive corre­lations among the basal area o f putatively N-frxing legumes (21), nodule biomass, and symbiotic B N F rates (Fig. S3). W e argue that these relationships emerged because we measured B N F across a secondary successional chronosequence where the number and size o f legumes varied. Both the sh ift in legume abundance through succession and the strength o f these corre­lations provide em pirical evidence fo r emerging theory and models that suggest tha t symbiotic B N F is up- and down-regu­lated by facultative mechanisms (25) o r species replacement (37) to meet N demand. The correlations also provide compelling evidence tha t symbiotic B N F may be a function o f legume basal area during succession and suggest a prom ising avenue fo r future w ork attem pting to scale symbiotic B N F rates from legume basal area estimates.

A lthough our estimates o f BN F were collected in one set o f sites, two lines o f evidence suggest tha t the low rates o f symbiotic B N F that we observed m ight be common in low land tropical forest. F irst, as mentioned above, legume abundance and basal area were s im ilar between our sites and other tropical forest sites around the w o rld (Table 1 and Fig. S2) (34), impljdng that the low rates o f B N F tha t we observed were no t because o f a lack o f legumes. Second, a reanalysis o f a recent lower global BN F es­tim ate tha t used a top-down ^®N isotope model approach (9) also im plied lower rates o f trop ica l forest BNF. V itousek et al. (9) estimated that preindustria l global B N F was 58 Tg N y“ ̂ (equal to 5.3 kg N ha“ ̂ y“ ̂ on vegetated land), much less than previous estimates o f 128-195 Tg N y“ ̂ (8,11, 17). Because tropical forest is believed to contribute a disproportionate amount to global BN F (11), we argue that fixation in this biome cannot remain high i f global B N F is one-half o r one-th ird the value suggested by p re­vious estimates. W e show this po in t by constraining two existing

global models o f B N F (11, 17) w ith our em pirica lly derived es­tim ate o f 5.7 kg N ha“ ̂y“ ̂ in tropical forest, while hold ing B N F constant in all other biomes. W ith tropical forest B N F con­strained, global B N F estimated by these models was reduced to the range predicted by the top-down isotope-based estimate (Fig. 3). A lthough we do not suggest tha t our em pirically derived estimate o f BN F represents the value for all tropical forests, i t is noteworthy that our measured rate is much easier to reconcile w ith 58 Tg N y“ ̂ global BN F than previous estimates (e.g., 25-30 kg N ha“ ̂ y “ ^) (F ig. 1). W e also note tha t i t is the firs t, to our knowledge, fie ld-based estim ate o f B N F tha t supports, by d i­rect measurement, the top-dow n ^®N isotope-based m odel estimate o f B N F (9). A ltho ug h some areas w ith exceptionally high rates o f B N F undoubtedly exist in trop ica l forest, these high rates o f BN F may either be transient (Fig. 1) (26) o r lack sufficient spatial extent to generate high rates o f BN F throughout the tropical rainforest biome.

If, as m ultip le lines o f evidence suggest, low rates o f B N F are common in tropical forest, this find ing w ould fundam entally alter our understanding o f bo th the tropical N cycle and the impact that humans have had on it. In undisturbed tropical forest, where relatively high soil N availability is common (38-42), lower rates o f B N F are consistent w ith theoretical tradeoffs in the energetics o f nutrient acquisition (37, 4 3 ^5 ). A lthough high B N F has been invoked as a driver o f those N-rich conditions, the maintenance o f such high rates in the face o f N abundance has always been para­doxical (25). The data that we present here support recent evidence suggesting that the m ajority o f BNF-derived N that enters tropical forest may do so episodically (i.e., during periods o f forest re­generation), when N demand and biomass accrual are high (26). Subsequent development o f N-rich conditions may, therefore, re­sult more from shifting patterns in nutrient lim itation and forest growth strategies than from chronically high rates o f BNF.

Low tropical forest B N F rates also im ply tha t recent human activity has enriched the tropical N cycle far more than p re­viously thought. Anthropogenica lly derived N deposition inputs in tropical regions are already sim ilar to (o r exceed) the rates o f B N F that we report here (15, 46). In areas deforested fo r agri­culture, massive per-area increases in N inputs are common because o f fe rtilize r application and/or the growth o f N-frxing crops— most notably, soy (47). Deforestation is o ften the result o f economically expedient (and po litica lly supported) agricul­tura l extensification (sensu lato, ref. 48), after which exhausted agricultural areas are abandoned to secondary succession.

To provide context fo r the importance o f B N F in evaluating the overall anthropogenic change in human vs. natural N inputs to the tropical forest biome, we compiled biome-wide atm o­spheric N deposition estimates (49), inorganic and organic fe r­tilize r use data (50), and an estimate o f agricultural N fixation based on crop area and jdeld statistics (51) w ith in the historical tropical rainforest biome (i.e., the tropical rainforest biome in the absence o f human activity) (52). M ost strikingly, this analysis shows the importance o f accurate B N F estimates in considering human changes to the tropical N cycle. In particular, anthropo­genic N sources only represented 31% o f to ta l nonanthropogenic N inputs using p r io r assumptions o f high forest B N F (Fig. 4). Using the lower B N F value indicated by our fie ld study, we suggest that humans have increased the amount o f reactive N entering the tropical rainforest biome by 134% (Fig. 4). W e note that our estimate o f anthropogenic N flux does no t include o r­ganic fertilizer, like manure, which recycles both endogenous and exogenous N w ith in ecosystems (assuming that most manure is not a net im port from other biomes). However, in Fig. 4, we depict biome-level organic fertilizer inputs to illustrate that ma­nure application in the tropics rivals the biome-wide forest B N F value derived from our estimate and that manure is an im portan t alternative to inorganic fe rtilize r manufactured by the H abe r- Bosch process (Fig. 4). The values fo r both agricultural BN F and

4 o f 6 I www.pnas.org/cg i/doi/10.1073/pnas.1320646111 Sullivan e t al.

A 1999 Model B 2009 Model

Original Revised50

40

30

10

0

73I

Original Revised

30

40

20 «

Global Tropical Global Tropical Forests Forests

Global Tropical Global Tropical Forests Forests

Fig. 3. Revising g loba l BNF m odels w ith o u r gap-based estim ate o f BNF in tro p ica l forests (5.7 kg N ha y ) places existing g loba l estim ates in th e range o f BNF measured by a to p -d ow n isotope m odel o f g loba l BNF (9) (dark b lue horizon ta l line, m ean; shaded b lue horizon ta l area, h igh and low estim ates). {A) By revising trop ica l BNF in th e o rig ina l 1999 m odel (11), w e reduced g loba l BNF fro m 11.8 to 8.5 kg N ha“ ̂ y “ \ (6) By revising trop ica l BNF in th e o rig ina l 2009 m odel (17), w e reduced g loba l BNF fro m 14.7 to 10.1 kg N ha“ ̂ y “ \ Dots represent means, w ith values no ted. Black ho rizon ta l bars w ith in box p lo ts represent th e m edian values, box p lo ts represent 1 SEM, and whiskers represent th e to ta l range o f estim ates. G lobal mean BNF estim ates using these tw o m odels rem ain substantia lly h ighe r th a n th e m edian va lue because o f h igh legum e-derived estim ates o f BNF in trop ica l savannas.

fertilizer use are undoubtedly conservative: although they represent com pila tions o f the curren tly available data (circa 2000), they do not integrate the fact that, over the past 15 y, both overall N fe rtilize r use and the extent o f soy cultivation (an im portan t source o f agricultural BNF) have expanded dramatically in tropical regions (6, 53).

Regardless o f the extent to which human activities have actu­ally perturbed the tropical N cycle, there is no doubt that, as the proportion o f anthropogenic N inputs increases relative to nat­ura l inputs, the region w ill continue to see significant ecological and socioeconomic impacts. In temperate regions, anthropogenic N inputs have contributed to shifts in species composition, even when below so-called critica l loads— thresholds below which inputs o f N are supposedly safe fo r ecosystems (54). G iven that tropical regions contain much o f the biological diversity on Earth (55), many tropical species may be vulnerable to the unintended, indirect effects o f increasing anthropogenic N. Similarly, as seen in much o f the temperate zone, increased anthropogenic N inputs could have positive effects on human health, but w ithou t effective management, they could also have profound negative effects (7). Low rates o f B N F in tropical regions add urgency to growing calls to manage increasing N inputs in tropical biomes and require local, regional, and in ternational policy instruments that, fo r the most part, are no t yet in place (6).

MethodsW e m easured BNF in 12 0.5-ha p lo ts (50 x 100 m) on th e Osa Peninsula, Costa Rica near th e Piro B iolog ica l S ta tion (8°24' N, 83°20' W ; 87 m above sea level) (56). SI M ethods, sections 1 and 2 p rov ide a d d itio n a l site d e ­scrip tion . W e measured sym bio tic , soil, and l it te r N fix a tio n fo u r tim es in 2012 and 2013, ca p tu ring all seasonal p re c ip ita tio n va ria tio n th a t th e sites experience— January, May, July, and O ctober. To measure sym b io tic BNF, w e m easured nodu le biomass du rin g th e early w e t season a t each site using SACS (28). SI M ethods, section 2 and Fig. S4 g ive a d d itio n a l deta ils . In to ta l, w e searched fo r nodules in ~1,500 5.5-cm-wide x 10-cm-deep cores. W e measured BNF rates on excised nodules, soil, and lit te r using th e acetylene reduction assay. SI M ethods, sections 3 and 4 give add itiona l details. W e ag­gregated th e 5- to 15-, 15- to 30-, and 30- to 50-y fo rest age classes in to the category o f secondary forest. W e measured prim ary fo rest tu rn o ve r (gap dy­namics) using tw o scenarios— low tu rn o ve r (150 y) and h igh tu rn o ve r (75 y) based on estimates in ref. 29. SI M ethods, section 5 details m easurem ent o f

BNF in each scenario. W e measured th e re la tionship betw een nodule biomass,

sym biotic BNF, and basal area o f pu ta tive N-fixing legumes using linear re­gression in th e statistical package R (57). W e constrained tw o existing spatially

exp lic it g lobal N2 fixa tion m odels (11, 17) by o u r em pirically derived estim ate o f trop ica l N2 fixa tion and com pared both previous and constrained models

w ith th e to p -dow n g lobal BNF measured in re f. 9. SI M ethods, section 6 gives add itiona l details on constra in ing th e models. W e quan tified th e exten t o f

anth ropogen ic impacts on th e trop ica l N cycle by com paring preindustria l N

70

60

C 50-z.U) 40

in3a

30

20

10

0

□ BNF■ P re ind u s tr ia l N d ep os ition H A g r ic u itu ra i BNF D in o rg a n ic fe r ti liz e r□ A n th ro p o g e n ic N d ep os ition

2009 BNF Rescaled TotalBNF anthropogenic

N sources

Manure

Fig. 4. The ra te o f BNF in trop ica l ra in fo res t genera ted using th e SACS approach is much low er th a n previous estim ates, and it im plies th a t th e hum an p e rtu rb a tio n to th e trop ica l N cycle is app ro x im ate ly fo u r tim es g rea te r th a n previous N fixa tio n estim ates w o u ld suggest. W e estim ated BNF using an existing (2009) m odel (17) th a t genera ted h igh estim ates o f BNF, and th e n , w e estim ated low e r rates o f BNF by dow nsca ling th e 2009 model (17) to o u r mean BNF estim ate o f 5.7 kg N ha“ ̂ y“ \ A n th ro po g e n ic N de­position was calculated as th e d iffe rence betw een to ta l trop ica l N deposition (49) and p re industria l N deposition (58). M anure was n o t tre a te d as an an­th ro p og e n ic N inp u t, because w e assumed th a t it is reg iona lly p roduced and th a t i t represents a recycling o f previous a n th rop o ge n ic o r n a tu ra lly fixed N. Flowever, m anure is a m a jo r source o f N to trop ica l ecosystems nearly equ iva len t to all o th e r a n th rop o ge n ic N sources com bined.

1 ^is1*^

Sullivan e t al. PNAS Early E d ition | 5 o f 6

inputs (BNF and N deposition) w ith recent estimates o f anth ropogen ic N inputs (postindustrial N deposition, agricu ltu ra l BNF, and agricu ltu ra l fe r t i l­ization) across to ta l historical trop ica l fo rest area. SI M ethods, section 7 gives add itiona l details.

ACKNOWLEDGMENTS. W e th a n k B. V ilchez and E. O rt iz fo r es ta b lish in g th e ch ronosequence p lo ts ; M . Lopez-M ora les , R. Cole, S. W e in tra u b , and

Vitousek PM, et al. (1997) Human alteration o f the global nitrogen cycle: Sources and consequences. Ecoi A p p i 7(3):737-750.Galloway JN, e t al. (2008) Transformation o f the nitrogen cycle: Recent trends, questions, and potentia l solutions. Science 320(5878):889-892.Rockstrom J, e t al. (2009) A safe operating space fo r humanity. Nature 461(7263): 472-475.Matson PA, Parton WJ, Power AG, Sw ift MJ (1997) Agricultural intensification and ecosystem properties. Science 277(5325):504-509.Clark CM, Tilman D (2008) Loss o f plant species a fter chronic low-level nitrogen de­position to prairie grasslands. Nature 451(7179):712-715.Austin AT, et al. (2013) Environment. Latin America's nitrogen challenge. Science 340(6129):149.Townsend AR, et al. (2003) Human health effects o f a changing global nitrogen cycle. Front Ecoi Environ 1 (5):240-246.Galloway JN, et al. (2004) Nitrogen cycles: Past, present, and fu ture. Biogeochemistry 70(2):153-226.Vitousek PM, Menge DNL, Reed SC, Cleveland CC (2013) Biological nitrogen fixation: Rates, patterns and ecological controls in terrestria l ecosystems. Philos Trans R Soc Lond 8 Biol Sci 368(1621):20130119.Reed SC, Cleveland CC, Townsend AR (2011) Functional ecology o f free-living n itro ­gen fixation : A contemporary perspective. A nnu Rev Ecoi Evol Syst 42:489-512. Cleveland CC, et al. (1999) Global patterns o f terrestria l biological nitrogen (N2) fix ­ation in natural ecosystems. Global Biogeochem Cycles 13(2):623-645.Dentener F, et al. (2006) Nitrogen and sulfur deposition on regional and global scales: A m ultim odal evaluation. Global Biogeochem Cycles 20:GB4003.Liu X, e t al. (2013) Enhanced nitrogen deposition over China. Nature 494(7438): 459-462.Howarth RW, et al. (1996) Regional nitrogen budgets and riverine N & P fluxes fo r the drainages to the North A tlan tic Ocean: Natural and human influences. Bio­geochemistry 35(1):75-139.Hietz P, e t al. (2011) Long-term change in the nitrogen cycle o f tropical forests. Sci­ence 334(6056):664-666.Jordan C, et al. (1982) The nitrogen cycle in a 'Terra Firme' rainforest on oxisol in the Amazon te rrito ry o f Venezuela. Plant Soil 67:325-332.Wang Y-P, Houlton BZ (2009) Nitrogen constraints on terrestria l carbon uptake: Im­plications fo r the global carbon-climate feedback. Geophys Res Le tt 36:L24403. te r Steege H, e t al. (2006) Continental-scale patterns o f canopy tree composition and function across Amazonia. Nature 443(7110):444-447.Crews TE (1999) The presence o f nitrogen fix ing legumes in terrestria l communities: Evolutionary vs ecological considerations. 6/ogeoc/iem/stry 46(1-3):233-246. Sylvester-Bradley R, De Oliveira LA, De Podesta Filho JA, St. John TV (1980) Nodula­tion o f legumes, nitrogenase activity o f roots and occurrence o f nitrogen-fixing Azospirillum spp. in representative soils o f central Amazonia. Agro-ecosyst 6(3): 249-266.Sprent Jl (2005) Nitrogen Fixation in Agriculture, Forestry, Ecology, and the Envi­ronment, eds W erner D, Newton WE (Springer, Berlin), pp 113-141.Barron AR, Purves DW, Hedin LG (2011) Facultative nitrogen fixation by canopy le­gumes in a lowland tropical forest. Oecologia 165(2):511-520.Cleveland CC, et al. (2010) Using indirect methods to constrain symbiotic nitrogen fixation rates: A case study from an Amazonian rain forest. Biogeochemistry 990-3): 1-13.Gehring C, Viek PLG, de Souza LAG, Denich M (2005) Biological nitrogen fixation in secondary regrowth and m ature rainforest o f central Amazonia. A gric Ecosyst Envi­ron 111:237-252.Hedin LG, Brookshire ENJ, Menge DNL, Barron AR (2009) The nitrogen paradox in tropical forest ecosystems. A nnu Rev Ecoi Evol Syst 40:613-635.Batterman SA, et al. (2013) Key role o f symbiotic din itrogen fixation in tropical forest secondary succession. Nature 502(7470):224-227.Davidson EA, et al. (2007) Recuperation o f nitrogen cycling in Amazonian forests fo llow ing agricultural abandonment. Nature 447(7147):995-998.Thompson SK (2002) Sampling (Wiley, New York).Hartshorn GS (1980) Neotropical forest dynamics. Biotropica 12(2):23-30.Boring LR, Swank WT, W aide JB, Henderson GS (1988) Sources, fates, and impacts of nitrogen inputs to terrestria l ecosystems: Review and synthesis. Biogeochemistry 6{2): 119-159.

S. A lva re z -C la re fo r assistance in th e f ie ld ; S. Castle, E. Prag, N. B oote , and A . G in te r fo r assistance in th e la b o ra to ry ; R. Cole fo r sharing soil chem ica l d a ta ; J. S p ren t and th re e ano n ym o us rev iew ers fo r com m en ts on o u r m a nu scrip t; and C. Nelson and D. A ff le c k fo r assistance w ith th e s tra tif ie d a d a p tiv e c lu s te r sam p ling des ign . G rants fro m th e A n d re w W . M e llo n F o u n d a tio n , th e N a tio n a l Science F o u n d a tio n (G ran t DEB- 0919080), th e Ecosystems M ission A re a o f th e US G eo log ica l Survey, and th e B lue M o on Fund su p p o rte d th is w o rk .

31. Silvester WB, Sollins P, Verhoeven T, Cline SP (1982) Nitrogen fixation and acetylene reduction in decaying conifer boles: Effects o f incubation time, aeration, and mois­tu re content. Can J For Res 12(3):646-652.

32. Furnkranz M, e t al. (2008) Nitrogen fixation by phyllosphere bacteria associated w ith higher plants and the ir colonizing epiphytes o f a tropical lowland rainforest o f Costa Rica. /SM f 7 2(5):561-570.

33. Yamada A, e t al. (2006) Nitrogen fixation by term ites in tropical forests, Thailand. Ecosystems 9(1):75-83.

34. Losos EC, Leigh EG (2004) Tropical Forest Diversity and Dynamism (University o f Chicago Press, Chicago).

35. Nardoto GB, et al. (2014) Basin-wide variations in Amazon forest nitrogen-cycling characteristics as inferred from plant and soil ^^N:^"‘n measurements. Plant Ecoi Divers 7(1-2):173-187.

36. Pons TL, Perreijn K, van Kessel C, Werger MJA (2007) Symbiotic nitrogen fixation in a tropical rainforest: ^^N natural abundance measurements supported by experi­m ental isotopic enrichment. New Phytol 173(1):154-167.

37. Menge DNL, Hedin LG (2009) Nitrogen fixation in d iffe ren t biogeochemical niches along a 120 000-year chronosequence in New Zealand. Ecology 90(8):2190-2201.

38. Jenny H (1950) Causes o f the high nitrogen and organic m atter content o f certain tropical forest soils. Soil Sci 69(1):63-70.

39. Vitousek GM, Sanford RL, Jr. (1986) Nutrient cycling in moist tropical forest. A nnu Rev Ecoi Syst 17(1):137-167.

40. Hedin LG, Vitousek PM, Matson PA (2003) Nutrient losses over fou r m illion years o f tropical forest development, fco /ogy 84(9):2231-2255.

41. Houlton BZ, Sigman DM, Hedin LG (2006) Isotopic evidence fo r large gaseous n itro ­gen losses from tropical rainforests. Proc Natl Acad Sci USA 103(23):8745-8750.

42. Brookshire ENJ, Gerber S, Menge DNL, Hedin LG (2012) Large losses o f inorganic ni­trogen from tropical rainforests suggest a lack o f nitrogen lim itation. Ecoi Le tt 15(1): 9-16.

43. Vitousek PM, Field CB (1999) Ecosystem constraints to symbiotic nitrogen fixers: A simple model and its implications. 6/ogeoc/iem/stry 46(1-3):179-202.

44. Vitousek PM, e t al. (2002) Towards an ecological understanding o f biological nitrogen fixation . Biogeochemistry 57-58(1):1-45.

45. Menge DNL, Levin SA Hedin LG (2008) Evolutionary tradeoffs can select against ni­trogen fixation and thereby m aintain nitrogen lim itation. Proc N a tl Acad Sci USA 105(5):1573-1578.

46. Boy J, Rollenbeck R, Valarezo C, W ilcke W (2008) Amazonian biomass burning-derived acid and nutrien t deposition in the north Andean montane forest o f Ecuador. Global Biogeochem Cycles 22:GB4011.

47. Macedo MN, et al. (2012) Decoupling o f deforestation and soy production in the southern Amazon during the late 2000s. Proc Natl Acad Sci USA 109(4):1341-1346.

48. M artine lli LA (2011) Block changes to Brazil's forest code. Nature 474(7353):579.49. European Commission-Joint Research Center (2014) EDGAR 4.2: Emissions Database

fo r Global Atmospheric Research. Available a t http7/edgar.jrc.ec.europa.eu/index. php. Accessed Gctober 28, 2013.

50. Potter P, Ramankutty N, Bennet EM, Donner SD (2010) Characterizing the spatial patterns o f global fe rtilize r application and manure production. Earth Interact 14(2): 1- 2 2 .

51. M onfreda C, Ramankutty N, Foley JA (2008) Farming the planet: 2. Geographic dis­tribu tion o f crop areas, yields, physiological types, and net primary production in the year 2000. Global Biogeochem Cycles 22:GB^022.

52. Ramankutty N, Foley JA (1999) Estimating historical changes in global land cover: Croplands from 1700 to 1992. Global Biogeochem Cycles 13:997-1027.

53. Townsend AR, Howarth RW (2010) Fixing the global nitrogen problem. Sci Am 302(2): 64-71.

54. Payne RJ, Dise NB, Stevens CJ, Gowing DJ; BEGIN Partners (2013) Impact o f nitrogen deposition a t the species level. Proc Natl Acad Sci USA 110(3):984-987.

55. Dirzo R, Raven PH (2003) Global state o f biodiversity and loss. A nnu Rev Environ Resour 28(1): 137-167.

56. Morales-Salazar M, et al. (2012) Diversidad y estructura horizonta l en los bosques tropicales del Corredor Biologico de Gsa, Costa Rica. Revista Forestal Mesoamericana Kuru 9(23): 19-28.

57. R Core Team (2012) R: A Language and Environment fo r Statistical Computing (R Foundation fo r Statistical Computing, Vienna).

58. Holland EA, Dentener FJ, Braswell BH, Sulzman JM (1999) Contemporary and pre­industrial global reactive nitrogen budgets. 6/ogeoc/iem/stry 46(1-3):7-43.

6 o f 6 I www.pnas.org/cgi/doi/10.1073/pnas.1320646111 Sullivan et al.

Supporting InformationSullivan et al. 10.1073/pnas.1320646111SI Methods

^ (1) Plot Description. M ean annual temperature at the P iro B i-l i J i ological Station is 26 °C, and mean annual prec ip ita tion is 3,450

mm (1). The area experiences a climate tha t can be divided in to three seasons: a short dry season between late December and A p ril, an early wet season between A p r il and September, and a late wet season between September and December. The 12 sites used in this study consisted o f fou r age classes: 5-15 y, 15- 30 y, 30-50 y, and prim ary forest. Each age class bad three replicates (n = 3). Slopes were <10% at all 12 sites. Each p lo t was subject to a census fo r tree size and species iden tification (or genus i f species iden tification was impossible) in 2010-2011.

(2) Estimating Nodule Biomass Using Stratified Adaptive Cluster Sampling. The stratified adaptive cluster sampling (SACS) de­sign generates an unbiased estimate o f the mean nodule biomass per site (2). Beginning w ith an in itia l sample, i t adaptively adds more samples from the adjacent neighborhood o f the original sample i f nodules are present, and the neighborhood grows un til no additional samples have nodules present. To use the SACS approach, we firs t iden tified a 10 x 70-m strip w ith in each plot, which we partitioned in to seven 10 x 10-m strata that consist o f Nc = 9,604 possible cores each (Fig. S4). In each stratum, we sampled 10 cores (mc), one at each 1 m along a transect on the centra l axis o f each stratum tha t is perpendicu lar to the long axis o f the to ta l area T. I f nodules were present in any o f 10 sampled cores along the centra l transect, we expanded our search at each p o in t tha t nodules were in it ia lly present u n til no nodules were present on all sides o f the netw ork o r we were physically obstructed from digging the hole (e.g., by a tree) (Fig. S4). As described by Thom pson (2), we defined the mass o f nodules in a core (w „) , weighted by the num ber o f networks in the sample, as

SM „

N „ ’

where M „ is the mass o f the nodules in the network and N „ is the number o f holes containing nodules in the network. Because our networks never crossed strata, we then estimated mean nodule biomass (fi) (2) using the equation

f i = Tj; V —T ^ ric ^

C=1 1 = 1

W e calculated the average mass o f nodules per area (Ma ; in grams meter“ ^) by multiplying f i by the area sampled w ith each core.

(3) Acetylene Reduction Assay w ith Nodules. In a Panamanian tropical forest, i t was shown that nodule biomass does no t change substantially seasonally (including in a dry season) (3); thus, we measured nodule biomass one tim e (during the early wet season) but assessed variab ility o f biological nitrogen fixation (BN F) rates (nodu le activ ity ) fo u r tim es th roughou t the year. W hen we measured nodule biomass using the SACS approach, we used the acetylene reduction assay (A R A ) on nodules excavated from the SACS sampling; in those samplings and subsequent samplings, we used between 6 and 10 samples o f A R A per site, and each A R A sample had one to five nodules (depending on nodule size) in each sample. W e perform ed the A R A using the fo llow ing p ro ­tocol. A fte r excavating nodules and gently separating them from

attached roots, we incubated them in a 50-mL clear acrylic tube fo r 1 h w ith a 10% acetylene atmosphere. A fte r incubation, 14-mL headspace samples were removed from tubes w ith a sampling syringe, placed in 10-mL Vacutainers (Becton-Dickinson), and returned to the U n ited States fo r analysis by G C using a Shimadzu GC-2014 equipped w ith a flam e io n iza tion detector. W e ac­counted fo r the ethylene produced from soil and litte r w ithou t acetylene exposure, the ethylene produced from tubes and Vacu­tainers, and the concentration o f ethylene w ith in our acetylene as well as the ethylene lost because o f photodegredation during transport. To convert acetylene reduction rates into BN F rates, we measured the uptake o f ^®N-Iabeled N 2 in nodules to generate a site-specific ethyIene:N conversion ratio o f 2.8:1 fo r symbiotic BNF. Although the A R A method requires the excision o f nodu­lated roots from the host plant, we are unable to assess the artifact o f this sampling method on actual rates o f B N F but instead, sought to minimize these artifacts by using a short (1 h) incubation— an approach that is frequently used (3, 4).

(4) ARA w ith Soil and Litter. W e measured free-liv ing soil B N F in in tact cores taken from the top 2 cm o f the m ineral soil but converted these rates to 10-cm estimates, because B N F is re la­tively constant in the top 10 cm o f a soil p ro file (5). Lea f litte r was collected by hand and placed in acrylic tubes fo r incubation, and area-based estimates o f leaf litte r B N F were estimated by co l­lecting all o f the litte r w ith in several 0.09-m^ quadrants at each site and each sam pling tim e, a llow ing fo r mass-to-area con­versions. B o th soil and li t te r samples were exposed to 10% acetylene atmosphere fo r 18 h. A fte r incubation, 14-mL head­space samples were removed from tubes w ith a sampling syringe, placed in 10-mL Vacutainers (Becton-Dickinson), and returned to the U n ited States fo r analysis by GC using a Shimadzu GC- 2014 equipped w ith a flame ionization detector. W e accounted fo r the ethylene produced from soil and litte r w ithou t acetylene exposure, the ethylene produced from tubes and Vacutainers, and the concentration o f ethylene w ith in our acetylene as w e ll as the ethylene lost because o f photodegredation during transport. To convert acetylene reduction rates in to B N F rates, we applied the theoretical conversion ra tio o f 3:1 (moles ethylene:moles N ) to free-liv ing B N F (5).

(5) Estimating BNF in Primary Forests w ith Disturbance. To estimate B N F in the low- and high-turnover scenarios, we assumed that mean to ta l B N F from each age class was constant and m ultip lied to ta l B N F by the number o f years in each category (and allowed the 5- to 15-y category to apply to the f irs t 5 y o f forest de­ve lopm ent). P rim ary fo rest rates were app lied to forest ages greater than 50 y.

(6) Constraining Global Models o f BNF to Low Tropical BNF. Toconstrain two existing spatially exp lic it global N 2 fixation models (6, 7), we used the fo llow ing equation:

mBNFadj = ^i=0

eBNF

niB N FXm BNFi

where niB N F j represents spatially explic it modeled N 2 fixation rates, niBNF represents the mean o f modeled trop ica l N 2 fixa­tion, eBNF represents our gap-filled estimate o f mean tropical N 2 fixation (5.7 kg N ha“ ̂ y“ ^), and mBNFadj represents to ta l tropical N 2 fixation constrained by our empirically derived estimate. M ode led trop ica l N 2 fixa tion was derived from ( i) the centra l

Sullivan e t al. w w w .pnas.org /cg i/content/short/1320646111 1 o f 4

N 2 fixation empirical model based on net primary productivity (6) applied to moderate resolution im aging spectroradiometer net prim ary productivity data (8) (m BNF equaled 25.1 kg N ha“ ̂y “ ^) and ( i i ) the Carnegie A mes Stan fo rd A pproach biogeochem i­cal process m odel (7) (m B N F equaled 32.1 kg N ha“ ̂ Y” ^)- T rop ica l forest pixels (n,rop) were defined using the U n ivers ity o f M ary land evergreen b road lea f forest land cover classifica-

g tio n (9).

(7) Quantifying the Extent o f Anthropogenic impacts on the Tropicai N Cycie. Historica l tropical forest extent (i.e., trop ica l forest extent in the absence o f human activity) was calculated by combining the tropical evergreen forest and tropical deciduous forest land cover classifications w ith in the global po tentia l vegetation data product from ref. 10. Preindustrial N deposition was estimated from ref. 11, whereas natural rates o f N fixation were derived from ref. 7. Rescaled to ta l natural tropical N inputs were estimated by ap­plying our updated estimate o f mean tropical forest N fixation across the to ta l historical tropical forest area. Tota l anthropo­

genic N inputs were defined to include postindustrial N deposition (calculated as the difference between total N deposition and pre­industrial N deposition) (12), agricultural N fertilization (13), and agricultural N fixation (14) inputs. W e also highlight the scale o f N inputs from manure additions (13), although they are not truly additional anthropogenic inputs, because most manure is locally and naturally derived and thus, represents within-biome N cycling. Postindustrial N deposition was estimated for total historical trop­ical forest areas using the E D G A R 4.2 emission database (15) generated w ith a global 3D chemistry transport model fo r the year 2008, the most recent year available. Total N fertilization and N inputs from manure application occurring w ith in historically trop i­cal forest areas were estimated from a fertilization and manure application database generated from year 2000 agricultural statistics data (13). Total agricultural N fixation was estimated by combining year 2000 crop area and jdeld data fo r all Food and Agricultural Organization-recognized crop types (14) w ith known crop-specific rates o f symbiotic and free-living N fixation (16, 17).

1. Keller AB, Reed SC, Townsend AR, Cleveland CC (2013) Effects o f canopy tree species on belowqround bioqeochemistry in a lowland w e t trop ical forest. So/7 Biol Biochem 58(1):61-69.

2. Thompson SK (2002) Sampling (Wiley, New York).3. Barron AR, Purves DW, Hedin LO (2011) Facultative nitrogen fixation by canopy

legumes in a lowland tropical forest. Oecologia 165(2):511-520.4. Batterman SA, e t al. (2013) Key role o f symbiotic din itrogen fixation in tropical forest

secondary succession. Nature 502(7470):224-227.5. Reed SC, Cleveland CC, Townsend AR (2008) Tree species control rates o f free-living

nitrogen fixation in a tropical rain forest, fco /ogy 89(10):2924-2934.6. Cleveland CC, et al. (1999) Global patterns o f terrestrial biological n itrogen (N2)

fixation in natural ecosystems. Global Biogeochem Cycles 13(2):623-645.7. Wang Y-P, Houlton BZ (2009) Nitrogen constraints on terrestrial carbon uptake:

Implications fo r the global carbon-climate feedback. Geophys Res Le tt 36(24):L24403.8. Zhao M, Running SW (2010) Drought-induced reduction in global terrestria l net

primary production from 2000 through 2009. Science 329(5994):940-943.9. FriedI MA, e t al. (2010) MODIS Collection 5 global land cover: A lgorithm refinements

and characterization o f new datasets. Remote Sens Environ 114(1):168-182.

10. Ramankutty N, Foley JA (1999) Estimating historical changes in global land cover: Croplands from 1700 to 1992. Global Biogeochem Cycles 13(4):997-1027.

11. Holland EA, Dentener FJ, Braswell BH, Sulzman JM (1999) Contemporary and pre­industrial global reactive nitrogen budgets. S/ogeoc/iem/stry 46(1):7-43.

12. Dentener F, e ta l. (2006) Nitrogen and sulfur deposition on regional and global scales: A m ultim odal evaluation. Global Biogeochem Cyc/es 20(4):GB4003.

13. Potter P, Ramankutty N, Bennet EM, Donner SD (2010) Characterizing the spatial patterns of global fertilizer application and manure production. Earth Interact 14(2):1-22.

14. M onfreda C, Ramankutty N, Foley JA (2008) Farming the planet: 2. Geographic distribu tion o f crop areas, yields, physiological types, and net primary production in the year 2000. Global Biogeochem Cycles 22(1):GB1022.

15. European Commission-Joint Research Center (2014) EDGAR 4.2: Emissions Database fo r Global Atmospheric Research. Available a t httpy/edgar.jrc.ec.europa.eu/index. php. Accessed October 28, 2013.

16. Smil V (1999) Nitrogen in crop production: An account o f global flows. Global Biogeochem Cycles 13(2):647-662.

17. Herridge DF, Peoples MB, Boddey RM (2008) Global inputs o f biological nitrogen fixation in agricultural systems. P/ant So/7 311 (1):1-18.

12

n Soiln Litter

10 ■ Sym biotic

>

'(0

O)

m

Primary Secondary

Forest type

Fig. S I. To ta l annual BNF inputs fro m soil, litte r, and sym biotic pools in p rim a ry and secondary fo res t on th e Osa Peninsula, Costa Rica. In th is study, w e define free -liv ing BNF as th e sum o f soil and lit te r BNF. The BNF ra te in secondary fo res t was calculated as th e mean o f sym biotic and free -liv ing BNF rates in each o f th e th re e secondary fo res t ages th a t w e measured: 5-15, 15-30, and 30-50 y. Error bars dep ic t ±1 SE o f mean to ta l N2 f ix a tio n in p rim a ry {n = 3) and secondary {n = 9) forest.

Sullivan e t al. vvvvw .pnas.org/cg i/content/short/1320646111 2 o f 4

Legume basal area (m^ ha'^) Legume basal area (%) Legume abundance (# ha'^) Legume abundance (%)

25 -

20 -

15 -

10 -

5 -

CTFS Osa

1001000

20 - 800

15 - 600

10 - 40400

5 - 200

CTFS Osa CTFS OsaCTFS Osa

Fig. S2. D is trib u tion o f legum e basal area and legum e abundance (bo th in abso lu te values and percent) in 5 C enter fo r Tropical Forest Science (CTFS) plots located around th e w o r ld (Table 1) and 12 secondary and p rim a ry fo res t p lo ts on th e Osa Peninsula, Costa Rica.

3

2.5

2= 0.91

1.5

1

0.5 ■ 15-30 y♦ 30-50 y* Primaty

00 5 10 15 20 25

D)

O!oE

7

6

5

= 0.77

3

2

1

00 0.5 1 1.5 2 2.5 3

CQoo!5E>0)

7

6

5

= 0.704

3

2

1

00 5 10 15 20 25

Putative N fixing tree basal area (m ha' Nodule biomass (g m' Putative N fixing tree basal area (m ha'

Fig. S3. Relationships be tw een (A) nodu le biomass and th e basal area o f legum e species p u ta tive ly capable o f fo rm in g nodules and fix in g N, (6) sym bio tic BNF (on a spatia l basis) and nodu le biomass, and (C) sym bio tic BNF rates (on a spatia l basis) and th e basal area o f legum e species p u ta tive ly capable o f fo rm ing nodules and fix in g N. A ll re la tionships are s ign ifican t (P < 0.05), and th e streng th o f each co rre la tion is described by values w ith in each panel.

Sullivan e t al. w w w .pnas.org /cg i/content/short/1320646111 3 o f 4

70 m

Eo

i I

Stratum

V

10 m

^ Primary core, nodules present

^ Secondary core, nodules present

Q No nodules present, 1° o r 2° core

9 m

8 m

7 m

6 m

5 m

4 m

3 m

2 m

1 m

Fig. S4. The SACS experim enta l design. (A) First, w e id e n tifie d a 10 x 70-m sam ple area. (6 and C) Second, w e sampled 10 cores a long th e centra l transect o f each stra tum . If nodules w ere present in 1 o r m ore o f 10 prim ary transect cores, (C) w e expanded th e sam pling design to include all samples in th e ne igh ­borhood o f th e o rig ina l core th a t had nodules present and stopped expand ing th e n e ighbo rhood w hen no m ore ad jacent samples had nodules present o r (D) w e w ere physically prevented fro m sam pling. The clusters o f nodules presented in C and D represent actual patte rns o f nodules th a t w e fo u n d on th e landscape.

Table S1. Soil to ta l and inorganic N, soil to ta l and extractab le P, and soil pH in th e surface soil (0 -1 0 cm) am ong th e sites occupying th e fo ur fo rest age classes o f th e Piro chronosequence

Forest a ge c lass ifica tio n

Soil p a ra m e te rs 5 -1 5 y 1 5 -3 0 y 3 0 -1 5 y P rim ary

T o ta l N (% )

In o rg a n ic N (m g k g “ ') T o ta l P (m g k g “ ') E x tra c ta b le P (m g k g “ ')

pH

0.37

2.57 (4.31) 499

0 .94 (0.84)

5.4

0 .36 (0.05)

3.31 (6.40) 522 (275.7) 3.90 (5.22)

5.2 (0.42)

0 .44 (0.12)

3.69 (3.82) 489 (277.8) 2 .54 (3.08)

5.4 (0.42)

0 .39 (0.04)

4 .13 (2.72) 665 (110.5) 1.56 (2.61)

5.4 (0.21)

Inorganic N was calculated as th e sum o f am m on ium and n itra te ; extractab le P was measured using d ilu te am m on ium flu o rid e extrac tion . pH was measured in 1:1 so iliw a te r slurry. Inorganic N and ex trac tab le P concentra tions represent th e average o f th e fo u r seasonal sam pling events. Values are mean ± 1 SD (n = 3 per age class).

Sullivan e t al. w w w .pnas.o rg /cg i/con ten t/short/13 20646111 4 o f 4