# Nitrogen-fixing trees inhibit growth of regenerating Costa Rican rainforests

See allHide authors and affiliations

Edited by Christopher B. Field, Stanford University, Stanford, CA, and approved June 30, 2017 (received for review April 28, 2017)

## Significance

Regrowing tropical forests are critical for global biodiversity conservation and carbon capture. Nitrogen availability often controls how fast these forests can regrow. Because nitrogen-fixing plants are the primary source of new nitrogen into these forests, one might expect that more nitrogen fixers lead to faster forest regrowth. However, here we show that nitrogen fixers actually slow forest regrowth. Their competitive influence on neighboring trees outweighs any growth enhancement from their nitrogen inputs at this site. Our results call for a more critical evaluation of how nitrogen fixers influence the surrounding forest, especially given the large uncertainty in global climate projections that hinges on the role of nitrogen fixers during tropical forest regeneration.

## Abstract

More than half of the world’s tropical forests are currently recovering from human land use, and this regenerating biomass now represents the largest carbon (C)-capturing potential on Earth. How quickly these forests regenerate is now a central concern for both conservation and global climate-modeling efforts. Symbiotic nitrogen-fixing trees are thought to provide much of the nitrogen (N) required to fuel tropical secondary regrowth and therefore to drive the rate of forest regeneration, yet we have a poor understanding of how these N fixers influence the trees around them. Do they promote forest growth, as expected if the new N they fix facilitates neighboring trees? Or do they suppress growth, as expected if competitive inhibition of their neighbors is strong? Using 17 consecutive years of data from tropical rainforest plots in Costa Rica that range from 10 y since abandonment to old-growth forest, we assessed how N fixers influenced the growth of forest stands and the demographic rates of neighboring trees. Surprisingly, we found no evidence that N fixers facilitate biomass regeneration in these forests. At the hectare scale, plots with more N-fixing trees grew slower. At the individual scale, N fixers inhibited their neighbors even more strongly than did nonfixing trees. These results provide strong evidence that N-fixing trees do not always serve the facilitative role to neighboring trees during tropical forest regeneration that is expected given their N inputs into these systems.

The past two decades have seen a dramatic increase in the appreciation for the role that tropical secondary forests play in local economies (1), species conservation (2), and climate change mitigation (3). As this interest has spurred new research into the dynamics of tropical forest regeneration, we are beginning to recognize the wide range of regeneration rates and trajectories that tropical secondary forests can exhibit (4). In addition to climatic drivers (4), soil N availability can regulate tropical forest regrowth (5, 6) and dictate how these forests respond to changing climatic conditions (7⇓⇓–10). The largest potential source of new N into tropical secondary forests is from symbiotic N-fixing plants (11, 12), which form specialized root nodules to house symbiotic bacteria that convert atmospheric N_{2} gas into plant-available forms of N. These “N fixers” can fix up to 150 kg N·ha^{−1}·y^{−1} (12), which becomes available to the surrounding ecosystem as N-fixer tissues return to the soil and the N in those tissues is mineralized. These N inputs from N fixers are thought to meet most of the external N demands of rapidly regenerating forests (13). However, the effect that these N fixers have on tropical forest regrowth depends on both their N inputs and how they influence the demographic rates of the neighboring trees around them.

The unique potential for N fixers to bring newly fixed N into the surrounding ecosystem (14) means that they might fertilize neighboring trees, but N fixers also compete with their neighbors for light and other resources. N fixers have, on average, higher tissue N content in their foliage than nonfixers (15⇓⇓–18). Litterfall and decomposition of this N-rich leaf tissue is the primary means by which N fixers fertilize the surrounding ecosystem. However, high N content can also fuel rapid growth of the N fixer itself (11, 13, 19), potentially increasing the N fixer’s competitive influence on its surrounding neighbors. If the fertilization effect is strong, N fixers might facilitate neighbors or at least inhibit neighbors less than nonfixers (hereafter, “weak inhibition”; Fig. 1 *A* and *B*). Alternatively, if their competitive effect outweighs their fertilization effect, N fixers might inhibit neighbors more than nonfixers do (20) (“strong inhibition”; Fig. 1 *C* and *D*).

How N fixers affect neighboring trees is especially important in Neotropical secondary forests, which have great capacity for carbon storage (3, 4), are often thought to be N limited (5, 6), and have high relative abundances of N fixers. In Neotropical forests, N fixers typically comprise ∼10% of all trees (compared with <1% at higher latitudes in North America) (21⇓–23), and commonly make up 30–55% of the forest basal area at some sites (19, 24) (Fig. 2 and Table 1), making their impact on neighboring trees critical to the growth of these forests.

Do N fixers promote or inhibit growth in regenerating tropical forests? We addressed this question at multiple spatial scales using 10–17 y of census data from eight 1-ha moist tropical rainforest plots in Northeastern Costa Rica—six regenerating forests ranging in stand age from 10 to 42 y old and two old-growth forests (19, 25, 26) (Table 1). First, we asked whether the abundance of N fixers affects forest growth at the 1-ha forest plot and 10 × 10 m subplot levels. Next, we analyzed the effects of N fixers on their neighbors at the individual scale—the scale where competition and facilitation interactions likely occur. Specifically, we asked how the makeup of a tree’s neighborhood—the percent of its crowding that comes from neighboring N fixers rather than from nonfixers—affects its growth, recruitment, and survival.

## Results

If N fixers promote forest growth at the 1-ha plot level, we would expect a positive relationship between the abundance of N fixers and the annual increase in tree basal area of a plot. However, we found that plots with more N fixers had lower overall growth (*P* < 0.0001; Fig. 2*A*) and lower nonfixer growth (*P* < 0.0001; Fig. 2*C*), even after accounting for variation in total plot basal area. A change in N-fixer prevalence from 10 to 35% of the plot’s basal area corresponded to a reduction of total annual growth from 2.2 to 0.6% and reduced nonfixer growth from 2.0 to 0.06%. To overcome potential confounding correlations between plot age, N-fixer abundance, and growth, we also assessed growth at the 10 × 10 m subplot scale within each plot. We found a nonsignificant negative trend between total growth and N-fixer prevalence (*P* = 0.12; Fig. 2*B*) and a significant negative correlation between nonfixer growth and N-fixer prevalence (*P* < 0.0001; Fig. 2*D*). These results suggest that N fixers are inhibiting, not facilitating, overall growth and growth of local nonfixers in our study region.

To gain a more mechanistic understanding of how individual N fixers drive plot-level growth patterns, we assessed how N fixers affect individual neighboring trees. This requires spatially explicit data on the demographic rates of individual trees over relatively long timescales. We estimated the degree of neighbor crowding that each individual experienced using a Neighborhood Crowding Index (NCI). A larger NCI (more crowding) could come from any combination of more neighbors, bigger neighbors, and closer neighbors. To estimate crowding from N fixers, we calculated the percent of each tree’s NCI coming from neighboring N fixers—a continuous scale from 0 (all of an individual’s neighbors are nonfixers) to 100% (all of an individual’s neighbors are N fixers). We then used hierarchical Bayesian models to examine how crowding from N fixers affected the growth, survival, and recruitment of each tree (both N fixers and nonfixers), after accounting for overall crowding and tree size (27). Based on established changes in tree demographic rates at this site (19), we ran our models independently for forest stands ≤25 y old (“young forests”) and >25 y old (“old forests”).

If N fixers facilitate or weakly inhibit their neighbors, we would expect tree demographic rates to increase with crowding from N fixers (Fig. 1 *A* and *B*). Our results showed the opposite trend. N fixers strongly inhibited their neighbors—exhibiting greater negative effects on their neighbors than nonfixers did (Figs. 3 and 4 and *SI Appendix*, Figs. S2–S4, resembling Fig. 1 *C* and *D*). In young forests, crowding from N fixers strongly inhibited all demographic rates of neighboring nonfixers and strongly inhibited the growth rates of neighboring N fixers (Fig. 3). In old forests, N fixers strongly inhibited both the growth and survival of neighboring N fixers (Fig. 4). This N-fixer inhibition effect was stronger on the growth of neighboring N fixers than neighboring nonfixers regardless of forest age (Figs. 3*B* and 4*B*) and on the survival of neighboring N fixers in old forests (Fig. 4*F*). However, N fixers more strongly inhibited the recruitment of neighboring nonfixers than neighboring N fixers in young forests (Fig. 3*D*).

## Discussion

Together, our individual-scale results and our 1-ha plot-scale findings show that N fixers in these forests inhibit their neighbors more than do nonfixers. It is important to note that the strong inhibition of N fixers we report is independent of the overall level of crowding and tree size. For example, an average-sized nonfixing tree (diameter at breast height [DBH] ∼ 13 cm) with an average amount of crowding (NCI ∼ 1,900) in a young forest stand would have a 43% lower expected growth rate if its neighbors were all N fixers than if its neighbors were all nonfixers. If this “average tree” is an N fixer itself, a neighborhood with all N fixers reduces its expected growth rate by over 60% compared with a neighborhood with all nonfixers.

The negative influence that N fixers have on their neighboring trees in our study region is likely due to two factors. First, high growth and survival rates of N fixers in these plots (19), and the high nutrient demand of N fixers (15⇓–17, 28), mean that N fixers likely cast more shade and take up more soil nutrients and water than do nonfixers. Second, the presumed facilitation of N fixation might not occur because nonfixers are not limited by N availability. It is also possible that facilitation might not occur because N fixers are not fixing much N, but we find this possibility less likely. The lower cost of acquiring N from the soil than from fixation (29) suggests that N fixers down-regulate fixation when soil N is available, and a number of recent studies are consistent with this idea (13, 24, 30). However, our observations from preliminary soil cores indicate that the N fixers in our plots commonly have active nodules, and theory suggests that even small amounts of N fixation can enhance the growth of neighboring nonfixers if they are N limited (31). This suggests that the N fixers in our study are not merely operating ecologically as nonfixers—they are bringing new N into these ecosystems—but rather that nonfixers in our study do not respond to this greater N availability.

Our results come from a single region with high annual rainfall, so they may not be ubiquitous across tropical forests. However, because ours is one of the few studies to assess the effect of N fixers on the growth of neighboring trees, statistical sampling suggests that our results are likely not rare. Beyond this sampling argument, several lines of evidence indicate that our findings might be common in moist tropical forests. First, the climate and soil type of our study area are commonly found in other moist tropical forest sites (32). Second, although we do not have rigorous soil N data from our plots, litterfall N and N transformation rates in our broad study area (La Selva Biological Station) are similar to those in many rainforests in the African, Asian, American, and Australian tropics (33, 34), indicating that N cycling at our site is representative of many moist tropical forests worldwide. Finally, the most likely mechanism for the strong competitive effects of N fixers that we found is high N-fixer growth rates, which are also common at other moist tropical forest sites (11, 13, 35).

Despite the similarities between our study site and many moist tropical forests, the heterogeneity in this biome (36) means that differences in local features, such as soil nutrient availability, may drive N fixers to have different effects on their neighbors in some sites. Although no other studies have assessed the effects of N fixers on the demographic rates of their neighbors in regenerating tropical forests, two previous studies have investigated how N fixers influence ecosystem-scale biomass accumulation in other regenerating moist tropical forests in Brazil and Panama (11, 13). Contrary to our 1-ha plot-scale findings, both of those studies showed that N fixers were correlated with total biomass accumulation, primarily due to N fixers’ own high growth rates. Although N fixers also grow faster than nonfixers at our study sites (19), we found that N-fixing trees inhibit biomass accumulation at the plot scale (Fig. 2*A*) because their inhibition of neighbors outweighs their own rapid growth.

Why might N-fixers inhibit their neighbors more in our sites than in sites in Brazil (11) and Panama (13)? The three studies differ in the primary N-fixing taxa (*Pentaclethra macroloba* here vs. *Inga* spp. in ref. 13 and a diverse group of legumes in ref. 11) and the age range of succession studied (our youngest sites are 10 y vs. 5 y in ref. 13 and 2 y in ref. 11). One of our study plots, Tirimbina (TIR), was dominated by *Inga* rather than *P. macroloba* N fixers yet still demonstrated the same patterns as our other plots—N fixers inhibited neighbors more than did nonfixers (*SI Appendix*, Fig. S7)—suggesting that species identity is not the primary driver of our results. Could the age range explain the discrepancy? In a site near ours, Gilman et al. (37) experimentally planted a diverse set of N fixers in fallow cattle pasture and found no positive influence of these N fixers on the recruitment and growth of neighboring trees during the first 5 y of succession. Their study suggests that N fixers do not facilitate their neighbors at earlier ages in these forests, but given the substantial differences between studies (e.g., naturally regenerating forest in ours vs. experimental planting in theirs), we cannot rule out the possibility that N fixers might have facilitated or weakly inhibited neighbors in earlier years in our plots. More likely, however, other site-specific factors like soil water, nitrogen, phosphorus, and molybdenum availability explain the discrepancy between inhibitory versus facilitative effects of N fixers in our region versus other sites. More broadly, N fixers may play different roles in the dynamics of dry forests (38), which cover 523 million hectares of the world’s tropics (39).

Current modeling efforts allow for high C-capturing potential in tropical secondary forests, but only if *N*-fixing trees relieve N limitation (8, 10). Based on the N inputs of N fixers into these systems, modelers may be tempted to use high N-fixer abundances in forests as an indicator of high growth and C-capturing potential, especially given that advances in remote sensing of tropical N fixers (40) may soon make abundance data much more readily available than direct data on N inputs. Our findings suggest that these modeling results might be misleading for some or even many moist tropical forest sites and that a more critical evaluation of N fixers’ effect on forest growth is needed to accurately predict the regeneration dynamics and future C sink of the world’s secondary tropical forests.

## Conclusions

The influence that N fixers have on the surrounding forest is a balance between their negative competitive interactions with neighboring plants and the facilitative effects of their N inputs into the surrounding ecosystem. Because their ability to bring N into ecosystems is both important and rare within the plant kingdom, it is easy to focus on the potential facilitative effects of N fixers. However, our results demonstrate that the competitive effects of N fixers on their neighbors can be sufficiently strong that N fixers inhibit tropical forest growth. Many Earth System models now incorporate dynamic N cycles (including N fixation) into estimates of future tropical forest carbon capture. As we refine how N fixers are incorporated into these models, our results highlight that we must consider that N fixers may have a negative influence on tropical forests’ ability to capture and store C in some sites. Given the large potential for C capture in regenerating tropical forests, improving our understanding of how N fixers influence this C-capturing potential is vital to our ability to predict future climate scenarios.

## Methods

### Plot and Census Data Description.

We studied eight 1-ha plots in a humid tropical rainforest in the Caribbean lowlands of northeastern Costa Rica, in and around La Selva Biological Station (10.4233°N, 84.022°W). All plots are within 15 km of each other, are similar in elevation (5–220 m), have similar topography and soil type, and experience similar climatic conditions. Similar sets of species dominate all of the plots, but species-relative abundances change with stand age throughout succession in our plots (25). *P. macroloba*, an N fixer, was the most common species across all plots, and *Inga cocleensis* and *Inga pezizifera* were the second and third most abundant N fixers in our dataset (*SI Appendix*, Table S4). The three most common nonfixing species across all of our plots were *Miconia affinis*, *Casearia arborea*, and the palm *Socratea exorrhiza*. Detailed descriptions of plot design and census methods are available elsewhere (19, 26, 41).

Each plot is 50 × 200 m, for a total area of 1 ha. Plots are broken up into 100 square subplots with 10 m sides. Six plots are located in naturally regenerating secondary forests that ranged in age from 10 to 25 y at the time of establishment. The remaining two plots are located in primary, old-growth forest that has remained undisturbed for at least 200 y. Within each plot, all adult trees ≥5 cm DBH were tagged, identified to species, measured for diameter at breast height (1.3 m above ground level), and mapped onto a plot-level *x*,*y* coordinate system using the subplot corners as reference points. Based on species identification, each tree was categorized as a putative N fixer or nonfixer based on methods in ref. 19. This designation followed a three-tiered approach in which species identifications were first checked for reports of nodulation in ref. 42 and the US Forest Service’s GRIN database. After these two N-fixer lists were exhausted, we assigned species as N fixers if they were in a genus with ≥60% congeners that were confirmed N fixers, as fixation is thought to be a trait primarily conserved at the genus level.

### Plot-Scale Effects of N Fixers.

To determine the effect of N fixers on plot- and subplot-level growth, we calculated the percent change in tree basal area (ΔBA) as the change in tree basal area divided by the total tree basal area for each plot or subplot over each census period and multiplied by 100. Changes in nonfixer basal area were scaled by total nonfixer basal area in the same fashion. We used generalized mixed linear regression to model changes in total basal area and nonfixer basal area in response to the proportion of a plot or subplot comprised of N fixers accounting for variation in total basal area and including a random plot effect.

### Individual-Scale Effects of N Fixers.

At the individual scale, we calculated absolute growth rate and survival of each tree over each annual census period. For recruitment, we calculated the frequency of recruits into each 10 × 10 m subplot over each annual census period. NCI was calculated as the squared DBH of each neighbor divided by the squared distance of that neighbor to the focal individual, summed for all neighbors within 10 m. We modeled the response of individual growth and survival, and subplot recruitment, to the NCI and proportion of NCI comprised of N fixers that each individual or subplot experienced using a hierarchical Bayesian approach. Each demographic process (growth, recruitment, and survival) was individually modeled as a response to the effects of NCI, the proportion of NCI made up of N-fixers, DBH (for growth and survival), and the interactions between these variables, as well as the plant’s fixer type (N fixer or nonfixer) and plot. We interpreted 95% credible intervals that did not overlap 0 as statistically significant.

For each tree in each census year, we indicated whether that individual had recruited into the adult dataset or suffered a mortality event. We calculated growth since the previous census for each tree that was present in consecutive census years. Recruitment and mortality were treated as binary variables for each individual in each census year. Growth rate was calculated as *i* in census *c*, *DBH*_{i,c} is the DBH of tree *i* in census *c*, and *t* is the time interval between censuses. Because census intervals varied slightly from year to year, *t* was calculated as the time between measurements of an individual tree in days. The NCI of each tree was calculated to represent the crowding that each individual experienced within a set radius around the individual’s stem. We calculated NCI as *i* is the sum of the squared DBH of all neighbors *j* divided by the squared distance of each neighbor *j* to individual *i* of which there are *n* number of neighbors within a set radius of 10 m. Using a radius of 10 m meant that all trees ≤10 m from the plot edge were excluded as focal trees from neighborhood-scale analyses as NCI could not be accurately calculated along the plot edge.

### Model Description.

We used a set of three Bayesian models to determine how tree size, fixation status, and neighborhood crowding influence the growth and survival of an individual tree, and the recruitment of individuals into 10 × 10 m subplots, in a given census year. Data for DBH and NCI were natural log-transformed, and data for the proportion of NCI comprised of N fixers were arcsine-square root transformed before analysis. Each demographic model estimated the effect of each covariate—ln(DBH), ln(NCI), the proportion of NCI comprised of N fixers (arcsin[^{2}, and ln(NCI) × (arcsin[*z*-transformed by subtracting the mean value for that variable and dividing by the SD. Within the model, each of these parameter estimates could vary based on whether the individual was an N fixer or nonfixer. We modeled each parameter estimate as a normal distribution with uninformative priors (mean and SD of 0 and 100, respectively), and all error terms associated with random effects (plot, species, and individual random effects) were modeled as gamma distributions with uninformative priors (shape and scale of 100 and 100, respectively). All models were run in the stan package of R version 3.2.2 (43).

Standardized absolute growth rate data (change in the DBH of a tree over the census period; ^{−1}) were modeled as a function of each of the six transformed covariates listed above along with random intercept effects for individual tree ID, species ID, and plot. Each parameter estimate could vary based on the N fixation status of the individual tree. The structure of the growth model was as follows:

where *G* and *E*(*G*) are the standardized growth rate of individual *i* of species *j* in plot *p* with fixation status *f* and its expected value, respectively. The intercept (_{1}–β_{6} were modeled for each covariate (described above) and allowed to vary based on the fixation status (*f*) of the individual.

Survival was modeled as a binary variable using a logit link, which could vary as a function of the six covariates used in the growth model with random species-specific and fixation-species intercepts (*i* and plot *p*

where *S* and *s* represent the survival and the probability of survival, respectively, of an individual (*i*) in a given time interval (*t*) with all subscripts the same as the growth and model above.

Recruitment was modeled as the frequency of individual trees recruiting into each 10 × 10 m subplot, that varied as a function of the standardized, transformed covariates: average NCI of trees in that subplot [ln(NCI)], the average proportion of NCI comprised of N fixers for all trees in the subplot [arcsin(^{2}, and ln(NCI) × the proportion of NCI comprised of N fixers [arcsin(

where *Rec* is the recruitment of an individual in a given census year with all subscripts the same as in the growth and survival equations above except for σ_{sp}, which represents the random effect of subplot.

## Acknowledgments

We thank J. Paniagua, B. Paniagua, and E. Salicetti for their work collecting field data and M. Uriarte, N. Schwartz, and A. Quebbeman for statistical advice. This work was supported by NSF Grants DEB-0424767, DEB-0639393, and DEB-1147429; the Andrew W. Mellon Foundation; the NASA Terrestrial Ecology Program; and the University of Connecticut Research Foundation (to R.L.C.).

## Footnotes

- ↵
^{1}To whom correspondence should be addressed. Email: bentonneiltaylor{at}gmail.com.

Author contributions: B.N.T. and D.N.L.M. designed research; B.N.T. and R.L.C. performed research; B.N.T. and B.B. analyzed data; and B.N.T., R.L.C., B.B., and D.N.L.M. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Data deposition: Data used in this study are available in the figshare data repository at https://figshare.com/articles/Supplement_1_Data_on_tree_dynamics_during_secondary_succession_and_wood_specific_gravity_in_northeastern_Costa_Rica_/3520553/1.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1707094114/-/DCSupplemental.

## References

- ↵
- ↵
- ↵.
- Pan Y, et al.

- ↵
- ↵.
- Davidson E, et al.

- ↵
- ↵.
- Hungate BA,
- Dukes JS,
- Shaw MR,
- Luo Y,
- Field CB

- ↵.
- Gerber S,
- Hedin LO,
- Keel SG,
- Pacala SW,
- Shevliakova E

- ↵.
- Gerber S,
- Hedin LO,
- Oppenheimer M,
- Pacala SW,
- Shevliakova E

- ↵.
- Wårlind D,
- Smith B,
- Hickler T,
- Arneth A

- ↵
- ↵.
- Binkley D,
- Giardina C

- ↵
- ↵
- ↵.
- Adams MA,
- Turnbull TL,
- Sprent JI,
- Buchmann N

- ↵.
- Nasto MK, et al.

- ↵
- ↵
- ↵.
- Menge DNL,
- Chazdon RL

- ↵
- ↵
- ↵
- ↵.
- Menge DNL, et al.

- ↵.
- Sullivan BW, et al.

- ↵.
- Chazdon RL, et al.

- ↵.
- Lasky JR,
- Uriarte M,
- Boukili VK,
- Chazdon RL

- ↵
- ↵.
- Sprent J,
- McKey D

- McKey D

- ↵
- ↵
- ↵.
- Menge DNL,
- Wolf AA,
- Funk JL

- ↵.
- McDade L,
- Hartshorn G

- ↵
- ↵
- ↵.
- Pearson H,
- Vitousek P

*Acacia koa*. Ecol Appl 11:1381–1394. - ↵
- ↵.
- Gilman A, et al.

- ↵.
- Ferreira da Silva A, et al.

- ↵.
- Bastin J-F, et al.

- ↵.
- Asner GP,
- Jones MO,
- Martin RE,
- Knapp DE,
- Hughes RF

- ↵
- ↵.
- Sprent JI

- ↵.
- R Core Team

*R: A Language and Environment for Statistical Computing*(R Foundation for Statistical Computing, Vienna), Version 3.2.2.

## Citation Manager Formats

## Article Classifications

- Biological Sciences
- Ecology