Introduction

Ghana is the second largest producer of cocoa in the world and over 6 million Ghanaians depend on the cocoa sector for their livelihood (Kolavalli and Vigneri 2011). In recent years, cocoa farmers in Ghana have been confronted with the challenge of yield decline and, in many growing regions, the average yield is 0.45 Mg cocoa beans ha−1, much lower than the potential achievable yields of 0.8–1.0 Mg ha−1 (Baffoe-Asare 2013). In addition to drought, one important cause of such yield decline is the depletion of soil nitrogen (N) due to continuous cropping with little or no N replenishment (Baah et al. 2011; Aneani et al. 2011). Soil N deficiency, causing stunted cocoa growth, fruit abortion, and poor pod yields (Tscharntke et al. 2011) could be prevented by the supply of mineral N fertilizers, but these are often unaffordable for smallholder farmers, who represent the majority of cocoa farmers in Ghana (Opoku-Ameyaw et al. 2012; Nunoo et al. 2014). Moreover, many remote regions of Ghana are not easily accessible and recent estimates indicate that only 20% of the 1.6 million hectares of cocoa farms in Ghana receive mineral fertilizers (IFDC 2012).

N2-fixing legume trees are common in tropical agroforestry systems and, if properly managed, they may provide an alternative to N fertilizers as a means for enhancing soil N availability to the benefit of the non-legume plants. In their review, Nygren et al. (2012) listed more than 15 tree species belonging to Fabaceae, present in humid or sub-humid agroforestry systems, whose N derived from atmosphere varied from 28 to 92%, with ranges for each species depending on the organ considered, the microbial strain, the tree age, the pruning frequency, the sampling time, and the reference species. Some of these trees, like Inga edulis and Gliricidia sepium, also provide the shade required by the cocoa tree. Nitrogen deriving from biological fixation (BN) is a cheap, practically unlimited, and environmentally friendly N source. Gliricidia sepium (Jacq. Kunth ex Walp.), hereafter gliricidia, is capable of forming a symbiotic association with N2-fixing rhizobia (Kinkema et al. 2006) and produces high-quality biomass for use as green manure (Subramanian et al. 2005).

To account for the potential of gliricidia in introducing N from BNF, 15N isotopic techniques are apparently the most recommended for field studies (Boddey et al. 2000). The acetylene reduction assay (ARA), although useful to assess the nitrogenase activity, is unsuitable to quantify the N fixed under field conditions (Chalk et al. 2017). A few studies have attempted to estimate N fixation by the ureide abundance in xylem sap (Unkovich et al. 2008), but the required relationship between %Ndfa and ureide abundance for gliricidia was not developed, yet. The 15N enrichment and the 15N dilution technique are difficult to apply under field conditions in view of problems related to the dilution of the labeled fertilizer N, with unlabeled soil mineral N that may lead to an unequal 15N enrichment in space and time (Unkovich et al. 2008). The 15N natural abundance method (15NNAM) is the best option for estimating N2 fixation in tree plantation and agroforestry systems as the soil available N is naturally enriched with 15N, allowing the quantification of N2-fixation integrated over one season or even over longer periods (Boddey et al. 2000; Unkovich et al. 2008; Khamzina et al. 2009).

The 15NNAM is based on the difference in δ15N values between two N sources (soil mineral N and atmospheric N2) for the plant (Unkovich et al. 2008). The 15NNAM requires one or more non-N-fixing reference plants that provide an integrated determination of the 15N of the soil mineral N available for plant growth over the duration of the study (Boddey et al. 2000). Such reference plants should have roots with similar vertical distribution of the N2-fixer plant and be far enough from the latter so that the residues of one type of plant do not affect the 15N abundance of the soil available N.

In this study, we have (i) molecularly characterized the bacterial communities of gliricidia root nodules, and assessed (ii) the fraction of N derived from the atmosphere present in shoots of gliricidia trees intercropped with cocoa, and (iii) the potential transfer of atmospheric N to the neighboring cocoa trees through the pruning of gliricidia shoots.

Methodology

Experimental sites

The experiment was carried out in two cocoa-growing regions of Ghana (West Africa). Three cocoa plantations (referred to as sites 1, 2, and 3), where gliricidia trees were present as an intercropped species, were sampled, both at the end of the dry (January 2015) and of the wet (September 2015) seasons. A fourth and a fifth site (site 4, 5) were sampled only at the end of the wet season. At sites 1–4, cocoa trees were planted with 3 × 3 m spacing while gliricidia trees were dispersed at variable distances from cocoa trees, from less than 5 to more than 10 m. At site 5, gliricidia, cocoa, and orange trees were not intercropped but grown as separate stands in neighboring plots with a spacing of 9 m × 9 m (gliricidia) and 3 m × 3 m (orange and cocoa).

The size of the plots ranged from 0.20 to 0.85 ha (Table 1). Sites slightly differed in type of management and intercropping characteristics (Table 1). All sites were located at an altitude between 200 and 250 m a.s.l. Sites 1, 2, and 5 are located in the Ashanti region and are part of the experimental farm of the Faculty of Renewable Natural Resource, Kwame Nkrumah University of Science and Technology (KNUST) in Kumasi (Table 2). Sites 3 and 4 are located in the eastern region and within the experimental farm of the Bunso Cocoa College in Bunso. Both regions form part of the semi-deciduous forest zone of Ghana (Aneani et al. 2011) and are characterized by a bimodal distribution of rainfall, with a main wet season from March to July and a minor wet season that starts in late August and lasts until November (Anim-Kwapong 2003; Partey et al. 2011) (Table 2). The soils in all sites have sandy loam texture. Soil N concentration ranged from 0.08 to 0.14% (Table 3). Gliricidia trees are normally pruned in December and the pruned shoots left beneath the trees as green mulch. The horizontal spread (maximum diameter of the soil projection of the canopy) of the gliricidia trees was measured at the end of the wet season in 2018 in sites 2, 3, and 4 on four randomly chosen trees per site and averaged 10.0 m (standard deviation, s.d. = 2.2 m) tree−1, a value in line with literature data (Elevitch and Francis 2006).

Table 1 Some management characteristics of the selected sites
Full size table
Table 2 Main climate characteristics, soil type, and geographic coordinates of study sites in the Ashanti and Eastern regions of Ghana
Full size table
Table 3 Main soil chemical characteristics and δ15N values from 0 to 45 cm
Full size table

Nitrogen derived from the atmosphere (Ndfa)

To assess the percentage of N derived from the atmosphere (% Ndfa), the 15N natural abundance method (15NNAM) was adopted. The technique is based on the comparison of the delta (δ) 15N of a plant that develops symbiosis with N2-fixing microorganisms with one or more reference plants, whose N derives entirely from soil N forms. The δ15N of the atmospheric N2 is defined as 0; the isotopic composition of other N sources is normalized to the atmospheric composition (Shearer and Kohl 1986). Within each site, three sub-plots (about 500–800 m2 each) were randomly selected, one gliricidia tree and two cocoa trees, one close (cocoa C), and the other distant (cocoa F) to the closest gliricidia tree were used for sampling. The means and standard deviations of the distance of selected cocoa C and cocoa F from the closest gliricidia tree were 5.48 ± 0.35 and 13.04 ± 0.77 m, respectively. Sixty to eighty young, fully-expanded leaves were sampled from current year shoots from each sampled tree. At site 5, three sub-plots were selected within separate stands of gliricidia, cocoa, and orange trees. Leaves of orange and shoot axes of gliricidia, in addition to cocoa and gliricidia leaves, were collected at site 5.

In September 2015, three soil samples from each sub-plot were collected from the 0–30-cm upper layer (Hartemink 2005), at least 5 m from the closest gliricidia tree (sites 1–4) and in the cocoa separate stand present at site 5. Soil samples were analyzed for chemical and physical properties as well as for δ15N, to check one of the requirements for applying the 15NNAM. Soil pH was determined in 0.01 M CaCl2 (1:2.5). Total N, δ15N, and organic C (after soil acidification using 6 M hydrochloric acid) were measured at the Isotope Laboratory of the Free University of Bozen-Bolzano, Italy, using an elemental analyzer coupled with a continuous flow isotopic mass spectrometer (Flash 2000 and Delta V, Thermo Scientific Instrument). The extractable magnesium (Mg) was analyzed at the agricultural chemistry laboratory of the Research Centre for Agriculture and Forestry of Laimburg according to Verband Deutscher Landwirtschaftlicher Untersuchungs-und Forschungsanstalten (Bassler et al. 1991). The “plant-available” potassium (K) and phosphorous (P) were extracted in calcium-acetate-lactate (CAL solution) according to the Austrian standards (ÖNORM L 1087 2012).

The sampled leaves were cleaned from dust and oven-dried at 60 °C for 72 h. The dried material was milled into a homogenous powder using a laboratory ball-and-capsule vibrating mill. About 2.0–2.5 mg of milled material was encapsulated in tin and analyzed for their percentage nitrogen (% N) and 15N (using the same instrument described above).

The percentage of nitrogen derived from the atmosphere (% Ndfa) in the gliricidia plant was estimated according to Shearer and Kohl (1986) and Boddey et al. (2000) using the formula:

$$ Ndfa=\frac{\delta^{15}{N}_{reference}-{\delta}^{15}{N}_{gliricidia}}{\delta^{15}{N}_{reference}-B}\bullet 100 $$

where δ15N reference is the δ15N of leaves from the non-N-fixing plant (cocoa for sites 1–4 and both cocoa and orange for site 5) and δ15N gliricidia is the δ15N value of gliricidia leaves or shoot axes. Theobroma cacao proved to be suitable as a reference plant in a study carried out by Nygren and Leblanc (2009), giving similar results of Ndfa as three other non-legume reference species. The B value represents the δ15N value of gliricidia when N2 is the sole N source for the tree. In our calculations, we used four B values that had been experimentally determined for Gliricidia sepium leaves or shoots, − 0.45‰ (Anhar 2005), − 1.45‰ (Ladha et al. 1993), − 1.11‰ (Hairiah et al. 2000), and − 2.07‰ (Nygren et al. 2000).

Assessment of growth and nitrogen uptake by gliricidia

To estimate the amount of nitrogen fixed by gliricidia trees, the biomass of shoots grown in 1 year and their N content were determined. This part of the study was carried out at site 1 only. In January 2015, six 5-year-old gliricidia trees, whose circumference at 20 cm from the ground varied from 17.9 to 22.7 cm, were randomly selected. The average number of branches per each tree was 13 (range 9–15). Five branches per tree were tagged and within each branch, the terminal point of previous season growth of each shoot was identified and marked. The average number of shoots per branch was 10 (range 7–14) and the number of leaves per shoot was 19 (range 17–27). In January 2016, the tagged branches were cut and the new growth separated into leaves and shoot axes, which were oven-dried at 70 °C for 72 h and weighed. The average dry weight (DW) of leaves and shoots present on the sampled branches was multiplied by the total number of branches per tree to estimate the total biomass of leaves and shoot axes per tree. Samples were analyzed for total nitrogen concentration and the total amount of N in leaves and shoot axes was calculated.

Analysis of root nodules and nodule bacteria

Gliricidia-nodulated root samples were collected from the study sites 1–3 at the end of the dry (January 2016) and wet (September 2016) season and only at the end of the wet season at site 4. Three gliricidia trees per site were sampled, with the exception of site 3 where only two trees were sampled. Under each tree, four holes (40 × 40 cm with 15 cm depth) were dug 45–60 cm distant from the trunk.

After the dry season, we did not find root nodules in any of the sites, whereas after the wet season, nodules were always present. When present, root nodules were visible with the naked eye. Roots collected from each hole were used to prepare a composite, representative root sample for each tree (10–20 g). Roots were cleaned from soil particles, left under shade for 6–12 h to remove excess water (only after the September sampling), placed in paper envelopes and wrapped in a polyethylene bag, transported to Italy in cooling flasks to avoid desiccation, and processed within 48 h.

To assess root nodules, root samples from sites 1–4 were washed with tap water and wiped. Sub-samples of about 1.8 g each were collected and analyzed under a dissecting microscope (Wild, Leica, Milan, Italy) to quantify the number of nodules.

A culture-independent method, PCR-DGGE, was used to identify the bacteria occurring in the nodules. Ten nodules were collected for each sample, re-hydrated in sterile distilled water, and externally sterilized by immersion in ethanol (70% v/v) for 2 min and then in sodium hypochlorite (3% v/v) for 4 min, followed by two rinses of 30 min in sterile water. Sterilized nodules were crushed with sterile plastic pestles in 150 μl of the “PowerSoil™ DNA Isolation Kit” extraction buffer (Mo-Bio Laboratories Solana beach, CA, USA). DNA was isolated according to the manufacturer’s instructions. For analysis of bacterial communities, the amplification of the variable region V3–V5 of the 16S rRNA gene was carried out using the primers 341F and 907R (Yu and Morrison 2004), as reported in Agnolucci et al. (2015).

For DGGE separation, the amplicons were analyzed using the DCode™ Universal Mutation Detection System BIORAD and a gel with a urea-formamide denaturing gradient ranging between 35 and 65%. A composite mix of bacterial 16S rRNA gene fragments from Ensifer meliloti (formerly Sinorhizobium meliloti) IMA N29., Bacillus sp. IMA CH19, Streptomyces spp. IMA W77, and Mesorhizobium ciceri IMA M620 were added as reference DGGE marker (M). Gels were run and visualized as described in Agnolucci et al. (2013). DGGE fragments were cut out from the gels for sequencing (Agnolucci et al. 2015). Sequences were analyzed using BLAST on the NCBI web and submitted to the European Nucleotide Archive, under the accession numbers from LT797814 to LT797824.

Statistical analysis

The effects of plant type (gliricidia, cocoa C, and cocoa F) and sampling season (after dry and wet periods) on δ15N and total N concentration of leaves sampled from sites 1–3 were verified using an ANOVA after having checked the assumptions of normal distribution and the homogeneity of variance. A one-way ANOVA was similarly used to assess differences in total N concentration and δ15N among gliricidia, cocoa C, and cocoa F at site 4 as well as among gliricidia organs, cocoa, and orange at site 5. The probability level of 5% was considered for the statistical significance; though where appropriate, p values less than 1% are stated. Mean separation was carried out using the Tukey’s Honestly Significant Difference (HSD) procedure at 5% level of probability. Statistical analysis was performed using StatGraphic Centurion XV (StatPoint, Inc.). Where shown, data are averages ± standard deviation (s.d.) of the three replicates per site and season.

Results

Nitrogen derived from the atmosphere

Nitrogen concentrations were significantly (p < 0.01) higher in gliricidia than in cocoa leaves, regardless of site, season, or distance between cocoa and gliricidia (Table 4, Fig. 1). With the exception of gliricidia and cocoa F at site 2, both cocoa and gliricidia leaves had higher N concentration after the dry than after the wet season (Table 4). At site 5, gliricidia leaves also had higher N concentrations as compared to orange leaves and gliricidia shoot axes (Fig. 1).

Table 4 N concentration (% N DW) of cocoa and gliricidia leaves after the dry and the wet seasons
Full size table
Fig. 1
figure 1

N concentration (% DW) and δ15N of selected gliricidia organs, cocoa leaves, and orange leaves sampled from separate stands (site 5). Bars indicate standard deviations

Full size image

The δ15N of total soil N in the five sites ranged from 7.0 to 7.5‰ (Table 3). The effects of season and plant type on leaf δ15N were different across sites (Table 5). With the exceptions of gliricidia and cocoa C at site 1, leaf δ15N was always higher after the wet than after the dry season (Table 5). The δ15N value of gliricidia leaves was lower than that of cocoa F for both sampling periods at site 2 and at sites 1, 4, and 5 after the wet season (Table 5, Fig. 1). With the exception of site 3 and the dry season sampling at site 1, leaves from cocoa C had lower δ15N than cocoa F, and were generally similar to gliricidia. Gliricidia leaves and shoot axes had similar δ15N at site 5, and both had lower δ15N than orange leaves (Fig. 1).

Table 5 δ15N (‰) of gliricidia and cocoa leaves after the dry and wet seasons
Full size table

% Ndfa by gliricidia could be assessed after the wet season at sites 1, 4, and 5, for both sampling periods at site 2, but neither sampling period for site 3, nor the dry period at site 1, when the δ15N of gliricidia and the reference plant did not significantly differ. The average of estimates of % Ndfa present in gliricidia leaves using four different B values varied from 22% in the sampling carried out after the wet season at site 2 to 50% in the sampling carried out after the wet season in site 4 (Fig. 2). At site 5, the choice of cocoa or of orange as reference trees affected the estimates of % Ndfa of gliricidia leaves, 25% and 39%, respectively. The use of four B values from the literature resulted in four estimates of % Ndfa of gliricidia leaves, whose coefficient of variation was 10% on average (Fig. 2). At site 5, shoot axes averaged 48 ± 5% of N from the atmosphere (data not reported in the figure).

Fig. 2
figure 2

% Ndfa in gliricidia leaves collected after the dry and the rainy season calculated using four B values retrieved from the literature (see legend). Reference plants were cocoa in sites 1, 2, and 4 (white symbols) and both cocoa (white symbols) and orange (black symbols) trees in site 5. The white empty circle indicates the average of each set of data and the vertical bars indicate the standard deviation

Full size image

Between January 2014 and January 2015, each gliricidia tree increased its shoot biomass by approximately 38 kg on average (DW). The shoot axis:leaf ratio was 7:1. Leaves had much higher N concentration than shoot axes, but due to their lower biomass, their N content was approximately 30% of the total shoot N (Table 6).

Table 6 Estimated annual shoot growth (DW = dry weight) of gliricidia trees and its N content. Data refer to the site 1 (mean ± standard deviation)
Full size table

Analysis of root nodules and nodule bacteria

No nodules were found after the dry season, while nodules were present after the wet season and their density was assessed (Table 7). All crushed nodules showed an intense red color due to the presence of leghemoglobin, which is a characteristic of effective N2 fixation.

Table 7 Number of nodules per gram of root fresh weight (mean ± standard deviation)
Full size table

DGGE analysis revealed the presence of Rhizobium tropici and Rhizobium etli at all the four sites. In addition, other sequences from the four sites showed 99% similarity with Rhizobium sp. strain 6B. Moreover, sequences affiliated to Sporosarcina thermotolerans strain CCNWQLS96 (100% similarity) and to Planomicrobium chinense strain RT4 (99% similarity) were retrieved from site 1 (Fig. 3).

Fig. 3
figure 3

DGGE analysis of bacterial communities occurring in Gliricidia sepium root nodules from sites 1–4. P, replicate plants; M, markers

Full size image

Discussion

In this study, we have quantified the N deriving from atmospheric N fixation in Gliricidia sepium trees, assessed the occurrence of effective nodules, and then molecularly characterized their microbiota. The number of effective nodules per gram of root ranged from 62 to 92, values that can be hardly compared with literature data as, to our knowledge, no studies have assessed root nodule numbers of legume trees in the field, while most of them have been carried out in vitro or in microcosm experiments, where the whole root systems were harvested and analyzed after short-time experiments, i.e., 3 to 13 weeks (Rosendahl et al. 1989; Tsai et al. 1998). Here, in order to identify the bacterial communities occurring in the nodules, a culture-independent approach, PCR-DGGE, was chosen, as it is able to overcome the problem of underestimation consequential to the constraints of culture media and cultivation conditions, as well as to detect the presence of bacteria in Viable But Not Culturable (VBNC) state (Agnolucci et al. 2015). The analysis of excised DGGE bands revealed sequences affiliated to R. tropici and R. etli. Rhizobium tropici has been reported as associated with nodules of bean and leguminous trees, including G. sepium (Martínez-Romero et al. 1991; Nick et al. 1999; Acosta-Durán and Martínez-Romero 2002), while R. etli has been retrieved from a broad number of host species (Hernandez-Lucas et al. 1995). At all four sites, we found one sequence showing 99% similarity with Rhizobium sp. strain 6B, which has been described as associated with Vachellia karroo in South Africa (Nxumalo et al. 2016).

The %Ndfa in gliricidia shoots varied among the sites: the lowest and the highest values were recorded after the wet season at site 2 (22%) and 4 (50%). Explaining the differences in the %Ndfa in gliricidia organs across sites is beyond the scope of the study, although we might speculate that the age of the trees (Leblanc et al. 2007) and the availability of soil-derived N may account for the differences. It is also known that, at least within a given range, increasing soil phosphorous (P) availability has a positive effect on N fixation rates (Isaac et al. 2011). Soil available P levels in the five sites of the present study are within typical range found in a survey carried out in cocoa plantations in Ghana (Afrifa et al. 2010), but likely lower than the optimal soil P levels for cocoa production. In their review on dinitrogen fixation in agroforestry system, Nygren et al. (2012) averaged data from several published studies and reported Ndfa values of 67% (± 13%) for gliricidia trees, while Apolinário et al. (2016) reported Ndfa of 55% for G. sepium grown in the State of Pernambuco, Brazil.

The effect of the sampling period on % Ndfa was only observed at site 2, where slightly higher values were found after the dry than after the wet period (Fig. 2). However, interestingly, we did not find root nodules after the dry season, similarly to what has been reported for other legume trees like Sesbania sesban (L.) Merr. and Leucaena leucocephala (Lam.) de Wit (Fownes and Anderson 1991; Dirzo et al. 2012). These two results could be regarded as conflicting. It should be considered, however, that root nodule sampling was carried out at the end of the dry period when soil moisture had reached its lowest value, while the δ15N of the leaves sampled in January likely reflected an average value of dinitrogen fixation for the entire dry period. In addition, it has been reported that root nodulation ceases following shoot pruning, which in our study was carried out in December, and that the trees remain for several weeks unnodulated after pruning (Nygren and Ramírez 1995; Nygren and Cruz 1998). We cannot exclude some tree-internal N remobilization from leaves developed in the wet period to those growing in the dry period. Our data, therefore, do not contradict the hypothesis that atmospheric N fixation in gliricidia trees is mostly active when water availability in the top soil is not limiting and rapid shoot growth occurs (Nygren and Leblanc 2009).

As stressed by multiple authors (Boddey et al. 2000; Nygren and Leblanc 2009), the estimation of the N contribution by legume trees to the N cycle in agroforestry systems under field conditions is a methodologically-challenging task. The 15N natural abundance techniques provides reliable estimates of biological N fixation when several conditions are met (Shearer and Kohl 1986; Ladha et al. 1993; Gehring and Vlek 2004), including the correct choice of the reference plant, the presence of soil δ15N higher than 5‰, and the choice of the organs to be sampled. We have not assessed the δ15N of the mineral N and are aware that some discrimination (2.4‰, Hobbie and Högberg 2012) against 15N occurs during the ammonification and nitrification processes, which explains the depletion of plant N of cocoa trees as compared to total soil N (Table 3). In our study, we have used cocoa trees as reference plants in four sites, and both cocoa and orange on site 5. As a reference plant, T. cacao alone gave similar estimates for N2 fixation of Inga edulis Mart. and G. sepium as a group of non-N2-fixing reference species in the study by Nygren and Leblanc (Nygren and Leblanc 2009). Both T. cacao and G. sepium have quite superficial root systems, mainly concentrated in the upper 0–30-cm soil layer, where also most soil available N is present (Schroth and Zech 1995; Rowe et al. 2001; Nygren et al. 2013). As far as reference plant choice is concerned, if we had used only orange trees—also known for a superficial root system—instead of cocoa trees at site 5, we would have obtained an additional increase in Ndfa by 13 and 11%, for leaves and shoot axes, respectively. We might then speculate that our estimates for the other sites, where only cocoa was used as a reference plant, are likely conservative. The choice of the B value also had an impact on the final assessment of % Ndfa; for instance, using either a B value of − 0.45‰ or of − 2.07‰ would result in % Ndfa estimates differing from a minimum of 6% (site 1) to a maximum of 11% (site 2, dry sampling, Fig. 2).

The fact that we did not find differences in the δ15N of cocoa and gliricidia at site 3 for both samplings, in spite of the presence of nodules of rhizobia on gliricidia roots, could be due to the regular supply of synthetic N fertilizers in this cocoa plantation, whose δ15N is similar to that of the atmosphere (Bateman and Kelly 2007).

To estimate the amount of N deriving from the atmosphere that can become available to cocoa trees due to the presence of legume trees, not only are reliable estimates of the % Ndfa needed, but it is also necessary to quantify the amount of total N present in the shoots that can potentially be removed by annual pruning. Such amounts depend on, among others, soil fertility, tree growth potential, and tree density. When gliricidia trees were grown at high tree density in hedgerows (2500 trees ha−1) in Northern Lampung (Sumatra), the shoot biomass that could be pruned annually amounted to more than 4 Mg ha−1 (DW) and its N content was equal to 104 kg ha−1 (Hairiah et al. 2000), an amount similar to what is reported by Apolinário et al. (2016) for N in annual litter (105–109 kg N ha−1) produced by gliricidia trees in an agroforestry system (spacing 10 × 0.5 m). Decreasing the pruning frequency increases the annual shoot biomass production and enhances the shoot axis:leaf biomass ratio (Marroquín et al. 2005).

When intercropped in cocoa plantations to provide shade, gliricidia trees are often planted at recommended spacing of 9 m × 9 m (Asare and David 2010). On the basis of shoot growth assessment (Table 6), the shoots (shoot axis and leaves grown in 1 year) of the resulting 124 gliricidia trees per hectare would therefore contain 78 kg N ha−1. Using the data of % Ndfa obtained for leaves on the sites 1, 2, 4, and 5 (Fig. 2) and for the shoot axes on site 5 (48%) and assuming that on all sites gliricidia trees had similar growth as on site 1, we can speculate that the annually produced shoots of gliricidia, theoretically suitable to become green manure after pruning, contain from 31 to 38 kg Ndfa ha−1 (using the lowest and highest % Ndfa, site 2 wet and site 4, respectively) (Fig. 4). These amounts certainly underestimate the whole tree N fixation of gliricidia as they do not take into account the N stored in roots and in stems.

Fig. 4
figure 4

Schematic of the transfer of the atmospheric N (Ndfa) fixed by Rhizobium tropici and R. etli associated with gliricidia root nodules to cocoa trees in intercropping systems through pruned shoots left on top of the soil as green manure. (1) Symbiotic N fixation by root nodules; (2) uptake of Ndfa from gliricidia trees and allocation to shoots; (3) Gliricidia shoots pruned and left at the soil surface decompose and release Ndfa that enters the soil; (4) potential direct Ndfa transfer from gliricidia roots to cocoa roots through mycorrhizal network; (5) Ndfa uptake by cocoa roots. Data are in kg N ha−1

Full size image

According to the recent review on mineral nutrition of cocoa (van Vliet and Giller 2017), the N removal by beans and husks average 30 kg N Mg−1. In Ghana, the national on-farm average cocoa bean yields are approximately 0.4 Mg ha−1 and as high as 3.36 Mg ha−1 if trees are properly managed (Aneani and Ofori-Frimpong 2013). It is therefore reasonable to expect net N removals ranging from 12 to 100 kg N ha−1 depending on the yields (Fig. 4). Such values are compatible with the recommended N-fertilizer rates in Ghana and Ivory Coast, often in the range of 58–80 kg N ha−1 (van Vliet and Giller 2017).

In our study, the benefit of gliricidia trees in terms of enhanced N availability for cocoa trees grown in the vicinity was supported by the evidence that the δ15N of cocoa C trees was lower than that of cocoa F trees growing further away from gliricidia trees. The fact that we have not found differences of δ15N between gliricidia and close cocoa trees is in line with Nygren and Leblanc (2009), who reported that leaves of Inga edulis trees had lower δ15N values than those of cocoa trees growing further away, but similar δ15N values to close cocoa trees. It should, however, be kept in mind that there are risks associated with an estimate of N transfer from legume to non-legume trees based solely on δ15N values, due to the unknown extent of the δ15N isotopic fractionation during the transfer process (Peoples et al. 2015). Similar δ15N values between a legume and a non-legume tree would in fact apparently suggest that the latter entirely use N from the legume. Gliricidia shoots used as green manure in intercropping systems under tropical climates rapidly decompose at the soil surface and release most of their N content within few months (Handayanto et al. 1994), which is then available for successive uptake by cocoa trees. Further investigation is, however, required to assess to which extent the transfer of the atmospheric N to cocoa trees occurs through decomposition of pruned material or by direct transfer from gliricidia to neighboring cocoa trees via a common mycelial network that, although not yet demonstrated for cocoa and gliricidia, is likely to occur in nature (Kurppa et al. 2010).

In conclusion, we have demonstrated that Gliricidia sepium trees are able to take advantage of the association with rhizobial symbionts, to fulfill, at least in part, the N needs of their rapidly growing shoots. In mixed-stand agroforestry systems, with intercropped gliricidia and cocoa trees, the amount of N derived from the atmosphere that could enter the soil if the pruned shoots are used as green manure could diminish the N-fertilizer needs by a minimum of 37% (using 31.4 kg ha−1 Ndfa and a fertilizer rate of 80 kg ha−1 N) to a maximum of 65% (using 39 kg ha−1 Ndfa and a fertilizer rate of 58 kg ha−1 N).