1. Introduction

Fine-root functional traits reflect strategic trade-offs in plants, balancing the efficient acquisition of limited resources with the metabolic expenditure required for root development and maintenance (Lugli et al., 2021). They are fundamental to numerous ecological processes and modulate how plants acclimate to shifting environmental conditions (Bardgett et al., 2014). Different combinations of traits may facilitate rapid nutrient uptake, or enhance defense and conservation of the obtained resources (Weemstra et al., 2016). Nevertheless, earlier research has predominantly depended on easily measurable “soft traits”, for instance those describing morphology and architecture, to establish links between plants and ecosystem functioning. In contrast, “hard traits” such as those related to root exudation are more challenging to quantify and yet, are often more directly tied to precise ecological functions (Belluau and Shipley, 2018; Freschet et al., 2018; Ros et al., 2018).

Root exudates, key products of photosynthetically-derived carbon, include numerous primary and secondary metabolites, which is termed “general root exudates” by Wen et al. (2022). The carbon release rate of general root exudates (RE) is usually expressed as the exudation rate of the total soluble organic carbon per root dry weight and time (Phillips et al., 2008). Root exudates are commonly considered a carbon investment by plants, which can enhance the release of mineral-bound nutrients, stimulate microbial activity, and increase the rate of organic matter mineralization and nutrient availability (McCormack and Iversen, 2019). However, this investment is cost-effective only when coupled with traits that ensure rapid nutrient uptake (e.g., high SRL). Otherwise, mobilized nutrients may be exploited by soil microbes or competing plants instead. This is why high RE is usually associated with acquisitive root traits; for instance, RE increases exponentially with specific root length (SRL) (Meier et al., 2020). This positive correlation is further supported by Guyonnet et al. (2018), who indicates that fine-root morphological traits can serve as effective proxies for exudation traits.

The secretion of root acid phosphatase (RAP), a component of root exudates, represents an efficient P-mining strategy (Lugli et al., 2020, 2021). Phosphatase enzymes, which are upregulated under high phosphorus (P) demanding plants, catalyze the hydrolysis of organic P, enabling “biochemical P mineralization”. This process separates P release from the breakdown of organic matter (Goll et al., 2017; Reichert et al., 2022). As a nitrogen-rich enzyme, the synthesis of RAP depends on root nitrogen metabolism (Wen et al., 2022) and requires P-rich ribosomal RNA (rRNA) (Elser et al., 2000), which signals that RAP production is closely linked both with nitrogen and P metabolism within the roots. However, the observed correlations between RAP and other root traits are variable across studies. For instance, while some investigations reported a strong linkage of RAP with root nitrogen concentration (Bi et al., 2023; Ao et al., 2025), others found significant correlations of RAP with root morphological traits such as SRL (Han et al., 2022). This variability suggests that the integration of RAP into belowground trait spectra may be context-dependent, potentially influenced by factors like plant community composition and mycorrhizal association. Therefore, examining how RAP is coordinated with a suite of root traits will help clarify these contrasting patterns observed across different studies.

Trait covariation patterns reveal critical insights into adaptive strategies employed by plants in changing habitats (Laughlin, 2014; Messier et al., 2017). Many earlier studies have focused on trait coordination within major economic dimensions, overlooking complex inter-trait relationships across dimensions (Kleyer et al., 2019; Wang et al., 2023). Plant trait networks (PTNs) provide a novel approach for capturing broad-scale relationships among multiple traits. They reveal overall adaptive strategies through network metrics, including degree, edge density, and modularity (Flores-Moreno et al., 2019; Li et al., 2024a, Li et al., 2024b). For detailed interpretations and functional significance of each metric, refer to Table 1. Modules within these networks often correspond to trait dimensions, and independent trait modules may enhance plants' capacity to adjust functions in varied environments (Kleyer et al., 2019).

Previous research has demonstrated that under nutrient stress (e.g., deficiency), trait integration tends to be stronger so as to enhance adaptive capacity under adverse conditions. In contrast, when nutrient availability increases, trait integration weakens (Gianoli and Palacio-López, 2009; Wen et al., 2019; Wang et al., 2021). Research has also shown that nutrient addition can reshape trait associations and increase the modularity of PTNs (Tao et al., 2022). Although these advances have improved our understanding of aboveground trait coordination, belowground trait networks, and particularly their responses to nutrient availability, remain largely unexplored. In subtropical evergreen broad-leaved forests, soil P availability is highly heterogeneous, forming localized “P-rich patches” that serve as critical nutrient hotspots (Hodge, 2004; Raven et al., 2018; Xia et al., 2020; Jiang et al., 2025). Nevertheless, how belowground trait covariation patterns relate to the utilization of nutrient patches in highly diverse and complex natural mature forests remains poorly understood.

To address these questions, we examined a suite of P-acquisition-related root traits across 17 co-occurring arbuscular mycorrhizal (AM) tree species within a subtropical evergreen broad-leaved forest as in Zhu et al. (2023). We constructed root trait networks based on 14 key traits and compared network topological parameters between control and P-rich patches. In contrast to Zhu et al. (2023), who focused on plastic responses of belowground foraging traits to P-rich patches, the present study aims to elucidate: (1) the associations of root exudation traits (RE and RAP) with other traits as well as their positions within the root trait network; and (2) how root trait networks change in P-rich soil patches compared to Control patches. We hypothesize that: (1) RE is closely associated with morphological traits and will cluster within root exploration trait module; whereas RAP, whose synthesis is inherently linked to root nutrient stoichiometry or metabolism, is associated with the chemical trait module; and (2) P addition will reduce the strength of root trait associations while increasing network modularity.

2. Materials and methods 2.1. Study site

This research was carried out within a natural evergreen broad-leaved forest located in the Gesikao Nature Reserve (26°9′N, 117°28′E) in Fujian Province, China. The site is characterized by a subtropical monsoon climate, with mean annual temperature of 19.5 ℃ and mean annual precipitation of 1700 mm. The soils are classified as Oxisol (USDA Soil Taxonomy, Soil Survey Staff, 2014). The stand elevation ranges between 200 and 300 m with a northeast aspect. Historically a bamboo (Phyllostachys edulis) plantation abandoned prior to the 1940s, the area transitioned to its current evergreen broad-leaved forest composition by the 1970s. The forest has remained undisturbed for approximately five decades.

2.2. Species selection and plot establishment

A 4-ha permanent plot (100 m × 400 m) was established in December 2019. We selected 17 AM tree species spanning 14 families and 17 genera (Table S1). For each species, between four and seven individuals with comparable height and diameter at breast height (DBH) were chosen, yielding 89 sample trees. To ensure independence, conspecific individuals were separated by at least 15 m.

2.3. Root bag preparation and phosphorus addition

To evaluate root foraging responses, we used a root bag technique that isolated roots of individual species in controlled soil environments. Soil was collected from the top 20 cm, homogenized, and sieved (5 mm mesh) to remove debris. Contour lines spaced 10 m apart were established along the slope, with 20 sampling points per line. The soil background values are shown in Table S2.

Each root bag received 3 kg of prepared soil in January 2020. Two treatments were applied: unfertilized Control patches and P-rich patches. For each tree, both treatments were replicated twice, resulting in 356 root bags. Lignified roots (5 mm diameter) were traced back to the target tree, and all fine lateral roots were removed. A 20 cm segment was then inserted into a nylon mesh bag (30 × 30 cm) designed with a top mesh size of 0.25 mm to allow infiltration of nutrients and water, and a bottom mesh size of 0.15 mm to prevent intrusion by roots of other species. This approach ensured that all subsequent lateral root growth within the bag represented new root proliferation in response to the experimental treatment. Following a light irrigation, a protective nylon mesh (40 cm × 40 cm, 0.15 mm openings) was placed over each root bag to function as a physical barrier. Subsequently, the native soil and litter were reapplied to cover the bags.

P fertilization commenced in June 2020, five months after root bag installation. Monthly applications of NaH₂PO₄·2H₂O solution were made for four months, delivering a total of 8 mg P per kg soil. According to research on subtropical trees by Liu et al. (2015), root growth enhancement was observed under elevated soil P conditions, specifically when plant-available P reached concentrations fourfold higher than native levels. Control patches received equivalent volumes of deionized water.

2.4. Exudate collection and harvest

After four months of treatment, root exudates were sampled in situ prior to harvest. Sampling was conducted over multiple days, grouping conspecific individuals per day. For most species, three individuals with robust root growth were selected; exceptions included Ilex formosana and Symplocos lancifolia (n = 2, see Table S1). One control and one P-rich bag per tree were randomly selected for exudate extraction.

Root exudates were collected using a syringe-based incubation system adapted from Phillips et al. (2008) with modifications to suit our experimental conditions. Forty-eight hours before sampling, each root bag was carefully opened and a segment of intact lateral roots was selected, rinsed with deionized water, and inserted into a pre-prepared 50 mL syringe. The syringe contained 80 g of sterile glass beads and 20 mL of carbon-free nutrient solution (0.5 mM NH₄NO₃, 0.1 mM KH₂PO₄, 0.2 mM K₂SO₄, 0.2 mM MgSO₄, 0.3 mM CaCl₂) to maintain ionic balance. The root insertion point was sealed with a butyl rubber septum, and the syringe was reburied and covered with native soil and litter. After 48 h of field incubation, exudates were collected by slowly injecting 50 mL of deionized water into the syringe using a pump. The eluate was filtered through 0.45 μm membranes, and 50 mL aliquots were stored at −20 ℃. Three blank samples without roots were processed simultaneously each day to account for background organic carbon.

Upon finishing exudate sampling, the lignified roots located at the bag openings were cut, and the harvested root systems were rapidly transferred to pre-labeled, sealed bags. These were then placed in insulated containers cooled with ice packs and swiftly delivered to the laboratory. Root bags from which no exudates were collected were processed following the same protocol.

2.5. Root processing

Roots were carefully cleaned with distilled water to remove adhering soil particles and debris. For each species, three individuals exhibiting vigorous root development were selected. From these trees, the root bags previously used for in situ exudate collection (one Control and one P-rich bag per tree) were subsequently utilized for measuring the RAP activity. The remaining two bags (one Control and one P-rich) were reserved for assessments of root morphology, architecture, proliferation, chemical composition, and biomass. Washed roots were sorted into absorptive (first-three orders) and transport roots. Only absorptive roots were analyzed in this study, as they are most responsive to nutrient changes and play a dominant role in nutrient acquisition (McCormack et al., 2015; Freschet and Roumet, 2017). Sub-samples were scanned at 300 dpi resolution using an Epson Perfection V370 scanner for morphological and architectural assessment. Both scanned and non-scanned root samples were packaged separately in paper envelopes, oven-dried at 65 ℃ to constant weight for biomass determination and chemical analyses. Those absorptive root specimens designated for root acid phosphatase (RAP) assay were maintained at −20 ℃ for short-term preservation. We acknowledge that this ex situ assay cannot capture rapid, protein synthesis-dependent responses and may underestimate the sustained secretory capacity of intact roots due to the lack of continuous carbon supply. However, as our goal was to compare the relative phosphatase potential among treatments rather than to quantify the absolute in situ secretion rates, this established method is appropriate and efficient for our purpose (Lugli et al., 2020; Han et al., 2022; Dallstream et al., 2023).

2.6. Root morphological, architectural, proliferation and chemical traits

Scanned root images were analyzed using WinRHIZO Pro 2009b software (Regent Instruments Inc., Canada) to quantify morphological parameters including root length, average diameter, volume, tip number, and fork count (Zhu et al., 2023). These measurements were used to calculate the SRL, root tissue density (RTD), specific root tip density (SRT), specific root fork density (SRF), root length growth rate (LGR) and root mass growth rate (MGR) (see Table S3).

For chemical analyses, dried root samples were ground to a fine powder using a ball mill. Root carbon (RC) and nitrogen (RN) concentrations were determined using an elemental analyzer (Elemental Vario EL Ⅲ, Germany) with 8–10 mg subsamples (Zhu et al., 2022). Root phosphorus (RP) concentration was measured using a modified protocol based on the forest soil total P determination method (LY/T 1232–1999). Briefly, 100 mg of ground root material was digested with 4 mL H₂SO₄ and 1 mL HClO₄ at gradual elevated temperatures until the solution turned clear. The digested solution was then diluted and analyzed using a flow analyzer (Skalar San++, Netherlands).

2.7. Mycorrhizal traits

Arbuscular mycorrhizal colonization (AMC) was assessed using a modified acid fuchsin staining method (Giovannetti and Mosse, 1980). Root samples preserved in FAA solution (formalin-aceto-alcohol; 90:5:5:5 ratio by volume of 70% ethanol, glacial acetic acid, 37% formaldehyde, and glycerin) were cleared with 10% KOH at 90 ℃ for 50–210 min (depending on species), acidified in 2% HCl, and stained with 0.01% acid fuchsin lactoglycerol solution overnight. Fifty randomly selected 1-cm root segments per sample were examined using a Leica DM4B anatomical microscope at 200× magnification. AMC was quantified following McGonigle et al. (1990) and Liu et al. (2015), representing the proportion of root length inhabited by fungal structures (hyphae, arbuscules, or vesicles).

Extraradical hyphal length (EHL) was quantified using the membrane filtration technique (Rillig et al., 1999). Fresh soil samples (4.0 g) were suspended in 100 mL deionized water with 12 mL of sodium hexametaphosphate solution (35 g L⁻¹), stirred vigorously, and allowed to settle. The supernatant was filtered through a 38-μm sieve, and the residue was resuspended and filtered onto 1.2 μm microporous membranes. Membranes were stained with 1% acid fuchsin and examined under a microscope (ICC50W, Leica) at 200× magnification. AM hyphae were distinguished from non-AM hyphae by their morphology and staining color (Miller et al., 1995): AM hyphae lack septa or have few irregular septa and are lightly colored (easily stained), whereas non-AM hyphae possess regular septa and are darkly colored (hardly stained). Extraradical hyphal density (EHD) was calculated as the EHL per unit length of absorptive roots within each root bag (m hyphae m⁻¹ root).

2.8. Root exudation and phosphatase activity

The concentration of total organic carbon (TOC) in root exudates was quantified with a TOC analyzer (TOC-L, Shimadzu). The RE was derived from the TOC released per gram of root dry mass per unit time (Sun et al., 2021; Chen et al., 2025).

The RAP was determined fluorometrically by employing 4-methylumbelliferone (MUB) as the substrate, in accordance with the protocols established by Peng and Wang (2016) and Lugli et al. (2020). Fresh absorptive root samples (approximately 30 mg) were incubated in 3.75 mL sodium acetate buffer for 30 min at 25 ℃ with shaking at 200 rpm. After clarification, 200 μL of root solution was mixed with 50 μL of MUB substrate solution (0–25 μM concentration gradient) in a 96-well black microplate. The plate was incubated in darkness at 25 ℃ for 3 h, and fluorescence was measured at excitation 365 nm/emission 450 nm using a microplate reader (Synergy H4, BioTek). RAP activity was expressed as nmol MUB released per unit root dry weight per unit time.

2.9. Statistical analyses

To satisfy the assumption of normally distributed residuals, all trait values were log₁₀-transformed. To assess potential associations with tree sizes, a linear mixed-effects model was employed, incorporating DBH as a covariate, with both species and individual trees as random effects. The analysis revealed no significant influence of DBH on any traits (P > 0.1, Table S4), and therefore, DBH was not considered in subsequent analyses. Trait measurements from individual trees were aggregated to obtain species-level means within each patch treatment. The variability in belowground traits across the 17 tree species was assessed using the coefficient of variation (CV) for each patch treatment. Differences in traits between patch types were assessed using paired sample t-tests.

To test Hypothesis 1 and Hypothesis 2, root trait networks were established through Pearson correlation coefficients using R v.4.5.1 and adopting a statistical significance threshold of P < 0.05 (Table S5) (Li et al., 2024a, Li et al., 2024b). Correlation computations were implemented with the “Hmisc” package (Harrell and Dupont, 2020). Visualization of the trait networks was generated employing the “igraph” package, wherein nodes correspond to traits (Csardi and Nepusz, 2005). Significant correlations are depicted as edges whose thickness scales with the absolute Pearson correlation coefficient values to visually represent association strength (Li et al., 2024a, Li et al., 2024b). Furthermore, the network parameters (Table 1) were quantified for the root trait networks for both Control and P-rich patches.

To assess the significance across distinct trait dimensions, both absolute and relative importance values were computed. The absolute importance refers to the average degree of a trait within its respective dimension, whereas the relative importance is calculated as this value normalized by the summed degrees across all traits (Wang et al., 2023). To compare the importance of exploration dimension traits versus chemical dimension traits across patches, independent sample t-tests were conducted.

3. Results 3.1. Interspecific variability of belowground traits

The variations in root exudation and mycorrhizal traits were greater than in most absorptive root traits (Fig. 1). Root exudation traits exhibited high variability, particularly with RAP coefficient of variation (CV) values exceeding 170% in Control patches and 140% in P-rich patches (Fig. 1b). RE exhibited a 16-fold variation in Control patches (CV = 73.84%), while a 32-fold variation was observed in P-rich patches (CV = 74.1%, Fig. 1a). EHD showed greater variability than AMC, with CVs of 94.2% in Control patches and 86.7% in P-rich patches (Fig. 1c and d). This suggests that EHD is a more sensitive indicator than AMC in capturing interspecific differences and environmental responses in mycorrhizal symbiosis.

Despite a significant increase in available soil P in P-rich patches relative to controls (P < 0.05, Fig. S2), the belowground traits of plants showed no significant differences between treatments in terms of inter-specific average values, other than a decrease in RE (Fig. 1a).

3.2. Interconnections between root exudation traits and other traits, as well as the positions of exudation traits within the root trait network

A root trait network was constructed based on the 14 traits (Fig. 2). RAP correlated negatively with RC, while EHD correlated negatively with SRT, SRF, LGR and MGR. These relationships remained consistent and were unaffected by P-rich soil patches (Fig. 2 and Table S5). The addition of inorganic P altered the covariance among traits, strengthening some correlations while weakening others. For instance, the covariation of RAP with RD and RP, as well as of RE with SRF, LGR, and EHD, was enhanced. In contrast, the covariation of RE with SRL and SRT, and of EHD with SRL and RTD, was reduced (Fig. 2). In the root trait networks, RE and RAP consistently clustered within distinct functional modules regardless of patch type (Control or P-rich) (Fig. 2). Specifically, RE belonged to the exploration trait module, which integrates morphological, architectural, and proliferation traits, whereas RAP pertained to the chemical trait module. This clear separation indicates their differential placements along the spectrum for resource acquisition strategies.

3.3. Shifts in root trait network structure

In the root trait network, the transition from Control patch to P-rich patch was characterized by a decrease in both edge density (ED) and average clustering coefficient (AC), while the diameter, average path length (AL), and modularity increased (Table 2). The top three traits in betweenness shifted from SRL, SRT, and LGR in the Control patches to SRF, RD, and RAP in the P-rich patches (Fig. 2 and Table S6). Furthermore, the absolute importance of the two trait dimensions differed significantly in Control patches, with the exploration dimension exhibiting markedly greater importance than the chemical dimension. In contrast, no significant difference was detected between the two dimensions in P-rich patches (Fig. 3). Overall, a consistent pattern was observed for the relative importance of the trait dimensions (Fig. S3).

4. Discussion

This study investigated the effects of P-rich patch treatment on plant belowground strategies by analyzing 14 traits related to absorptive roots, mycorrhizae, and root exudates within the framework of root trait networks. The main findings include: (1) Root exudation traits (RE and RAP) exhibited substantial interspecific variation and could be assigned to distinct functional modules: RE was assigned to the exploration trait module, and RAP to the chemical trait module, which supported the first hypothesis. (2) P addition reduced the strength of correlations among root traits and increased the network modularity, which was consistent with the second hypothesis. These findings are further discussed below.

4.1. Functional roles of root exudation traits in the trait network

Root exudates are critical for activating recalcitrant P pools by chelating and hydrolyzing organic P (McCormack and Iversen, 2019). In this study, RE corelated positively with morphological and/or architectural traits, as well as with proliferation traits, across both patch types (Fig. 2), consistent with findings from Meier et al. (2020) and Guyonnet et al. (2018). This pattern, however, contrasts with reports of weak or negative RE-morphology associations in other studies (Sun et al., 2021; Zheng et al., 2025). Such discrepancy suggests that the linkage between RE and root morphology is likely context-dependent, influenced by plant life form, mycorrhizal association, and environmental setting.

In our study of co-occurring AM-associated trees in a subtropical forest, the integration of RE into the exploration trait module implies that a coordinated “fast foraging” strategy, which combines rapid soil exploration via fine roots with local rhizosphere enhancement via exudates, representing a synergistic approach for acquiring heterogeneously distributed soil P. Thus, while a universal tight linkage between RE and root morphology may not hold across all plant communities, our results demonstrate its ecological relevance within the specific context of complex, mature subtropical evergreen broad-leaved forests, where intense belowground competition may favor such integrated carbon-investment strategies.

Roots with high SRL, SRT, and SRF typically exhibit smaller diameters, lower cortex-to-stele ratios, and reduced mycorrhizal colonization rates (Kong et al., 2014; Brundrett and Tedersoo, 2018; Ma et al., 2018). These findings indicate that such plants may reduce carbon investment in AM fungi while allocating more carbon to root exudation (Dallstream et al., 2023). Ultimately, this aligns with the negative correlation between RE and EHD (Fig. 2b), thereby supporting the notion that plants dynamically adjust carbon allocation to different belowground strategies under carbon limitation (Raven et al., 2018; Chen et al., 2025).

The high variability of RE (CV > 70%) and its significant reduction across species in P-rich patches (Fig. 1a) indicate that RE is strongly sensitive to carbon costs. The synthesis and the release of root exudates account for a substantial portion of photosynthates, consuming as much as one-third of total fixed carbon (Liese et al., 2018). Therefore, plants may shift from a carbon-costly exudation strategy to a more economical direct inorganic P uptake strategy when P limitation is alleviated (Raven et al., 2018; Zhu et al., 2023). In contrast, RAP showed no significant change under P-rich conditions, possibly because its response primarily targets organic P forms, or because the added P dosage did not reach its threshold of response (Allison and Vitousek, 2005; Nannipieri et al., 2011).

In contrast to RE, RAP was consistently grouped within the chemical trait module (Fig. 2), reflecting a “biochemical mining” strategy tightly linked to root nutrient metabolism and stoichiometry. RAP synthesis relies on P-rich ribosomal RNA (Elser et al., 2000), which was reflected by its positive correlation with RP (Fig. 2b). Furthermore, the consistent negative correlation between RAP and RC (Fig. 2) indicates a fundamental trade-off in belowground carbon allocation. High RC typically signifies greater investment in root structural components such as lignin, cellulose, and recalcitrant secondary metabolites, which enhance tissue longevity and defense but require substantial carbon construction costs (Guo et al., 2004; Terzaghi et al., 2013). In contrast, RAP synthesis represents a direct, high-turnover metabolic investment for organic P acquisition (Meeds et al., 2021). The observed trade-off may thus reflect a strategic divergence wherein species either prioritize carbon allocation to durable root structures (high RC) or to rapid, enzyme-mediated P mining (high RAP). No significant association was found between RAP and SRL or SRF in this study, which contrasts with the findings of Han et al. (2022). This discrepancy could arise from variations in species richness and the composition of mycorrhizal associations. All trees examined in this study were AM-associated, whereas Han et al. (2022) included four symbiotic types, which suggests that different symbiotic strategies may lead to variation in RAP association patterns (Yaffar et al., 2022).

We also found that AMC did not correlate significantly with other traits, and it was therefore excluded from the trait network (Fig. 2 and Table S5). Cohen and Havlin (2010) suggested that if a trait exhibits strong connectivity within a subnet or “module” but minimal connections with other traits in the network, it may represent an independent functional dimension. This perspective aligns with the results of Liang et al. (2023), which similarly reported no correlation between AM fungal abundance (i.e., the ratio of AM fungal DNA copy number to root tissue DNA) and RD or SRL. This shows that the previous assumption of generally low mycorrhizal colonization intensity in thin roots does not always hold. In contrast, EHD demonstrated higher variability and environmental responsiveness (Fig. 1c and d). AMC primarily reflects intraradical colonization structures with limited interspecific variation, whereas EHD directly represents the spatial extension and resource acquisition functions of hyphae (Miller et al., 1995; Smith et al., 2011). Regulated by multiple factors including fungal species, soil nutrients, and host photosynthetic carbon supply (Bever et al., 2001; Hodge, 2004; Cheng et al., 2016; Sheldrake et al., 2018), EHD is more suitable for reflecting functional differences in mycorrhizal symbiosis (van der Heijden and Scheublin, 2007; Cusack et al., 2021). Notably, most mycorrhizal and absorptive root traits, including EHD and AMC, showed no systematic response across species to the P-rich patch treatment (Fig. 1). This finding challenges the conventional view that low P availability enhances mycorrhizal dependence (Treseder, 2004; Ma et al., 2021) and suggests that P availability in the soil is not the sole factor regulating mycorrhizal symbiosis (Guilbeault-Mayers and Laliberté, 2024). Overall, host-specific strategies may play a more critical regulatory role (Zhu et al., 2023).

4.2. Changes in root trait network structure and their functional implications under P addition

Although most traits did not show systematic changes at the species level in response to P addition (Fig. 1), the root trait network exhibited structural shifts, with reduced connectivity and increased modularity in P-rich patches (Table 2). This apparent paradox can be explained by species-specific idiocratic responses to P-rich patches. As demonstrated in Zhu et al. (2023), different tree species exhibited divergent belowground strategies: some altered only one or a few traits, others showed coordinated multi-trait adjustments, while a subset remained unresponsive. Such interspecific variations in trait responsiveness, rather than uniform shifts across all species, reconfigure the correlational structure among traits at the network level. Therefore, even without consistent directional changes in trait means, the reorganization of trait associations and the emergence of more independent functional modules can drive the observed shifts in network topology.

Changes in trait network parameters can be used to quantify interdependencies among multiple traits (Flores-Moreno et al., 2019). In our study, Control patches exhibited higher ED and AC compared to P-rich patches, but shorter diameter and AL (Table 2). This indicates that Control patches have stronger trait interdependencies. High trait interdependencies enhance trait coordination and integration, thus enhancing the efficacy of resource acquisition and mobilization (Li et al., 2021; Rao et al., 2023). This pattern aligns with the expectation that resource limitation intensifies selective pressures for efficient acquisition (Reich, 2014; Liu et al., 2019). The observed tight coordination in the trait network may therefore enhance the efficacy of belowground foraging when P is scarce.

However, establishing and maintaining trait connections can be costly for plants, as such correlations rely on specific structural and physiological foundations (Flores-Moreno et al., 2019; He et al., 2020). When P availability increases, plants may minimize the costs in trait coordination by weakening overall trait correlations and increasing modularity (Table 2). This reduced covariation among root traits may also facilitate the emergence of novel trait combinations, thus enabling functional diversification and enhanced resource acquisition efficiency (Laughlin, 2014; Wang et al., 2021). Our findings support this view by demonstrating that, from Control to P-rich patches, correlations between some traits were weakened while others became strengthened (e.g., RAP-RD and RE-SRF; Fig. 2). Moreover, modular structures support functional and strategic diversity, improve adaptive flexibility under environmental fluctuations, and promote resource partitioning and the maintenance of species diversity (Kleyer et al., 2019; Li et al., 2022; Wang et al., 2022). In P-rich patches, the decreased overall connectivity (reflected by lower AC) and increased independence among traits (reflected by larger Diameter and AL) indicate a shift toward higher modularity. This modular organization not only enhances the adaptive potential of the trait network in fluctuating environments but also allows plants to adopt more flexible strategies for efficient inorganic P acquisition. The partial autonomy among modules also helps buffer the entire system against localized functional disruptions (Barabasi and Oltvai, 2004). Thus, plants exhibit stronger trait connectivity in Control patches but higher modularity in P-rich patches. These contrasting strategies represent divergent functional adaptations that likely evolved under varying nutrient selection pressures.

We observed that the top three traits in betweenness shifted from SRL, SRT, and LGR in Control patches to SRF, RD, and RAP in P-rich patches (Fig. 2 and Table S6). In Control patches, SRL, SRT, and LGR reflect a resource acquisition strategy primarily driven by morphological, architectural and proliferation traits. Despite lower soil nutrient availability, plants can achieve efficient nutrient acquisition through coordinated expression of these traits (Li et al., 2024a). The significantly higher absolute and relative importance of the exploration trait dimensions compared to the chemical trait dimension in Control patches provides further evidence for this interpretation (Figs. 3 and S3). When P availability increased, the immediate P demand was alleviated, thus enabling plants to allocate resources more evenly between exploration traits and chemical traits, resulting in a nonsignificant divergence between these two trait dimensions (Figs. 3 and S3). In P-rich patches, the trait network became more centralized, with SRF, RD, and RAP exhibiting high betweenness, indicating their critical bridging roles in maintaining connections among subnetworks. The removal of such highly connected traits could lead to network fragmentation (Burton et al., 2020; Guo et al., 2021; Li et al., 2024b; Ye et al., 2024). This increased centrality of SRF, RD and RAP suggests that under relaxed P limitation, communication between the exploration and chemical strategic dimensions becomes particularly important. These traits do not merely correlate with others; they act as integrative hubs that coordinate the otherwise decoupled modules, likely enabling plants to fine-tune the balance between soil exploration and biochemical processing in a heterogeneous resource environment. Thus, despite the increase in network modularity, a certain degree of communication and coordination among different functional modules remains necessary through these key connector traits (Kleyer et al., 2019; Li et al., 2021). This pattern may represent an adaptive shift in plant strategy from “holistic coordination” in Control patches to “modular flexibility” in P-rich patches, where strategic coherence is maintained through a few pivotal traits that bridge different functional dimensions.

The modular trait network identified in this study provides a structural basis for co-existing tree species to express diverse P acquisition strategies. In P-rich environments, various tree species will not adopt convergent adaptive strategies. Instead, originally strong trait correlations may decouple, facilitating the formation of more independently plastic and internally cohesive functional modules. For example, low-SRF species compensate for reduced root absorptive surface area through greater reliance on mycorrhizal fungi to a extend P uptake beyond the rhizosphere depletion area (Smith and Read, 2008), while high-SRF species can use relatively recalcitrant P pools by integrating root morphology/architecture-exudation traits (Keiluweit et al., 2015). Roots in P-rich patches generally exhibit more economical P acquisition approaches, since the carbon expenditure per unit of P acquired declines with increasing soil P availability (Raven et al., 2018). Tree species incur varying carbon costs for their expressed P-acquisition strategies. Consequently, their belowground foraging traits demonstrate diverse plastic responses to P-rich patches (Zhu et al., 2023). Previous reports of functional divergence in P acquisition strategies among woody plants (Batterman et al., 2018; Guilbeault-Mayers et al., 2020) indicate a capacity for complementary exploitation of distinct soil P sources, opening the possibility for greater community adaptability especially in P-limited tropical forests (Batterman et al., 2018; Dallstream and Soper, 2024). Our study provides further supporting evidence for these perspectives from the standpoint of trait modularity and response independence.

5. Limitations and implications

This study has three main limitations: (1) The P-patch treatments were only applied at two levels, and they were limited to more readily available inorganic P. This means that they were unable to capture nonlinear relationships (e.g., saturating responses) between trait responses and P availability, nor did they reflect foraging responses to more complex inorganic and organic P forms. For example, mycorrhizal colonization usually increases under low P availability but is suppressed when P is abundant or severely deficient (Wen et al., 2020). (2) The study was limited by its local scale and focus on a small number of AM tree species (17), so how the traits of other mycorrhizal types (e.g., ECM vs. AM) respond to P-rich patches remains unknown. (3) The short-term experimental duration may have been insufficient for inducing detectable root morphological modifications. Instead, rapid adjustments were observed in physiological traits. Extending the observation period and diversifying P-rich patch types, species coverage, and mycorrhizal types must be priorities for future studies.

Nevertheless, the present study integrated root exudation traits, mycorrhizal traits, and traditional absorptive root traits to construct a root trait network, revealing key mechanisms through which P availability regulates plant belowground strategies. First, it established stable associations between RE and the exploration trait module, as well as between RAP and the chemical traits module. These associations indicate that root exudation traits co-vary with specific functional dimensions across varying P environments. Second, the study uncovered a structural reorganization in the trait network, with a shift from high integration in Control patches to high modularity in P-rich patches, thus reflecting a strategic transition toward modular independence, where functional trait groups can vary more flexibly, potentially enhancing plasticity in responding to soil resource heterogeneity. Finally, the study demonstrated that most belowground traits exhibited no systematic change among species in response to soil P concentration, which goes to highlight the critical role of species-specific strategies in P acquisition. These findings deepen our understanding of belowground functional diversity and P utilization strategies in subtropical forest plants as well as offer novel perspectives on species coexistence mechanisms from the standpoint of trait modularity and functional differentiation.

Acknowledgements

This work was supported by the National Natural Science Foundation of China (grant number 32301356), and the Special Project for Guiding Science and Technology Development of Local Government by the Central Government of China (grant number 2022L3009).

CRediT authorship contribution statement

Liqin Zhu:Writing – original draft, Methodology, Investigation, Formal analysis, Data curation, Visualization, Funding acquisition. Xiaodong Yao:Resources, Investigation. Xiaohong Wang:Resources, Investigation. David Robinson:Writing – review & editing. Weile Chen:Writing – review & editing. Tingting Chen:Investigation. Qi Jiang:Investigation. Linqiao Jia:Investigation. Ailian Fan:Investigation. Guangshui Chen:Writing – review & editing, Supervision, Methodology, Investigation, Funding acquisition, Conceptualization.

Data availability

The data that support the findings of this study are available in the data storage community of Plant Diversity in the Science Data Bank (Science DB): https://doi.org/10.57760/sciencedb.j00143.00128.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.pld.2026.01.002.