Apparent pH-TAN operational breakpoint links duckweed-microbiome destabilization and nitrogen fate shift in ammonia-rich swine digestate

Anaerobic digestion is widely used for swine manure management because it reduces organic load and enables energy recovery, but the liquid digestate after digestion still contains high concentrations of ammoniacal nitrogen and other residual pollutants, requiring further polishing before discharge or reuse. Duckweed-based treatment has therefore attracted increasing attention as a low-energy post-treatment strategy because duckweeds combine rapid vegetative propagation, high nutrient uptake capacity, and the ability to convert wastewater-derived nutrients into harvestable biomass. Previous studies have demonstrated that duckweed can effectively recover nitrogen and phosphorus from swine wastewater and anaerobically treated swine effluent, supporting its potential as a nature-based platform for nutrient recovery and polishing. However, for clean-water applications, the key challenge is not only nutrient removal efficiency, but also whether treatment remains stable enough to prevent residual reactive nitrogen from persisting in the treated water. Treatment performance remains highly variable across digestate types and operating conditions, indicating that the underlying stressors and system constraints are not yet fully resolved^1,2,3.

Among the factors limiting duckweed-based remediation of digestate, ammonia toxicity is likely one of the most important. In ammonia-rich waste streams, the biologically relevant stressor is not merely total ammoniacal nitrogen (TAN), but the free-ammonia (FA, NH₃ -N) fraction, which increases with pH and temperature and can exert disproportionately strong inhibitory effects on aquatic plants. Earlier work has shown that high ammonium or ammonia exposure suppresses duckweed growth, damages photosynthetic performance, and alters carbon-nitrogen balance, while more recent evidence indicates that reducing pH can substantially alleviate duckweed growth inhibition in high-nutrient agricultural wastewater, emphasizing the importance of NH₃ -related stress rather than TAN alone^4,5. These findings imply that duckweed digestate-polishing systems may not respond linearly to ammonia-related physicochemical conditions, but may instead exhibit a threshold-like transition across a pH-dependent ammonia-speciation regime.

At the same time, duckweed-based remediation is increasingly recognized as a plant-microbiome process rather than a plant-only process. Duckweed hosts structured bacterial assemblages that share key features with terrestrial phyllosphere-associated communities, and experimental studies have shown that environmental bacterial communities can markedly enhance or suppress duckweed growth^6,7. More importantly, recent work has demonstrated that synthetic bacterial communities (SynCom) can be established for duckweed and that aquatic plant growth-promoting bacteria can improve host performance, indicating that microbiome composition is functionally relevant to duckweed productivity and stress responses^8,9,10,11,12. Under high-FA conditions, host stress, physicochemical disruption, and microbial restructuring are therefore likely to jointly determine whether nitrogen is biologically retained or remains in residual dissolved pools in the treated water.

Despite these advances, two major gaps remain in the context of swine digestate treatment. First, most previous studies have evaluated duckweed systems primarily in terms of bulk nutrient removal, whereas the effects of high FA on duckweed–microbiome interactions and microbiome assembly remain poorly understood. Second, nitrogen removal is often reported only as a decrease in aqueous nitrogen concentration, which does not distinguish whether nitrogen is assimilated into duckweed biomass, retained in microbial pools, transformed through nitrification- or denitrification-related pathways, or left as residual dissolved nitrogen and other loss fractions. This distinction is especially important under high-FA stress, where host physiological impairment and microbiome reshaping may jointly redirect nitrogen fate rather than simply lower overall removal efficiency^2,3,10. A framework that links FA stress, treatment stability, microbial nitrogen-transformation potential, and nitrogen partitioning is therefore still lacking.

In this study, we used a duckweed-based swine digestate polishing system to examine treatment stability and nitrogen fate across an experimentally imposed pH–TAN regime in which calculated FA covaried with pH. We first identified an apparent pH–TAN operational breakpoint and then compared predefined low-FA and high-FA conditions to characterize associated changes in duckweed growth and physiology, bulk-water physicochemical conditions, bacterial community diversity and composition, selected nitrogen-cycling gene abundance, and ¹⁵N-resolved nitrogen partitioning. We further tested whether a pre-constructed SynCom, Chlorella sp., or their combination could partially restore treatment performance under the high-FA operating condition. We hypothesized that the transition across the pH–TAN regime would be associated with coordinated changes in duckweed physiological performance, bulk-water physicochemical conditions, bacterial community composition, selected nitrogen-cycling gene abundance, and nitrogen partitioning, whereas the intervention treatments would partially shift these responses toward those observed under the low-FA condition. By integrating threshold analysis, host physiological responses, microbiome ecology, functional gene quantification, isotope-based nitrogen partitioning, and path-based construct analysis, this work aims to move beyond conventional nutrient-removal reporting and establish an operational framework for improving the reliability of duckweed-based treatment of ammonia-rich wastewater. Although the present study was conducted in swine digestate, the framework may also be relevant to other high-FA wastewaters in which treatment performance depends on the coupling among host stress, microbiome stability, and nitrogen transformation. Understanding how FA regulates treatment stability and nitrogen fate is therefore critical for advancing duckweed-based technologies toward reliable clean-water applications.

Identification of an apparent pH–TAN operational breakpoint

Duckweed responses across the experimental gradient showed a clear threshold-type pattern rather than a gradual linear decline. As illustrated in the experimental framework (Fig. 1a, b), Experiment I comprised seven pH-adjusted treatments with comparable TAN concentrations, corresponding to calculated FA concentrations of approximately 0.8–17.0 mg NH₃-N/L (Table 1). Four representative levels were designated as L-FA, M-FA, H-FA, and VH-FA for descriptive comparison, while the full seven-level gradient was used for breakpoint analysis. Across the lower portion of the gradient, biomass accumulation and RGR remained comparatively stable, whereas visible growth suppression and physiological deterioration emerged at higher pH and calculated FA conditions (Fig. 1c–e). Segmented regression fitted to replicate-level responses using the initial calculated FA concentration as the operational predictor supported a nonlinear transition in duckweed performance across the experimentally imposed pH–TAN gradient (Fig. 2a–d). Breakpoint estimates for RGR, Fv/Fm, MDA, and root length clustered within a similar region of the gradient, corresponding to a calculated FA concentration of approximately 8.5 mg NH₃-N/L. Because the gradient was generated primarily through pH adjustment while TAN remained comparatively similar, we define this transition as an apparent pH–TAN operational breakpoint rather than an FA-specific toxicity threshold. The synchronous shifts in growth, photosynthetic, oxidative, and root traits were consistent with a transition from a relatively stable physiological condition to a destabilized duckweed response across the tested pH–TAN regime.

This threshold is mechanistically important because the relevant stressor in ammonia-rich digestate is not simply TAN, but the un-ionized NH₃ fraction, which increases with pH and temperature and is more biologically disruptive than NH₄⁺. Previous studies have shown that duckweed can effectively recover nutrients from swine wastewater, but its performance depends strongly on wastewater chemistry and ammonia speciation. In particular, lower pH can alleviate duckweed inhibition in high-nutrient wastewater, while earlier toxicity studies demonstrated that duckweed growth declines strongly as NH₃ increases under alkaline conditions. Together, these findings indicate that pH-dependent ammonia speciation is biologically relevant to duckweed performance in ammonia-rich digestate, while the present experimental design does not permit the independent contributions of pH and FA to be separated^2,5.

The physiological pattern observed here is also consistent with ammonia-induced stress acting through coupled photochemical impairment and oxidative damage. The decline in Fv/Fm together with the rise in MDA beyond the identified operational transition (Fig. 2b, c) suggests that once FA exceeded the tolerance window, duckweed was no longer able to maintain normal photosystem II efficiency and membrane stability. Similar responses have been reported for Lemna minor exposed to elevated ammonium, where growth inhibition was accompanied by stronger oxidative stress and altered antioxidant metabolism. In this context, the breakpoint identified here likely marks the transition from metabolic adjustment to stress-dominated destabilization¹³.

Notably, root length displayed a threshold response comparable to whole-plant growth (Fig. 2a, d), which is especially relevant because duckweed roots form the immediate interface between the host and the digestate environment. Beyond the identified operational breakpoint, reduced root development coincided with conditions that may have been less favorable for maintaining the attached microbial habitat. Thus, the threshold identified here should be interpreted not only as a host growth limit, but also as the onset of duckweed system destabilization, where host performance and microbiome-supporting capacity begin to deteriorate together^9,10. Overall, these results identify an apparent pH–TAN operational breakpoint under the present digestate conditions, corresponding to a calculated FA concentration of approximately 8.5 mg NH₃-N/L. Because calculated FA covaried strongly with pH in the gradient design, this value should not be interpreted as a universal or FA-specific toxicity threshold. Instead, it marks a condition-specific transition within the experimentally imposed pH–TAN regime that was associated with duckweed physiological destabilization. Calculated FA is therefore retained as a descriptor of ammonia speciation and for defining the experimental L-FA and H-FA conditions, whereas the breakpoint itself is referred to throughout the manuscript as an apparent pH–TAN operational breakpoint.

H-FA impairs duckweed performance and alters bulk-water physicochemical conditions

H-FA caused a marked decline in duckweed performance relative to the L-FA. Biomass decreased from 0.26 to 0.14 g DW/reactor, while RGR fell from about 0.0182 to 0.0082/d (Fig. 3a, b). Root length was reduced from roughly 2.68 to 1.31 cm, and frond number declined from around 162 to 101 (Fig. 3c, d), indicating that under the H-FA condition, growth inhibition was expressed at both whole-plant and root levels. In parallel, Fv/Fm decreased from about 0.81 to 0.66 (Fig. 3e), whereas MDA increased from 5.2 to 12.0 nmol/g FW (Fig. 3f), showing that H-FA impaired photochemical efficiency and intensified oxidative damage. Antioxidant enzyme activities also shifted substantially under H-FA, with SOD, CAT, and POD all higher than in the L-FA treatment (Fig. 3g–i), suggesting activation of stress-defense responses but incomplete maintenance of physiological homeostasis. Representative phenotypes further confirmed this transition, with H-FA plants showing stronger chlorosis and poorer root development than L-FA plants (Fig. 3j). Together, these results indicate that H-FA did not merely suppress growth, but induced coordinated physiological deterioration involving reduced productivity, impaired photosynthesis, and elevated oxidative burden, consistent with previous reports of ammonium/ammonia toxicity in duckweed^4,14,15.

H-FA was also associated with marked changes in bulk-water physicochemical conditions. Throughout the incubation, bulk-water DO remained lower under H-FA than under L-FA, reaching approximately 2.7 mg/L by day 15 compared with about 3.8 mg/L under L-FA (Fig. 3k). In parallel, bulk-water pH remained higher in the H-FA reactors, increasing from approximately 7.93 to 8.16 mg/L, whereas the L-FA treatment increased only from approximately 7.00 to 7.24 mg/L (Fig. 3l). These measurements characterize reactor-scale aqueous conditions and should not be interpreted as direct measurements of the physicochemical environment at the duckweed root surface. Nevertheless, the concurrent differences in bulk-water oxygen availability and pH provide relevant environmental context for the observed changes in duckweed performance and microbial community composition. Because root-scale microsensor measurements were not performed, oxygen and pH gradients at the root–water interface remain unresolved^16,17,18. Thus, high FA likely imposed not only direct host stress but also a less favorable physicochemical context for root-associated microbial functioning.

Duckweed-surface-associated bacterial abundance decreased from approximately 8.33 to 7.60 log₁₀ 16S rRNA gene copies/g FW under H-FA (Fig. 3m). In contrast to this whole-surface molecular estimate, fluorescence microscopy specifically visualized root colonization and showed visibly reduced bacterial coverage on the root surface under H-FA (Fig. 3n). Because duckweed-associated bacteria are functionally important for host growth and stress performance, this result suggests that H-FA weakened the ability of the host to support an attached microbial community^19,20,21. Together, these results show that HN was associated with impaired duckweed performance, altered bulk-water physicochemical conditions, and lower duckweed-surface-associated bacterial abundance.

High free ammonia destabilizes treatment-associated microbiome organization

High free ammonia markedly shifted the duckweed-associated bacterial community toward a less diverse and less densely colonized state. Under H-FA, the Shannon index declined from 4.17 to 3.46, and Chao1 richness decreased from 258 to 198 (Fig. 4a, b). In the PCoA, L-FA and H-FA samples formed two clearly separated clusters with no overlap among replicates, and the first two axes explained 42.3% and 18.7% of the community variation, respectively (Fig. 4c). The clear separation indicates that high FA generated a distinct microbiome state rather than a weakened version of the L-FA community. In plant-associated microbiomes, comparable stress-driven divergence has been reported under drought and salt stress, where altered host traits and niche conditions drive selective enrichment of a new set of root-associated bacteria rather than uniform taxon loss^22,23,24.

Taxonomic profiles were consistent with this community-level shift. LN showed higher representation of Pseudomonas, Rhizobium, Hydrogenophaga, Flavobacterium, Rhodococcus, and Comamonas, whereas Bacillus, Exiguobacterium, Cyanobacteria, and Paenibacillus signatures were comparatively enriched under HN in the LEfSe analysis (Fig. 4d–f). Such treatment-associated taxonomic turnover is consistent with selective environmental and host filtering reported for plant-associated microbiomes under abiotic stress²⁵.

Taken together, these results show that H-FA was associated with marked restructuring of the duckweed-associated bacterial community, as reflected by reduced alpha diversity, clear compositional separation, taxonomic turnover, and lower root-associated bacterial abundance. The key conclusion is therefore that the high-FA condition selected a compositionally distinct and less diverse bacterial assemblage rather than that it demonstrably disrupted ecological interaction networks. From a treatment perspective, these compositional changes provide a community-level context for the accompanying changes in nitrogen-cycling gene abundance and nitrogen partitioning described below.

H-FA decouples nitrogen transformation and biological recovery pathways

In Experiment II, the measured aqueous NH₄⁺-N concentration represented in Fig. 5a decreased from approximately 56.65 to 10.50 mg N/L by day 15, whereas 39.09 mg N/L remained under HN (Fig. 5a). NO₃⁻-N reached 17.78 mg N/L under LN but only 4.50 mg N/L under HN (Fig. 5c), while final TN concentrations were 28.56 and 60.76 mg N/L, respectively (Fig. 5d). These profiles indicate that the high-FA treatment was associated with reduced conversion and removal of aqueous nitrogen. These shifts indicate that elevated FA did not simply slow TN removal, but constrained the coupled conversion of ammonium into downstream oxidized and assimilated N pools, thereby retaining more reactive N in the water phase. This interpretation is consistent with previous evidence that duckweed can efficiently recover ammonium-rich nitrogen from swine wastewater under favorable conditions, whereas excessive ammonium/ammonia stress impairs plant carbon–nitrogen metabolism, and previous studies have shown that FA can inhibit nitrifying microorganisms, including nitrite oxidizers; however, nitrite-oxidation-specific genes were not quantified in the present study. Accordingly, the observed NO₂⁻-N and NO₃⁻-N patterns are interpreted as treatment-level changes in aqueous nitrogen dynamics rather than as direct measurements of nitrite-oxidation activity^14,26,27.

Consistent with the aqueous-N profiles, the 16S-normalized DNA-level abundances of several measured nitrogen-cycling genes were lower under HN than under LN (Fig. 5e–i). The 16S-normalized relative representation of amoA, nirS/nirK, and nosZ within the bacterial DNA pool decreased under HN, whereas ureC remained comparatively less responsive. These qPCR data indicate treatment-associated differences in the genetic potential represented by the targeted N-cycling genes and should not be interpreted as direct measurements of transcriptional activity or process rates. Although all assays are reported on the same 16S-normalized copy-number basis, treatment effects were interpreted primarily within each gene because primer-specific amplification characteristics and biological gene-copy variation can influence direct comparisons among different functional genes. This interpretation is consistent with evidence that FA suppresses microorganisms involved in biological nitrogen removal and that the composition and abundance of N-cycling guilds are closely linked to ecosystem functioning^28,29.

The ¹⁵N tracer analysis provided direct evidence that tracer-N partitioning among the three directly measured compartments differed markedly between LN and HN (Fig. 5i and Table 2). Under LN, 37.0% of the initial tracer N was recovered in the plant-associated pool and 15.5% in the microbial-associated pool, whereas 29.0% remained in the residual dissolved pool. Under HN, plant-associated recovery decreased to 14.5% and microbial-associated recovery to 8.5%, while the residual dissolved fraction increased to 57.0%. Thus, HN was associated with a clear redistribution of tracer N away from biological retention in plant- and microbial-associated pools toward persistence in the residual dissolved pool^30,31. Overall measured tracer-N recovery was comparable among treatments (79.0%–81.5%; P > 0.05), indicating similar mass-balance closure across treatments. Because gas-phase ¹⁵NH₃, ¹⁵N₂, and ¹⁵N₂O were not directly quantified, interpretation was restricted to the three directly measured tracer-N pools.

Bacterial–microalgal bioaugmentation improves treatment resilience under H-FA

Functional bacterial and microalgae-containing interventions partially improved the HN-impaired duckweed treatment system, with the combined HN + B + M treatment consistently showing the strongest recovery under the present operating configuration. Under HN, duckweed biomass remained at only 0.15 g DW/reactor, whereas HN + B and HN + M increased it to 0.19 g DW/reactor, and HN + B + M further restored it to 0.23 g DW/reactor (Fig. 6a). A similar trend was observed for photosynthetic performance and oxidative stress, with Fv/Fm rising from 0.67 under HN to 0.79 in HN + B + M and MDA declining from 11.7 to 5.9 nmol/g FW (Fig. 6b, c). In parallel, bulk-water DO increased, and pH shifted downward toward the LN range in the algal treatments (Fig. 6d). Because the Chlorella-containing reactors were gently aerated, these physicochemical improvements should be interpreted as the combined outcome of algal presence and the associated aerated operating condition rather than as a microalgae-only effect. These changes indicate that the interventions did not simply stimulate stressed fronds; they rebuilt the physicochemical envelope in which duckweed and its associated microbiota interact. In duckweed, host benefit can emerge from interactions among microbiome members rather than from single strains in isolation, making community reassembly itself a functional trait of the rescue process^32,33.

The intervention-associated shift extended to bacterial community composition. In the PCoA, HN + B and HN + M occupied intermediate positions between HN and LN, whereas HN + B + M was positioned closest to the LN cluster (Fig. 6e). The taxa heatmap showed a similar tendency, with several LN-associated genera, including Pseudomonas, Rhizobium, Hydrogenophaga, Flavobacterium, and Comamonas, showing partial recovery in the intervention treatments, particularly under HN + B + M (Fig. 6f). These results indicate partial recovery of community composition under the intervention treatments, but do not by themselves demonstrate restoration of microbial interaction networks or ecological stability. Thus, the rescue process did not simply improve duckweed growth under stress, but reassembled a more connected and potentially more functional microbiome. Recent work in plant microbiomes has shown that stress-alleviating communities are assembled through joint host-environment filtering rather than passive recolonization, and that root-network complexity tracks the abundance of keystone taxa and ecological functioning^34,35. The movement from the HN network toward the LN-like network is best read as recovery of community organization, not merely recovery of taxon richness.

Importantly, this structural recovery translated into partial restoration of nitrogen-processing capacity. Relative to HN, all rescue treatments improved inorganic-N conversion and N allocation, with HN + B + M again showing the strongest effect across Figs. 5 and 6h. By day 15, residual NH₄⁺-N decreased from 39 mg/L under HN to 11 mg/L in HN + B + M, while TN decreased from 60.5 to 30.0 mg/L (Fig. 5a, d). In parallel, the 16S-normalized DNA-level abundances of several selected nitrogen-cycling genes, including amoA and nosZ, increased under HN + B + M relative to HN (Fig. 5e–i), consistent with partial recovery of the measured nitrogen-cycling genetic potential. In parallel, plant-associated tracer-N recovery increased from 14.5% under HN to 32.0% under HN + B + M, microbial-associated recovery increased from 8.5% to 16.5%, and the residual dissolved fraction declined from 57.0% to 32.5% (Fig. 5i and Table 2). Together, these results show that the intervention treatments partially restored biological tracer-N retention while reducing the proportion remaining in the residual dissolved pool, with HN + B + M showing the largest shift toward the LN partitioning pattern. This view is consistent with evidence that plant-microbe interactions regulate dissolved-N removal and denitrification-related functions in vegetated treatment systems, and that algal-bacterial consortia enhance ammonium and TN conversion through complementary roles of algal assimilation and bacterial denitrification^36,37.

Among the three interventions, HN + B + M showed the greatest overall recovery across several measured endpoints, including duckweed performance, bacterial community composition, selected nitrogen-cycling gene abundance, aqueous nitrogen concentrations, and tracer-N partitioning. However, the present design was not a fully factorial experiment for testing statistical interaction between bacterial and microalgal interventions, and the microalgae-containing treatments were additionally associated with gentle aeration. Therefore, the stronger performance of HN + B + M cannot by itself be interpreted as evidence of synergy or functional complementarity between the bacterial consortium and Chlorella. We instead interpret HN + B + M as the combined intervention configuration that produced the response pattern most similar to the LN treatment under the conditions tested. A similar mechanistic direction has been reported in macrophyte-based treatment systems, where plant-associated oxygen diffusion and rhizosphere retention alter nutrient transformation, and in algal-bacterial systems, where consortium-level interactions improve nitrogen conversion beyond single-component systems^38,39.

Integrated interpretation of treatment-associated responses

Across the measured endpoints, the HN treatment consistently differed from LN at multiple organizational levels. HN was associated with reduced duckweed growth and photosynthetic performance, greater oxidative stress, lower bulk-water DO and higher pH, reduced bacterial alpha diversity and marked shifts in community composition, lower 16S-normalized DNA-level abundance of several selected nitrogen-cycling genes, and redistribution of tracer N away from plant-associated and microbial-associated pools toward the residual dissolved pool. Together, these observations indicate coordinated treatment-associated changes across duckweed physiology, reactor-scale physicochemical conditions, bacterial community composition, selected nitrogen-cycling functional potential, and nitrogen partitioning. Such coordination is ecologically plausible because plant physiological status and environmental stress can influence the recruitment and composition of associated microbiomes, as demonstrated in recent studies linking host developmental and stress responses with microbiome assembly^25,40. However, the concurrent responses observed here should not be interpreted as demonstrating a fixed directional cascade among the measured components. Most microbiome, qPCR, and isotope measurements were obtained at terminal sampling points, and the experimental replication was not designed to resolve temporal causality among host, environmental, microbial, and nitrogen-fate responses.

The microbial and nitrogen data provide complementary but distinct levels of evidence for the HN-associated response pattern. Reduced alpha diversity, clear PCoA separation, and taxonomic turnover support restructuring of bacterial community composition under HN, but do not demonstrate changes in ecological interaction strength or network stability. In parallel, the lower DNA-level abundance of amoA, nirS/nirK, and nosZ indicates reduced genetic potential associated with the selected ammonia-oxidation- and denitrification-related functions examined in this study, rather than direct suppression of transcriptional activity or process rates. Previous studies have shown that elevated FA can inhibit nitrifying microorganisms, including nitrite oxidizers²⁷; however, nitrite-oxidation-specific genes were not quantified here, and the present results therefore do not directly resolve nitrite-oxidation activity. The molecular patterns instead coincided with slower NH₄⁺-N depletion, lower NO₃⁻-N accumulation, and higher residual TN under HN. Independently, the ¹⁵N tracer experiment showed that plant-associated tracer-N recovery decreased from 37.0% under LN to 14.5% under HN and microbial-associated recovery decreased from 15.5% to 8.5%, whereas the residual dissolved tracer-N fraction increased from 29.0% to 57.0%. Thus, the strongest isotope-based evidence for altered nitrogen handling under HN is the significant redistribution of tracer N among the three directly measured plant-associated, microbial-associated, and residual dissolved pools, rather than inference regarding unmeasured nitrogen-loss pathways.

The intervention treatments showed partial movement toward the LN response pattern across several independent measurements. Relative to HN, the interventions improved duckweed performance, shifted bacterial community composition, partially increased the abundance of several selected nitrogen-cycling genes, reduced residual aqueous nitrogen, and increased biological tracer-N retention. HN + B + M showed the largest overall recovery across these endpoints and produced the treatment profile most similar to LN. Similar system-level responses are biologically plausible because host-associated microbial communities can contribute to aquatic-plant performance through community-level interactions, while algal–bacterial systems can jointly influence oxygen availability, nutrient assimilation, and nitrogen conversion^30,32,33. Nevertheless, the stronger performance of HN + B + M in the present study does not by itself demonstrate synergy or functional complementarity between the bacterial consortium and Chlorella. The experiment was not designed as a fully factorial interaction test, and the microalgae-containing treatments were gently aerated, whereas non-algal treatments were manually mixed. The HN + B + M response should therefore be interpreted as recovery under the combined bacterial-microalgal, aerated treatment configuration rather than as evidence for a defined synergistic mechanism.

Taken together, the results can be summarized as three experimentally observed response patterns: a low-FA reference condition, an H-FA-associated altered condition, and an intervention-associated partial-recovery condition. Under LN, duckweed performance and biological tracer-N retention were comparatively high, bacterial diversity was greater, and less tracer N remained in the dissolved phase. HN was associated with impaired duckweed physiology, altered bulk-water DO and pH, bacterial community restructuring, lower abundance of several selected nitrogen-cycling genes, and greater persistence of tracer N in the residual dissolved pool. The intervention treatments partially shifted several of these endpoints toward the LN pattern. This synthesis integrates independently measured host, physicochemical, microbial, and nitrogen-partitioning responses without assigning a statistically demonstrated causal order among them. Previous studies showing that plant physiological status can shape associated microbial communities and that plant-, algal-, and bacterial-mediated processes jointly affect nitrogen transformation in aquatic systems provide an ecological basis for considering these responses together^25,32, but the directionality and strength of such coupling remain unresolved in the present experiment.

From an engineering and clean-water treatment perspective, the apparent pH–TAN operational breakpoint identified here provides a practical indicator of declining duckweed-system performance under comparable digestate conditions. Maintaining the system below the apparent destabilization range through approaches such as pH management, dilution, staged loading, or upstream ammonia recovery may help preserve duckweed performance and limit the accumulation of residual dissolved nitrogen. This operational interpretation is consistent with the broader sensitivity of ammonia-rich biological treatment processes to the FA fraction, particularly where nitrifying microorganisms are involved²⁷. When elevated-FA episodes occur, biological intervention may provide an additional stabilization option, although the individual contributions of bacterial inoculation, microalgae, and aeration require further separation. Accordingly, the main practical implication of this study is not that FA alone determines treatment outcome, but that the combined pH–TAN operating regime, together with calculated FA as an ammonia-speciation descriptor, can provide useful warning information for declining biological performance and altered nitrogen partitioning in ammonia-rich digestate polishing.

Several limitations define the scope of this interpretation. First, calculated FA covaried strongly with pH across the experimental gradient; therefore, the observed transition is referred to as an apparent pH–TAN operational breakpoint rather than a pH-independent FA toxicity threshold. The corresponding calculated FA concentration should be interpreted only as the location of this operational transition along the experimentally imposed ammonia-speciation gradient. Factorial experiments independently varying TAN and pH will be required to distinguish their individual and interactive effects. Second, bulk-water DO and pH characterize reactor-scale aqueous conditions and were not direct measurements of physicochemical gradients at the duckweed root surface. Root-scale microsensor measurements will therefore be necessary to determine whether and how local O₂ and pH conditions contribute to the observed host and microbial responses. Third, because Chlorella-containing treatments were gently aerated whereas non-algal treatments were manually mixed, the present experiment cannot fully separate microalgal effects from aeration-associated changes in DO, pH, ammonia volatilization, or nitrogen transformation. Future experiments should include aerated non-algal controls and identical mixing regimes across treatments. Finally, the ¹⁵N experiment directly quantified tracer N only in the plant-associated, microbial-associated, and residual dissolved pools. Because gaseous ¹⁵N species were not measured, the isotope-based interpretation was restricted to treatment differences among these three measured compartments and does not resolve specific gaseous or other unmeasured nitrogen-loss pathways.

In summary, this study identifies an apparent pH–TAN operational breakpoint associated with the stability of duckweed-based treatment of ammonia-rich swine digestate and the partitioning of reactive nitrogen in the treated water. This transition corresponded to a calculated FA concentration of approximately 8.5 mg NH₃-N/L under the tested conditions but should not be interpreted as an FA-specific toxicity threshold. Beyond this operational transition, the system showed impaired duckweed performance, altered bulk-water physicochemical conditions, restructuring of the bacterial community, lower abundance of several selected nitrogen-cycling genes, and redistribution of tracer N away from plant-associated and microbial-associated pools toward the residual dissolved pool. The intervention treatments partially shifted several of these responses toward the low-FA reference condition, with HN + B + M showing the greatest overall recovery under the tested operating configuration (Fig. 7). These findings support an integrated operational interpretation of pH–TAN-associated treatment instability and recovery, while the causal contributions of pH, ammonia speciation, host physiology, microbial responses, and nitrogen fate remain to be resolved experimentally.

Digestate collection and pretreatment

Solid-liquid separated swine digestate effluent was collected from a commercial pig farm in Yantai, Shandong Province, China. The manure treatment process at the farm consisted of pig house flushing collection, mechanical solid-liquid separation, anaerobic digestion, and digestate storage. The liquid fraction after anaerobic digestion was used in this study. Solid-liquid separation was conducted using a screw-press separator (LX-180, China). Digestate samples were collected from the outlet storage tank, transported to the laboratory in sterile polyethylene containers within 4 h, and stored at 4 °C. All experiments were initiated within 48 h of sampling.

Upon arrival, the digestate was allowed to settle for 24 h at room temperature, and the supernatant was passed through a 100-mesh nylon filter to remove residual particles. The pH was adjusted to 7.0 ± 0.1 with 1 mol/L HCl or NaOH before use. For sterile treatments, the pretreated digestate was autoclaved at 121 °C for 20 min, followed by sterility verification on nutrient agar plates.

The raw digestate was characterized prior to preparation of the experiment-specific working media. The measured characteristics of the raw digestate were pH 7.84 ± 0.06, EC 6.72 ± 0.18 mS/cm, NH₄⁺-N 412.6 ± 18.4 mg/L, TN 587.3 ± 21.7 mg/L, TP 68.4 ± 3.9 mg/L, and COD 2145.0 ± 96.3 mg/L. Working digestate was subsequently prepared separately for Experiments I and II according to their respective experimental designs, as described below.

Duckweed, microalga, and bacterial rescue inocula

Healthy fronds of Spirodela polyrhiza were obtained from a laboratory stock culture originally collected from Guangzhou, China. Before the experiment, fronds were rinsed thoroughly with deionized water and acclimated for 7 d in diluted Hoagland nutrient solution under controlled conditions. Uniform, visually healthy fronds without chlorosis or root damage were selected for all experiments.

Chlorella sp. was obtained from the Institute of Nanfan and Seed Industry, Guangdong Academy of Sciences stock and cultivated in BG11 medium at 25 °C under a 16:8 h light/dark cycle. Cells in the exponential phase were harvested by centrifugation at 5000 × g for 10 min, washed twice with sterile deionized water, and resuspended before inoculation. The initial algal dosage was set at 0.10 g/L dry cell weight (DCW).

The bacterial rescue inoculum used in this study was a pre-constructed five-member synthetic consortium (SynCom) selected before the present experiment on the basis of biosafety, high-FA tolerance, complementary plant growth-promoting functions, and potential roles in nitrogen transformation and microbiome restoration. The SynCom comprised Bacillus velezensis, Sphingomonas sp., Paenibacillus sp., Priestia megaterium, and Rhodococcus erythropolis. The strain-screening, compatibility evaluation, and functional prevalidation used to assemble this SynCom were completed prior to this study and were therefore not included as experimental components of the present work. For inoculum preparation, each strain was cultured separately in nutrient broth at 30 °C and 180 rpm to logarithmic phase, collected by centrifugation (6000 × g, 10 min), washed twice with sterile saline, and resuspended to OD₆₀₀ = 0.8–1.0 (approximately 10⁸ CFU/mL). Equal volumes of the five component strains were mixed immediately before use, and the SynCom was added to rescue reactors at a final density of 10⁶ CFU/ mL.

Free-ammonia calculation, total ammoniacal nitrogen (TAN) was defined as the sum of ionized ammonium-N (NH₄⁺-N) and un-ionized ammonia-N (NH₃-N), expressed on a nitrogen-mass basis. Free ammonia (FA, NH₃-N) was calculated from TAN, pH, and temperature according to the ammonium–ammonia equilibrium:

where T is the absolute temperature (K). TAN, pH, and temperature were measured at each sampling time, and FA was recalculated accordingly. Throughout the manuscript, TAN and individual dissolved inorganic nitrogen species (NH₄⁺-N, NH₃-N, NO₂⁻-N, and NO₃⁻-N) are reported on an N-mass basis. TAN is reported as mg N L⁻¹, whereas FA is reported as mg NH3-N L⁻¹, which denotes the mass concentration of nitrogen present as un-ionized NH3.

Experimental design and reactor operation

Experiment I: FA-gradient setup and breakpoint estimation. Experiment I was designed to characterize duckweed responses across an increasing FA gradient and to estimate the apparent FA-associated breakpoint under the tested pH–TAN conditions. For Experiment I, a dedicated working digestate was prepared from the pretreated raw digestate before establishment of the FA gradient. The resulting working medium had an initial TAN of approximately 142 mg N/L, with the detailed measured chemistry given in Table 1. The seven treatments were prepared from this common working medium, and the FA gradient was then generated primarily by pH adjustment while maintaining comparable TAN concentrations across treatments. Thus, the approximately 142 mg N/L values reported in Table 1 refer specifically to the initial TAN of Experiment I and should not be interpreted as the NH₄⁺-N concentration of the separate Experiment II reactors. Seven gradient treatments (G1–G7) were established using the same batch of digestate, with initial calculated FA concentrations of approximately 0.8, 2.0, 4.0, 6.5, 9.0, 13.0, and 17.0 mg NH₃-N/L, respectively (Table 1). For descriptive purposes, representative levels within this gradient were designated as low FA (L-FA; G1), moderate FA (M-FA; G3), high FA (H-FA; G5), and very high FA (VH-FA; G7), while G2, G4, and G6 served as intermediate gradient levels to improve resolution of the FA-response relationship. The FA gradient was generated primarily by pH adjustment while maintaining comparable TAN concentrations and otherwise consistent temperature, light regime, and duckweed inoculum among treatments. Experiment I was operated for 10 d and was used to characterize nonlinear duckweed responses and estimate the apparent transition in growth and physiological performance across the tested FA gradient.

Experiment II: intervention design. Based on the breakpoint identified in Experiment I, Experiment II compared one permissive L-FA condition and one inhibitory H-FA condition and then evaluated bacterial and microalgae-containing interventions under high FA. Five treatments were established: LN (L-FA + duckweed), HN (H-FA + duckweed), HN + B (H-FA + duckweed + SynCom), HN + M (H-FA + duckweed + Chlorella sp.), and HN + B + M (H-FA + duckweed + SynCom + Chlorella sp.). Because Chlorella-containing reactors were gently aerated to prevent algal sedimentation and maintain gas exchange, while non-algal reactors were manually mixed, the effects of HN + M and HN + B + M were interpreted as microalgae-containing aerated interventions rather than microalgae-only effects. This experiment was therefore designed to evaluate whether these biological intervention configurations could improve duckweed performance, microbiome organization, nitrogen-transformation potential, and residual dissolved nitrogen under H-FA stress, while recognizing aeration as a co-occurring operational factor in the algal treatments.

Experiment II was prepared separately from Experiment I using an experiment-specific working digestate. Accordingly, the aqueous nitrogen concentrations shown in Fig. 5 represent measurements from the Experiment II reactors and should not be directly compared with the TAN concentrations of the Experiment I FA-gradient treatments reported in Table 1. All HN-based intervention treatments were established from the same HN working medium before addition of the corresponding biological interventions.

Reactor operation and sampling

Both experiments were conducted in transparent glass reactors containing 2.0 L of working solution. Duckweed was inoculated at 2.0 g fresh weight per reactor. For microalgal treatments, Chlorella sp. was added at 0.10 g/L dry cell equivalent. For bacterial rescue treatments, the SynCom was added at 10⁶ CFU/mL. Reactors were maintained at 25 ± 2 °C under cool-white LED illumination with a 16:8 h light/dark cycle and a light intensity of 120 μmol photons/ m² s.

Microalgal treatments were gently aerated through sterile air stones to prevent sedimentation and maintain gas exchange, whereas the other reactors were gently mixed twice daily by hand. Water samples were collected on days 0, 2, 4, 6, 8, and 10 in Experiment I and on days 0, 3, 6, 9, 12, and 15 in Experiment II. Unless otherwise stated, terminal samples collected on day 10 (Experiment I) and day 15 (Experiment II) were used for microbial community, qPCR, and isotope-based nitrogen fate analyses.

Bulk-water physicochemical analyses

At each sampling time, pH, temperature, and DO were measured immediately using a multi-parameter probe (HQ40d, Hach, USA). NH₄⁺-N was determined by the indophenol blue method, NO₂⁻-N by N-(1-naphthyl)-ethylenediamine spectrophotometry, NO₃⁻-N by ultraviolet spectrophotometry, and TN and TP after alkaline persulfate digestion. COD was measured by the potassium dichromate method according to standard methods.

Bulk-water DO and pH were monitored to characterize reactor-scale physicochemical conditions during incubation. These measurements represent conditions in the bulk aqueous phase and were not interpreted as direct measurements of oxygen or pH at the duckweed root surface. Root-associated bacterial abundance and colonization were assessed separately as biological characteristics of the root-associated microbial community. Because microsensor measurements at the root-water interface were not performed, no direct inference regarding root-scale O₂ or pH gradients was made.

Duckweed growth and physiological analyses

Duckweed biomass was determined gravimetrically. Fresh biomass was recorded after gently blotting fronds with filter paper, and dry biomass was obtained after drying at 65 °C to constant weight. Relative growth rate (RGR) was calculated as:

where W₀ and W_t are the initial and final dry biomass, respectively, and t is the incubation time (d). Frond number was counted manually, and root length was measured from 20 randomly selected fronds per replicate using a digital caliper. Representative phenotypes were documented photographically.

For physiological assays, duckweed tissues were collected at the designated terminal time points, immediately frozen in liquid nitrogen, and stored at −80 °C. Maximum quantum yield of photosystem II (Fv/Fm) was measured using a PAM fluorometer (PAM-2500, Walz, Germany) after 20 min of dark adaptation. Oxidative stress and antioxidant responses were evaluated using malondialdehyde (MDA), superoxide dismutase (SOD), catalase (CAT), and peroxidase (POD). Frozen tissue (0.2 g) was homogenized in 50 mmol/L phosphate buffer (pH 7.8) and centrifuged (12,000 × g, 20 min, 4 °C). MDA was quantified by thiobarbituric acid reaction, SOD by nitro blue tetrazolium inhibition, CAT by H₂O₂ decomposition at 240 nm, and POD by guaiacol oxidation.

Duckweed-associated bacterial analyses

Duckweed-associated microorganisms were collected from the combined frond and root surfaces at the terminal sampling points. Fresh duckweed was immersed in sterile phosphate-buffered saline (PBS) and subjected to mild sonication (40 kHz, 2 min) to detach microorganisms from the frond and root surfaces. The resulting suspension was centrifuged, and the microbial pellet was used for DNA extraction and subsequent molecular analyses. Thus, unless otherwise specified, the term “duckweed-associated microbiota” in the microbial community and gene-abundance analyses refers to microorganisms recovered collectively from both frond and root surfaces rather than from roots alone.

Total duckweed-surface-associated bacterial abundance was quantified by qPCR targeting the bacterial 16S rRNA gene. In parallel, bacterial colonization specifically on duckweed roots was visualized by fluorescence microscopy after staining fresh root samples with SYTO 9 (Invitrogen, USA). Images were recorded under identical exposure settings and were used qualitatively to compare root-surface bacterial colonization among treatments.

Microbial community profiling

Microbial DNA was extracted using the DNeasy PowerWater Kit (QIAGEN, Germany) according to the manufacturer’s instructions. The V3–V4 region of the bacterial 16S rRNA gene was amplified using primers 338F/806R and sequenced on an Illumina MiSeq PE300 platform (Majorbio Bio-Pharm Technology Co., Ltd., Shanghai, China). Raw reads were processed in QIIME2, and taxonomic assignment was performed against the SILVA database.

Alpha diversity indices, including Shannon diversity and Chao1 richness, were calculated in QIIME2. Beta diversity was evaluated by principal coordinates analysis (PCoA) based on Bray–Curtis dissimilarity, and treatment differences in community composition were tested by PERMANOVA. Relative abundance profiles at the phylum and genus levels were used to characterize compositional shifts, and differentially represented taxa were identified by LEfSe using an LDA threshold of 2.0. These analyses were used to evaluate treatment-associated changes in bacterial diversity and community composition without inferring ecological interaction strength or network stability from the limited number of replicates.

Quantification of nitrogen-cycling genes

To evaluate the DNA-level genetic potential of selected nitrogen-cycling functions, qPCR was performed for amoA, nirS, nirK, nosZ, and ureC using duckweed-associated microbial DNA, and the primer sequences and annealing temperatures are listed in Table S1. Each 20 μL reaction contained 10 μL SYBR Green Master Mix, 0.4 μL of each primer (10 μmol/L), 1 μL template DNA, and 8.2 μL nuclease-free water. Amplification was conducted on a CFX96 Touch Real-Time PCR Detection System (Bio-Rad, USA) with an initial denaturation at 95 °C for 3 min, followed by 40 cycles of 95 °C for 15 s, primer-specific annealing for 30 s, and 72 °C for 30 s. Standard curves were generated using serial dilutions of plasmids carrying the corresponding target fragments. Standard curves were generated using serial dilutions of plasmids containing the corresponding target fragments and fitted as C_t = m × log10(copy number) + b. Amplification efficiency was calculated as E = [10^−1/m − 1] × 100%. The standard-curve equation, coefficient of determination (R²), amplification efficiency, linear quantitative range, and lower quantitative limit for each assay are provided in Table S2. Sample copy numbers were obtained by interpolation from the corresponding standard curve without extrapolation beyond the validated quantitative range.

Because total duckweed-associated bacterial abundance differed among treatments, normalization to bacterial 16S rRNA gene copy number was used to express the abundance of each target nitrogen-cycling gene relative to the size of the bacterial DNA pool in the same sample. This normalization was selected specifically to reduce the influence of treatment-associated differences in total bacterial abundance and to distinguish changes in the relative representation of functional genes within the bacterial community from changes driven simply by differences in overall bacterial biomass. It was therefore used for a study-specific analytical purpose rather than solely because 16S rRNA gene normalization is commonly applied in the literature. Target-gene abundances are reported as copies per 10⁶ bacterial 16S rRNA gene copies, with the target gene and 16S rRNA gene quantified from the same DNA extract. Because bacterial taxa differ in 16S rRNA operon copy number, these normalized values are interpreted as relative representation within the bacterial DNA pool rather than exact cell-normalized abundances. Furthermore, because the measurements were DNA-based rather than RNA-based, they represent DNA-level genetic potential and should not be interpreted as transcriptional activity or process rates.

¹⁵N tracer and nitrogen partitioning. A parallel ¹⁵N tracer experiment was conducted to quantify the distribution of ammonium-derived N among the major measurable nitrogen pools under the L-FA, H-FA, and intervention conditions. Five treatments were included: LN, HN, HN + B, HN + M, and HN + B + M, with three biological replicates per treatment. ¹⁵NH₄Cl was added as the labeled nitrogen source while maintaining the same TAN background as in the corresponding unlabeled reactors. The initial ammonium pool was enriched to approximately 10 atom% ¹⁵N, corresponding to an initial tracer-N input of 20.0 mg N/reactor for mass-balance calculations.

At the end of the incubation, tracer-derived N was quantified in three directly measured compartments: duckweed biomass, microbial biomass recovered from root-associated and suspended material, and the residual dissolved N pool. Duckweed tissues were thoroughly rinsed, dried, weighed, homogenized, and analyzed for total N and ¹⁵N enrichment by elemental analyzer–isotope ratio mass spectrometry (EA–IRMS). Microbial-associated N was recovered by gently detaching root-associated and suspended biofilm material and collecting the resulting biomass on pre-combusted glass-fiber filters. The remaining filtered liquid phase was used for determination of residual dissolved tracer N after the corresponding chemical conversion procedure.

Isotopic enrichment was corrected for background ¹⁵N abundance and expressed as atom% excess (APE):

where atom%¹⁵N_sample is the measured isotopic abundance of a given N pool and atom%¹⁵N_background is the corresponding background isotopic abundance determined from unlabeled material. Tracer-derived N recovered in each compartment was calculated as:

where N_total,pool is the total N mass of the corresponding plant, microbial, or dissolved pool; atom%¹⁵N_pool is its measured isotopic abundance; atom%¹⁵N_initial is the isotopic enrichment of the initial labeled ammonium pool; and atom%¹⁵N_background is the background abundance used for isotope correction.

Tracer-N partitioning for each measured compartment was expressed relative to the initial tracer-N input:

Measured tracer-N recovery was calculated as the sum of tracer N recovered in the three directly quantified compartments:

Measured recovery was used to evaluate overall mass-balance closure. Any difference between the initial tracer-N input and the sum of the three directly measured pools was treated only as an unassigned mass-balance residual and was not analyzed as a separate nitrogen-fate compartment. All tracer-N partitioning and recovery calculations were performed independently for each biological replicate before treatment means and standard deviations were calculated.

Statistical analysis

All experiments were conducted with three biological replicates, and results are presented as mean ± standard deviation unless otherwise stated. One-way ANOVA followed by Tukey’s post hoc test was used to compare treatment means at individual time points. Two-way ANOVA was used for repeated measurements involving treatment and time.

Threshold responses in Experiment I were evaluated using one-breakpoint segmented linear regression, with the initial calculated FA concentration of each gradient treatment used as the predictor variable. For each response variable, the segmented model was expressed as:

where X represents initial calculated FA, ψ is the estimated breakpoint, and (X−ψ)+ equals zero when X≤ψ and X−ψ when X>ψ. Breakpoint estimates were obtained from the fitted segmented models. Because FA covaried with pH across the experimental gradient, the estimated breakpoint was interpreted as a condition-specific FA-associated transition rather than a pH-independent FA toxicity threshold.

Microbial beta diversity was analyzed by Bray–Curtis-based PCoA and tested using PERMANOVA, and LEfSe was used to identify discriminative taxa. All remaining conventional statistical analyses were conducted in SPSS 25.0 or R as specified above. Differences were considered significant at P < 0.05.