Supported by the National Natural Science Foundation of China’s Central South China Sea Basin Scientific Expedition Research Shared Voyage (Voyage No.: KK2102-2), from July 27 to August 20, 2021, seawater samples at vertical depths of five stations in the central SCS basin were collected along the 113^°E meridian for denitrification incubating experiments and determinations of bio-chemical parameters ( Figure 1 ). Seawaters were sampled with a Seabird SBE-911 plus conductivity-temperature-density ( Figure 2A ) and Rosette system. Samples for analysis of DO, dissolved inorganic nitrogen (DIN), SPM and POC were collected from surface to 1500 m layers. Determinations of denitrification potentials and the abundance of functional genes were targeted from 300 m to 1500 m range ( Table 1 ), the depths of which covering the typical DO minimum zones in the SCS basin (Li and Qu, 2006; Liu et al., 2011).

In order to facilitate the following discussion, the sampling water columns from surface to 1500 m were divided into three layers in our study according to literature definition, i.e., surface waters (0~100 m), subsurface waters (100~600 m) and intermediate waters (600~1500 m) (Gong et al., 1992; Li and Qu, 2006; Chou et al., 2007).

The collection of DO was directly transferred from rosette-sampler to a 120 mL grinding glass bottle and then sealed and stored after overflowing 3 times the volume, with the concentration being measured onboard by the standard Winkler method (Hansen and Koroleff, 1999). 3~5 L of seawater was filtered with a pre-combusted (450°C, 4 h) and weighted glass fiber membrane (Advantec GF75, 0.3 μm pore size). SPM and POC were collected from each sampling layer, and the filters were frozen at −20°C. Back to land-based laboratory, the membranes were dried and weighed at 60°C to obtain the SPM content. After removing the inorganic carbon by fumigation with concentrated hydrochloric acid, the POC content was determined by an elemental analyzer (FlashSmart NCS, Thermo Scientific), with the detection limit of 0.1 μmol C of this instrument. While the filtrate of each layer seawater was collected into an acid-cleaned and sample rinsed high-density polyethylene bottle (Nalgene) and stored frozen (−20°C). After returning to land laboratory, the concentrations of nitrate (NO₃-N) and nitrite (NO₂-N) within the filtrates were determined by colorimetric method via continuous-flow Auto Analyzer III (AA3, BRAN-LUEBBE, Germany) (Kirkwood et al., 1996) (NO_x was used to represent the sum of the two concentrations in this study). The detection limits for the measurement of NO₂-N and NO₃-N were 0.05 μmol/L and 0.2 μmol/L, respectively (hydro-chemical measurements data are available in the Supplementary Data Sheet ).

The stations mapping, statistical analysis, and data visualization were performed using ODV 7.5, SPSS 26.0, and Origin 10.0 software, respectively. In this study, the Shapiro-Wilk test was applied to assess the normality of experimental data, while Levene’s test was used to evaluate the homogeneity of variances. The significant differences between simulated incubation groups, the spatial distribution of environmental factors and the correlation coefficients of regressions were all tested via one-way ANOVA. When either test indicated a violation of the ANOVA assumptions (p < 0.05), significance was evaluated using the Kruskal-Wallis test, followed by the Wilcoxon test for post-hoc comparisons. Independence of observations was guaranteed by the randomized block design and confirmed by plotting residuals against predicted values; no discernible patterns or temporal autocorrelation were detected. Outliers with studentized residuals > 3 were examined and retained only when measurement errors could be ruled out. A significant threshold of p < 0.05 was used in all analyses. Correlations between variables were determined using Principal Component Analysis (PCA), Redundancy Analysis (RDA), and Spearman’s rank correlation analysis. To account for multiple comparisons, false discovery rate (FDR) adjustment was applied using the Benjamini-Hochberg procedure to all pairwise correlation tests, with an FDR-adjusted q-value threshold of 0.05 considered statistically significant.

The DO profiles of five sampling stations showed a consistent vertical pattern ( Figures 2B, C ). In the surface waters, there was a sharp DO decline to 120~130 μmol/L with a saturation of about 50%. In the subsurface waters, DO concentrations and saturations continued to decrease with depth but gradually. The typical DO minimum occurred in intermediate waters at 700~800 m, with the lowest DO concentrations among stations ranging between 78.8~90.6 μmol/L and the saturation at 25.3~28.7% level. Below this depth, DO exhibited a gradual increasing trend. The vertical DO pattern observed in our cruise was consistent with the previous reports (Gong et al., 1992; Li and Qu, 2006; Liu et al., 2011). Horizontally, both DO concentrations and saturations displayed an overall increasing trend from north to south among five stations.

The NO_x concentrations increased with depth, ranging from 0.00 to 13.83 μmol/L in surface water and 8.98 to 24.93 μmol/L in subsurface water, both showing a spatial variance of higher in the north and lower in the south. While in the intermediate layers, NO_x varied between 21.97 and 31.25 μmol/L with less variation across depths, and no significant differences were observed among sampling stations ( Figure 2D ) (ANOVA test, p > 0.05).

During the sampling period, the SPM content across all stations ranged from 0.10 to 0.86 mg/L, with a pulse increase in each water layer ( Figure 2E ). An obvious SPM peak in the surface layer appeared at 75~100 m depths, with a content between 0.41~0.79 mg/L. In subsurface layer, a slight increase was observed at 500~600 m, with the content between 0.28~0.69 mg/L. In the intermediate layer, the peak appeared at depths of 800~1000 m, showing a larger variation among stations (0.22~0.87 mg/L). The POC concentrations in water column ranged from 0.08 to 1.17 μmol C/L. Vertically, all stations showed higher POC in surface waters and lower contents in subsurface and middle layers ( Figures 2F, G ). Although less pronounced than SPM, slightly elevated POC were still observed at depths of 500~600 m and 700~800 m across stations. The increased loading both of SPM and POC in the subsurface and intermediate waters implies potential contributions from external inputs within these two layers. Horizontally, both SPM and POC concentrations among stations generally exhibited a decreasing trend from north to south.

To ensure the reliability of denitrification rate measurements and exclude possible misinterpretation caused by detecting bias, data validation was conducted based on formula (1) in Section 2.3.4. If N₂ only derived from microbial denitrification in the incubation system, the metabolism rates (R) calculated from r ₂₉ and r ₃₀ should be coincident. Regression analysis revealed that the N₂ production rates calculated from r ₂₉ and r ₃₀ of all incubations followed a 1:1 linear relationship (R² = 0.997, p < 0.001) ( Figure 3A ). This confirms that the detected ²⁹N₂ and ³⁰N₂ in the incubation system were exclusively derived from denitrifying yield (Hamersley et al., 2007; Holtappels et al., 2011).

Under in-situ DO condition of the seawaters, the N₂ production above detection limit only accounted for 28.6% of all incubations. The detected rates (referred to as R_in-situ hereafter) ranged from 0.20 ± 0.08 to 1.07 ± 0.23 nmol N₂/L/d ( Table 2 ), and all the N₂ productions observed were confined in the intermediate layer ( Figure 3B ). Among stations, the average R_in-situ within 600~1500 m depths differed significantly (ANOVA test, p < 0.05), exhibiting a decreasing trend from north to south ( Figure 3D ). The highest mean R_in-situ was observed at Sta. S-7 (0.41 ± 0.13 nmol N₂/L/d), while the lowest level occurred at Sta. S-23 (0.03 ± 0.01 nmol N₂/L/d).

In contrast, significant N₂ production under deoxygenating amended was much more pervasively observed, accounting for 65.7% of incubations. The measured potential denitrification rates (referred to as R _potential hereafter) ranged from 0.16 ± 0.07 to 32.8 ± 11.2 nmol N₂/L/d, higher by 1~2 orders of magnitude than R_in-situ ( Table 2 ). Except for the 300 m and 500 m at Sta. S-7 and Sta. S-37, detected potential denitrification all occurred in the intermediate waters ( Figure 3C ). Notably, much higher R _potential compared to other sampling depths was observed at 700 m and 800 m across all stations, suggesting some favorable conditions for denitrification performance within these two water layers. Similar to the spatial pattern of R_in-situ , the R _potential showed a marked horizontal variability by decreasing from north to south also ( Figure 3D ). The mean R _potential across 600~1500 m depths was highest at Sta. S-7 (17.5 ± 6.7 nmol N₂/L/d) and lowest at Sta. S-23 (1.20 ± 0.51 nmol N₂/L/d). It implies that specific mechanisms within the basin may sustain microbial denitrification potential.

As key functional genes involved in microbial denitrification, both narG and nirS genes were detected through the 300~1500 m depths during our cruise ( Figures 4A, B ), which provided a biological basis for the prevalent denitrifying potential observed in the SCS waters. The abundance of narG and nirS ranged from 0.55×10⁵ to 12.5×10⁵ copies/L and 0.14×10⁴ to 5.51×10⁴ copies/L, respectively. There is a significant linear correlation between these two genes ( Figure 4C ), suggesting a synergistic operation of nitrate reduction (narG-mediated) and nitrite reduction (nirS-mediated) in shaping denitrification potential of the basin waters, while nitrate reduction likely plays a dominant role. Vertically, both the genes exhibited depth-dependent variations, with peak abundances focused on 700~1000 m depths across stations ( Figures 4A, B ). Horizontally, there was also a notable north-south gradient, with Sta. S-7 and Sta. S-37 bearing higher abundance compared to the other three stations ( Figure 4D ). The spatial distributions of these two functional genes further indicate probable differences in the community structure of denitrifying microorganisms across the study area.

Characterization of NO₃-N uptake kinetic experiments revealed a linear increase of ²⁹N₂ production rates (r ₂₉) against the corresponding K¹⁵NO₃ in-spike gradients at each targeted sampling depth ( Figure 5 ). According to formula (7) in Section 2.3.5, the apparent first-order kinetic coefficients (k) of denitrifying metabolism for 700 m and 1000 m at Sta. S-7 and Sta. S-30 were calculated ( Table 3 ). The k values at both depths in Sta. S-7 were significantly higher than those at Sta. S-30, indicating spatial variability in the NO₃-N utilization capacity of denitrifying microorganisms. If the measured R _potential in each depth is considered as the NO₃-N saturated denitrification rate [i.e., R _max in formula (7)], then the corresponding half-saturated constants (K _m) are derived ( Table 3 ). In contrast to the coefficient k, the K _m values varied non-significant (p > 0.05) among four depths, with an average of 2.41 ± 0.19 μmol/L, which suggests a consistent affinity of potential denitrifying metabolism to NO₃-N substrate in the basin waters.

During sampling period, detectable microbial denitrifying was pervasively observed in the subsurface and intermediate waters across the SCS basin, which have rarely been reported in previous studies. Although the R_in-situ incubated under ambient DO was relatively weak and the gene abundances were much lower, the denitrification potential rates R _potential of the basin could reach a high level comparable to those measurements in oxygen-deficient waters (8~200 nmol/L/d; Jensen et al., 2009; Dalsgaard et al., 2013; Chang et al., 2014). Since lack of hypoxic condition in the water column, denitrifying metabolism was considered as negligible (Liu et al., 2002; Yang et al., 2017). That means unrecognized mechanisms remain to be revealed regarding the fixed-N removal within the basin. To better elucidate the inducing factors of observed potential denitrification, statistical analyses are conducted on the distributions and correlations among hydro-chemical parameters, incubation rates, and functional genes. PCA analysis reveals that the environmental factors and denitrification functions in the intermediate waters of northern stations (Sta. S-7 and Sta. S-37) significantly differed from other layers, characterized by low DO, high SPM loading, elevated denitrification rates and high functional gene abundances ( Figures 6A, B ). RDA and Spearman analyses further indicate marked negative correlations of denitrification rate and gene abundance against DO and its saturation, while denitrification rate correlates positively with both SPM and POC contents ( Figures 6C, D ). These findings highlight the dominant roles of DO and particulate matter in inducing the activity of denitrifying metabolism in the SCS basin waters.

Traditionally, microbial denitrification, as an anaerobic metabolic process, was believed to occur only under strictly hypoxic conditions (DO < 2 μmol/L) (Codispoti et al., 2001; Devol, 2008). Therefore, coastal hypoxic waters, pelagic ODZs and marine sediments have been well documented as primary regions for denitrification performance (Lam and Kuypers, 2011; Devol, 2015). However, the ambient DO level at subsurface and intermediate waters of the basin during cruise ranged between 80~120 μmol/L, far above typical hypoxic threshold. And even historical minimum DO records were as high as 65~75 μmol/L (Liu et al., 2011). Then there comes a query, that is whether the R_in-situ detected in this study resulted from oxygen depletion by microbial respiration in the incubation period? In previous report, the heterotrophic bacteria in the intermediate SCS waters were counted at about 2×10⁶ cells/L (Li et al., 2015). Adopting the representative microbial respiratory rate of 0.3 fmol O₂/cell/d (Robinson, 2008), a simple arithmetic suggests that bacterial oxygen consumption over the 48-hours incubation would only be < 0.1 μmol/L, which is negligible compared to the ambient DO contents. And in fact, we also selected the 800 m and 1000 m of Sta. S-7 and measured the remaining DO concentrations within aluminum foil bags to be about 68 μmol/L after incubations. This reckoning thus rules out experimental interference from bacterial respiration and indicates the adaptation of denitrification to the oxygenated water environments. Actually, not a few denitrifying bacteria in nature exhibit facultative anaerobic metabolism, capable of tolerating a broad range of DO levels (Zumft, 1997). Earlier laboratory experiments have demonstrated that strains such as Pseudomonas nautica, Pseudomonas stutzeri, and Paracoccus denitrificans could turn to express denitrifying function when DO saturations below 70% (Bonin et al., 1989; Körner and Zumft, 1989). Based on incubations of coastal Namibia and Peru hypoxic waters, Kalvelage et al. (2011) revealed that nitrate reductase activity (mediating the NO₃-N → NO₂-N step in denitrification) persisted even at about 30 μmol/L oxygen level. In addition, some research also confirmed that in fluctuating redox environments (e.g., tidal zone) the DO tolerance threshold of denitrification could elevate to about 90 μmol/L (Gao et al., 2010). Intriguingly, N-cycling genera, including Halomonas, Idiomarina and Pseudoalteromonas, which have been identified with high denitrification efficiency and aerobic denitrifying potential, were recently isolated from the deep SCS basin (Wang et al., 2019; Chen et al., 2024). These findings provide critical biological evidence supporting the observed denitrification activity within the basin waters.

Apart from the adaptation of denitrifying bacteria to O₂ tolerance, SPM in the water column is also one of the important ways to induce denitrification in non-hypoxic environments. Because of its large specific surface area and organic-rich composition, SPM is beneficial for the attachment and growth of aquatic microorganisms (Zhang et al., 2016; Bianchi et al., 2018; Balmonte et al., 2021; Wan et al., 2023). It is easy to generate hypoxic structures inside particles with severe aerobic respiration, thus creating ideal micro-niches for anaerobic metabolism by bacteria such as denitrification (Ploug, 2001; Xia et al., 2021). This phenomenon has been widely documented. For instance, laboratory simulations demonstrated active denitrification within organic polymer aggregates secreted from phytoplankton even under ambient DO levels higher than 100 μmol/L (Klawonn et al., 2015; Stief et al., 2016). Fuchsman et al. (2019) reported that spring phytoplankton blooms in the Black Sea basin triggered vigorous denitrification within the oxygenated subsurface layer (DO in the range of 30~50 μmol/L) due to elevated SPM inputs. Similarly, in the well-oxygenated Beibu Gulf, Zeng et al. (2018) observed pronounced denitrification in the bottom nepheloid layers driven by sediment resuspension. The detected high denitrification potential in our study consistently coincided with elevated SPM loads in sampling depths, which means SPM, as a carrier, might also play a pivotal role in activating denitrification metabolism within the SCS waters.

Obviously, both DO and SPM act as critical factors in inducing denitrifying function of the SCS seawaters. Regarding the incubation measurements, the Spearman correlations and significance levels of these two factors against R_in-situ as well as R _potential showed no marked difference ( Figure 6D ), remaining it still difficult to discern the dominance between DO and SPM. While in terms of functional genes, DO exhibited a more significant correlation compared to SPM ( Figure 6D ). Furthermore, both narG and nirS abundances positively correlated with R_in-situ and R _potential strongly ( Figure 6D ). Based on the above analyses, it could be speculated that low DO level likely prior induces the functional genes expression associated with anaerobic denitrifying metabolism in the basin waters. Then SPM further triggers enzymatic activity by providing available metabolic niches, and ultimately performs high denitrification potential. Since there was no pronounced correlation of narG/nirS ratio against either DO or SPM according to Spearman analysis ( Figure 6D ), the response of denitrifying functional structure to these two environmental factors remains unclear and deeper explorations are still necessary.

SPM not only provides favorable micro-environments for microbial N removal under non-hypoxic conditions, but also contributes the carbon source for heterotrophic metabolisms such as denitrification. Therefore, SPM serves as an important link for coupling micro-scale carbon and nitrogen cycles in aquatic ecosystems (Bianchi et al., 2018; Wan et al., 2023). In order to further elucidate the influence of SPM content and its composition on the denitrification function of the SCS basin, the intermediate water body (600~1500 m) with prominent potential denitrifying activity is focused for discussion below.

Linear and exponential fitting regressions were applied to analyze relationships of SPM and POC against to R_in-situ , R _potential, narG and nirS gene abundances, respectively ( Figure 7 ). The results showed that increased SPM and POC levels significantly enhanced functional genes expression and denitrification potential, consistent with observations across diverse aquatic systems, including rivers, estuaries, coastal zones, and pelagic oceans (Zeng et al., 2018; Fuchsman et al., 2019; Balmonte et al., 2021; Xia et al., 2021; Wan et al., 2023). It reflects a common modulatory role of particulate matter on denitrification and implies that particle-associated denitrifying bacteria may contribute more substantially to denitrification potential than their free-living counterparts within the basin waters (Jia et al., 2016; Zhang et al., 2016; Zeng et al., 2018; Li et al., 2021). Comparing the regressions among several scenarios, it is found that, on the one hand, both SPM and POC fit better with R _potential and nirS abundance than fittings with R_in-situ and narG abundance (ANOVA test, p < 0.01). Since the interface of particles usually promotes the metabolic activity of microorganisms (Li et al., 2015; Klawonn et al., 2015; Balmonte et al., 2021; Wan et al., 2023), denitrification potential will exhibit stronger dependence on particulate matter and organic carbon under deoxygenated conditions. It suggests that when O₂-inhibition relieved, the particulate matters will play a dominant role in modulating spatial pattern of denitrifying process. Thermodynamically, the Gibbs free energy of narG-mediated NO₃-N to NO₂-N reduction (△G ⁰ = -244 kJ) is greater than that of nirS-mediated NO₂-N to NO reduction (△G ⁰ = -371 kJ) (Lam and Kuypers, 2011), which means a more electron donor (carbon source) demand of the latter step and thus explains the observed disparities in gene-specific regressions. On the other hand, exponential fittings in the regression analyses generally show superior goodness compared to linear ones (ANOVA test, p < 0.05), indicating complex nonlinear relationships between particulate matter and denitrification potential as well as functional genes expression. Notably, the influence of particulate matter on the denitrifying function is not solely mediated by its abundance. Previous studies have emphasized that particle size, compositional heterogeneity and organic carbon source collectively influence microbial denitrification activity on particles (Klawonn et al., 2015; Jia et al., 2016; Zeng et al., 2018; Fuchsman et al., 2019).

Worth mentioning, the high particulate loading observed in the intermediate waters during cruise was not an accident. According to former estimations, up to 7×10¹⁰ tons of terrigenous sediments annually from Luzon island in the east, mountainous rivers of Taiwan in the northeast, Pearl river runoff in the north, Red river in the northwest, Mekong river in the southwest as well as Malay Peninsula in the south, can be imported into the SCS basin (Liu et al., 2013; Schroeder et al., 2015). Driven by mid-depth circulation, these terrestrial materials are transported and dispersed for a long distance across the basin, resulting in widespread high particulate export features in the middle and deep of SCS waters (Liu et al., 2013; Schroeder et al., 2015; Wei et al., 2017; Ma et al., 2017). Moreover, during the summer southwest monsoon, intensified upwelling systems in the SCS (Chao et al., 1996; Liu et al., 2002) would enhance sediment resuspensions over the shelf and stimulate primary productivity in upper layers (Dippner et al., 2007; Liu et al., 2007). Concurrently, hydrological dynamics such as mesoscale eddies and typhoon events further amplify the vertical and lateral transport of SPM and POC into the deep basin (Wang et al., 2003; Kao et al., 2010). Sediment trap data also corroborated that summer SPM fluxes in the intermediate waters (1000~2000 m) increase prominently, accompanied by 1~2 fold elevations in POC input and organic matter degradability (Lahajnar et al., 2007). It can be inferred that favorable hydrodynamic conditions which sustain particulate and organic matters supply, likely contribute to maintaining the strong denitrification potential in the intermediate of SCS basin.

By comparison, it is found that the spatial trend of R _potential as well as average functional genes abundances observed in our cruise is in good agreement with the pattern of sinking particle plumes, which were depicted by Schroeder et al. (2015) covering the whole SCS basin ( Figure 8A ). Synthesizing the above discussions, a preliminary conceptual framework of coupling between denitrification and SPM in the SCS basin is then proposed, as schematically described in Figure 8B . That is, the particulate matters from terrestrial or benthic sediments are priorly imported and dispersed across the basin-wide seawater driven by hydrodynamics, accompanied by the attachment of plentiful microorganisms. When transported to the intermediate waters, low ambient DO stimulates the expression of anaerobic denitrifying function and also facilitates hypoxic micro-niches creation in particles. Finally, the bacterial denitrifying metabolism is fueled with sufficient organic carbon supply. Although needing to be confirmed, the coupling framework provides a novel perspective for unraveling the mechanisms driving microbial N removal in non-hypoxic marginal basins.

To further assess the importance of this particle-loaded N removal to nutrient cycling in the SCS, we compare measured denitrification with major microbial N-transformations of the basin. When integrating R_in-situ and R _potential throughout the intermediate water column (600~1500 m) among five sampling stations, average fluxes of 175.1 μmol N/m²/d and 6.39 mmol N/m²/d are obtained, respectively. The depth-integrated R_in-situ is comparable to the import flux from N₂ fixation in SCS upper euphotic zone, while the depth-integrated R _potential is 1~2 orders of magnitude higher (Lu et al., 2019; Liu et al., 2023; Tian et al., 2025). Besides, both R_in-situ and R _potential detected in our incubation assays are within the range of reported nitrification rates in SCS subsurface layer (0.1~100 nmol/L/d; Xu et al., 2018). It implies that the microbial denitrifying function discovered should be ecologically meaningful, which might have a basin-scale impact on NO₃-N budget balance and nutrient cycling patterns of SCS. Given our study relying on only a single cruise, basin-wide investigations covering seasonal and interannual variations remain extremely limited. How the potential denitrification would affect primary productivity in the upper SCS is still unknown, requiring more robust observations and model verification in future research.

The K _m constant (2.25~2.68 μmol/L) of potential denitrification in this study estimated via ¹⁵NO₃-N gradient incubation is well consistent with literature reports. For example, laboratory-isolated strains such as Paracoccus denitrificans, Pseudomonas chlororaphis, and Pseudomonas aureofaciens exhibited K _m values of 5.0 μmol/L, 1.7 μmol/L and 1.8 μmol/L, respectively (Parsonage et al., 1985; Christensen and Tiedje, 1988). Similarly, Jensen et al. (2009) reported a K _m of 2.9 μmol/L for denitrification in the hypoxic Denmark Mariager Fjord. While Dalsgaard et al. (2013) measured a value of 2.5 μmol/L in the oxygen-deficient Gotland Basin of Baltic Sea. This result indirectly reflects the potential denitrification observed in the SCS to be typical anaerobic denitrifying metabolism, with nitrate reductase exhibiting strong substrate affinity for NO₃-N. Comparing with the NO₃-N concentration in the intermediate waters (20~29 μmol/L), it further confirms that substrate availability is sufficient for denitrification performance. This is also consistent with the insignificant correlation between incubated denitrification rates and NO₃-N concentration ( Figure 6D ).

Different from the constant K _m, the apparent first-order kinetic coefficient k varied markedly among sampling layers ( Table 3 ). Additionally, there appears strong linear correlations of k with both SPM and POC contents in corresponding layers ( Figures 7E, J ). As mentioned in the methods, coefficient k in our discussion reflects the apparent activity of nitrate reductase. Although it could not rule out the contribution of change in denitrifying bacterial community composition to the variation of enzymatic activity (Li et al., 2015, Li et al., 2021), the correlations between k and SPM as well as POC verified a dominant role of particulate matter content in mediating the potential denitrification in our study region. Indeed, the “interfacial effect” between particle-water interface favors substrate diffusion and migration more effectively, thereby stimulating higher metabolic activity for particle-attached bacteria compared to those free-living ones (Liu et al., 2013; Zhang et al., 2016; Li et al., 2021).