The study region is placed in Xincun Bay, which is located in the southeast coast of Lingshui County, Hainan Island, South China Sea ( Figure 1 ). This bay is an almost entirely closed bay with only one narrow channel connecting to the South China Sea in the southwest. According to previous surveys, the Xincun seagrass bed is a mixed seagrass bed, which occupies an area of approximately 175 ha and is mainly distributed in the shallow waters of southern Xincun Bay (Huang et al., 2006; Huang et al., 2019). Seagrass species in this bay include Enhalus acoroides, Thalassia hemprichii, Cymodocea rotundata, Halodule uninervis, and Halophila ovalis, with E. acoroides and T. hemprichii as the dominant species (Huang et al., 2006; Huang et al., 2019).

According to the abundance and distribution of seagrass, samples for analysis were collected at stations S1, S2, and S3 ( Figure 1 ) during low tide in summer (August 2018) and winter (January 2019). At each station, one to three samples were collected depending on the abundance and distribution of sampling organisms. Seagrass (E. acoroides and T. hemprichii) and macroalgae Ulva lactuca were collected by hand and washed with distilled water to remove attachments. For seagrass epiphytes, we applied the commonly used scraping method (i.e., removed carefully from the surface of seagrass leaves using a scalpel blade), because of the wideness of seagrass leaves: from 0.8 to 2.1 cm width (Huang et al., 2019), and the relatively high biomass of epiphytes (Cui et al., 2021b). Suspended particulate organic matter (SPOM) samples were obtained by filtering surface seawater onto pretreated (heated 2h at 450°C) Whatman GF/F filters. For sediment organic matter (SOM) samples, only the upper 1cm of the surface sediment layer was collected using a spade. For macroinvertebrates, crustaceans were collected by trammel nets, gastropods were sampled by hand, and polychaetes were sampled by sieving sediments through the 5mm mesh screen. Live invertebrates were put into containers with filtered seawater and kept for 24h to evacuate gut contents. Afterward, the gastropods and crustaceans were sorted, identified, and taken muscle. All the samples were placed in plastic polyethylene bags and stored in a freezer at -20°C (< 1 week) for their transportation to the laboratory.

In the laboratory, samples were freeze-dried at -40°C for 48h, and grounded into a fine homogeneous powder. Before stable isotope analysis, epiphytes, SPOM, and SOM were acidified to remove the effect of inorganic carbon on the δ¹³C ratio of these samples. SPOM samples were acidified by exposure to hydrochloric acid (HCl) vapor for 24h. Epiphytes and SOM were acidified with a 1 mol/L solution until there were no bubbles. Subsequently, all acidified samples were rinsed with distilled water, freeze-dried again, and stored in the centrifuge tube for subsequent analysis.

Homogenized powder samples and SPOM were transferred to tin capsules, and δ¹³C, δ¹⁵N, and δ³⁴S values of samples were measured using a continuous-flow isotope-ratio mass spectrometer (Delta V Advantage, Thermo Fisher Scientific, Waltham, MA, USA) attached to an elemental analyzer (Flash EA 1112, Thermo Fisher Scientific, Milan, Italy). δ¹³C, δ¹⁵N, and δ³⁴S values were expressed in ‰ relative to the standard reference materials (Vienna Pee Dee Belemnite for δ¹³C values, atmospheric N₂ for δ¹⁵N values and Vienna Canyon Diablo Troilite for δ³⁴S values) using the standard δ notation:

where X is δ¹³C, δ¹⁵N or δ³⁴S ratio, and R is ¹⁵N/¹⁴N, ¹³C/¹²C or ³⁴S/³²S. The detection limits for δ¹³C, δ¹⁵N and δ³⁴S values were<0.2‰,<0.2‰, and<0.3‰, respectively.

Fatty acids were extracted using a method modified from (Folch et al., 1957). Briefly, a 2:1(v/v) chloroform and methanol mixed solution was used to extract the lipids of each sample (about 0.2-1mg), and a 0.9% sodium chloride solution was used to separate the lipids solution from the mixed solution. The sample was then concentrated using a rotary evaporator to obtain total lipid. The resulting total lipid samples were methylated with 0.5mol/L sodium hydroxide-methanol solution, derivatized with boron trifluoride-methanol solution, extracted with hexane and stratified with saturated sodium chloride to obtain fatty acid methyl ester (FAMEs) samples. To quantification the fatty acid, mixtures of FAMEs: internal standard methyl nonadecanoate (1:1) were run on a Gas Chromatograph (GC-7890B, Agilent Technologies, Inc. USA) attached to a Mass Spectrometer Detector (MSD-5977A, Agilent Technologies, Inc. USA). Helium was the carrier gas, and the thermal gradient was set as from 125°C to 227°C at 2°C min^-1. Fatty acid profile identifications were performed by comparison to relative retention times of a known standard (GAQSIQ, 2008).

The contribution proportion of each source to macroinvertebrate diets was estimated based on δ¹³C, δ¹⁵N or δ³⁴S values using the Bayesian stable isotope mixing model SIMMR (Parnell and Inger, 2019), which is a package in R (Team, 2019). Here, the food sources were divided into four groups mainly including seagrass (including E. acoroides and T. hemprichii), epiphytes (including epiphytes attached to E. acoroides and T. hemprichii), macroalgae (U. lactuca), and SPOM. SOM was excluded from SIMMR since that SOM is a mixture of other primary food sources (i.e., seagrass, epiphytes, macroalgae, and POM) (Xu et al., 2018). The δ³⁴S values of SPOM were not detected due to the relatively small amount of collected material and were not used in this model. The trophic enrichment factors (TEFs) in this Bayesian mixing model were δ¹³C of 0.4 ± 1.3‰ and δ¹⁵N of 2.3 ± 1.61‰ for snails, bivalves and polychaetes (Mascart et al., 2018; Débora et al., 2020), and δ¹³C of 0.5 ± 0.8‰ and δ¹⁵N of 3.2 ± 0.26‰ for crustaceans (Mittermayr et al., 2014; Riccialdelli et al., 2017; Beesley et al., 2020). This SIMMR model was based on normal (100,000) interactions, and outputs of the contributional proportion were expressed as mean value, 2.5%, 25%, 50%, 75% and, 99% confidence intervals (CI).

To explore whether the variation of multiple stable isotope ratios (δ¹³C, δ¹⁵N, and δ³⁴S) was greatest amongst food sources (i.e., seagrass, epiphytes, macroalgae, SPOM, and SOM) or between seasons (i.e. summer and winter), all data were firstly checked the normality and/or homogeneity of variance using the Shapiro-Wilk test and Levene’s test, respectively. Subsequently, a two-way ANOVAs was performed. Meanwhile, a post-hoc test (Tukey’s HSD) was also applied in order to identify the specific differences among food source types. Furthermore, we also performed a one-way ANOVA or Mann-Whitney U-test to evaluate the seasonal changes of each food source in detail. Similarly, two-way ANOVAs were performed in order to test the significant differences amongst consumer groups (snails, crustaceans, bivalves, and polychaetes) and between seasons. The variations in the fatty acid markers of food sources and consumers were tested by a distance-based permutational analysis of variance (PERANOVA). This analysis used Euclidean distance resemblance calculated from untransformed data. In addition, to visualize multivariate patterns we generated a generalized discriminant analysis using Canonical Analysis of Principal Coordinates (CAP) (Anderson and Willis, 2003). To explore which fatty acid markers contributed most to the differences among food sources, we correlated the fatty acid data with the canonical axes. ANOVA analyses were performed with SPSS version 22 (IBM Corporation, Chicago, IL, USA), PERMANOVA and CAP were performed with PRIMER v6 software (Clarke and Gorley, 2006).

The main food sources in Xincun seagrass bed displayed a wide range of δ¹³C and δ³⁴S values ( Figure 2 ), which correspondingly ranged from -18.6 ± 0.9‰ for winter SPOM to -8.4‰ for summer macroalgae ( Appendix S1-Table A1 ), and from 5.6 ± 2.5‰ for winter SOM to 17.4 ± 0.2‰ for winter macroalgae, respectively. In contrast, the δ¹⁵N values showed a narrow range, ranging from 5.5 ± 0.2‰ for winter SPOM to 8.6 ± 0.4‰ for summer macroalgae. Two-way ANOVA analysis showed distinct differences among the five food sources for δ¹³C, δ¹⁵N, and δ³⁴S values, while the clear distinction between seasons could only be found in δ¹³C and δ³⁴S values ( Table 1 ). Post-hoc test indicated that the difference in δ¹³C values among the food sources was driven by the OM (organic matter, including SPOM and SOM), which differed from the seagrass, epiphytes, and macroalgae, respectively ( Figure 2 and Appendix S1-Table A2 ). Whereas, significant differences in the δ¹⁵N values only existed in seagrass versus macroalgae, epiphytes versus SPOM, and macroalgae versus OM, respectively. By contrast, all other food sources were significantly different for δ³⁴S values.Multiple stable isotope (δ¹³C, δ¹⁵N and δ³⁴S) values of macrobenthos were significantly different among groups (snail, crustaceans, bivalves and polychaetes, Two-way ANOVA, δ¹³C: F₃ = 19.56, p = 0, δ¹⁵N: F₃ = 20.90, p = 0, δ³⁴S: F₃ = 10.22, p = 0), whereas there were no significant differences between seasons (Two-way ANOVA, δ¹³C: F₁ = 0.33, p = 0.565, δ¹⁵N: F₁ = 0.03, p = 0.861, δ³⁴S: F₁ = 8.19, p = 0.01). In contrast to the food sources, macrobenthos had a reduced range in the δ¹³C and δ³⁴S values, which ranged from -15.8 ± 0.2‰ (crustaceans Calappa xp in winter) to -7.3 ± 0.5‰ (snail Notosinister subaura in winter) for δ¹³C values and from 8.8 ± 1.9‰ (polychaetes Arenicola cristata in summer) to 18.3 ± 0.5‰ (crustaceans Calappa xp in winter) for δ³⁴S values, redpectively. However, for δ¹⁵N values, macrobenthos showed a larger range than that of food sources, ranging from 8‰ (snail Notosinister subaura in summer) to 12.1 ± 0.7‰ (crustaceans Menippe rumphii summer).

A total of twenty-one fatty acids were used for the profiles of food sources and macrobenthos ( Appendix-Table A3 ). For food sources, percentages of saturated fatty acids accounted for greater than 50% of total fatty acids composition. PERANOVA of the total fatty acid data showed significant differences among food sources and between seasons ( Table 1 ). CAP on total fatty acids showed that seagrass was significantly distinct from epiphytes and OM along the first axis, and the epiphytes were clearly separated from the macroalgae and OM along the second axis ( Figure 3A ), while SPOM had a greater overlap with SOM. When only poly-unsaturated fatty acids were considered ( Figures 3B , 4 ), Fatty acids 18:2n6c and 18:3n3 for seagrass, 18:1n9c and 16:1n7 for epiphytes, and 18:1n9t, 20:1, 20:5n, 22:6n3, 18:3n6, and 14:1n5 for OM, were significantly higher than those of the other food sources. Macroalgae showed some certain degree of dispersed fatty acid profiles with relatively high levels of 15:1n5 in summer, and 18:1n9c in summer and winter, although these fatty acids also showed similarly high levels in seagrass and epiphytes.

For consumers, the percentages of saturated fatty acids still accounted for a larger proportion (55-75%) than unsaturated fatty acids (26-47%), but it was lower than that of food sources ( Appendix-Table A3 ). Significant differences in total fatty acids were also evident among groups ( Table 1 ), while differences between seasons were not significant. Especially, for poly-unsaturated fatty acids, gastropods, crustaceans, and polychaetes had a high level of 18:1n9c and 20:5n3, while bivalves contained a high level of 16:1n7 and 22:6n3. In addition, the levels of 18:2n6c and 18:3n3 in snails and polychaetes were higher than those of other macrobenthos.

The SIMMR mixing models showed that the contribution of the food sources to most macrobenthos was relatively similar between seasons in this seagrass bed ( Figure 5 ). For the snail Notosinister subaura and the polychaete Arenicola cristata, seagrass contributed the highest to their diet, with the corresponding mean value of 30-35% and 29-31%, respectively, while epiphytes contributed subordinately, with the corresponding mean value of 25-26% and 24-29%, respectively. For the crustaceans, SPOM contributed significantly to Calappa sp and Menippe rumphii in summer, while macroalgae and SPOM were the main contributors to their diets in winter. The diet of Portunus sanguinolentus and Thalamita crenata had significantly seasonal variation, with a high contribution of epiphytes to P.sanguinolentus (67%) and macroalgae (42%) to T.crenata in summer, as well as the high contribution of macroalgae to their diets in winter ( Figure 5 ). While for the bivalves Perna viridis and Scapharca subcrenata, SPOM showed a consistently high contribution to their diets in both seasons (mean value, 34-44% and 33-41%, respectively).

In the present study of tropical primary produces, the main variance of food sources for stable isotopes and fatty acids was amongst source types rather than seasons, except for the obvious seasonal variation in δ¹³C values of macroalgae. In particular, the significant difference in δ¹³C values between seagrass and OM allows this indicator to distinguish between vascular and organic matter. However, the wide range of δ¹³C (from -14‰ to -7.9‰) and δ¹⁵N (from 4.1‰ to 8.8‰) values for epiphytes overlapped with that of seagrass, which suggested the difficulty in using these biomarkers to identify the diets of consumers. The epiphytes attached to seagrass are mainly composed of bryozoans, diatoms Cocconeis scutellum, red algae, brown algae, small crustose coralline alga Pneophyllum lejolisii, bacteria, and fungi (Kharlamenko et al., 2001). The variation of environmental factors caused by stations (e.g. salinity and nutrient availability) and seasons (e.g. temperature and light) may lead to differences in the composition of epiphytes, and thus result in a large isotopic range (Nichols et al., 1985; Gacia et al., 2009; Liu et al., 2017). Similarly, the wide range of δ¹³C values for macroalgae, overlapped with that of seagrass, might also be caused by these variations in environmental factors. This is mainly because the macroalgae Ulva sp. have been proven to be capable of both C3 and C4 photosynthesis (Valiela et al., 2018a), which may be affected by environmental factors (Xu et al., 2012). However, the δ ¹⁵N values of seagrass and macroalgae were significantly different, which indicated that we can use this indicator to distinguish macroalgae from seagrass. This may be reasonable because macroalgae are generally considered to assimilate significantly sufficient amounts of dissolved inorganic nitrogen from highly nutritious seawater columns in a shorter time than seagrass and OM (Gartner et al., 2002; Thornber et al., 2008), and previous study have shown that this area was subject to considerable anthropogenic nitrogen source input (Liu et al., 2020a).

Sulfur isotope was considered to be the most likely biomarker to effectively distinguish food sources since different food sources have distinct sulfur sources (Moncreiff and Sullivan, 2001). In the present study, significant differences in the δ³⁴S values of food sources among food sources helped distinguish food sources. In particular, the obvious difference in δ³⁴S values between macroalgae and epiphytes can make up for the limitation in the use of δ¹³C and δ¹⁵N values. Algae usually obtain more sulfur from seawater sulfate (δ³⁴S =20‰) than seagrass, which results in enriched δ³⁴S values (Moncreiff and Sullivan, 2001). In our study, the δ³⁴S values of macroalgae were the most enriched, which was similar to those of macroalgae in other seagrass beds (Holmer and Nielsen, 2007; Oduro et al., 2012). However, similar to δ¹³C and δ¹⁵N values, the δ³⁴S values of epiphytes and seagrass also exhibited significant overlap. This obvious overlap might be attributed to the difference in the composition of epiphytes among stations or the resuspension of benthic microalgae in some stations (Jorge and Van Beusekom, 1995; Kasim and Mukai, 2006), which led to great changes in the sulfur isotopes of epiphytes, ranging from 5.8‰ (close to the δ³⁴S values of 6.8‰ SOM in this area) to 15.5‰. In addition, significant seasonal differences in δ³⁴S values of seagrass and epiphytes may also increase the overlapping probability of their δ³⁴S values. Therefore, the combination of multiple stable isotopes cannot completely distinguish the food sources in this area.

In comparison to the multiple stable isotope data, the fatty acid compositions of food sources are generally considered to be significantly different (Crawley et al., 2009; Madgett et al., 2019). In our study, all food sources were rich in saturated fatty acids (61%-77%), especially 16:0, which was similar to the research results in other seagrass beds (Crawley et al., 2009; Park et al., 2013). In contrast, the unsaturated fatty acid compositions among food sources were clearly distinct. The much higher level of 18:2n6c and 18:3n3 in seagrass, 16:1n7 in epiphytes and 18:1n9t, 20:1, 20:5n3, 22:6n3 18:2n6t,18:2n6 and 14:1n5 in OM can effectively separate seagrass, epiphytes, and OM from each other. It is worth noting that 20:5n3 and 22:6n3 were also abundant in epiphytes. The unsaturated fatty acids 16:1n7, 20:5n3, and 22:6n3 are generally biosynthesized by diatoms (Coelho et al., 2011) and have been used as benthic diatom biomarkers (Kharlamenko et al., 2001; Belicka et al., 2012; Jankowska et al., 2018). Therefore, diatoms mainly originated from the resuspension of SOM might have an important effect on the component of epiphytes, which can also explain the wide range of δ¹³C values for epiphytes and the overlap in δ¹⁵N and δ³⁴S values for epiphytes and SOM in this area. In addition, the fatty acid compositions of macroalgae were also similar to that of seagrass and epiphytes, with a high level of 18:2n6c and 18:3n3. Such fatty acid composition characteristics of seagrass and macroalgae were also evidenced in other temperate bays (Meziane and Tsuchiya, 2000; Lebreton et al., 2011). The most likely explanation was that macroalgae and seagrass are in the same lineage with the same suite of photosynthetic pigments and similar biochemical pathways (Galloway et al., 2012). These characteristics limit us to distinguish seagrass and macroalgae based on their specific fatty acid compositions.

Despite the combination of multiple stable isotopes and fatty acid biomarkers can make a clear distinction for food sources, caution is still required with interpretations of field-based biomarker studies, as some factors such as growth rates, metabolism, and mixed diets will affect the isotope values and fatty acid compositions of consumers (Hanson et al., 2010; Pecquerie et al., 2010). In this study, the accuracy of using multiple stable isotopes and fatty acid biomarkers to identify the diets of different trophic groups was significantly different. The filter-feeder bivalves are the only trophic group, of which diets can be accurately identified by the stable isotopes and fatty acid biomarkers. This has been proved by the similar δ¹³C values for these bivalves (P. sanguinolentus and S. subcrenata) and OM, and the high level of OM FA markers 18:2n6c, 22:6n3, and 20:5n3 for these bivalves, which all indicated the significant OM contribution to their diets. However, for the grazer snail N. subaura, its δ¹³C, δ¹⁵N, and δ³⁴S values significantly overlapped with that of seagrass and epiphytes. This indicated that their main food sources were more likely to be seagrass, which was also confirmed by the SIMMR results. Accordingly, the seagrass FA marker 18:2n6c in the snail was indeed higher than that of other consumers, whereas, the most abundant fatty acid composition in the snail was OM indicators 20:5n3 and 22:6n3, which may mislead us to identify the diet of this snail. The most likely reason is that there are some transformations in the process of assimilation of fatty acids by the snails. Evidence to date indicates that some primary consumers, such as herbivores, have the ability to convert C18 fatty acids to 20:5n3 and 22:6n3 (Caramujo et al., 2008; McLeod et al., 2013), therefore a high level of 20:5n3 and 22:6n3 in these consumers may be transformed by the seagrass biomarker 18:2n6c. Similarly, although the δ¹³C, δ¹⁵N, and δ³⁴S values of deposit feeder polychaete A. cristata were more similar to that of seagrass and epiphytes, the level of 18:2n6c in the polychaete was not clearly higher than that of other unsaturated fatty acids (i.g., 20:5n3 and 20:6n3), which indicated that the polychaete may also have the ability to convert these fatty acids.

Most crustaceans are known to be omnivores or generalized predators and have mixed diets, therefore there may be great obscure when using δ¹³C and δ¹⁵N values to identify their diets (Reaka-Kudla, 2001). In this study, the δ¹³C values of predators P. sanguinolentus and M. rumphii overlapped more with those of OM in both seasons, which is often mistakenly believed that their carbon sources were mainly OM in both seasons (Fan et al., 2011; Du et al., 2019). The mixing model also showed a high POM contribution to their diets. Whereas, their δ¹³C values were also located in the middle of OM and epiphytes, which also indicated they might feed on OM and epiphytes together. This possibility has also been confirmed by their fatty acid composition. The high level of OM FA marker 18:2n6c in summer, and OM FA markers (20:5n3, 22:6n3, and 18:2n6t) and epiphytes FA markers (18:1n9c and 16:1n7) in winter indicated that their food sources were OM in summer and OM and epiphytes in winter, respectively. Some researchers also suggested that OM and epiphytes had a high contribution to the predator crustaceans (Ning et al., 2019; Paar et al., 2019).

Similarly, it is also difficult to accurately distinguish the diets of omnivores crab C. sp. and T. crenata because of their wide range of δ¹³C and δ¹⁵N values. The SIMMR results showed the relative average coverage contribution of each food source to C. sp. and the high contribution of macroalgae to T. crenata. Relatively higher levels of epiphytes FA marker 16:1n7 and OM FA markers 20:5n3 and 22:6n3 in these omnivores indicated a mixed diets contribution of epiphytes and OM. Nevertheless, the significantly high level of FA marker 18:1n9 in the omnivores during summer suggested that their chief food source was probably macroalgae. Therefore, it is necessary to further distinguish epiphytes from macroalgae by the δ³⁴S values. Considering the trophic fraction of δ³⁴S values was from 0 to 1.9 (McCutchan et al., 2003; Mittermayr et al., 2014), and the higher δ³⁴S values of macroalgae were higher than that of these omnivores. Therefore, the food sources of these omnivores were more likely to be epiphytes and OM.

Overall, for diverse primary producers in our tropical seagrass bed food webs study, multiple stable isotopes usually change obviously among primary producer species and seasons within a region. This characteristic would affect the application of multiple isotopes to identify the food sources of dispersed consumers. Whereas, for snails, polychaetes and bivalves, multiple isotopes usually could discriminate their food sources, because limited mobile ability induced their diets to be often relatively single and fixed within a period and region (Fukumori et al., 2008; Tue et al., 2012; Cui et al., 2021b). By contrast, fatty acids generally provide clearer separation of primary producer species than multiple stable isotopes by providing specific FA biomarkers for each food source. This characteristic would allow FAs to identify the food sources of crustaceans with complex diets. However, the specific FA marker of each species might be the same due to they might have the same suite of photosynthetic pigments and similar biochemical pathways (i.g., seagrass and macroalgae in this study) (Galloway et al., 2012), and its dynamic is always linked with the metabolic conditions of the consumers (i.g., snails in this study). Therefore, using both types of biomarkers together may reduce the uncertainty caused by natural variation and thus strengthen the separation of food sources and the diets of different consumer species. Further study is needed to apply multiple isotopes and fatty acids into the mixing model to quantify the diet contribution of consumers. In addition, in our study, most consumers have a high level of diatom biomarkers, which indicated microphytobenthos might also be their important food source. Therefore, further study in seagrass ecosystems is also needed to consider microphytobenthos as an independent carbon source.