The Xisha Islands, in the central South China Sea, are derived from coral reefs and comprise over 40 islands, reefs and cays, including the Dongdao Atolls, the Huaguang Atolls, the Xuande Atolls, the Yongle Atolls, and some smaller islands (Zhao et al., 2017; Ding et al., 2019; Zhao et al., 2019). The Qilianyu Islets and Cays (16°59′ N, 112°18′ E) of the Xisha Islands comprise an arced reef flat, which extends in a NW-SE direction and curves to the NNE (Shen et al., 2017). Eight herbivorous reef fish species common in the Qilianyu Islets and Cays were collected by SCUBA in June 2020 ( Figure S1 ). Fishes were captured by driver with spear guns between the hours of 18:00 and 20:00. All fishes were collected along the shallow waters around the coral reef regions (water depth below 15 m). The details of the fishes sampled in this study, along with their phylogenetic classification and feeding strategies, are shown in Table 1 . The feeding strategy of each species was determined based on the previous published studies. All fishes collected were kept at -4°C and transported to the laboratory, where they were dissected using sterile scissors. In the dissecting process, all fishes were made sure to have food item in the intestines. The anterior intestinal contents were separated and divided into two subsamples. One subsample was used for microscopic examination, and the another was used for 18S rRNA high-throughput sequencing analysis. A small piece of white muscle tissue near the dorsal fin was sampled from each fish and used for stable isotope analysis.

Intestinal contents of each fish species were collected for dietary assessment. Intestinal contents of individual fish were spread on glass culture dish and put under a microscope (Zeiss SteREO Discovery.V20, Germany). Photographs of the food items were taken using the microscope. Macrophytes species (Macroalgae and seagrasses) were only observed here, however, quantity analysis of the macrophytes composition wasn’t concerned. Macrophytes species were identified according to their morphological traits (Titlyanov et al., 2017; Huang, 2018).

The total DNA of intestinal contents (0.2 g sample) of individual fish was extracted using a QIAamp^® Fast DNA Stool Mini Kit (Qiagen, Germantown, MD, USA) according to the manufacturer’s protocols. The total DNA recovered from the same fish species was pooled together for PCR amplification which targeted the V4 region of the eukaryotic 18S rRNA gene using primers 528F (5’-GCGGTAATTCCAGCTCCAA-3’) and 706R (5’-AATCCRAGAATTTCACCTCT-3’), where the barcodes were an eight-base sequence unique to each sample (Cheung et al., 2010). Amplicons were extracted from 2% agarose gels and purified using the AMPure XP Beads (Beckman, Agencourt, USA). These purified amplicons were pooled in equimolar paired-end sequences (2 × 250) on an Illumina platform (Gene Denovo Co., Guangzhou, China). Raw reads for each fish species were deposited into the NCBI Sequence Read Archive (SRA) database with the accession number PRJNA742779. The representative sequences affiliated to the phyla Chlorophyta, Ochrophyta and Rhodophyta obtained in this study were deposited in the GenBank database under accession numbers MZ481947-MZ481964, MZ481932-MZ481946 and MZ481965-MZ482014, respectively.

Raw reads were filtered using FASTP (V0.18.0) to obtain high quality clean reads according to the following rules: (1) remove reads containing more than 10% of unknown nucleotides and (2) remove reads containing less than 50% of bases with a quality (Q-value)>20 (Chen et al., 2018). Paired-end clean reads were merged as raw tags using FLSAH (V1.2.11) with a minimum overlap of 10 bp and mismatch error rates of 2% or less (Magoc and Salzberg, 2011). Noisy sequences of raw tags were then filtered using QIIME (V1.9.1) (Caporaso et al., 2010) under specific filtering conditions (Bokulich et al., 2013) to obtain high quality clean tags. Clean tags were searched against the reference database (http://drive5.com/uchime/uchime_download.html) to perform reference-based chimera checks, using the UCHIME algorithm (Edgar et al., 2011). All chimeric tags were removed to finally obtain effective tags for further analysis.

The effective tags were clustered into operational taxonomic units (OTUs) of ≥ 97% similarity using UPARSE (V9.2.64) (Edgar, 2013). A dominant sequence was selected within each cluster as a representative sequence. The representative sequences were then picked-out to annotate taxonomic assignments using the RDP classifier (V2.2) (Wang et al., 2007) based on the SILVA database (V132) (Pruesse et al., 2007). The OTU sequences belonging to class Actinopterygii were discarded in further analysis. Moreover, if an OTU sequence could not be assigned to any phylum in the SILVA database, it was subjected to a further BLAST search (http://blast.ncbi.nlm.nih.gov/Blast.cgi). The taxonomic classification was then finally confirmed by its similarity to a BLAST result.

Fish muscle tissue were dried at 60°C to a constant weight. Samples were then ground to a fine, homogeneous powder using an automatic sample grinder (Jxfstprp-24, Jingxin Co., Shanghai, China). Samples were taken for carbon and nitrogen stable isotope analysis using a continuous-flow isotope ratio mass spectrometer (Finnigan MAT 253, Thermo Scientific, USA) coupled to an elemental analyzer (Flash EA 1112, Thermo Scientific, USA). The C and N isotope ratios were determined as δ¹³C and δ¹⁵N, respectively, according to the following formula:

where R is the corresponding ratio ¹³C/¹²C or ¹⁵N/¹⁴N, R_sample is measured for fish, and R_standard is an international standard (Pee Dee Belemnite for C isotopes and atmospheric N₂ for N isotopes).

Bar charts of relative abundances of eukaryote sequences found in eight fish intestines and a biplot of stable isotope signatures (δ¹³C and δ¹⁵N) of fish muscles were constructed using Origin 2018 software (OriginLab Co., Northampton, MA, USA). A phylogenetic tree of 18S rRNA sequences related to the phyla Chlorophyta, Ochrophyta and Rhodophyta was constructed using MEGA X and the neighbor-joining algorithm, and the maximum composite likelihood method with bootstrap analyses for 1,000 replicates (Kumar et al., 2018). A heat map of the relative abundance of the 18S rRNA sequences belonging to the phyla Chlorophyta, Ochrophyta and Rhodophyta was individually generated using the OmicShare tools (http://www.omicshare.com/tools). Dominant OTUs from the heat maps were selected to analyze taxonomic assignments using BLAST research. Based on the relative abundance of macroalgae sequences in eight fish intestines, data were transformed as square root with Bray-Curtis similarity resemblance. Then cluster analysis of the Bray-Curtis similarity was done with the Primer 5 software. Statistical differences about the stable isotope values of δ¹³C and δ¹⁵N among the different eight fishes were analyzed, with p<0.05 considered significant.

Intestinal contents of the eight fish species collected were analyzed by microscopy. After thorough washing, anterior intestinal contents of fishes were examined and a high proportion of flocculent detritus was found. Coral sand was only detected in the intestinal contents of C. carolinus and S. argenteus. Importantly, macrophytes (macroalgae or seagrasses) were easily observed in fish intestines ( Table 2 ). The number of macrophyte species consumed by the eight herbivorous reef fishes varied. S. argenteus consumed the highest number of macrophyte species (six species), followed by N. unicornis (four species). However, only a few macrophyte species could be identified morphologically, while other fragments belonging to Chlorophyta, Ochrophyta or Rhodophyta were too small and were difficult to identify. The green algae Valonia ventricosa and Halimeda sp. were detected in the fish intestines of N. unicornis and S. argenteus, respectively. The brown alga Turbinaria ornata and the red alga Acanthophora spp. were observed in the intestinal contents of N. unicornis and S. punctatissimus, respectively. In addition, the seagrass Halophila ovalis was identified as a food source of S. argenteus.

The dietary spectra of the eight fishes were analyzed with 18S rRNA gene high-throughput sequencing at the phylum and class levels ( Figure 1 ). The 15 most abundant phyla accounted for 88.2%–99.3% of the total sequences found in the eight fish species’ intestines ( Figure 1A ). There were marked differences in the relative abundance of the dominant phyla in each species sample. Porifera was the most abundant taxon in C. carolinus, which was further identified as Demospongiae (sponges) ( Figure 1 ). The dominant phyla in K. cinerascens were Bacillariophyta and Porifera. Ochrophyta, Streptophyta and Bigyra were significantly more abundant in K. vaigiensis, S. argenteus and S. punctatissimus, respectively. However, Cnidaria, mainly Anthozoa (corals), were much higher in S. puellus, N. brevirostris, N. unicornis and C. carolinus ( Figure 1 ). Sequences belonging to Streptophyta were identified as H. ovalis, a seagrass species which occurred in the intestines of C. carolinus, N. brevirostris, S. argenteus and S. punctatissimus.

Sequences referring to the phyla Chlorophyta, Ochrophyta and Rhodophyta found in the eight fishes were further analyzed ( Figure 2 ). A total of 18, 15 and 47 OTUs were observed belonging to Chlorophyta, Ochrophyta and Rhodophyta, respectively ( Figure 2A ). A phylogenetic tree of the observed OTUs was constructed and is shown in Figure 3 . Chlorophyta OTUs were most diverse in K. vaigiensis and S. argenteus (≥10 OTUs). The highest numbers of Rhodophyta OTUs (23 OTUs) were observed in K. cinerascens and K. vaigiensis. Ochrophyta OTUs were most abundant in K. vaigiensis (11 OTUs). The relative abundance of macroalgae found in fish intestines are shown in Figure 2B . K. vaigiensis consumed the greatest abundance of macroalgae (76.9%), followed by K. cinerascens (14.5%). N. unicornis preferred brown algae, while S. punctatissimus consumed more green algae. The diet of S. argenteus contained 2.9% green algae, 0.8% brown algae and 1.9% red algae.

The differences in macroalgal composition between the intestinal contents of the eight fish species were analyzed ( Figure 4 and Table 3 ). The species composition of green algae or red algae in the eight herbivorous fishes diets was different ( Figures 4A, C ). Regarding the green algae, Otu000030, identified as Dictyosphaeria cavernosa, was abundant in K. vaigiensis. Otu000055, which was very similar to Ulvella leptochaete, was a dominant component in S. argenteus and K. cinerascens. Both Otu000067 and Otu000102 were affiliated to the genus Cladophora. The occurrence of Otu000067 was higher in K. cinerascens, while Otu000102 dominated in N. brevirostris and S. argenteus. Regarding the red algae ( Figure 4C ), Otu000056 dominated in S. argenteus and showed a high similarity to Spyridia filamentosa. Otu000077 was related to Centroceras hyalacanthum, and was a major component in S. argenteus and S. punctatissimus. Otu000094 matched with Ceramium sinicola and was relatively abundant in K. cinerascens. Otu000200 and Otu000260 were relatively abundant in K. vaigiensis and were similar to Peyssonnelia rumoiana and P. armorica, respectively. Otu000535 was identified as the red alga P. rosenvingei and was very abundant in C. carolinus. The intestinal contents of C. carolinus also contained a dominant OTU (Otu000554), which was matched with Pneophyllum conicum. The dominant OTUs in N. unicornis included Otu000309 and Otu000874, which were similar to Chondrophycus cf. undulates and C. sinicola, respectively. Regarding the brown algae ( Figure 4B ), Otu000009, related to Lobophora variegate, was dominant in the intestinal contents of all the fishes, except S. argenteus. Otu000014 was also abundant in most fishes, especially S. argenteus in which it was most abundant. Cluster analysis indicated that macroalgae consumption from different fish of the same species always grouped together with each other. However, there was a low similarity among the fishes (<52%), only N. brevirostris and N. unicornis group at a high similarity (50.8%) ( Figure 5 ).

Stable isotope values of δ¹³C and δ¹⁵N from the eight herbivorous fish species are shown as a bi-plot ( Figure 6 ). The δ¹³C values across species ranged from -19.2 to -12.3‰. The δ¹³C values were more enriched in C. carolinus (-12.3 ± 0.8‰), which was significant different with other five species, except for N. unicornis (-13.4 ± 2.6‰) and K. vaigiensis (-13.8 ± 1.5‰) ( Figure S2 ). However, S. punctatissimus and N. brevirostris had lower enrichment of δ¹³C values, with -19.2 ± 0.9‰ and -18.0 ± .02‰, respectively. The δ¹⁵N values of fishes varied from 5.1 to 8.3‰. K. vaigiensis (8.3 ± 0.5‰) had significantly higher δ¹⁵N values than others, except for K. cinerascens (p>0.05). N. unicornis showed the lowest δ¹⁵N values (5.1 ± 0.6‰) and exhibited significant different from most species, but not for S. punctatissimus and K. cinerascens. K. cinerascens, S. puellus and S. argenteus had generally similar δ¹³C and δ¹⁵N values with no significant difference. The differences in δ¹⁵N observed in the eight fishes indicated that N. unicornis occupied a relatively low trophic position.

Microscopic examination and 18S rRNA high-throughput sequencing analyses were used to investigate the dietary composition of eight herbivorous reef fish species off the Xisha Islands, China. There were some similarities in the results of the two methods. For example, the seagrass H. ovalis was observed in the intestinal contents of S. argenteus based on microscopic examination, and high-throughput sequencing analysis also found that this seagrass (Liliopsida) dominated the intestinal contents of S. argenteus. Microscopic examination revealed coral sand in the intestinal contents of C. carolinus and S. argenteus. High-throughput sequencing analysis also showed a high abundance of the class Anthozoa (corals) in C. carolinus and S. argenteus.

However, diet composition and breadth may vary according to the different approach used (Nalley et al., 2021). In this study, 18S rRNA high-throughput sequencing analysis provided higher taxonomic resolution of dietary composition, compared with microscopic examination. For example, some taxa that were found to dominate the intestinal contents using molecular analysis, such as Porifera, Cnidaria, Bacillariophyta were not detected by microscopic examination. The possible reasons were as follows. First, small organisms or digested organisms are too difficult to identify using a microscope (Dromard et al., 2015; Kume et al., 2021). In this study, microscopy detected a large amount of detritus in the fish intestinal contents. Dromard et al. (2015) using microscopy, also observed a high proportion of detritus in the intestines of Scaridae. Second, high-throughput sequencing is highly sensitive and can detect traces of DNA in mixed samples. This molecular technique can discriminate OTUs with only a single nucleotide variation within the targeted amplified region (Albaina et al., 2016). Third, the number of sampled fishes in this study was small so that some taxa in fish intestine were not observed using the microscope. For example, previous studies have shown that S. puellus can consume sponges (Hoey et al., 2013). While microscopic examination did not find any sponge remains in fish intestinal contents, Demospongiae (sponges) were identified as the main food for S. puellus using high-throughput sequencing. Leray et al. (2015) also showed that pyrosequencing analysis could broaden the recognition of the food webs of coral-dwelling predatory fish and achieved unprecedented taxonomic resolution of their diets. Nalley et al. (2022) reported that the resolution of metabarcding is greater than visual observation; metabarcding can identify specific taxa within the categories of turf algae that were targeted by herbivorous coral reef fishes, while visual inspection was limited to the identification of turf algae.

Based on microscopy, 18S rRNA high-throughput sequencing and stable isotope analyses, we suggested that K. vaigiensis, N. unicornis and S. argenteus showed a high consumption potential of macroalgae from the coral reef ecosystem. Similarly, previous studies have also shown that K. vaigiensis and N. unicornis were highly effective in consuming macroalgae, especially the brown macroalgae (Lefevre and Bellwood, 2011; Streit et al., 2015; Puk et al., 2016). Similarly, we found that K. vaigiensis and N. unicornis favored brown macroalgae. However, in the northern GBR, Choat et al. (2002) demonstrated the dominant macroalgae genera in the diet of N. unicornis and K. vaigiensis were Turbinara and Dictyota, while Cvitanovic and Bellwood (2009) found that K. vaigiensis in the central region of the GBR preferred the brown macroalgae Sargassum. In this study, K. vaigiensis favored Lobophora and Dictyota, but N. unicornis targeted more on Lobophora and Turbinara. The diet difference of fishes between the GBR and the Xisha Islands may be related with occurrence of the dominant macroalgae in the habitat. Previous study showed that Sargassum was one of the abundant species in nearshore reefs of the GBR (McCook, 1997). However, Sargassum was seldom founded in the Xisha Islands. As for S. argenteus, in the Seychelles of the Indian Ocean, S. argenteus removed a range of substrate algae, including turf algae, macroalgae, seagrass and epiphytic algae (Ebrahim et al., 2020). Here, we also showed S. argenteus can consume macroalgae and seagrass. Furthermore, S. argenteus was supposed to focus more on red and green macroalgae in the northern GBR (Hoey et al., 2013). Here, S. argenteus also was found to favor red and green macroalgae. Of course, macroalgal grazing selectivity of herbivorous fishes can vary with location, season, food availability and according to the other fish species present (Puk et al., 2016).

In addition to identifying the diets of the eight herbivorous reef fishes studied, our results demonstrated the ecological importance of these fishes in removing macroalgae from coral reefs. Herbivorous fishes are broadly classified into four functional groups: excavators, scrapers, grazers, and browsers (Hoey and Bellwood, 2009). Excavators, scrapers and grazers generally consume small macroalgae and algal turfs, while browsers target large, erect macroalgal species and play a critical part in controlling macroalgal spreading (Hoey and Bellwood, 2009; Michael et al., 2013; Ebrahim et al., 2020). Previous studies have shown that C. carolinus, K. cinerascens, K. vaigiensis, N. brevirostris, N. unicornis and S. argenteus in the Moorea, the GBR or other coral reef islands belong to the browser group (Puk et al., 2016; Sura et al., 2021) and this was confirmed in our study. Based on microscopic examination, N. unicornis can consume V. ventricosa and T. ornate, while S. argenteus can consume Halimeda sp. ( Table 2 ). According to the 18S rRNA high-throughput sequencing analysis, we found that C. carolinus, K. cinerascens, K. vaigiensis, N. brevirostris and N. unicornis consumed the brown alga L. variegate, and that K. vaigiensis and N. unicornis foraged especially heavily on this macroalgae. However, S. argenteus fed more on D. linearis. K. vaigiensis played an important role in removing the green alga D. cavernosa, while S. argenteus preferred to feed on the red algae S. filamentosa and C. hyalacanthum. K. vaigiensis also preferred to consume Peyssonnelia spp. The macroalgae mentioned above all belonged to either the upright calcareous or fleshy macroalgae. These two macroalgae groups are problematic as they probably inhibit coral settlement, while the crustose calcareous algae and algal turfs have only minor negative effects on coral settlement (Diaz-Pulido et al., 2010). Overgrowth of Lobophora and Dictyota is widely supposed to reduce coral settlement (Foster et al., 2008; Diaz-Pulido et al., 2010; Evensen et al., 2019; Vieira, 2020) and the spread of Dictyota can even cause coral disease outbreaks (Brandt et al., 2012). Unfortunately, Lobophora and Dictyota, are the dominant species on Indo-Pacific coral reefs (Cardoso et al., 2009; Titlyanov et al., 2017). We found that the nitrogen signature (δ¹⁵N) of N. unicornis was significantly lower than most of the other herbivorous fishes, indicating a high consumption potential of macroalgae. In general, we suggested that K. vaigiensis, N. unicornis and S. argenteus were the key browsers removing macroalgae from reefs off the Xisha Islands. This result is similar to that of a previous view, which recommended that K. vaigiensis, N. unicornis and Siganus canaliculatus were the predominant removers of macroalgae, the “true macroalgae browsers” of coral reefs (Puk et al., 2016). K. vaigiensis and N. unicornis were commonly considered as important macroalgal consumers in the coral reefs of the GBR (Verges et al., 2012).

Herbivorous fishes that eat the entire macroalgal thallus have a great functional impact on macroalgae removal from coral reefs than species that only consume algal fronds (Streit et al., 2015). In this study, K. cinerascens, K. vaigiensis, N. unicornis and S. punctatissimus were found to feed on the entire macroalgal thallus by microscopic examination ( Table 2 ). Streit et al. (2015) also showed that K. vaigiensis and N. unicornis bit on the entire macroalgal thallus in almost 90% of bites. Body size, tooth shape, and feeding behavior of different herbivores may result in these functional differences (Streit et al., 2015). S. punctatissimus can fed on Acanthophora sp., while Acanthophora can inhibit the successful settlement of coral planulae (Vermeij et al., 2009). The red alga, P. conicum is widely distributed on Indo-Pacific coral reefs and can overgrow and kill live corals (Antonius, 2001). Coincidentally, we found that C. carolinus consumed P. conicum. In addition, herbivorous fishes can selectively remove some macroalgae including Ulvella, Cladophora, Sphacelaria, Spyridia, Centroceras, Ceramium and Neosiphonia, which often occur in the Xisha Islands (Titlyanov et al., 2017; Huang, 2018). The spread of macroalgae can negatively affect coral reefs through the inhibition of coral fecundity and growth, the reduction of coral larval settlement and recruitment and the increasing prevalence of coral diseases (Dell et al., 2020). Burkepile and Hay (2008) identified herbivore species richness and feeding complementarity as key factors in the effective suppression of the spread of macroalgae in a coral ecosystem. In summary, we recommend reducing fishing mortality for key browser species, such as K. vaigiensis, N. unicornis and S. argenteus, in order to control the spread of macroalgae on the reefs off the Xisha Islands. Gowri et al. (2016) recommended a ban on fishing herbivorous fishes during the times of spawning aggregations, and a ban on the use of small-mesh gill nets in coral management of India. Dell et al. (2020) also encouraged the government to protect and monitor the key browsers in the coral reefs. Some countries, such as Belize, even have implemented bans on fishing herbivorous fishes (Kramer et al., 2017).