P. haitanensis samples were collected from farming regions in southern and northern China in November 2019. Two farms near coastal water were located in Rizhao City (35°42′N, 119°50′E), Shandong Province, China; and Ningde City (26°85′N, 120°03′E), Fujian Province, China. Three sampling points were designated at each farm; each sampling point was approximately half nautical mile apart. A total of 18 samples were obtained. Each sample contained approximately 100 P. haitanensis thalli. The thalli were placed in sterile bags containing seawater from the same sampling location and transported to the laboratory within 8 h in a low-temperature (4°C) box. Parts of the thalli were cleaned to extract metagenomic DNA, and the rest were stored in a freezer at −80°C.

To remove loosely associated microbes and other attachments, 20 thalli from each sample were randomly selected and washed three times with sterilized seawater (Burke et al., 2009). Then, the clean thalli were cut into fragments of approximately 3 mm × 3 mm using sterilized blades and cleaned again with sterilized seawater. These thalli were used as material for metagenomic DNA extraction. All steps were performed on a clean bench. Metagenomic DNA was extracted using the DNeasy PowerSoil Kit (QIAGEN, Germany), according to the manufacturer’s instructions (Yan et al., 2019; Ahmed et al., 2021). The concentration of the extracted DNA was determined using the NanoDrop ND-2000 spectrophotometer (NanoDrop, Wilmington, DE, United States), and the mass was evaluated using a 1% agarose gel. To minimize extraction bias, the three replicates from each sampling point were pooled, resulting in six DNA samples that were denoted as RZ-A, RZ-B, RZ-C, ND-A, ND-B, and ND-C.

Construction of a metagenomic DNA library and sequencing of paired ends (2 × 150 bp) were performed by Shanghai Persenor according to the standard protocol.¹ Each sample contained approximately 30 GB of raw reads. The 3′ and 5′ end adapters were removed using Trim Galore v0.5.0,² and reads with length < 20 bp were removed. Paired reads were mapped to the P. haitanensis genome (GCA_008729055.1) using Bowtie2 v2.3.2.³ To minimize the impact of host contamination, we filtered out read sequences aligned to the P. haitanensis genome. The PCR replicates were removed using FastUniq v1.1 (Xu et al., 2012). The contamination rate (%) of clean reads was assessed by remapping the reads to the host genome. Non-pareil v3.30 was used to assess the average coverage of each metagenome (Rodriguez-R et al., 2018). The clean reads of each sample were assembled using SPAdes v3.13.0 (metaspades model, kmers were 21, 33, and 55) (Bankevich et al., 2012). Then, we merged all of the unassembled reads and co-assembled them. Finally, all contigs were moved to a single file for subsequent analysis. The assembly results were statistically analyzed using QUAST v4.5 (Gurevich et al., 2013). Contigs with lengths of less than 1,000 bp were filtered. Finally, to check the assembly effect and quality of the data, clean reads were realigned to contigs.

To observe the microbial community, Kaiju v1.5.0 (Menzel et al., 2016) was used for taxonomic assignment based on the clean reads of each sample. Non-redundant NCBI BLAST was selected as the taxonomic annotation database. The rarefaction curves for each metagenome were plotted using R language v4.0.1 [vegan package, rarecurve(), step = 2,000] based on the Kaiju results.

Binning was performed using MetaBAT2 v2.12.1 (Kang et al., 2019), MaxBin2 v2.2.4 (-max_iteration 5) (Wu et al., 2016), and CONCOCT (Alneberg et al., 2014). The “bin_refinement” function of MetaWRAP v1.1.8 (Uritskiy et al., 2018) was used to optimize the draft genomes (complete > 50%, contamination < 10%). The dRep v2.5.4 (Olm et al., 2017) was used to remove the redundant MAGs.

Using the Genome Taxonomy Database R89 (GTDB) (Parks et al., 2018), taxonomy was assigned to MAGs using GTDB-Tk v1.1.0 (Chaumeil et al., 2020). The “pipe” and “appraise” functions of SingleM v0.13.2⁴ were used to scan for single copy marker genes in both the MAGs and the sequenced raw reads to estimate the proportion of marker genes recovered in the MAGs. The “genome” function of coverM v0.5.0⁵ was used to evaluate the relative abundance of MAGs in each sample based on read alignments. MAGs with a cumulative abundance > 1% (sum of each MAG abundance in each sample) were included in the heatmap visualized using R language (packages including pheatmap and RColorBrewer), as shown in Supplementary Figure 1.

We filtered out MAGs with quality < 40% (calculated as “Quality = Genome completeness (%) - [5 × Contamination (%)]”). A concatenated marker gene tree was inferred using the 190 MAGs after quality control, together with 231 available IMG/M⁶ Pyropia isolated genomes (by 29 October 2020) that were ≥ 50% complete and contained ≤ 10% contamination. For comparison, taxonomy was reassigned to the previously Pyropia-associated genomes using the GTDB-R89 database (Supplementary Table 1). Thereafter, the bacterial tree was inferred using GTDB-Tk and embellished using ITOL (Letunic and Bork, 2021). As no archaeal genome was isolated and submitted to public databases before this study, phylogenetic distance (PD) and phylogenetic gain (PG) for the bacterial tree were calculated using GenomeTreeTk v0.0.41⁷ to determine the degree to which the phylogenetic diversity from the current study was added to the tree.

The annotation and annotation functions of EnrichM v0.5.0⁸ were used to annotate MAGs with the Kyoto Encyclopedia of Genes and Genome (KEGG) Orthology database (Kanehisa et al., 2021). Genes that were not annotated were assigned as a “hypothetical protein”. Considering that we obtained incomplete microbial genomes, to assign a pathway/module to MAG, the appropriate KEGG pathway/module needed to be ≥ 60% complete, and all key enzymes must be present (as defined in the literature) (Engelberts et al., 2020). Because the KEGG modules had no auxin biosynthesis module and only a partial cobalamin synthesis module, we created specific modules (Supplementary Note 1). The results of the feature annotation are shown in the phylogenetic tree with KO numbers (Supplementary Figures 2–7).

After the quality control step, 209 GB of data was obtained. Approximately 100% of the reads had a base quality greater than 99% (quality score > 20). The rate of clean reads mapped to the host genome was less than 0.1%.

On average, 47% of the reads were assigned to the taxonomic database. A total of 174 phyla were detected, in which five phyla with relative abundances of over 1% (Figure 1A). A total of 3,433 genera were identified, including 15 genera with relative abundances of more than 1% (Figure 1B). The rarefaction curves (Figure 1C) and non-pareil curves (Figure 1D) indicated that the sequencing data obtained in this study covered more than 95% of the whole microbiome.

Dereplication of the MAGs reduced the total number to 202 (75.05 ± 24.95% completeness, 4.74 ± 4.74% contamination), including 1 archaeal MAG-3 belonging to the phylum Thaumarchaeota (Crenarchaeota as per GTDB) and 201 bacterial MAGs belonging to eight phyla, including Actinobacteriota (12 MAGs), Bacteroidota (100 MAGs), Bdellovibrionota (5 MAGs), Chloroflexota (6 MAGs), Myxococcota (4 MAGs), Patescibacteria (15 MAGs), Proteobacteria (54 MAGs), and Verrucomicrobiota (5 MAGs) (Supplementary Table 2). All dominant taxa with a relative abundance of > 1% were detected in the MAGs. The MAGs contained an average of 3,466 ± 2,747 genes, of which 30.38 ± 20.24% were hypothetical proteins (Supplementary Table 2). Moreover, the analysis of single-copy marker genes presented in the microbial community and MAGs using SingleM showed that 202 MAGs represented 65.2% of the sequenced microbiome. The above results suggested that we obtained a representative dataset of the Pyropia microbiome and were able to obtain the metabolic potential of all dominant taxa.

A concatenated genome tree was constructed from 231 Porphyra/Pyropia-isolated microbial genomes from the IMG/M database and 190 MAGs after quality control in this study (Figure 2). Although previous genomes were widely distributed across the phylogenetic tree, the 190 MAGs from this study still provided a phylogenetic gain (the additional branch length contributed by a set of taxa) of 50% for the bacteria (Table 1).

We searched the MAGs for the six known prokaryotic carbon fixation modules in map00720 (Kanehisa et al., 2021), including the Wood-Ljungdahl (WL) pathway (M00377), reductive citric acid (rTCA) cycle (M00173), 3-hydroxypropionate/4-hydroxybutyrate (HP-HB) cycle (M00375), 3-hydroxypropionate (3-HP) cycle (M00376), dicarboxylate/4-hydroxybutyrate (DC-HB) cycle (M00374), and the Calvin Benson Bassham (CBB) cycle (M00165) (Supplementary Table 3). Only the CBB cycle was observed in MAG-99 and MAG-191 (family Rhodobacteraceae) (Supplementary Figure 2). Although the rTCA and the 3-HP cycles had more than 60% module integrity in some MAGs, both lacked the key enzymes ATP-citrate lyases (aclAB) and malonyl-CoA reductase (mcr), respectively (Zarzycki et al., 2009; Hugler and Sievert, 2011). The other three carbon fixation modules (HP-HB, DC-HB, and WL) were < 60% complete in the MAGs.

Nitrate can be reduced to ammonium via assimilatory nitrate reduction (ANRA) and dissimilatory nitrate reduction (DNRA) (Devol, 2015). ANRA includes assimilatory nitrate reductases (nasA or narB) and ferredoxin-nitrite reductase (nirA), while DNRA includes nitrate reductases (narGHI or napAB) and nitrite reductases (nirBD or nrfAH) (Fiore et al., 2010). The nasA and narB were identified in Acidimicrobiia (3 MAGs), Alphaproteobacteria (12 MAGs), Gammaproteobacteria (8 MAGs), Anaerolineae (1 MAG), and Bacteroidia (9 MAGs) (Figure 3). Meanwhile, nirA was identified in the family DEV007 (2 MAGs). For DNRA, narGHI or napAB were identified in the genera Halocynthiibacter (MAG-90), Arenitalea (MAG-201), and Colwellia (MAG-91). Four phyla, Proteobacteria (20 MAGs), Bacteroidetes (8 MAGs), Actinobacteria (3 MAGs), and Chloroflexota (1 MAG), contained nirBD. On the other hand, denitrification comprises four steps: nitrate reduction (narGHI or napAB), nitrite reduction (nirK or nirS), nitric oxide reduction (norBC), and nitrous oxide reduction (nosZ) (Fiore et al., 2010; Figure 4). The nirK was identified in MAG-3 (family Nitrosopumilaceae) and MAG-201. Furthermore, MAG-201 had norBC and nosZ, whereas nosZ was also identified in MAG-91. The nxrAB encode nitrite oxidoreductase, which catalyzes the nitrite oxidation (Sohaskey and Wayne, 2003). Both genes were identified in MAG-90. Except for the two genes (amoBC) of ammonia oxidation identified in MAG-3, KEGG Orthologys (Kos) associated with nitrification (Kanehisa et al., 2021) were absent from other MAGs, as were the KOs involved in nitrogen fixation (Supplementary Figure 3).

One of the most important mechanisms that allow marine organisms to cope with low orthophosphate (Pi) stress is the utilization of Dissolved Organic Phosphorus (DOP) by alkaline phosphatase (AP, phoABD) (Lin et al., 2016). The AP genes were identified in 136 MAGs, including Actinobacteria (8 MAGs), Bacteroidetes (86 MAGs), Bdellovibrionota (5 MAGs), Chloroflexota (2 MAGs), Chloroflexota (1 MAG), Myxococcota (1 MAG), Patescibacteria (3 MAGs), Proteobacteria (26 MAGs), and Verrucomicrobiota (5 MAGs) (Figure 3 and Supplementary Figure 4).

Microorganisms can metabolize the host taurine via ATP-binding cassette (ABC) transporters (tauACB), and taurine is subsequently converted to sulfite via taurine dioxygenases (tauD) (Rambo et al., 2020; Figure 5). The tauD was identified in MAGs belonging to Actinobacteria (8 MAGs), Bacteroidota (1 MAG), Chloroflexota (2 MAGs), Alphaproteobacteria (9 MAGs), and Gammaproteobacteria (4 MAGs), of which tauACB transporters were also identified in Actinobacteriota (6 MAGs), Chloroflexota (2 MAGs), Alphaproteobacteria (5 MAGs), and Gammaproteobacteria (3 MAGs) (Figure 3). In addition to the cleavage of taurine, sulfite may also originate from sulfide by dissimilatory sulfite reductase (dsrAB) (Engelberts et al., 2020; Figure 5). Both genes were identified in MAG-68 (Gammaproteobacteria) (Supplementary Figure 5). In turn, sulfite can be oxidized to sulfate via sulfite dehydrogenase/oxidase (sorAB/SUOX/soeABC) (Rambo et al., 2020). The SUOX/soeABC were identified in all phyla except Bdellovibrionota, Patescibacteria, Verrucomicrobiota, and Crenarchaeota (Figure 3). The assimilatory sulfate reduction module (M00176), which can reduce sulfate to sulfide (Kanehisa et al., 2021), was found in the phyla Actinobacteria (4 MAGs), Bacteroidetes (47 MAGs), Chloroflexota (5 MAGs), Thaumarchaeota (1 MAG), Myxococcota (3 MAGs), Proteobacteria (9 MAGs), and Verrucomicrobiota (4 MAGs) (Figure 3). The sulfur oxidation system (SOX) is a pathway that includes soxAB and soxXYZ (Rambo et al., 2020). These genes were found in five Alphaproteobacteria and six Gammaproteobacteria MAGs.

In the present study, potential cobalamin producers in MAGs were identified in different lineages, including the family Cellvibrionaceae (1 MAG), Flavobacteriaceae (1 MAG), Rhizobiaceae (2 MAGs), Rhodobacteraceae (13 MAGs), SZUA-35 (1 MAG), UBA10066 (1 MAG), and UBA6668 (3 MAGs) (Figure 3). Except for MAG-116, MAG-184, and MAG-48, 19 MAGs had the potential to synthesize cobalamin via both aerobic and anaerobic routes (Supplementary Figure 6). In addition to de novo biosynthesis, cobalamin may also originate from the lower axial ligand conversion of pseudocobalamin (Hoffmann et al., 2000). The bluB, which encodes for 5,6-dimethylbenzimidazole synthase (DMB), was present in the families Rhodobacteraceae (6 MAGs), Rhizobiaceae (2 MAGs), Nitrosopumilaceae (1 MAG), and Flavobacteriaceae (1 MAG). Amidohydrolase (CbiZ), and cobinamide-phosphate synthase (CbiB) can remodel pseudocobalamin into cobalamin by replacing adenine with DMB (Yi et al., 2012). These two genes were only identified in MAG-179 (family UBA11606).

The btuB, which encodes for TonB-dependent outer membrane cobalamin receptor and transporter (Noinaj et al., 2010), was identified in the phyla Bacteroidetes (85 MAGs), Bdellovibrionota (3 MAGs), Proteobacteria (19 MAGs), Myxococcota (2 MAGs), and Verrucomicrobiota (4 MAGs) (Supplementary Figure 6). However, few cobalamin biosynthesis genes were found in these MAGs. Moreover, 157 MAGs lacked metE (B₁₂-independent methionine synthase), while 115 MAGs only included metH (B₁₂-dependent methionine synthase) (Supplementary Table 3).

We searched all the MAGs for the four known 3-indoleacetic acid (IAA) biosynthesis pathways: the indole-3-acetamide (IAM), indole-3-pyruvate (IPyA), tryptamine (TAM), and indole-3-acetonitrile (IAN) pathways (Supplementary Note 1). The metabolic potential of the IAM pathway was present in the families Granulosicoccaceae (3 MAGs) and UBA11606 (2 MAGs) (Figure 3). The TAM pathway was present in 104 MAGs, including the phyla Actinobacteria (8 MAGs), Bacteroidetes (47 MAGs), Bdellovibrionota (1 MAG), Chloroflexota (4 MAGs), Myxococcota (2 MAGs), Patescibacteria (8 MAGs), Proteobacteria (33 MAGs), and Verrucomicrobiota (1 MAG) (Figure 3). However, for the IPyA pathway, the gene encoding the key enzyme (indolepyruvate decarboxylase, ipdC) (Duca et al., 2014) was absent in all MAGs (Supplementary Table 3). Furthermore, although some MAGs (7 MAGs in order Chitinophagales) were contained YUCCA (encoding the homologs of indole-3-pyruvate monooxygenase) (Mano and Nemoto, 2012), while Tam1 and IL4I1 (encoding aminotransferases and L-amino-acid oxidase, respectively) were absent (Supplementary Table 3). The enzymes encoded by Tam1 and IL4I1 can transform tryptophan into IPyA (Duca et al., 2014). More than 60% of the genes necessary for carrying out the IAN pathway were found in 11 MAGs; however, indoleacetaldoxime dehydratase (CYP71A13) and myrosinase (E3.2.1.147) were absent (Supplementary Figure 7). These two genes are related to the conversion of indole-3-acetaldoxime (IAOX) and glucobrassicin to IAN (Zhang et al., 2019).

Microorganisms can promote plant growth indirectly by converting the ethylene precursor 1-aminocyclopropane-1-carboxylate (ACC) to α-ketobutyrate and ammonia by ACC deaminase (acdS) (del Carmen Orozco-Mosqueda et al., 2020). The acdS was identified in the phyla Actinobacteria (4 MAGs), Bacteroidetes (77 MAGs), Proteobacteria (7 MAGs), and Chloroflexota (1 MAG) (Figure 3), of which 46 MAGs are also potential IAA producers. MAG-144 may contain both TAM and IAM pathways (Supplementary Figure 7).

The products of photosynthesis by plants and algae, such as rhamnose, xylose, and glucose, can provide a carbon source for epiphytic microorganisms (Bruckner et al., 2011; Rigonato et al., 2012; Thapa and Prasanna, 2018). Microorganisms that fix carbon make epiphytic microorganisms less dependent on photosynthetic products and allow them to set a niche for the colonization of other microbes (Rigonato et al., 2012; Thapa and Prasanna, 2018; Flores-Nunez et al., 2020). However, it is still unclear whether all epiphytic microbes are dependent on the carbon provided by Pyropia. Here, only two members of Rhodobacteraceae (MAG-99 and MAG-191) fix carbon using the CBB cycle (Supplementary Figure 2), suggesting that epiphytic microorganisms were largely heterotrophic need exogenous carbon to support their life activities. In addition, host-acquired autotrophic carbon from microbial fixation could also contribute to overall high environmental tolerance, particularly to ocean warming (Bell et al., 2018). However, the relative contribution and mechanisms of microbial carbon fixation to marine algal nutrition should be studied further.

Except for Cyanobacteria, plant cannot fix nitrogen on their own, and the process of nitrogen fixation usually requires the assistance of plant symbiotic microorganisms, such as rhizobia (Cheng, 2008; Kuypers et al., 2018). In our study, no genes associated with nitrogen fixation were found in MAGs, suggesting that the P. haitanensis epiphytic microbiomes are unlikely to catalyze the conversion of airborne N₂ to NH₃. Within the nitrate reduction reaction, microorganisms can convert nitrate and nitrite to ammonium through ANRA and DNRA (Devol, 2015). A recent study showed the presence of ammonium transporters in P. yezoensis (Kakinuma et al., 2017), implying that microbial ammonium may also be trade with the host. Excess nitrite has been shown to have toxic effects on organisms, such as affecting the proton permeability of cell membranes (Vadivelu et al., 2006; Zhou et al., 2010) and inhibiting photosynthesis (Singh et al., 2007). The microbial consumption of nitrite also reduces the risk of excessive nitrite (Vaucheret et al., 1992; Duncanson et al., 1993). Nitrate or nitrite can also be converted to nitrogen oxides (N₂O and NO) via denitrification (Fiore et al., 2010). In addition to being released into the environment, NO is an effective signaling molecule in plant-rhizobacteria interactions, promoting plant growth and development in an auxin-dependent manner (Lamattina et al., 2003; Molina-Favero et al., 2008). N₂O is a potent greenhouse gas and ozone-depleting substance, and N₂O emissions can cause atmospheric pollution (Ravishankara et al., 2009). Hence, MAG-91 and MAG-201 may also contribute to the reduction of N₂O emissions to the environment.

Phosphorus is a central component of nucleic acids and phospholipids and plays a central role in the production of chemical energy (Forlani et al., 2008). However, algae grown in coastal regions may experience Pi starvation because of the rapid consumption of Pi for photosynthesis (Lin et al., 2016). Moreover, anthropogenic N also increases the N:P ratio in receiving waters (Zhang et al., 2007; Zhou et al., 2017), thus limiting the Pi of surrounding organisms. Although there are other forms of phosphorus sources in the ocean, organisms preferentially utilize Pi as they can be taken up directly (Huang et al., 2005; Lin et al., 2016). The major mechanism by which marine phytoplankton and bacteria can convert DOP to bioavailable Pi is the induction of alkaline phosphatase (phoABD) (Suzumura et al., 2012; Lin et al., 2019). Here, phoABD were identified in 136 MAGs (Supplementary Figure 4). DOP hydrolysis has costly energetic metabolic activity because of the stable C-P bonds (Lin et al., 2016). In terrestrial plants, while both plant and microbial phosphatases are efficient in releasing Pi from soil organic phosphorus, there is evidence that microbial enzymes show a higher efficiency of phosphorus release (Tarafdar et al., 2001). Therefore, it is possible that the “extra” microbial Pi can be assimilated to support host metabolism and growth (Jones, 1972; Cole, 1982; Currie, 1990; Hoppe, 2003).

Taurine, dietary fibers, polyunsaturated fatty acids, and sulfated polysaccharides are enriched in Pyropia (Cao et al., 2016; Cho and Rhee, 2020). Our results showed the potential of 16 MAGs to obtain foreign taurine via ABC transporters and convert it to sulfite via dioxygenases (Supplementary Figure 5). Sulfate produced during dissimilatory sulfate oxidation and dissimilatory thiosulfate oxidation may provide the raw material for assimilatory sulfate reduction for cysteine (Takahashi et al., 2011) and other cell material biosynthesis. Similar to the nitrogen cycle, genes required for certain reactions in the sulfur cycle have been identified only in specific microbial lineages, implies that they may play prominent roles in nitrogen/sulfur cycling within the Pyropia phycosphere.

In many organisms with metH, vitamin B₁₂ usually acts as an enzyme cofactor participating in catalyzing primary biochemical reactions (Sanudo-Wilhelmy et al., 2014), such as amino acid (Banerjee and Matthews, 1990) and DNA synthesis (Blakley and Barker, 1964). Currently, cobalamin is thought to be synthesized by a relatively small set of prokaryotes, and algae may obtain cobalamin from these associated microbes (Giovannoni, 2012; Lynch and Neufeld, 2015). Our previous study revealed six potential cobalamin producers, all of which belong to Alphaproteobacteria (Wang et al., 2021). Here, 22 MAGs in five phyla were considered potential producers of cobalamin through aerobic or anaerobic pathway (Supplementary Figure 6). This result suggested that vitamin B₁₂ producers were not restricted to a particular microbial lineage. Another source of cobalamin is the substitution of the adenine lower axial ligand of pseudocobalamin by DMB (Helliwell et al., 2016). Pseudocobalamin is a vitamer that can be synthesized in cyanobacteria (Hoffmann et al., 2000). Furthermore, amidohydrolase (cbiZ) and cobinamide-phosphate synthase (cbiB) can replace the adenine lower axial ligand with DMB (Yi et al., 2012). In this study, 10 MAGs were identified to have DMB synthesis genes, and one MAG (MAG-179) was identified to have remodeled pseudocobalamin genes. These MAGs may play a key role in maximizing the effect of cobalamin production (Lu et al., 2020). In terms of potential cobalamin consumption, we found 113 MAGs with the TonB-dependent cobalamin transporter gene (btuB) and 115 MAGs with the B₁₂-dependent methionine synthase gene (metH). The MAGs with btuB had low complete of the cobalamin synthesis pathway (Supplementary Figure 6). This phenomenon was also presented in soil cobalamin research (Lu et al., 2020); our study was more specific by presenting at the individual level. Both btuB and metH were identified in more than half of the MAGs, suggesting that cobalamin producers play important roles in the microbial community (Banerjee and Matthews, 1990; Lu et al., 2020).

Auxin is one of the important plant hormones that influence plant growth and development (Vanneste and Friml, 2009; Bidon et al., 2020). In the rhizosphere, the potential to synthesize auxin is considered the main characteristic of plant growth-promoting rhizobacteria (PGPR) (Spaepen et al., 2007). A previous study showed that co-culture with IAA producers can positively affect the growth of P. yezoensis (Matsuda et al., 2018). Furthermore, bacterial IAA is also reported to regulate microbial physiology, including overcoming stress, regulating cellular processes, and providing a carbon and nitrogen source (Bianco et al., 2006; Tromas and Perrot-Rechenmann, 2010; Van Puyvelde et al., 2011). Here, the IAM and TAM pathways were present in 109 MAGs, including diverse microbial lineages (Supplementary Figure 7). Both pathways are also widely distributed in rhizobacteria (Zhang et al., 2019). The IPyA pathway was usually discovered in plant growth-promoting rhizobacteria (Fu et al., 2012); however, the key gene (ipdC) was absent in the MAGs (Supplementary Figure 7). Notably, although the IPyA and IAN pathways lack genes encoding key enzymes, they both have a high degree of integrity in some MAGs (Supplementary Table 3). Therefore, it does not exclude the possibility that bacteria can synthesize IAA from intermediates of other microbial lineages (Zhang et al., 2019), even of host (Amin et al., 2015).

Ethylene is associated with plant senescence, abscission, flower development, and pathogen defense signaling (Lin et al., 2009; Van de Poel et al., 2015). Ethylene biosynthesis is induced under diverse stress conditions, thereby inhibiting plant growth (Abeles et al., 2012). In this process, ACC is an important intermediate (Lin et al., 2009). A previous report showed that exogenous addition of ACC can increase ethylene production and promote the formation of spermatia and zygotospores in P. yezoensis (Uji et al., 2020). In the rhizosphere, microbial ACC deaminase can metabolize ACC exuded by plants, resulting in a reduction or stalling of ethylene biosynthesis inside the host owing to a decrease in the precursor (Fu et al., 2012), and it is also another feature of PGPR (Wang et al., 2007; Bulgarelli et al., 2013). In this study, the 89 MAGs possessed the ACC deaminase gene, of which 46 MAGs also possessed IAA synthesis potential (Supplementary Figure 7). These MAGs with “dual identity” may be advantageous for future screening of beneficial bacteria (Lugtenberg and Kamilova, 2009).

Although archaea are extremely widespread in the biosphere (Vaucheret et al., 1992), their potential role in macroalgae is rarely mentioned (Trias et al., 2012; Karray et al., 2017; Ghosh et al., 2019), especially in Pyropia. In this study, we obtained an archaeal draft genome (MAG-3) belonging to the Nitrosopumilaceae family. Our result filled a gap in the understanding of the archaeal genome in the epiphytic microorganisms of Pyropia. Further functional analysis of MAG-3 revealed its potential role in MAG-3 ammonia oxidation (Supplementary Figure 3), nitrite reduction, DMB biosynthesis, and assimilatory sulfate reduction (Supplementary Figures 5, 6). However, there may be some other functional features that have not been identified owing to the incomplete genome.

We also evaluated the differences between microbial communities though the abundance of MAGs (Supplementary Figure 9). Regarding inorganic salt metabolism, the relative abundance of MAGs with DOP metabolic potential was higher in the ND group (RZ vs. ND: 56.6% vs. 78.31%); in contrast, the relative abundance of MAGs with sulfur and nitrogen metabolic potential was higher in the RZ group (S: 61.58% vs. 52.93%; N: 17.16% vs. 15.52%). In terms of phytohormone anabolism, the relative abundance of MAGs with IAA synthesis potential exceeded 60% in both groups (RZ vs. ND: 77.58% vs. 61.76%); the relative abundance of MAGs with metabolic ACC potential was higher in the ND group (35.78% vs. 53.05%). Both environment and algae are capable of influencing the community structure of epiphytic microorganisms, so we speculate that the reason for these differences may be the result of a combination of environmental and algal effects (Yan et al., 2019; Wang et al., 2020; Ahmed et al., 2021).

It should be emphasized that our results and conclusions were based on bioinformatics methods. Although we recovered 65.2% of the sequencing community, the remaining 35.8% of unbinned sequences may contain functions that we have not characterized. In addition, bioinformatics-based predictions may include incomplete metabolic pathways because of incomplete genome analysis. We believe that future studies aimed at understanding Pyropia as a holobiont should also move toward the retrieval of host genes and genomes so that the metabolic potential of Pyropia and its epiphytic microorganisms can be assessed for metabolic complementarity.