This study was carried out on an east-facing flat rocky platform at Coalcliff, New South Wales, Australia (-34.248°, 150.978°) and lasted for 16 months. At this site, thalli of Hormosira were ~15cm in length, forming a bed of 100% canopy cover in interspersed patches of ~150m². The experiment was setup in October 2018 and consisted of 25 randomly selected 50x50cm plots, ~3m apart, randomly assigned to five experimental treatments that were spatially interspersed on the rocky platform. Disturbance treatments included plots where Hormosira canopy was 1) completely (100% removal, zero density of Hormosira) or 2) partially removed (50% removal and 43.5 ± 3.2 ind/m² SE of Hormosira density), 3) undisturbed Hormosira control plots (0% removal), 4) procedural control plots where the canopy was disturbed through physically handling and shaking H. bansksii, simulating the manual disturbance done in the disturbance treatments, but not removed (PC, both control treatments with Hormosira density of 87 ± 6.4 ind/m² SE), and 5) plots with bare rock where Hormosira was not present (n=5 for each treatment).

The epilithic microbial community was sampled during low tides, two (Time 1: 6/11/2018), three (Time 2: 26/11/2018), twenty-eight (Time 3: 19/5/2019) and sixty-one weeks (Time 4: 8/1/2020) after the initial disturbance. Sampling times were chosen to determine short (T1 and T2) and long term (T3 and T4) responses of both Hormosira and the epilithic bacterial community. Due to the harsh environmental conditions of the intertidal, It was expected that monitoring at a short term would provide us with valuable information on the immediate changes of bacterial communities post-disturbance. Comparatively, long-term sampling it was thought that due to Hormosira’s known slow growth rate, potential recovery to produce new canopy cover would only occur after approximately ~ 1 year post disturbance (as seen in Pocklington et al., 2019; Cameron et al., 2021). At each time, an epilithic bacterial sample was taken by swabbing the substratum within a 5x5 cm quadrat at each plot for 30 seconds using a sterile cotton swab (Marzinelli et al., 2015). To ensure that the swabbed area of the substratum was only sampled once during the experiment, each plot was divided into four quarters using the diagonals and the 5x5cm quadrat was placed in each quarter once. In plots where Hormosira was not removed (i.e., control and procedural control) or was only partially removed (50% cover), a second bacterial swab sample was also taken from the surface of Hormosira laminae (similar area as above) to compare understory epilithic and host-associated epiphytic bacterial communities. Prior to swabbing, benthic or algal surfaces were rinsed with 0.22 µm filtered seawater to remove any unattached epibionts. Swabs were stored immediately inside sterile cryogenic tubes and liquid nitrogen, transported to the University of New South Wales, Sydney; and kept at -80°C until DNA extractions were performed. Unsuccessful processing during molecular analysis, due to contamination during DNA extractions and/or low PCR amplification, resulted in different number of samples among times (epilithic: Time 1: n = 25; Time 2: n = 24; Time 3: n = 19; Time 4: n = 19; and epiphytic: Time 1: n = 14; Time 2: n = 12; Time 3: n = 10; Time 4: n = 8).

Following the initial disturbance, the i) percent canopy cover of Hormosira and ii) organisms comprising the benthic understory (i.e., the substratum area that was swabbed) in each plot were monitored through photographic techniques at the same sampling times as the microbial swabs. Hormosira canopy cover was quantified (as plots with 50% density removal might still have high canopy cover) via the online machine learning engine, CoralNet, which performs a semiautomatic point annotation, based on specified classifying labels (Beijbom et al., 2015). These classifying labels were verified through the Collaborative and Automated Tools for Analysis of Marine Imagery classification scheme (CATAMI; v.1.2, Althaus et al., 2015) and included point annotations for seven taxa/groups: Hormosira, calcareous encrusting tube worms (TW), articulated calcareous red algae (ACR), brown laminate macroalgae (BLM), turf algae (TA), grazers (i.e., gastropods), calcareous crustose algae (CCA: healthy and BCCA: bleached) and two abiotic variables, sand and rock (examples of each category can be seen in Figure S1 ).

DNA was randomly extracted from each swab sample using the Powersoil DNA Isolation Kit (Mo Bio Laboratories #12888-100) following the manufacturer guidelines. DNA extracts were stored in a freezer at -20°C until PCR amplification with primers 341F (5’ -CCTACGGGNGGCWGCAG- 3’) and 805R (5’ -GACTACHVGGGTATCTAATCC- 3’) which target the V3-V4 regions of the 16S rRNA gene (Klindworth et al., 2013). Agarose gel electrophoresis, Nanodrop 1000 and the Qubit 2.0 fluorometer (Thermo Fisher Scientific) were used to check the quality and quantity of the amplicons before being sent for sequencing in an Illumina MiSeq 2000 Platform at the Ramaccioti Centre for Genomics.

Gene sequence reads were quality filtered without specified maximum expected errors (as suggested by Prodan et al., 2020) and maximum truncation lengths assigned to forward and reverse reads independently. Trimming parameters for maximum truncation were decided upon inspection of the quality error plots of both reads with any base pairs with a median Q score below 30 removed. Using the untrimmed sequences, the maximum error rates were calculated and included in the DADA2 denoising model. Sample inference was performed using the DADA2 denoising algorithm and all resulting denoised paired reads were merged to form unique amplicon sequence variants (ASV). These unique ASV sequences were used to construct an abundance per sample table and subsequently used to detect and remove chimeric sequences (consensus method; Callahan et al., 2016). An initial taxonomic assignment to the ASV table without chimeric sequences was done using SILVA v. 138.1 (Quast et al., 2013) to increase the detection of chloroplasts and mitochondrial ASV. These were removed from the ASV table, and a second taxonomic assignment was done using the Genome Taxonomy Database, which provides a higher taxonomic resolution (GTDB, Parks et al., 2022). The DADA2 pipeline was performed using R v.4.1.1 and the dada2 package v.1.26 (Callahan et al., 2016). An average of 76.3%± 0.65 SE of total reads was kept at the end of the bioinformatic pipeline.

Two separate data subsets were created from the original bacterial ASV table: (1) samples taken exclusively from biofilms in the benthic substrate (hereafter Dataset 1, 87 samples in five disturbance treatment levels) and (2) samples taken from host-associated epiphytic bacterial communities on the surface of Hormosira and samples from the nearby understory epilithic bacterial community within the same plot (hereafter Dataset 2, 44 samples from treatments only including 50% and 0% Hormosira removal treatments). Statistical analyses of each of the two data subsets were performed separately. Singletons and sequences with low abundance (i.e., <0.01% of the total) were removed and data sets were normalized independently using DESeq2’s median of rations method to account for heterogeneous library sizes (DESeq2, package phyloseq v.1.40.1; Love et al., 2014). Abundance tables for both analyses were independently square-root transformed before statistical analyses (Swift et al., 2023). An ASV accumulation curve was also constructed for each data subset to evaluate the sampling effort to effectively describe the overall bacterial community in the experiment (package vegan v.2.6-4, Oksanen et al., 2013).

A total of 6,418 microbial ASVs across 81 samples were included in the analysis and sampling effort was found to be adequate for this community ( Figure S2A ). All alpha diversity indices show a significant effect of disturbance treatment, independent of sampling time ( Figure 1 and Table S1 , Richness, F_4,76 = 15.68, p<0.001; Shannon-Wiener diversity: F_4,76 = 24.2, p<0.001; evenness: F_4,76 = 20.95, p<0.001). Control plots (0% removal and PC) had higher bacterial diversity (p<0.032) and evenness values (p<0.010) compared to plots where Hormosira had been completely cleared. However, there was no difference between these controls and the 50% removal treatments (see post hoc tests in Table S2 ). Bacterial richness (decrease of 59 ± 2.7% SE; p<0.008), diversity (decrease of 27 ± 2.2% SE; p<0.001) and evenness (decrease of 15 ± 1.8% SE; p<0.023) was lower in bare rock plots compared to all other disturbance treatments ( Figure 1 and post hoc tests in Table S2 ). Bacterial richness, diversity and evenness were positively correlated with canopy cover of Hormosira ( Figure 2 ).

There was an interaction between disturbance and sampling time on the structure (i.e. composition and relative abundance of ASVs) of understory epilithic bacterial communities ( Figure 3 and Table S3 , PERMANOVA, F_12,69 = 1.22, p<0.001). Structural intragroup variability among disturbance treatments was found (PERMDISP, Table S3 : F_4,76 = 3.35m, p=0.035), but only between the procedural controls and other disturbance treatments (see post hoc tests Table S4 ). Thus, this structural intragroup variability was not found to be a driver of the differences found with the PERMANOVA. The epilithic bacterial community structure on bare rock plots consistently differed from all other treatments at all sampling times, except T3, when they did not differ from plots with 50% removal and procedural controls ( Figure 3 and Table S5 ). Bacterial community structure in 100% Hormosira removal plots was distinct from those on bare rock and other treatments at most sampling times ( Figure 3 , post hoc tests in Table S5 ), except for T3 when they did not differ from 50% removal and controls (p>0.09). No differences in epilithic bacterial community structure were found at any time point between control and 50% removal plots ( Figure 3 and Table S5 ).

A differential abundance analysis was used to identify epilithic bacterial ASVs that differed in abundance between Hormosira disturbance treatments and across time. For this first model, 40 bacterial ASV were identified to be affected by these factors, but only the 10 most abundant (0.06%< and >0.005%) and prevalent ASVs (>34% present across all samples) were selected. Results from this analysis show that soon after the removal of Hormosira (T1, November 2018-1), treatments with bare rock and 100% removal had similar abundances of bacterial groups such as Verrucomicrobiae (i.e., genus Rubritalea) and Bacteroidia (i.e., genera Winogradskyella, Olleya and an unidentified ASV of the family Flavobacteriaceae), but with lower relative abundance compared to other treatments ( Figure S3 ). After T1, all of these ASV decreased in abundance across the following sampling times regardless of the Hormosira disturbance treatment, while other groups had a more complex temporal abundance pattern. For example, one ASV assigned to the genus Vibrio and one to unidentified Pirellulales, had a higher abundance in the control group compared to other removal treatments, but this pattern was not constant through time ( Figure S4 ). Comparatively, another ASV assigned to the genus Vibrio, a Cetobacterium and an unidentified Phormidiaceae (Cyanobacteria), had a higher abundance in 100% Hormosira removal plots than in those assigned to other treatments, but only at some sampling times ( Figure S5 ).

There were differences in epilithic bacterial community structure over time for treatments with 0% and 50% removal, as well as procedural controls (PERMANOVA, Table S3 and post hoc tests, Table S7 ). The differential abundance analysis focusing on the most prevalent ASV, showed a similar pattern with some groups increasing in the last two sampling times across all treatments, except bare rock (e.g., Unidentified genus of Pirelullales, Figure S4 ) while other ASV decreased substantially in the latter time points (e.g., genus Olleya, unidentified genus of Flavobacteriales, genus UBA9320 and one Vibrio, Figures S3–5 ). Relative abundances of the ten selected ASV remained constant through time and generally with lower abundances in bare rock compared to other Hormosira removal treatments.

The effect of the experimental disturbance to the canopy Hormosira changes through time ( Figure S6 and Table S8 ; interaction, F_12,76 = 19.79, p<0.001). After the initial disturbance, a lower canopy cover was predictably found in the plots with complete removal of Hormosira and bare rock compared to other disturbance treatments, a trend that continued throughout the experiment ( Figure S6A and see post hoc tests in Table S9 , overall t ratio<-3.04, p<0.03). However, treatments with complete removal of Hormosira experienced a rapid recovery in Hormosira cover from the time of disturbance to the end of the experiment (increase of 68 ± 8% cover from T1) ( Figure S6B and post hoc tests in Table S10B , t ratio< -4.47, p<0.001). No differences in canopy cover were found between plots with 50% removal and both control treatments at any stage of the experiment ( Figure S5A and Tables S9–10 ).

There were no significant differences in the covers of other understory macrobenthic components across disturbance treatments, although values of most groups (BLM, ACR, TW, TA and CCA) were generally lower in bare rock plots ( Figures S7–9 ). The exception to this pattern were grazers, which were observed to have higher coverage values in bare rock plots compared to other treatments, and the amount of sand coverage which was higher in plots with 50% and 0% Hormosira removal compared to bare rock ( Figures S7A–B ). The dbRDA, marginal test and vector fit determined that Hormosira canopy cover was the main and most consistent driver of the structure of the understory epilithic bacterial communities across all sampling times albeit explaining a small amount of the total variation ( Figure 4 , and Table S11 ; marginal tests: p<0.015; envfit: r² >0.82). Other understory benthic components such as rocky substrate, turfing algae, grazers, tube worms, brown laminate macroalgae, bleached and unbleached CCA, were found to influence the overall structure of the epilithic bacterial community across more than one sampling time (marginal tests: p<0.040) but to a lesser magnitude ( Table S11 , envfit: r²<0.43, p<0.006) or limited to specific sampling times (e.g., CCA, T3:May 2019, marginal tests: p=0.001 and envfit: r2 = 0.86, p=0.02).

A total of 4519 bacterial ASVs across 44 samples were included in the second dataset that focused on a comparison between host associated epiphytic bacteria in Hormosira and the understory epilithic bacterial communities. These ASVs provided a robust representation of the sampled communities ( Figure S2B ).

There was a significant interaction among disturbance, sampling time and substratum type on bacterial richness ( Figure S10A and Table S12 , F_3,40 = 3.13, p=0.05). Pairwise contrasts showed that bacterial richness was higher on rocky substrata than on the surface of Hormosira, consistently across sampling times and disturbance levels (summary of pairwise comparisons in Table S13 ). Likewise, bacterial diversity (F_1,42 = 116.33, p< 0.001) and evenness (F_1,42 = 67.35, p<0.001) were higher on rocky substrata than on Hormosira ( Figures S10B, C and Table S12 ), but they did not vary with time or disturbance level.

Disturbance (F_1,42 = 1.45, p<0.001), sampling time (F_3,40 = 3.08, p<0.001) and substratum type (F_1,42 = 31.26, p<0.001) independently influenced the structure of the whole bacterial community (PERMANOVA, Figure 5 and Table S14 ), despite variability among samples within each substratum types ( Figure 5 and Table S14 , PERMIDISP, F=66.76, df=1 and p=0.001). A total of 630 ASVs were found to differ significantly in relative abundance between the understory substrata and Hormosira surface samples. Analysis of the most abundant and prevalent bacterial ASVs revealed that the surface of Hormosira had higher relative abundances of the classes Bacteroidia (i.e., identified genera: Maribacter, Haliscomenobacter, Unidentified Saprospiraceae and Unidentified Bacteroidia), Verrucomicrobiae (i.e., genus Rubritalea), Alphaproteobacteria (i.e., genera Sulfitobacter and Octadecabacter), Gammaproteobacteria (i.e., Unidentified genus) and Aciidomicrobia (i.e., MedAcidi-G1) compared to the adjacent understory substrata ( Figure S11 ). A full list of bacterial genera that differed between rocky substrata and Hormosira samples is presented in Table S15 .

Understory epilithic microbial communities are key components of biodiversity in seaweed-dominated systems, influencing the recruitment of seaweeds, invertebrates and regulating species interactions (Underwood, 1978; Pedicini et al., 2023; and Hawkins & Hartnoll, 1983). Understanding how and under which conditions these microbial changes may cascade through the system is critical to better predict and manage responses to environmental disturbances. Here, we found that bacterial changes were not related with the assemblage of understory macroorganisms, suggesting a direct effect of the macroalgae canopy removal.

Full, but not partial, removal of the habitat-forming seaweed Hormosira banksii had a strong impact on canopy cover and caused changes in the understory epilithic bacterial communities that persisted for ~1.5 years, despite the appearance of new Hormosira recruits. These fully cleared plots had the lowest bacterial richness (20-30% lower), diversity (10-14% lower), evenness (7-10% lower) and a different community structure compared to partially cleared (50% density) or undisturbed plots. Hormosira may facilitate the recruitment and development of macroalgae, including conspecifics, and other benthic macroorganisms through stress amelioration and a nursery effect given by its canopy cover (Cameron et al., 2021; Lewis et al., 2021), and likely have a similar effect on the bacterial communities beneath them. Complete removal of canopies may have influenced the structure of the epilithic bacterial community through a variety of mechanisms including through changes on the transfer rates of nutrients (e.g.DOC, labile sugars or hydrophobic molecules), allelochemical agents (e.g. antibacterial agents) or free-living microorganisms from the diffusive boundary layer of the blade or holdfast to the understory substratum (Pfister et al., 2019; Elsherbini et al., 2023).

In contrast to fully cleared plots, partial reductions of Hormosira density (i.e., 50% removal) had little effect on the understory epilithic bacterial communities. In these plots, canopy cover was still relatively high throughout the experiment (above 73% of cover) and probably offered a similar benthic understory environment to the canopy on undisturbed plots. Thus, a moderate thinning of canopies would have limited effects on the structure of the epilithic bacterial community since it would not significantly modify their chemically mediated effects nor their influence on environmental conditions. As shown in other systems (e.g., Cline et al., 2014 or Butitta et al., 2017), shifts between alternative states are often regulated by critical thresholds. Although very few studies have investigated transitions between alternative states on intertidal rocky shores, Rindi et al. (2018) showed that the shift in dominance from canopy to turf-forming macroalgae on a Mediterranean rocky shore was initiated by exceeding a reduction of 75% of the canopy cover. The responses of the microbial community to the total and 50% removal of Hormosira, generating covers of ~10% and ~73%, respectively, may suggest that the decline of Hormosira could cause marked changes to epilithic bacterial communities only when exceeding a critical threshold of cover.

Most of the epilithic bacterial groups on understory substrata are common in coastal waters (Freitas et al., 2012; Handley & Lloyd, 2013), sediments (e.g. genus Rubritalea, Lee et al., 2016) and below macroalgal canopies in temperate rocky systems (e.g. class Bacteroidia, Elsherbini et al., 2023). Fully cleared plots were characterized by an overall decrease in the relative abundance of ASVs assigned to the class Bacteroidia (i.e., genera from the family Flavobacteriaceae including Winogradskyella, Olleya, and an unidentified genus), Verrucomicrobiae (genus Rubritalea), Planctomycetes (unidentified genus of the order Pirellulales) and Gammaproteobacteria (genus Vibrio) to values more similar to those found in bare rock plots. The only exceptions to this pattern were the higher relative abundances of a Vibrio (three weeks post-disturbance) and a Cyanobacteria (seven months post-disturbance; ASV assigned to the family Phormidiaceae) and the lower abundance of a Cetobacterium in fully cleared plots compared to bare rocks. However, the fluctuations in relative abundance among predominant epilithic bacterial groups were not consistent over time. This can be attributed to the influence of seasonal environmental factors, such as temperature and nutrient concentration, which likely exert their effects with a certain degree of stochasticity (Antunes et al., 2019; and Caruso, 2020).

The functional relevance of prevalent bacterial groups that had lower relative abundances in cleared plots throughout the experiment is still understudied (Dogs et al., 2017). However, it is likely that most of these groups play an important role in carbon cycling as copiothrophic bacteria (e.g., family Flavobacteriaceae; Elsherbini et al., 2023 or the genus Vibrio, Nelson et al., 2013) and their abundance may be reflective of the organic enrichment produced by Hormosira on the understory substratum (Anderson, 2016). Future research is needed to further understand the functional implications that the loss these macroalgae may have on intertidal rocky systems.

Changes in the bacterial communities caused by the experimental reduction of Hormosira density and cover did not appear to affect the subsequent recovery of the macroalgae, which occurred through the recruitment of juveniles. Indeed, Hormosira cover in fully cleared plots increased to an average of 83 ± SE 2.7% after ~16 months since the start of the experiment ( Figure S6 , a total increase of 75 ± SE 6% of coverage across all 0% density plots), but the space in the plots was mainly occupied by juvenile individuals which did not form a true canopy. The relatively slow recovery of a full canopy by Hormosira post-disturbance is consistent with other experimental studies, with recovery to full adult size (>80 mm) expected to occur within 2 to 5 years although this is tightly linked to the type and magnitude of the disturbance (Keough and Quinn, 1998; Underwood, 1998; Schiel and Taylor, 1999; Lilley and Schiel, 2006; Pocklington et al., 2019; Cameron et al., 2021; Lewis et al., 2021). This may explain why bacterial communities in these plots remained different throughout the entire experiment. However, it could also be predicted that the nursery effects of Hormosira will be restored following the growth of these juvenile sporophytes, possibly facilitating the recovery of the structure of the original microbial biofilm (Park et al., 2011).

Recent studies on epilithic bacterial communities have shown strong effects of urbanisation on their composition, which in turn negatively influenced recruitment of habitat-forming macroalgae (Pedicini et al., 2023). In our study, however, Hormosira recruitment was observed across all cleared plots suggesting that observed changes in the understory bacteria did not affect recruitment of Hormosira. This could be due to the persistence of a potentially “core” component of the microbial community in the remnant biofilm which may play a critical role in influencing the rate of Hormosira recruitment. A list of the most prevalent bacterial ASVs (frequency> 95%) across disturbance treatments (0, 50 and 100% Hormosira removal) and sampling times is given in Table S16 . These prevalent bacterial taxa could potentially represent a part of this “core” bacterial community; however, this finding still remains speculative. The prevalence of these taxa might not be directly linked to a functional role that is essential to Hormosira recruitment or survival after a disturbance, but rather, a process of colonization occurring homogenously across the intertidal rocky habitat (Burke et al., 2011 and discussed further below).

Little is known about the mechanisms and factors that influence the establishment and structure of understory epilithic bacterial communities, for example whether bacteria that colonize the substratum are sourced from the water column or from nearby macro-organisms such as the canopy-forming macroalgae. Here, Hormosira-associated epiphytic bacterial communities were found to be mostly distinct and less diverse compared to understory epilithic bacterial communities, suggesting that few components of the understory microbial communities are supplied by Hormosira. Many recent studies have found that epiphytic microbes associated with macroalgae differ from the surrounding environment such as water and rocks (Lachnit et al., 2011; Roth-Schulze et al., 2016; Weigel et al., 2022). Genera such as Rubritalea, Maribacter, Octadecabacter and Haliscomenobacter were enriched on Hormosira and have also been found associated with other subtidal and intertidal macroalgae where they play an important role in algal growth, nutrition and resistance to stress (Lin et al., 2018; Quigley et al., 2018; Samo et al., 2018; Capistrant-Fossa et al., 2021) The structure of epyphtic bacterial communities could thus be determined by selective processes driven by Hormosira (e.g., production of secondary metabolites by Hormosira, Egan et al., 2013 and Roth-Schulze et al., 2016) It has also been suggested that the key level at which to address the assembly of bacterial communities may not be at a taxonomic (i.e., ASV) level but rather at the level of core functional genes (Burke et al., 2011). In this model, bacterial communities may originate from the water column (e.g., via local colonization by incoming tidal waterflows) and communities are subsequently shaped by neutral/random processes and competition with other microbes, retaining a core residual function independent of taxa (e.g. competitive lottery model, Burke et al., 2011). Functional genes associated with macroalgal recruitment have been found in many bacterial taxa, for example the N-acylhomoserine lactone gene (AHL) that has been associated with enhanced settlement of Ulva zoospores through quorum sensing (Joint et al., 2002). Characterisation of core functional genes in the epiphytic versus epilithic communities may thus provide an interesting avenue for future research.