Relationships between microbiome and host performance
Mesocosm experiment
Hormosira individuals of similar frond size (mean diameter: 7.8 +/- S.E 2.2 cm, length: 11.2 +/- S.E 0.81 cm; N = 120) were haphazardly collected by carefully detaching their holdfast from intertidal rocks during low tide at Cronulla beach, Sydney, Australia (34°03’22.8”S 151°09’19.7”E) on 4 December 2018. Hormosira were transported in seawater within one hour to the flow-through aquarium facility at the Sydney Institute of Marine Science (SIMS) where they were thoroughly rinsed in autoclaved filtered (0.2 µm) seawater to remove fouling organisms and loosely associated microorganisms. Hormosira individuals were then placed into 12, 2-litre tanks (10 individuals 2.5cm apart per tank) by cable-tying their holdfasts to weighted mesh (e.g.[47]) and left overnight to acclimatize. Tanks were supplied by filtered (5 µm), UV-treated, running seawater. The following day, algae were rinsed with autoclaved filtered seawater (AFSW) and randomly assigned to one of three antibiotic combinations (AB1, AB2 or AB3; see details in Table S1 and S2) or an AFSW control. Algae were left to soak in the different antibiotic treatments or in AFSW (control) for 24 hours, then rinsed with AFSW and placed back into the 12 independent tanks (n = 3 tanks per treatment; 10 individuals per tank) with UV treated, 5 µm filtered seawater.
One Hormosira individual was randomly selected and removed from each tank prior to the addition of any antibiotics (day 0), just after their treatment with antibiotics (day 1) and every second day for 1 week and on day 12 to characterise surface-associated microbiome and host performance (see details in Supplementary Information). Temporal sampling allows for the detection of unwanted direct effects as it is predicted that if antibiotics had any direct (e.g. chemical) effects on the host, such effects would be immediate (see Li et al 2022). On the other hand, if effects on host are microbially-mediated, it is predicted that there would be a lag between the application of the treatment and the response of the host (i.e. time for the microbiota to respond to the treatment and consequentially elicit a response on the host). Whilst temporal sampling provides evidence of direct effects if present, it is fundamentally important to also use suitable controls, i.e. in this case multiple antibiotics with different MOAs and chemical structures.
To characterise the associated microbiome, a consistent area of the algal surface (23 +/- SE 2.7 cm2; adjacent to, but independent from the area where photosynthesis was measured, see below) was swabbed for 30 seconds with sterile cotton swabs, which were immediately placed in sterile cryogenic tubes in liquid nitrogen and then stored at -80°C until DNA extraction (as per [48]). Host performance was assessed by quantifying the maximum photosynthetic quantum yield (after individuals were dark-adapted for 15 min) using a Pulse Amplitude Modulated (PAM) fluorometer (WALZ, Germany), a metric of host performance previously used for macroalgae that generally reflects poor host condition under stress [8, 49].
Field experiment
Forty-two Hormosira individuals of similar size (approximately 6.8 +/- 2.3 cm in diameter and 8.5 +/- SE 1.3 cm in length) approximately ~ 1m apart were tagged at Cape Banks, Sydney, Australia (33°59’55.3” S 151°14’53.6” E) during low tide on 13 May 2018. Algae were rinsed with AFSW for 1 minute and then subjected to one of 6 different treatments: 1) the antibiotic combination AB2 (chosen due to its effect on the microbiome and host performance in the mesocosm experiment above) applied once for 2 hours during exposure at low tide, 2) Povidone Iodine 10%w/v (‘iodine’ hereafter) applied once for 2 hours (IO), 3) iodine applied every ~ 2 days for 15 minutes (IM), 4) a procedural control for the treatment application once (IMO; applying AFSW once for 2 hours), 5) a procedural control for the continuous treatment application (IMC; applying AFSW for 15 mins every ~ 2 days), 6) control (C, undisturbed individual; see details in Supplementary Table S1).
Half of the surface area of each individual (~ 12 +/- 2.7 cm2) was swabbed for 30 seconds to quantify the surface-associated microbiome as described for the mesocosm experiment. Swab controls (i.e. used similarly to those for swabbing alga except no algae was swabbed) were collected. The other half was swabbed for bacterial culturing (details below); these swabs were stored in 1.5ml Eppendorf filled with 1ml of AFSW and placed in ice until arrival to the laboratory. Additional swab controls were collected to test for potential contamination in cultures due to the placement in AFSW. Samples were collected every 2nd day for a total of 46 days (see Supplementary Table S2 for specific dates).
Bacterial isolation and culture
For both the laboratory and field experiments, swabs were in 1.5 ml Eppendorf tubes filled with 1ml of AFSW were immediately taken to the laboratory for spreading on plates. Briefly, the suspended swab was vortexed and 10 ul of the solution was then plated on standard half strength marine broth agar (MP Biomedicals LLC). Individual morphologically distinct colonies were picked after 24hrs. These colonies were suspended in liquid marine broth and left to grow for 48 hrs at room temperature on a shaker plate. 10 ul of the liquid culture was then plated as above on a new sterile agar plate. Colonies were replated until only 1 morphological form remained which was then extracted and sequenced (details below). Isolates were stored in glycerol at -80oC until resuspension in marine broth for inoculations (see Supplementary Information for growth conditions).
Isolates were sequenced (details below) and the identity of the taxa were matched with those from the mesocosm and field experiment. Microbial amplicon sequence variant (hereafter, “ASV”) abundance was correlated with host photosynthetic yield to identify isolates for the inoculation experiments. Two strains were used for inoculations: Vibrio genomosp (F10 str. 9ZC157) and Vibrio chagasii (str ECSMB14107). V. genomosp was chosen as its increase in abundance post-antibiotic treatment was directly correlated with poor host performance (PAM) and had the greatest change in total abundance over the experiment (Figure S1). V. chagasii was chosen as it was phylogenetically similar to V. genomosp but remained at a constant abundance throughout the experiment and was thus uncorrelated to variation in host performance. It was therefore used as a control inoculant (Figure S1).
Testing for direct effects of components of the microbiome
Inoculation experiment
Hormosira individuals of similar length (7.5cm +/- SE 1.2; N = 110) were haphazardly collected as described above from the rocky shore at Cape Banks on 25 January 2021 and transported to the SIMS aquarium within an hour, where they were rinsed with AFSW. Hormosira individuals were then placed into 15ml falcon tubes (1 individual per tube) and attached via their holdfast to a stainless-steel rod at the bottom of the tube (Fig. S3). Each tube was individually fed filtered (50 µm) seawater through aquarium airlines from four replicate header tanks. Once all samples were attached in the aquarium, they were left overnight to acclimatise. On the next day, all tubes were rinsed with AFSW and the algae inside was randomly assigned to one of seven treatments (Table S1):
1) Microbiome disruption with antibiotic combination AB2 applied once for 25 hours (as in the mesocosm experiment; Table S1);
2) Microbiome disruption as in (1) followed by inoculation with bacteria whose relative abundances correlated with poor host condition in the experiments above (Vibrio genomosp, see Fig S1; cell density 5×108 CFU);
3) Microbiome disruption as in (1) followed by inoculation with bacteria unrelated with host performance in the experiments above (negative control: Vibrio chagsii sp.; see Fig S1; cell density 5×108 CFU);
4) Microbiome left undisturbed (AFSW) for 24 hours followed by inoculation with V. genomosp as in (2);
5) Microbiome left undisturbed (AFSW) for 24 hours followed by inoculation with the negative control strain as in (3);
6) Procedural control, AFSW for 25 hours;
7) Control (undisturbed algae).
Inoculations were done by immersing Hormosira for 1 hour within the high-density cell culture which allowed sufficient time for the inoculated bacteria to attach to algal surfaces (See supplementary figure S1). This timing was confirmed by leaving individuals within high cell density washed cultures for varying times before being swabbed as above. DNA was extracted and custom qPCR primers for the inoculants were used to determine abundance (see Supplementary Information for further details)
Samples were collected following the same timepoints as in the previous experiment: 5 individuals were collected at day 0 before treatments were applied and 3 independent individuals were randomly selected from each treatment on days 1, 2, 5, 8 and 12, at the same time of day on each sampling occasion. These time-points and the overall duration of this experiment were based on results from the previous experiments above which showed effects on microbiome and hosts within ~ 3–5 days (see Results) and our capacity to maintain independent replicate algae for each treatment x time combination. Surface-associated bacteria and algal photosynthetic efficiency was then characterised for each sample as described above (See mesocosm experiment above).
DNA extraction and sequencing
For characterisation of microbial communities in all experiments, microbial DNA was extracted from each swab sample in a randomised order using a PowerSoil DNA Isolation kit (Qiagen) following the manufacturers protocol. DNA extracts were quantified using spectrophotometry (NanoDrop 1000) and stored at -20oC until sequencing.
The extracted DNA samples were amplified with Polymerase Chain Reaction (PCR) using the 16S rRNA gene primers 341 (F) (5’- CCTACGGGNGGCWGCAG-3’) and 805(R) – (5’-GACTACHVGGGTATCTAATCC-‘3), covering the V3-V4 regions of the bacterial and archaeal 16S rRNA gene ([50]). The PCR conditions involved a pre-heating step to 95 oC for 3 min followed by 35 cycles of 95oC for 15 s, 55oC for 1 min and 73oC for 30 s. Both positive (with known DNA sequence) and negative controls (nuclease-free water, control swabs) were used. The negative controls did not amplify DNA, suggesting no contamination on swabs/materials or during extraction and amplification. Agarose gel electrophoresis and Nanodrop 1000 were used to ensure the quantity and quality of the amplicons before they were sent for sequencing via the Illumina MiSeq 2000 platform at the Ramaciotti Centre for Genomics (UNSW, Sydney).
Bioinformatics
Raw sequencing data was quality filtered using Trimmomatic [51] with a sliding window trim of 4:15 base pairs (bp) and removal of sequences with < 36bp. Paired-end reads were merged with a minimum length of 400bp and maximum of 500bp using USEARCH [52]. UNOISE was then used to remove chimeras and produce amplicon sequence variants (ASVs), i.e., operational taxonomic units at a unique sequence level (0% distance) (Edgar, 2017). USEARCH was used to map the original reads back to ASVs, generating a table of 5812, 7637 and 9138 ASVs for the mesocosm, field and inoculation experiments, respectively. ASV sequences were searched with BlastN against the SILVA SSU Ref NR99 database for taxonomic classification to classify and remove chloroplasts, GTDB was then used for taxonomic assignment. Singletons and low abundance taxa (< 0.01% of reads) were removed from the dataset for statistical analyses, resulting in 4112, 6488 and 7028 ASVs (mesocosm, field and inoculation experiments respectively).
Estimation of absolute bacterial abundance
Total abundance of the 16S rRNA gene was quantified for each sample by qPCR using the primers 341F/805R [53]. Gene amplification and analysis were performed using the QuantStudio 3 thermocycler (Thermo Fisher with PrimeTime® Gene Expression Master Mix, Integrated DNA Technologies) and associated software. The reaction conditions for amplification of DNA were 56oC for 2 min, 95 oC for 10 min and 40 cycles of 95 oC for 15 s and 60 oC for 1 min. The final gene copy number per sample was corrected for the total extraction volume, the surface area and the dilution factor and DNA yield per sample (see [54]for further details on the normalisation) and were used to estimate absolute abundances of ASVs and inoculants.
Statistical Analyses
To account for uneven sequencing depth among samples, data were normalised using the counts per million reads method in the R package DESeq2 [55]. This resulted in a total of 4,423,807, 10,443,709 and 9,921,043 sequences for the mesocosm, field experiment and inoculation datasets, respectively.
To test for effects of microbiome manipulations on host photosynthetic efficiency, we used a linear model with the factors treatment (fixed) and time (fixed, crossed). For the first mesocosm experiment where multiple Hormosira individuals were sampled from each tank, ‘tank’ was fitted as a random factor nested within treatment using the lme4 R package (v4.0.3). To meet the model’s assumptions, data were square root transformed. p-values were obtained using the anova function within the car package in R with significance being tested using F-tests, or likelihood ratio tests for the mixed model that included tank as a random effect.
Alpha diversity measures of richness (i.e., number of unique sequences) and Simpson’s diversity index were calculated using the ‘vegan’ R package [56] and differences between treatment (fixed) and time (fixed, crossed) were examined using a linear model in the R GAD package. Where appropriate, tank was included as a random factor.
To determine differences in the structure of the associated bacterial assemblages, the normalised ASV data were analysed using permutational multivariate analysis of variance (PERMANOVA) ([57]) in the R vegan package ([56]), with the factors treatment, time and tank (mesocosm experiment only) as above. Similarity matrices were calculated using Bray-Curtis measure on square-root transformed data and visualised through non-metric multi-dimensional scaling (nMDS) ordinations. We also calculated and plotted mean Bray-Curtis similarities to the control (undisturbed algae) treatment. To determine which bacterial taxa’s abundance differed among treatments and times, we used multivariate generalised linear models (GLMs) using the R package ‘mvabund’ [58], assuming a negative-binomial distribution to account for over-dispersion. In order to determine how the total abundance of the two inoculants changed over the course of the experiment, linear models were fitted with the factors treatment (fixed) and time (fixed, crossed) in R (v4.0.3 ).