Fresh diatom cultures were established from water samples (surface water plankton tow) collected from the Port Hacking oceanographic station, which is located on the south-eastern Australian coast near Sydney (34° 05.00 S, 151° 15.00 E). Port Hacking is a long-running time-series site within the Integrated Marine Observing System (IMOS) National Reference Station (NRS) network (Brown et al., 2018). Diatoms were isolated using drawn out Pasteur pipettes (micropipettes) under light microscopy (maximum magnification 200X). Extracted cells were initially grown in 24 well plates containing sterile seawater diluted 50:50 with f/2 medium (Andersen et al., 2005). Once cultures were established, they were transferred into 50 mL culture flasks (Falcon) and maintained in a constant temperature incubator at 21°C, with a light intensity of 60–70 μE m⁻² s⁻¹ and a 12:12 light/dark cycle. In addition to its global distribution in planktonic communities, we chose this diatom species due to its relatively large size (~50 μm) and because it was non-chain-forming, making it suitable for accurate flow cytometry enumeration.

For taxonomic identification, diatom cells were grown for 7 days, centrifuged at 2000 g for 5 min. DNA was extracted from the resultant pellet using a DNeasy PowerWater Kit following the manufacturer’s instructions (Qiagen). The full 18S rRNA gene was then amplified using the 18SmoonF and 18SmoonR primer set (Supplementary Table S1), with the following PCR cycle conditions: 94°C for 10 min; 30 cycles of: 94°C for 1 min, 55°C for 2 min, and 72°C for 3 min; followed by a final extension at 72°C for 10 min and cooling at 4°C. PCR were performed using a Thermal Cycler (Veriti 96-well Fast thermal Cycler, Applied Biosystems) and the product sequenced using Sanger sequencing (Australian Genome Research Facility; AGRF). The resulting sequence was compared to the National Centre for Biotechnologies Information (NCBI) database using the online BLASTn tool.

Transmission Electron Microscopy (TEM) was also used to unequivocally identify the diatom to the genus level based on distinct morphological features (pseudonodulus, central annulus, labiate processes, and areolation pattern). For TEM examination, cultures were preserved in Lugol’s iodine prior to cleaning using the method of Hasle and Fryxell (1970). Once cleaned, 3 μL of culture were placed on formvar-coated copper grids and loaded into a FEI Tecnai T20 TEM (LaB6), operated at a high tension of 120 kV and equipped with a Gatan 894 CCD 2 k × 2 k camera. Frustule characteristics and morphometrics were quantified using Image J1¹ and guided by Schmidt (1878) and Hasle and Syvertsen (1996).

A total of ~150 bacteria strains were isolated from the microbiome of Actinocyclus sp. and other diatom strains (all isolated from the environment at the same time from the same plankton tow sample) 20 days after the establishment of the cultures (Supplementary Table S2). Each bacterial strain was isolated and grown to purity by repeated plating on 1 and 10% Marine Agar (Difco) incubated at 21°C in the dark.

For identification, each bacterial isolate was grown for 24 h in 100% Marine Broth (Difco Marine Broth 2,216) at 21°C, under 160 rpm shaking. DNA was then extracted and the 16S rRNA gene amplified by PCR using the 27F and 1492R primer set (~1,500 bp, Supplementary Table S1), which was performed using the following PCR cycle conditions: 95°C for 10 min; 30 cycles of: (i) 95°C for 30 s, (ii) 50°C for 30 s, and (iii) 72°C for 90 s; followed by a final extension at 72°C for 10 min. Samples were then Sanger sequenced (Australian Genomic Research Facility). Sequences were quality trimmed using Geneious software (v2022.1.1) and aligned using the MUSCLE alignment method. Bacteria were identified to the species level if their sequences shared >99% similarity with an entry in BLASTn database, and otherwise identified to the genus level. For comparison with the microbiome data, bacterial DNA was also sent to AGRF for Illumina sequencing on a MiSeq platform targeting the V1-V3 region of the 16S rRNA gene (~ 490 bp, Supplementary Table S1). Sequences were processed as per the microbiome analysis below with the BLAST tool used to match the bacterial isolates ASVs against the Actinocyclus microbiome with a > 97% percent identity threshold.

To characterise the bacterial component of the Actinocyclus sp. associated bacterial assemblages (free-living and attached bacteria), 16S rRNA amplicon sequencing was used. A 7-day old Actinocyclus sp. culture was inoculated at a starting concentration of 5,000 cells mL⁻¹ and sampled at days 2, 4, 6, 8 and 12. At each time point, four replicates of 50 mL were filtered through 0.22 μm pore-size polycarbonate membrane filter (Millipore). DNA was subsequently extracted from filters using the Dneasy Power Water Kit (Qiagen) and quantified with a Nanodrop spectrophotometer (Thermo Scientific). Samples were then sequenced using the Illumina MiSeq platform targeting the V1-V3 region of the 16S rRNA gene (Supplementary Table S1). A negative control (DNA extraction with no sample) was also sequenced.

All sequences were processed through the DADA2 pipeline [https://github.com/martinostrowski/marinemicrobes/tree/master/dada2 (accessed on 31 May 2021)]. Briefly, the DADA2 R package was used to remove primers, filter and trim quality terminal error (16S rRNA trunc lengths R1 = 270, R2 = 250), denoise, merge, and remove chimeras from the dataset (Callahan et al., 2019). ASVs were assigned using a naïve Bayes classifier based on SILVA 138.1 and a bootstrap cut-off >50% (Wang et al., 2007; Yilmaz et al., 2014). ASV abundance tables were then normalised using the cumulative sum scaling approach implemented in the metagenomeSeq R package which account for uneven sequencing depths (Paulson et al., 2013).

Axenic cultures of Actinocyclus sp. (Figure 1B) were generated (n = 3) using the protocol described by Shishlyannikov et al. (2011) with minor modifications. Approximately 50 mL of each mid-late exponential phase Actinocyclus sp. culture was gravity-filtered through a 5 μm pore-size polycarbonate membrane filter (Millipore). The filter was then transferred into sterile seawater and the cells were gently resuspended by inverting, before the addition of 20 μg mL⁻¹ Triton X-100 (Sigma Aldrich). After 1 min of Triton X-100 exposure, the suspension was filtered through a new 5 μm filter and rinsed with sterile seawater before being transferred to fresh f/2 media containing a cocktail of antibiotics, including: ciprofloxacin (Sigma Aldrich; final concentration 20 μg mL⁻¹), ampicillin (Sigma Aldrich; final concentration 100 μg mL⁻¹), gentamycin (Sigma Aldrich; final concentration 67 μg mL⁻¹) and streptomycin (Amresco; final concentration 50 μg mL⁻¹). Cells were then incubated under regular growth conditions for 7 days. After that time, 1 mL of culture was transferred into fresh f/2 media. To verify that the cultures were axenic, we followed the procedure of Larsson et al. (2022), whereby treated cultures were considered axenic only after: (1) no bacterial growth was identified when 1 mL of algal culture was inoculated into 100% Marine Broth for 7 days; (2) drop-plating of algal culture onto 100% Marine agar plates resulted in no bacterial growth; and (3) no bacterial population was present in algal cultures stained with SYBR Green (Sigma-Aldrich S9430) and enumerated with flow cytometry. Finally, DNA from these cultures (n = 3) was also extracted using Dneasy Power Water Kit (Qiagen), the V1-V3 region of the 16S rRNA gene was amplified and sequenced on a MiSeq platform (2 × 300 bp) (Illumina) at AGRF. All reads were identified as chloroplast, confirming the Actinocyclus sp. cultures were indeed axenic.

To ensure that differences in growth observed between cultures of Actinocyclus with its whole microbiome and axenic cultures were not due to the stress induced by the axenic protocol (cf. Figure 1C), cultures were acclimated for ~12 transfers (i.e., ~6 months) before starting the co-culture experiments. Furthermore, we also performed a “reseeding” experiment, whereby we added the initial microbial consortia back into the axenic cultures (Supplementary Figure S1). After growing axenic cultures, and cultures with the whole microbiome side by side for 7 days, the cultures with the whole microbiome were pooled together and 100 mL filtered through a 5 μm pore-size polycarbonate membrane filter (Millipore); the filtrate, containing the bacteria, was collected in a sterile bottle for use in subsequent steps. The remaining cultures with the whole microbiome (not filtered) were enumerated using flow cytometry and diatom cells inoculated to 5,000 cells mL⁻¹ into fresh f/2 media (n = 4). Filtrate was diluted to 1 × 10⁶ cells mL⁻¹ in axenic cultures inoculated to 5,000 cells mL⁻¹ into fresh f/2 media (‘reseeded’ diatom) to match the density of bacteria present in the culture with the whole microbiome at day 0 (n = 4). The remainder of axenic cultures were diluted into fresh media to 5,000 cells mL⁻¹, which served as controls (n = 4). Each culture was subsampled every two days, whereby 200 μL aliquots were fixed with glutaraldehyde (2% final concentration, G6257 Sigma-Aldrich) and both diatom and bacterial cells were quantified using flow cytometry as described below. The axenic cultures were checked for contamination throughout the course of the study by (1) drop-plating on 100% Marine agar plates and (2) flow cytometry (following staining with SYBR green).

A 200 μL aliquot was sampled every 2 days and cells were enumerated using a flow cytometer (CytoFLEX LX, Beckman Coulter). The samples were analysed at a flow rate of 30 μL min⁻¹ for diatom and 25 μL min⁻¹ for bacteria. For diatom enumeration, samples were fixed with glutaraldehyde (2% final concentration) and incubated at 4°C for ~20 min. For bacterial enumeration, samples were first fixed with glutaraldehyde and after 20 min stained with SYBR Green I at 4°C for 15 min in the dark (final concentration 1:10,000, Life Technologies). Fluorescence was detected with blue laser (488 nm) excitation and a combination of 690–50 and a 525–40 nm bandpass filter. Diatom populations were identified and quantified according to chlorophyll-a (B690-50-A) vs forward scatter area (FSC-A), and SYBR Green fluorescence vs side-scatter area (SSC-A) for bacteria (CytExpert v2.4 software). To ensure accuracy of the gating settings and volume of samples processed between sampling time-points, fluorescent QC beads were used (B53230, CytoFLEX Daily QC Fluorospheres) and the flow cytometer regularly calibrated following the manufacturer’s instruction.

Out of the 150 bacteria isolated, we pre-selected 30 candidates for further examination based on their taxonomy and their likelihood to interact with diatoms. We subsequently performed a pre-screening experiment of their effect on Actinocyclus growth using co-culture plate assays. Each of the 30 co-culture was established in 1 mL volume in 48 well-plates (for details on co-culture preparation refer to the paragraph below). Plates were incubated as described above and algal growth monitored every 2 days for 16 days. Algal growth was estimated based on chlorophyll-a fluorescence (455 nm excitation, 680 nm emission) measured on an infinite M1000 pro plates reader (Tecan). Fluorescence was acquired from the bottom of the plate with 16 reads per well, a flash frequency of 400 Hz, 20 μs integration time, no lag time and 10 ms settling time. This process allowed us to narrow down the number of bacterial candidates to 15 (Supplementary Table S3), and to preselect strains which have a high likelihood to enhance the growth of Actinocyclus sp. for further experiments.

Each subsequent co-culture was established in a 70 mL tissue culture flask (Falcon) with a final volume of 50 mL (n = 4). For all experiments, bacteria were freshly plated from glycerol stock onto 100% Marine Agar plates 3–4 days prior to the start of the experiment and were grown from single colonies in Marine Broth. After 6 h, 100 μL of bacterial culture was transferred into fresh 100% Marine Broth and grown overnight (12 h). Bacteria were centrifuged at 1400 rpm for 6 min using an Eppendorf centrifuge 5424R. The bacterial pellet was then washed with sterile f/2 media twice, centrifuged (1,400 g for 6 min) and reconstituted in 1 mL of media. Bacteria were then diluted 1:10 in sterile media and a 200 μL aliquot was fixed with glutaraldehyde and stained with SYBR Green for enumeration via flow cytometry as described above. Diatoms were inoculated from 7-day old cultures into fresh media (n = 4) to a density of ~5,000 cells mL⁻¹ and bacterial isolates were added to the axenic diatom cultures at a standardized cell density of 1,000 cells mL⁻¹ (based on flow cytometry cell counts) to achieve a final ratio of 1:5 bacteria:diatom. Diatom and bacteria co-cultures were then incubated at 21°C, at 60–70 μE m⁻² s⁻¹ light intensity and gently shaken daily for 16 days. Axenic and non-axenic (culture with the whole microbiome) diatom cultures as well as bacteria in f/2 media incubated under the same conditions acted as controls (n = 4 for each). Over the course of the co-culture, 200 μL aliquots were sampled and fixed with glutaraldehyde (2% final concentration) every two days and both diatom and bacterial cells were quantified using flow cytometry as described above. The axenic culture controls were also checked for contamination throughout the course of the co-culture by (1) drop-plating on 100% Marine agar plates and (2) flow cytometry (following staining with SYBR green). The 15 co-cultures were run in three different batches, each with its own set of axenic and whole microbiome controls (n = 4).

Statistical analyses were performed using R (v4.1.3), Excel and SPSS (v26, Armonk, NY: IBM Corp). In order to test if the diatom associated bacterial assemblages differed between time points, a Bray-Curtis matrix was built and a permutational multivariate analysis of variance (PERMANOVA) was performed using the adonis function from the ‘vegan’ R package (Bonferroni-Hochberg adjusted p < 0.05).

For the co-culture experiments, data were assessed for normality and homogeneity of variance using Shapiro and Levene’s tests before repeated-measure ANOVAs were performed. Differences in cell numbers between the co-culture and the axenic control at each time point were assessed using a Simple Main Effect test (Quinn and Keough, 2002).

To evaluate the growth-promoting effect of each bacterial strain on the diatom, specific growth rates (μ) were calculated from cell counts (cells mL⁻¹) during the exponential phase (corresponding to the steepest part of the natural log-transformed cell abundance curves) of each co-culture and standard deviation of μ was calculated for the biological replicates (n = 4 unless otherwise indicated):

where x1 and x2 are the cell concentrations at time 1 (t1) and time 2 (t2).

A growth enhancement percentage from μ was also determined as: (μ_co-culture / μ_axenic)/μ_co-culture. Shapiro and Levene’s tests were used to assess the specific growth rate normality and homogeneity of variance and a one-way ANOVA followed by a Dunnett’s test (glht function from the ‘multicomp’ R package) were used to compare μ of every co-culture to their respective axenic controls. Additionally, we also quantified the total diatom cell concentration and the growth-promoting effect of each bacterial strain over time. To do this, we subdivided the co-cultures into three time-periods: days 0–6, days 8–12 and days 14–16 and a cell density index (CD) was then calculated for each bacterium for each of the three time-periods. Within each period, the CD index was determined by calculating the area under the curve (AUC; using the AUC() function from the DescTools R package) of each co-culture and dividing by the AUC of the axenic culture:

The 1900 bp 18S rRNA gene sequence from the targeted diatom (Figure 1B) showed a 99.8% similarity to Actinocyclus sp. strain CNS00114 (Accession No. MW750345) and a 99.2% similarity to Actinocyclus curvatulus strain AWI 85 (Accession No. X85401). Whilst neither of these referenced strains provided suitable morphological evidence to confirm their species identification, the genus identity of the targeted strain was also confirmed by TEM to be Actinocyclus. Cells were cylindrical in valve view (~20–25 μm diameter), having both pseudonodulus and central annulus present, a marginal ring of labiate processes and striae that were aligned to the central stria between the fascicle edges (Figure 1A).

The presence of the Actinocyclus sp. microbiome increased the maximum cell concentration of the diatom by 2-fold by day 12 of the experiment, relative to the axenic control (Simple Main Effect test, p < 0.05, Figure 1C; Supplementary Table S4). However, the cell abundance significantly increased above the control on day 8, and remained significantly higher until the end of the experiment (day 16, Figure 1C).

The composition of the bacterial community associated with the Actinocyclus sp. strain, as determined by 16S rRNA gene amplicon sequencing, significantly changed over the course of the experiment (PERMANOVA and all pair-wise comparison, p < 0.05, Figure 2; Supplementary Table S5). A total of 638 ASVs belonging to two phyla, Bacteroidetes and Proteobacteria, were identified across all sampling times. Proteobacteria was the most abundant phylum (~55%) during the early time-points (day 2, 4 and 6), whilst the later time points (day 8 and 12) were dominated by Bacteroidetes (~75%) (Supplementary Table S6). Amongst the Proteobacteria, the relative abundance of Alphaproteobacteria (primarily Rhodobacterales) was relatively stable (~15%) throughout the diatom growth cycle, with the exception of day 6, when their relative abundance decreased (~4%). Conversely, the relative abundance of Gammaproteobacteria (i.e., Pseudomonadales and Enterobacterales) was higher during the early time points (~43%) and decreased in later time points (~11%) (Supplementary Figure S2; Supplementary Table S7). On the other hand, the Bacteroidetes (i.e., Flavobacteriales) relative abundance increased from ~42% within the first three time points to ~74% in the later ones (Supplementary Figure S2; Supplementary Table S7).

The 15 bacterial isolates chosen for co-culture experiments belonged to Alpha- and Gammaproteobacteria, including the families Rhodobacteraceae (three Shimia, one Pseudophaeobacter and three Thalassobius), Pseudoalteromonadaceae (two Pseudoalteromonas), Alteromonadaceae (one Alteromonas), Oceanospirillaceae (one Marinomonas) and Vibrionaceae (four Vibrio; Supplementary Table S3). Within the Actinocyclus sp. microbiome, six ASVs, that exactly matched the 16S rRNA gene sequences of our selected bacterial isolates, represented up to 13.7% of the communities at day 2 with an average of 8.7% across all time points (Supplementary Table S8).

The exponential phase of the axenic Actinocyclus cultures lasted for 4 days, between day 2 and 6. This length and timing were identical when Actinocyclus was in co-culture with 11 of the 15 bacterial strains tested (including all members of the Rhodobacteraceae, Pseudoalteromonadaceae, Alteromonadaceae and Oceanospirillaceae). However, the presence of all Vibrio sp. isolates delayed the onset of the exponential phase by 2 days (starting at day 4), and Vibrio sp. strain b31 extended the length of the phase by a further 2 days (total of 6 days) (Figure 3; Table 1).

The addition of the bacterial isolates significantly affected the specific growth rate of Actinocyclus sp. (One-way ANOVA, F_15,55 = 14.72, p < 0.001; Figure 4; Table 1; Supplementary Table S9) with 80% of the bacterial isolates enhancing the growth rate of Actinocyclus relative to the axenic control in the exponential phase. Overall isolates belonging to the Rhodobacteraceae enhanced the specific growth rate of Actinocyclus the most, with growth rates increasing by 25–56% in the presence of these bacteria, with the presence of Thalassobius sp. strains having the greatest effect (~ 53%). Bacterial isolates within the Vibrionaceae had less effect on Actinocyclus sp. specific growth rate (~16% increase) with Vibrio sp. strain b9 having no significant effect (Figure 4; Table 1). Interestingly, whilst the addition of most isolates resulted in a smaller growth rate compared to the cultures with the whole microbiome, the presence of three of the Rhodobacteraceae isolates (Shimia thalassica b25, Shimia sp. b15 and Thalassobius sp. b30), Pseudoalteromonas phenolica strain b20 and Vibrio sp. strain b11 resulted in a similar specific growth rate of Actinocyclus sp. to that observed in the whole microbiome cultures (Figure 4; Table 1).

Although some bacteria induced an Actinocyclus sp. growth rate that was similar to the cultures with their entire microbiome, this proxy alone does not adequately describe the Actinocyclus growth dynamics in the cultures. Indeed, Actinocyclus numbers in the cultures with the whole microbiome reached levels that were on average 2 times higher than the Thalassobius sp. b30 and V. fortis b11 co-cultures, and these growth enhancements were maintained for longer (Supplementary Figure S4). This observation also applies to the co-cultures, for example Pseudophaeobacter sp. strain b23 did not alter the diatom’s growth rate, but significantly increased cell abundance compared to the axenic control (Figures 3, 4; Table 1; Supplementary Table S4). Thus, the specific growth rate of Actinocyclus sp. in the presence of the different bacterial isolates was not always reflective of changes in the diatom cell abundance and does not necessarily capture the differences in growth dynamics.

To capture the multifaceted growth dynamics induced by specific bacteria in co-culture with Actinocyclus, we quantified differences in diatom cell abundance through time between each co-culture and the axenic controls. Each of the 15 bacterial isolates tested enhanced the abundance of Actinocyclus sp. during at least one time point in the co-culture experiments (Simple Main Effect test, p < 0.05, Supplementary Table S4). However, the magnitude and duration of these effects were highly variable between the different bacteria tested (Supplementary Figures S4, S5). Members of the Rhodobacteraceae family elicited the largest increase in diatom abundance, with 58% more cells than the control on day 8. Within the Rhodobacteraceae, Thalassobius sp. b37 triggered the largest increase in cell abundance, with its presence resulting in a 2-fold increase in Actinocyclus sp. cell numbers (Figure 5A). The increase in cell numbers caused by the Rhodobacteraceae isolates were generally sustained, with significant effects observed on average for 10 days (Simple Main Effect test, p < 0.05, Supplementary Table S4). In comparison, the presence of Pseudoalteromonadaceae and Alteromonadaceae had a more modest effect on Actinocyclus sp. growth, with only a 21% increase in diatom cell numbers observed on average (Figure 5B).

In contrast to the sustained positive effects observed amongst Rhodobacteraceae, Pseudoalteromonadaceae, and Alteromonadaceae, isolates from the Vibrionaceae and Oceanospirillaceae families only had a positive effect on the number of Actinocyclus cells during later time points of the experiment (Figure 5C). More specifically, their most pronounced effects were observed on days 14 and 16 of the experiment (28% increase in cell numbers on average), but this enhancement only lasted 3 days on average. Notably, 5 isolates belonging to the Vibrio and Marinomonas genera initially impaired diatom growth (Figure 5C). It is also worth noting that whilst some of our isolates had identical (100% match) 16S rRNA gene sequences (Thalassobius sp. isolates b30-b35-b37 and Pseudoalteromonas phenolica isolates b18-b20), they did not all induce the same growth dynamics in co-culture with Actinocyclus (Figure 5).

The growth enhancement observed in Actinocyclus sp. was part of a reciprocal interaction, because all bacterial strains tested reached significantly greater cell numbers when co-cultured with Actinocyclus relative to controls (Supplementary Figure S6). The only exception was A. mediterranea, which decreased through time when in co-culture (Supplementary Figure S6).

Given temporal differences in the growth enhancement effect of each bacterial isolate tested, we quantified the difference in Actinocyclus cell numbers between each co-culture and the axenic controls during three distinct time-period windows (day 0–6, day 8–12, and day 14–16) and calculated the growth rate and maximal cell abundance for each period (Supplementary Table S11). Our results confirmed that although all tested bacteria significantly increased the cell abundance of Actinocyclus in at least one of the three time periods (Simple Main Effect test, p < 0.05. Supplementary Table S12), their maximal effects did not occur in the same period (Supplementary Figure S7; Supplementary Table S13). Most Rhodobacteraceae and Pseudoalteromonadaceae isolates increased the diatom abundance the most between days 8–12, with a maximum cell density index of 1.6 in co-culture with Thalassobius sp. (b35 and b37), whilst members of the Vibrionaceae mostly increased algal cell numbers between days 14–16 (Figure 6; Supplementary Tables S13, S14).

Out of the 15 isolates tested, 6 had a 100% identity match to ASVs within the microbiome of Actinocyclus sp. including Shimia (isolates b14 and b15), Pseudophaeobacter (isolate b23), Pseudoalteromonas (isolates b18 and b20) and Alteromonas (isolate b34). However, although those specific ASVs were found in the Actinocyclus microbiome, there was no link between the timing of the growth benefit on Actinocyclus and the relative abundance of these bacterial isolates within the microbiome (Supplementary Figure S8).

The Actinocyclus sp. bacterial community was dominated by members of the Bacteroidetes (mainly Cryomorphaceae) as well as α- and γ-Proteobacteria, which is largely consistent with the microbiome composition of other diatom species (Grossart et al., 2005; Buchan et al., 2014; Ajani et al., 2018; Stock et al., 2019; Koester et al., 2022). Our results also indicate that the Actinocyclus sp. bacterial community undergoes significant compositional changes throughout the diatom’s growth cycle, which is also consistent with observations in other diatom species (Moejes et al., 2017; Crenn et al., 2018). The early stages of the diatom’s growth cycle were characterised by a higher relative proportion of Proteobacteria (~55%), whilst Bacteroidetes relative abundance increased (40 to 75%) within the later time-points. Similar patterns have also been observed in various diatom species, both in culture conditions (Moejes et al., 2017), and during blooms in natural environments (Teeling et al., 2012; Needham and Fuhrman, 2016). The identity of bacteria that dominate diatom microbiomes at a given time may be strongly influenced by the photosynthates exuded by the diatom, which are known to differ between growth stages (Barofsky et al., 2009; Sarmento et al., 2013; Olofsson et al., 2022). For example, cells in exponential phase and later stages tend to actively release a higher proportion of high molecular weight compounds (e.g., carbohydrates such as polysaccharides) (Myklestad, 1974; Biddanda and Benner, 1997; Granum et al., 2002), which Flavobacteria are known to effectively metabolise (Cottrell and Kirchman, 2000; Buchan et al., 2014). Members of Flavobacteriaceae effectively remineralise organic matter (Kirchman, 2002), which may explain why they accounted for a larger proportion of the bacterial community at later growth stages, when nutrients (N, Si, P) become limited. It is therefore likely that the shift from a dominance of Proteobacteria during the early growth phases of Actinocyclus, to Flavobacteria during later stages, is governed by a changing pool of chemicals released by the diatom.

Interestingly, peaks in the relative abundance of ASVs within the Actinocyclus microbiome that were identical to the tested isolates did not always align with the timing of the growth benefits observed in co-cultures (Supplementary Figure S8). Although these isolates demonstrated beneficial effects on Actinocyclus’s growth in co-culture, it is possible that interactions between multiple members of the microbiome may further influence the growth benefit observed in the whole microbiome cultures.

Consistent with observations in other diatom species (Grossart et al., 2005; Sison-Mangus et al., 2014; Windler et al., 2014; Amin et al., 2015), we found that when in presence of its full bacterial consortium, the abundance of Actinocyclus sp. was double that of the axenic cultures, emphasising the strong positive influence that bacteria can have on diatom growth (Amin et al., 2012). This was reinforced by the observation that most of the bacterial isolates enhanced the specific growth rate of the diatom with several growing as fast as the cultures with the whole microbiome. However, although the overall growth response of Actinocyclus to the different bacterial isolates tested here was almost entirely positive, none of them reached the maximum cell abundance of the cultures containing the full microbiome, which emphasises the cumulative effect of the microbiome as a whole. Interestingly, even if the maximum cell numbers reached or the specific growth rate between the controls and the co-cultures were similar, the difference in cell abundance through the growth of the diatom significantly changed and was bacteria-specific. This is consistent with a recent study which demonstrated that within 256 diatom-bacteria dual co-cultures (in well-plates), multiple bacterial isolates had different effects on the growth of the 8 diatoms tested and this effect was variable depending on the growth stage of the algae (Deng et al., 2022).

In general, strains belonging to the Rhodobacteraceae had the strongest positive impact on Actinocyclus sp. cell numbers (especially Thalassobius sp.), with greatest levels of growth promotion occurring between days 8 to 12. This concurs with various studies that have shown that members of the Rhodobacteraceae can enhance algal growth (Amin et al., 2015; Durham et al., 2015; Johansson et al., 2019). For example, Sulfitobacter sp. increases cell division in the diatom Pseudo-nitzschia multiseries, through the production of the plant growth-promoting hormone indole-3-acetic acid (IAA) (Amin et al., 2015), and Ruegeria pomeroyi enhances the growth of Thalassiosira pseudonana, through the provision of vitamin B₁₂ (Durham et al., 2015). On the other hand, most of the strains belonging to Gammaproteobacteria, including members of the Alteromonas, Pseudoalteromonas, Vibrio and Marinomonas genera, had the greatest effect on Actinocyclus sp. cell numbers during days 14 to 16. Whilst members of the Alteromonadaceae and Vibrionaceae family are often attracted to diatom metabolites (Sarmento et al., 2013; Taylor and Cunliffe, 2017; Raina et al., 2022), they have been demonstrated elsewhere to have negative or neutral effects on diatom growth (Kim et al., 1999; Aharonovich and Sher, 2016; Wang et al., 2016). However, Alteromonas bacteria can have a positive effect on diatom growth and possess genes involved in vitamin synthesis (B₁, B₇ and B₁₂) and degradation of hydrogen peroxide (catalase enzyme), which can reduce oxidative stress related to photosynthesis (Garcia et al., 2017; Baker and Kemp, 2020). Additionally, recent whole-genome analyses also reported that many Gammaproteobacteria (i.e., Alteromonas, Pseudoalteromonas and Marinomonas genera) possess genes involved in phytoplankton-bacterial interactions, such as the production of vitamin B₁, plant growth promoting hormones or siderophores, suggesting that they may also promote diatom growth (Zoccarato et al., 2022).

The time-dependent effects reported here could indeed be explained by the provision of different molecules by bacterial associates. In culture during early growth phases, diatom cells are not likely to be limited in vitamins or trace metals, but they could benefit from the production of specific secondary metabolites such as growth-promoting hormones. Towards later growth-phases, bacterial provision of remineralized nutrients (e.g., phosphate, nitrogen), as well as trace metals and vitamins that would become limited in the culture medium (Christie-Oleza et al., 2015, 2017), may contribute to enhanced diatom growth. In addition, diatoms may also benefit from bacteria through the removal of toxic compounds that build-up in later growth stages, such as hydrogen peroxide (H₂O₂) (Morris et al., 2011; Coe et al., 2016). Whilst further analyses, coupling chemical and biological measurements will be required to tease-apart the suite of specific mechanisms likely responsible for the time-dependent impact of different bacteria on diatom growth, our results provide an intriguing new insight into the multifaceted influence of diatom associated bacterial communities on the growth of their hosts.