Bacterial strains used in this study are listed in Table 1. S. salivarius strains were grown as described previously (Couvigny et al., 2015b). F. nucleatum DSM 20482 was grown in brain heart infusion broth supplemented with 0.25% L-glutamic acid. Erythromycin (5 μg/ml) or kanamycin (1000 μg/ml) were added to the medium as required. PCRs were performed using Phusion^® high-fidelity DNA polymerase (New England Biolabs, Ipswich, MA, United States). The primers were purchased from Eurofins MWG Operon (Germany) and are listed in Supplementary Table S1. When necessary, PCR products and DNA restriction fragments were purified using QIAquick kits (Qiagen). Plasmids were purified using the QIAprep Miniprep kit (Qiagen).

Mutants with altered auto-aggregation properties were isolated using Todd Hewitt broth with 0.5% yeast extract with a low (0.03%) concentration of agar (Mercier et al., 2000, 2002; Llull et al., 2005). In these conditions, the extent of free displacement of bacteria in the environment depends on the viscosity of the semi-liquid (SL) medium.

Mutagenesis with pGhost9::ISS1 was performed essentially as described previously (Guedon et al., 2001). Briefly, cells containing pGhost9::ISS1 were grown overnight at 30°C in M17 medium in the presence of erythromycin. Stationary-phase cultures were diluted 1:1000 in fresh M17 broth without erythromycin, incubated for 150 min at 30°C and then at 38°C for 150 min. Samples were then diluted and plated at 38°C. Clones that grew on M17 medium containing erythromycin were selected as transposon mutants. Approximately 10,000 mutants were pooled and the mixture was deposited on the surface of 20 flasks containing SL medium and erythromycin. After 140 h of growth, 40 samples at the ends of the roots were removed and streaked out on agar plates to isolate single colonies of auto-aggregation mutants. The pGhost9::ISS1 insertion site was identified by cloning and sequencing of the chromosomal junctions as described previously (Couvigny et al., 2015b).

Streptococcus salivarius spontaneous mutants JIM9443 and JIM9447 were sequenced using the Illumina HiSeq 2000 system at I2BC¹, with around 7 million paired-end 100 base-long reads. Strains JIM9442, JIM9445, JIM9446, and JIM9448 were sequenced using the SOliD technology 4 platform at MetaGenoPoliS², with around 5 million single 50 base-long reads. SOLiD reads containing adapter and/or barcode fragments with a mean quality value <20 and low-quality reads (with 3 or more “N”) were discarded. Clean reads were mapped to the reference JIM8777 strain using Bowtie (Langmead et al., 2009) and variations were detected using the Tablet software (Milne et al., 2013). The Illumina and SOLiD sample sequences are available in the SRA database (project ID PRJEB23560).

Mutant derivatives of strain JIM8777 were constructed by exchanging the coding sequence of a target gene (sequence between the start and stop codon) for an erythromycin or kanamycin resistance cassette devoid of a transcriptional regulatory region (e.g., promoter or terminator), as described previously (Couvigny et al., 2015b). After peptide-induced transformation, integration of the antibiotic cassette at the appropriate location was verified by PCR (Supplementary Table S1).

Adhesion of S. salivarius strains to HT-29 human epithelial cell lines (colon adenocarcinoma; ATCC HTB-38) and MUC2 proteins (M2378, Sigma) was performed as described previously (Couvigny et al., 2017). Experiments were repeated at least three times, each in duplicate.

Transmission and scanning electron microscopy (TEM and SEM, respectively) experiments were conducted as described previously (Couvigny et al., 2017). Images were acquired and analyzed at the MIMA2 microscopy and imaging platform³.

The ability of bacterial cells to auto-aggregate and co-aggregate was assessed by a spectrophotometric procedure (Handley et al., 1987). Stationary-phase cells from overnight cultures were centrifuged, washed twice with PBS and resuspended in aggregation buffer to an optical density at 600 nm (OD₆₀₀) of 0.5 ± 0.05 (Burton et al., 2013). To determine percentage auto-aggregation, bacterial suspensions were placed in a glass test tube, vortexed, and the suspensions immediately transferred to a cuvette at room temperature (20 ± 1°C). The OD₆₀₀ of the bacterial suspensions was monitored at various time points and the percentage auto-aggregation was expressed as follows: (1 - OD_t/OD_t0) × 100, where OD_t represents the OD₆₀₀ of the suspension at time t (4, 20, or 24 h) and OD_t0 the OD₆₀₀ at time zero. To evaluate the impact on auto-aggregation of various pre-treatments of bacteria, cell suspensions prepared in aggregation buffer as described above were treated with EDTA (2 mM, instantaneous), Ca²⁺ (CaCl2, 2 mM, instantaneous), proteinase K (0.1 M, 2 h, 37°C), trypsin (0.1 M, 2 h, 37°C) and by heat (85°C, 10 min). Proteinase K- and trypsin-treated cells were harvested by centrifugation and the pellets washed twice in aggregation buffer before the auto-aggregation assay. To determine percentage co-aggregation, suspensions of S. salivarius strains were combined with an equal volume of a test strain (F. nucleatum) or aggregation buffer (as a control) and incubated at room temperature, without agitation, for at least 6 h. The percentage co-aggregation was calculated using the following equation: 100 × [OD₆₀₀ (S. salivarius control + F. nucleatum control)/2 - OD₆₀₀ (S. salivarius + F. nucleatum)]/OD₆₀₀ (S. salivarius control + F. nucleatum control)/2, where OD₆₀₀ (S. salivarius control + F. nucleatum control) represents the OD₆₀₀ of control tubes containing only the S. salivarius or F. nucleatum strain at time t and OD₆₀₀ (S. salivarius + F. nucleatum) represents the OD₆₀₀ of the mixture of both suspensions at time t. Experiments were repeated at least three times, each in duplicate.

Biofilms were grown in microscope grade μclear^® base 96-well plates (Greiner Bio-one, France) and analyzed by confocal laser scanning microscopy (CLSM) as described previously (Couvigny et al., 2015b). Briefly, surface-associated bacteria were fluorescently labeled with 5 μM Syto9 (Invitrogen, France) and analyzed under a Leica SP2 CLSM at the MIMA2 platform. Cells labeled with Syto9 were exposed to an argon laser beam, set to 20% of its maximal power, at 488 nm. The resulting fluorescence was collected in the 500–550 nm range with a photomultiplier. Images (512 pixels × 515 pixels) were recorded through a water immersion long distance objective (NA = 0.8) at micrometer intervals through the thickness of the biofilm. Three independent experiments were performed for each strain. Three-dimensional projections of the biofilm structures were reconstructed using the Easy 3D function of the IMARIS 7.0 software (Bitplane, Switzerland). Extraction of quantitative biofilm geometric descriptors from the CSLM series was performed with the PHLIP Matlab-based image analysis toolbox (Mueller et al., 2006).

Variables are expressed as the means and standard deviations of results obtained from three (auto-aggregation, co-aggregation, and adhesion assays) or two (CLSM analysis) independent experiments. Comparisons between groups of variables were analyzed using the student’s t-test. GraphPad Prism version 6 (GraphPad software Inc., La Jolla, CA, United States) was used for statistical analyses. A P-value < 0.05 was considered to be statistically significant.

Twenty-eight representative strains of S. salivarius were screened for auto-aggregation (Figure 1A and Table 1). Most of the S. salivarius isolates tested under our culture conditions formed aggregates, although their auto-aggregation ability was found to be highly strain specific (ranging from 9 to 56%). We chose one of the most strongly aggregating strains, JIM8777, for further study. The chromosome sequence of JIM8777 has already been determined (Guedon et al., 2011).

To characterize the environmental factors influencing JIM8777 auto-aggregation and to clarify the nature of the interactions, we evaluated the impact of various treatments on auto-aggregation levels. Auto-aggregation was most efficient at pH 8 (Figure 1B). We also found that addition of Ca²⁺ led to an increase in auto-aggregation, whereas heat-treatment, addition of EDTA, proteinase K or trypsin led to a reduction in auto-aggregation (Figure 1C), suggesting that proteins are involved in these cell-to-cell interactions.

To identify JIM8777 genes involved in auto-aggregation, we first isolated spontaneous mutants, which exhibited increased or decreased auto-aggregation phenotype. For this, we used a strategy initially developed for selection of mutants with alterations in their ability to form chains. This method is based on differential sedimentation of bacteria in semi-liquid (SL) agar medium according to chain-forming capacity (Mercier et al., 2000, 2002; Chapot-Chartier et al., 2010). In these conditions viscosity of SL medium restricts free displacements of bacteria and sedimentation represents the main means of displacement for bacteria which do not have properties of autonomous motility. When colony of non-chain-forming bacteria initiates from one cell, simultaneous growth and sedimentation results in streaked colony (Llull et al., 2005). Contrariwise, chain-forming bacteria in these conditions are trapped in extracellular agar matrix and do not sediment, consequently forming round colonies. It is possible to observe and to select mutants escaping such immobilization as they form faster sedimenting “roots” from round colonies (Mercier et al., 2002).

We hypothesized that similarly to chain-forming bacteria, the auto-aggregates of JIM8777 strain will counteract sedimentation, resulting in formation of round colony. In keeping with this reasoning, we found that the auto-aggregates of the JIM8777 strain became immobilized in the SL medium (Figure 2A, left panel, black arrow). After 140 h of growth, we observed the appearance of “roots” and “caps” from these round colonies, presumably formed by faster- and slower-sedimenting spontaneous mutants, respectively (Figure 2A, left and middle panels, red and blue arrows). Samples of the roots and of the caps were then removed and streaked out on agar plates to isolate single colonies of SL medium sedimenting mutants. Analysis of bacterial growth in liquid medium confirmed that selected mutants exhibited altered auto-aggregation phenotypes (representative images of the two groups of mutants are presented in Figures 2B,C). Compared to the parental strain JIM8777, the slower-sedimenting mutants (such as JIM9443) showed a hyper-auto-aggregating phenotype in liquid medium, whereas the faster-sedimenting mutants (including JIM9445) showed diminished auto-aggregation capacity (Figures 2B,C).

We then performed a similar experiment in which bacterial cells containing the whole genome-wide random insertion library of S. salivarius (obtained after pGhost9::ISS1 mutagenesis) was inoculated on the top of a flask with SL medium. After 140 h, the faster-sedimenting “roots” appeared and were collected for further analysis (Figure 2A, right panel). This experiment led to the isolation of 40 additional faster-sedimenting mutants.

The gene mutations potentially associated with the altered sedimentation and auto-aggregation phenotypes were mapped by whole genome sequencing of 6 spontaneous mutants and cloning and sequencing of the chromosomal DNA fragments flanking the insertion site of ISS1 from 11 insertion mutants. Altogether, 12 genes were identified from the two independent experiments (Figure 2D and Table 2). Two different spontaneous mutations were identified in the asp1 gene, as well as two different ISS1 insertions in the genes cshA, secA2, gtfG and gtfH. Our independent selection of several mutations in the same gene confirms the importance of these genes in the auto-aggregation phenotype.

To confirm the phenotypes of the bacterial mutants and the correlation between sedimentation in SL conditions and auto-aggregation in liquid medium, we constructed a series of mutants in the JIM8777 background carrying deletions of the genes in which we identified the point mutations associated with altered sedimentation (Table 1). We found that deletion of the asp1, asp2, bglB, cshA, gtfE, gtfG, gtfH, liaR, secA2, and srtA genes, point mutations in which induced a faster-sedimenting phenotype in SL conditions, led to a decreased auto-aggregation phenotype in liquid medium (Figure 2E). As cwpB and cwpK, two genes coding for enzymes predicted to synthetize cell wall polysaccharides (WPS), are located in the same operon (Figure 2D), we chose to include only one of these genes (cwpK) for further analyses and confirmed that the ΔcwpK mutant showed an increased auto-aggregation phenotype (Figure 2E). Altogether, our results confirmed that deletion of the genes identified as having an impact on bacterial sedimentation also affected the auto-aggregation properties of S. salivarius. This confirms that our method, based on differential sedimentation in SL conditions, provided an efficient tool for the specific isolation of mutants with altered auto-aggregation properties.

We investigated the ability of the auto-aggregation mutants to adhere to the human colonic epithelial cell line HT-29 (Figure 2F), to the mucin 2 (MUC2) glycoprotein (Figure 2G) and to the human pathogen F. nucleatum (co-aggregation) (Figure 2H). In comparison to the wild-type JIM8777, 5 out of 10 of the deletion mutants with altered auto-aggregation capacity (Δasp1, Δasp2, ΔgtfE, ΔsecA2, and ΔsrtA) displayed a significant (P < 0.005) reduction in adhesion to both HT-29 and MUC2 (Figures 2F,G). Moreover, the cwpK (increased auto-aggregation) and srtA (decreased auto-aggregation) deletion strains showed a drastic reduction (>90%; P < 0.0001) in co-aggregation with F. nucleatum compared to the wild-type strain (Figure 2H). The co-aggregation ability of the ΔgtfG strain was also altered but to a lesser extent (50% reduction). Similar results were obtained with the spontaneous mutants asp1_spont1 (JIM9445), asp2_spont (JIM9447), cwpB_spont (JIM9442) and cwpK_spont (JIM9443) (data not shown).

Cell aggregation and biofilm formation are interconnected bacterial properties (Sanchez et al., 2010; Couvigny et al., 2017). We evaluated whether the auto-aggregation factors we had identified also contributed to biofilm formation by analyzing biofilms formed by JIM8777 derivatives and comparing them with those of the wild-type strain by confocal microscopy. After 5 h of sessile growth, all mutants were able to form biofilms under the conditions employed. JIM8777 formed a relatively dense and filamentous biofilm that covered the entire available surface (Figure 3). Biofilms formed by the mutants differed in their architecture from JIM8777 (P < 0.05 compared to wild-type for at least one geometric descriptor of the biofilm structure). With the exception of strain ΔgtfG, all non-auto-aggregative strains (asp1_spont1,Δasp1,Δasp2, ΔbglB,ΔgtfE,ΔgtfH,ΔsrtA) developed compact and uniform biofilms. Hyper-auto-aggregative mutants (cwpB_spont, cwpK_spont and ΔcwpK) formed excessively filamentous biofilms compared to wild-type. This result indicates that the S. salivarius factors implicated in cell-to-cell interactions in planktonic conditions also influence the morphology and spatial organization of bacterial biofilm communities.

To provide insight into the role played by the products of the cwpK and asp1 genes in auto-aggregation, we evaluated the cell surface morphology of the corresponding mutants by scanning and transmission electron microscopy (SEM and TEM, respectively). We observed by SEM that the JIM8777 cells exhibited a rough and irregular surface that was partially covered by fibril-like structures (Figure 4A and Supplementary Figure S1). The ΔcwpK mutant displayed a smooth surface with fibril-like structures, whereas the Δasp1 mutant exhibited a rough surface without fibrils. Furthermore, the ΔcwpK Δasp1 mutant exhibited an additive phenotype, with both a smooth and bald surface. These findings suggest that surface roughness is linked to WPS biosynthesis involving CwpK, whereas Asp1 may be responsible for the presence of the extracellular fibrils.

TEM of osmium tetroxide-stained cells of the wild-type strain revealed the presence of an electron dense layer outside the cell wall, potentially corresponding to WPS (Figure 4B). The overall morphology of JIM8777 and Δasp1 was similar, however, the electron-dense layer was completely absent in micrographs of the ΔcwpK and ΔcwpK Δasp1 strains. A similar result was obtained with the spontaneous cwpK_spont (JIM9443) mutant, confirming that the point mutation resulted in loss of function (Supplementary Figure S2).

We have shown previously that the SecA2/Y2 transported glycoproteins form fibril-like structures that can be observed outside of the cell wall of ruthenium red-stained cells (Couvigny et al., 2017). Accordingly, here we found that the fibrillar layer (mainly composed by SRR glycoproteins) was greatly diminished in the Δasp1 and ΔcwpKΔasp1 mutants (Figure 4C), as well as in the asp1_spont1 and ΔsrtA strains (Supplementary Figure S2). Altogether, these results indicate that CwpB and CwpK are involved in WPS production and show that Asp1 and SrtA play a role in fibril detection, and therefore in the SRR glycoprotein synthesis pathway. It should also be noted that the surface glycoprotein layer appeared denser in the cwpK mutant background (i.e., ΔcwpK and ΔcwpKΔasp1), suggesting that WPS may hamper glycoprotein formation, exposure, or detection.