CS29544 was cultured in brain heart infusion (BHI) broth (Becton Dickinson), pH 7.38 at 37°C overnight aerobically with agitation (250 rpm) unless specified. CS29544 was enumerated and spread-plated on BHI agar plates following serial dilution in 1 × phosphate buffered saline (PBS; Dulbecco's Formula), pH 7.4. Escherichia coli was cultured in lysogeny broth (LB, Miller's formula) at 37°C overnight aerobically with agitation (250 rpm) unless specified. When necessary, ampicillin or chloramphenicol were added to BHI or LB at a final concentration of 100 and 35 μg/mL, respectively. To test for motility, CS29544 were grown on 0.4% agar composed of 3 g/L beef extract, 10 g/L Bacto peptone, 5 g/L sodium chloride (BPN) supplemented with 1% of 2, 3, 5-triphenyltetrazolium chloride (redox indicator) or observed microscopically by wet mount.

Stationary phase CS29544 cultures, grown in 10 mL BHI at 37°C, were held statically at room temperature (~25°C) for 6 h to allow autoaggregation. The change in optical density at 600 nm was gently measured at 30 min intervals for 2 h followed by 1 h intervals until 6 h. Autoaggregation was reported as the maximum percent autoaggregation. Typically the endpoint was used, and calculated by Equation (1).

Several additional autoaggregation assays were conducted with modifications after the growth of CS29544 in 10 mL BHI, including static incubation at different temperatures (4 and 37°C), the addition of 50 mM EDTA or PBS, and before and after blending at “whip” speed for 30 s (BL113SG; Black and Decker). Furthermore, autoaggregation assays were completed with CS29544 following growth in 10 mL BHI at different pH values (pH = 4, 5, 6, 7.38, and 8) or incubated anaerobically (90% N₂, 5% CO₂, 5% H₂). Finally, autoaggregation assays were run with CS29544 following growth (10 mL) in different media, including Miller and Lennox LB formulations (LB10 and LB5, respectively), tryptic soy broth (TSB), and BPN broth.

Stationary phase CS29544 and nonautoaggregating clonal variant (CV) cultures (described below) grown in BHI were mounted and held statically at room temperature for 1 h to allow autoaggregation. Still images were taken every 10 s for a total of 1 h by a stationary DSLR camera (Rebel T2i; Canon) with an intervalometer. Images (360 frames) were stitched together to create a video file with 24 frames per second. An additional time lapse video was constructed as previously described with the CS29544 and flagellum competition assays (described below) with still images taken every 20 s for a total of 6 h. Images (1,080 frames) were stitched together to create a video file with 72 frames per second.

Stationary phase CS29544 cultures autoaggregated for 2 h. Then, two separate 100 μL (1%; v/v) aliquots, one from the top fraction of the CS29544 culture and one from the bottom fraction (autoaggregating control) were passed into two fresh tubes of 10 mL BHI broth and incubated as described above. Successive passages following autoaggregation continued until autoaggregation was arrested. Two independent nonautoaggregating variants were isolated and characterized.

Genomic DNA was isolated from CS29544 and CVs using the UltraClean® Microbial DNA Isolation Kit (MoBio Laboratories) according to manufacturer's instructions. High-quality genomic DNA libraries were prepared and sequenced using the Illumina platform by the DNA Services group affiliated with the Roy J. Carver Biotechnology Center at the University of Illinois-Urbana Champaign. Paired-end libraries of CS29544 and CVs were prepared with the TruSeq Genomic DNA Sample Prep Kit (length: 200–600 bp) and sequenced on a HiSeq2500 with the TruSeq SBS Sequencing Kit v1 producing a mean read length of 160 nt. An additional mate-pair library of CS29544 was prepared with the Nextera Mate Pair Library Sample Prep Kit (length: 3–8 kbp) and sequenced on a MiSeq V3 with the MiSeq 600-cycle Sequencing Kit v3 producing a mean read length of 300 nt. Paired-end reads were imported into CLC Genomics Workbench v7.5, and quality and adapter trimmed using default settings. Mate-pair reads were quality and adapter trimmed using Cutadapt (Martin, 2011) and an in-house Perl script provided by the Roy J. Carver Biotechnology Center. Processed mate-pair reads were imported into CLC Genomics Workbench and de novo assembled using default parameters, and only contigs larger than 1,000 bp were kept. The paired-end reads were mapped to the CS29544 de novo assembly and putative single nucleotide polymorphisms were identified (>90% frequency) using the Basic Variant Detection tool with default parameters confirmed by targeted Sanger sequencing using an ABI 3730XL capillary sequencer (Life Technologies).

Targeted gene disruptions (flhA, fliG, motAB, fliC, and flaA) were constructed in the wild-type CS29544 using the lambda Red recombinase system (Cherepanov and Wackernagel, 1995; Datsenko and Wanner, 2000). All bacterial strains, plasmids, and primers used in this study are listed in Tables 1, 2. Briefly, linear DNA fragments were amplified by PCR with pKD3 DNA using the target gene specific primer set (60 bp) and appropriate experimental conditions. CS29544 containing the pKD46 plasmid were grown in 10 mL of LB containing 10 μg/mL of ampicillin and 10 mM L-arabinose at 30°C overnight aerobically with agitation (250 rpm). CS29544 pKD46 electrocompetent cells were transformed with 500 ng of the purified linear DNA fragment. The FRT-Cm^r-FRT cassette in the recombinant mutants was cured by transformation and subsequent removal of the temperature-sensitive flippase (FLP) recombinase helper plasmid (pCP20). The double gene knockout (fliC and flaA) was constructed as described above for the CS29544 ΔflaA strain. Gene disruptions were confirmed by junction fragment PCR using the appropriate primer sets and experimental conditions.

A fliC complementation vector was constructed by GenScript. Briefly, a 1,011 bp sequence, containing the fliC coding sequence and native promoter, was obtained from the publically available CS29544 genome (NCBI Reference Sequence: NZ_CP011047.1). The entire DNA fragment was synthesized and cloned into the pET-11a vector with the restriction enzymes BglII and BamHI. The cfliC vector was electroporated into E. coli Top10 and subsequently electroporated into the CS29544 ΔfliC and ΔflaAΔfliC strains using LB broth. Putative complements were grown in BHI or on motility agar plates supplemented with 50 μg/mL ampicillin. Restoration of wild-type function was assessed by autoaggregation assays, motility assays, microscopy, and flagella harvest as detailed above and below.

The presence of extracellular flagella of CS29544, gene knockout, and complementation strains were determined by a combination of imaging techniques. Log or stationary phase CS29544, gene knockout, and complementation cultures were stained using a crystal violet-based flagella stain (Hardy Diagnostics) according to manufacturer's instructions. Stains were visualized using a light microscope at 1,000 × total magnification (BA210; Motic). Images were captured with a 2-megapixel Motic camera.

Several overnight colonies of CS29544, gene knockout, and complementation strains were gently lifted from BHI agar plates and suspended in phosphate buffered Karnovsky's fixative containing 2% glutaraldehyde and 2.5% paraformaldehyde. Transmission electron microscopy (TEM) was completed by the Beckman Institute's Microscopy Suite at the University of Illinois-Urbana Champaign. Briefly, the samples were stained with 2% uranyl acetate for 1 min and visualized using a CM200 LaB6 transmission electron microscope (FEI Co.). TEM was conducted at 120 kV and images were captured with a 2 k × 2 k digital camera (Tietz; Gauting; Germany). Several locations on the grids were examined, and the pictures were representative of the whole sample.

The extracellular protein fraction of CS29544, gene knockout, and complementation strains was harvested by differential centrifugation (DePamphilis and Adler, 1971). Bacteria were cultured in two baffled flasks each containing 500 mL of BHI and incubated overnight at 37°C with agitation (250 rpm). Stationary phase cultures (1 L total) were centrifuged at 3,220 × g for 10 min at 4°C. Bacterial pellets were resuspended in a total of 250 mL 0.1 M Tris-HCl, pH 7.8, and blended at room temperature for 30 s at “whip” speed. Blended suspensions were centrifuged at 12,000 × g for 10 min at 4°C. The supernatant was further ultracentrifuged at 55,000 × g for 1 h at 4°C. Protein pellets were resuspended in a total of 1 mL 0.1 M Tris-HCl, pH 7.8, containing 50% glycerol (v/v, protein storage buffer) and stored at −20°C. Total protein was quantified with the Bradford Assay (BioRad Laboratories) and visualized with SDS-polyacrylamide gel electrophoresis. Typical flagellum protein recovery was 0.5–0.7 mg/mL from 1 L of cell mass (~10¹² cells).

The putative FliC (28.9 kDa) band from CS29544 flagellum preparation was gel-excised and treated in-gel with trypsin (G-Bioscience) by the DNA Services group affiliated with the Roy J. Carver Biotechnology Center at the University of Illinois-Urbana Champaign. Protein was digested at a ratio of 1:20 (trypsin:protein) in 25 mM ammonium bicarbonate at 55°C for 30 min. Following lyophilization, peptides were analyzed by liquid chromatography-mass spectrometry. A total of 1–2 μg of digested peptides were loaded into a Dionex Ultimate 3000 RSLCnano connected directly to a Thermo LTQ-Velos-ETD Pro Mass Spectrometer (Thermo Fisher Scientific). Peptide were run on an Acclaim 300 C18 nano column (Thermo Fisher Scientific) using a gradient of 100% A (water + 0.1% formic acid) to 60% B (acetonitrile + 0.1% formic acid) at a flow rate of 300 nL/min. Raw data were collected by Xcalibur (Thermo Fisher Scientific) and processed with an in-house Mascot Distiller and Mascot Server (Matrix Science) and identified with the NCBInr database.

The secondary structure of the CS29544 FliC protein was predicted from the amino acid coding sequence (NCBI Reference Sequence: NZ_CP011047.1) using the Iterative Threading Assembly Refinement (I-TASSER) method (Yang and Zhang, 2015) with default parameters. The I-TASSER method is publically available at http://zhanglab.ccmb.med.umich.edu/I-TASSER/, accessed 10/15/2015. The theoretical secondary structures of FliC were visualized and modified using UCSF Chimera v.1.10.2 (Pettersen et al., 2004), publically available at https://www.cgl.ucsf.edu/chimera/. The hydrophobicity index of the primary FliC amino acid sequence was determined using the ProtScale tool from the ExPASy Bioinformatics Resource Portal (Gasteiger et al., 2005), publically available at http://www.expasy.org/, accessed 06/01/2016. The hydrophobicity index was calculated using the Kyte and Doolittle amino acid scale (Kyte and Doolittle, 1982) with a window size of 15 amino acids.

A 3 mL aliquot of stationary phase CS29544 culture was mixed with detached flagellar pieces at a concentration of 0.1, 1, 5, 10, or 20 μg/mL of total flagellum protein and autoaggregated for 6 h. Controls included adding 20 μg/mL of bovine serum albumin (BSA) or equal volume of protein storage buffer (no protein).

Polyvinyl chloride tubing (PVC; 0.318 cm outer diameter; 0.159 cm inner diameter; U.S. Plastic Corporation) was cut into 5 cm long pieces (external surface area ~5.15 cm²) with a sterile blade. The PVC tube pieces were disinfected and submerged in 70% ethanol for 10 min and aseptically dried. Two PVC tube pieces were aseptically transferred to a 15 mL centrifuge tube containing 10 mL of BHI supplemented with 50 μg/mL ampicillin for the ΔfliC/cfliC strain. Each experimental vessel was inoculated with 1% (v/v) stationary phase CS29544, ΔmotAB, ΔfliC, and ΔfliC/cfliC cells and incubated vertically at 37°C aerobically with agitation (250 rpm) for 24 h. After incubation, each PVC tube piece was transferred with sterile forceps and washed three times in 5 mL PBS, pH 7.4. After washing, each PVC tube piece was placed into 30 mL PBS, pH 7.4, containing 3 g of autoclaved 0.1 mm diameter glass beads (Research Products International). Biofilms were subsequently disrupted by vortex at maximum speed for 1 min. Bacterial biofilm populations were enumerated by serial dilution (10⁻¹–10⁻³) in 0.1% peptone water, spread-plated (0.1 mL in triplicate) on BHI agar plates, and incubated at 37°C overnight aerobically.

Autoaggregation and biofilm formation assay results were from a minimum of three independent replicates. The differences between the mean maximum percent autoaggregation due to temperature, redox balance, pH, blending, and various media; the differences between biofilm formation due to the presence of FliC were determined with SAS® (Version 9.4; SAS Institute) using the generalized linear model. When statistical significance was observed (P < 0.05), a post-hoc mean separation was run using Tukey's Honest Significance Difference test which controlled for unequal sample sizes. All data satisfied the assumptions of normality and homogeneity of variance.

Stationary phase BHI cultures of CS29544 autoaggregated at 25°C within 30–60 min (Figure 1A). Following 6 h static incubation, the maximum mean autoaggregation percentage was 70.3 ± 2.2%. Various growth media and physiological conditions were tested to better understand CS29544 autoaggregation. Autoaggregation of BHI-grown cells at 37°C was significantly higher when statically incubated at 37°C than at 4 or 25°C (P = 0.0022; Figure 1B) or in the presence of EDTA (P = 0.0092; Figure 1C) compared to the PBS-added control. Although the maximum mean autoaggregation percentage was lower for cells grown in a reduced environment, the difference was not significant (P = 0.3003, Figure 1D). Autoaggregation following overnight growth in BHI was significantly higher than overnight growth in LB10, LB5; TSB, and BPN (P = 0.0203; Figure 1E). The difference in autoaggregation was not readily explained by the presence of salt, extracts, protein sources, or inclusion of phosphates but rather the initial pH of the media. Therefore, CS29544 was grown in BHI with varying initial pH values (range pH 4–8). CS29544 did not grow at pH 4, but its maximum growth was not affected by BHI from pH 5–8 (Figure 1F). As predicted, the maximum mean autoaggregation percentage decreased with decreasing pH (P < 0.0001; Figure 1F). The maximum mean autoaggregation percentage in BHI at an initial pH 6 was 51.9 ± 1.6%, which was slightly lower than the maximum mean autoaggregation percentage observed in the various media tested. During this initial characterization, an abolishment of autoaggregation was only achieved when grown in BHI at pH 5; however, this observation did not clearly point to a specific mechanism. Since the maximum mean autoaggregation percentage never reached 100%, we hypothesized that there might be a nonautoaggregating subpopulation of CS29544 mediated by an identifiable genetic variation.

Following five successive passages selecting against autoaggregating CS29544, we were able to isolate two independent nonautoaggregating CVs of CS29544 (Figures 2A,B; see Movie S1 in the Supplementary Material). De novo assembly of the mate-pair library preparation from wild-type CS29544 was used as the reference genome for comparative genomic analysis with the nonautoaggregating CVs (2.10 and 3.6) assemblies. One unique single nucleotide polymorphism was detected in each of the nonautoaggregating CV. Strain 2.10 contained a putative deletion of two consecutive base pairs (GC), while strain 3.6 contained a putative deletion of a single base pair (C). Further corroboration by Sanger sequencing revealed the two nonautoaggregating CVs contained frameshift mutations, and NCBI BLAST analysis revealed that these mutations were located in the open reading frames of two flagellum proteins, FlhA (2.10) and FliG (3.6). Comparison with the full length wild-type FlhA and FliG proteins (692 and 340 amino acids, respectively) revealed that the frameshift mutations results in truncated proteins (157 and 183 amino acids, respectively). Strains carrying these variant mutations are referred to as flhA_CV (2.10) and fliG_CV (3.6), respectively. Accordingly, we assessed the flagellation of the wild-type CS29544, flhA_CV, and fliG_CV with motility assays and microscopy (Figure 2C). Both the flhA_CV and fliG_CV strains were nonmotile and aflagellate by staining and TEM (Figure 2C). As a result, we constructed a variety of gene knockouts to determine if structurally intact and functional flagella were required for autoaggregation in CS29544.

Published C. sakazakii genomes, available from NCBI, have annotated some 40 genes related to their flagellum's structure, function, and regulation. Therefore, gene knockout strains (Table 1) were constructed to disrupt the structure and function of the CS29544 flagella (see Figures 3A–C for a simplified diagram of the bacterial flagella outlining our knockout strategy). Two basal body proteins, FlhA and FliG (same genes as the nonautoaggregating CVs) were targeted. Previously, FlhA truncation mutants in Campylobacter jejuni resulted in aflagellate cells lacking flagellar components past the inner membrane (Abrusci et al., 2013). FliG, along with FliN and FliM, forms the C ring of the basal body (Zhao et al., 2014). Proton-driven conformational changes in the MotA and MotB stator (motor) proteins (Kojima and Blair, 2001) directly interact with the C terminus of the FliG protein of the C ring to provide rotation to the flagella (Irikura et al., 1993; Lloyd et al., 1996). Although FliG is integral for the flagellum function, its necessity for assembly is disputed (Irikura et al., 1993; Lloyd et al., 1996). It was hypothesized that the disruption of motAB would render the cells nonmotile while retaining the structural components. Additionally, the disruption of flhA and fliG would block the early assembly of the flagella and the cells would be aflagellate. Based on the published CS29544 genome, CS29544 has redundant filament proteins; therefore, both FliC and FlaA single and double gene knockouts were constructed. Finally, the function of the flagella was disrupted by targeting the two motor proteins, MotA and MotB.

The ΔflhA, ΔfliG, ΔfliC, and ΔflaAΔfliC strains did not autoaggregate and were aflagellate by staining and TEM and nonmotile (Figures 4A–C). Conversely, the ΔmotAB and ΔflaA strains remained autoaggregative and had visible flagella by staining and TEM (Figures 4A–C). These two strains differed in their motility: ΔflaA was motile while ΔmotAB was not motile. Therefore, we concluded that motility was not required for autoaggregation in CS29544. Based on the phenotypes of the structural gene knockouts, we hypothesized that filaments composed of FliC, not FlaA, were required for autoaggregation. Upon mechanical removal of the filaments, only the autoaggregating CS29544, ΔmotAB, and ΔflaA strains possessed a dense 28.9 kDa protein band (Figure 5). The band was confirmed as FliC by a total of five peptides with individual ion scores of 44 and a Mascot ion score of 1,191 (data not shown). Furthermore, the ΔfliC strain did not contain a 50.1 kDa band, which is the predicted size for FlaA. Autoaggregation phenotype, motility, and flagellation were restored in the ΔfliC/cfliC and ΔflaAΔfliC/cfliC complementation strains (Figures 6A–C). Finally, we confirmed a loss of autoaggregation in CS29544 after mechanical removal of flagella from 79.4 ± 4.4% to 2.2 ± 2.2% with no reduction in cellular viability (data not shown).

Visual analysis of the CS29544 TEM image revealed supercoiled flagella linking several bacteria together in a cluster (Figure 7A). Therefore, it was hypothesized that autoaggregation in CS29544 proceeds as flagella from one cell becomes entangled with the flagella from neighboring cells. To test this hypothesis, mechanically detached flagella pieces were added to stationary phase CS29544 cultures. Autoaggregation of CS29544 was not affected by adding 0.1 or 1.0 μg/mL of detached flagella, 20 μg/mL BSA, or the no-protein control (PSB; Figure 7B). Autoaggregation was prevented when 20 μg/mL of detached flagella were added. The addition of detached flagella altered the manner by which the CS29544 autoaggregated. The controls autoaggregated as before by forming small cell flocs that settled to the bottom of the tube. In the presence of 5 and 10 μg/mL of detached flagellum flocs did not form, rather a large mass of cells settled gradually to the bottom of the tube (see Movie S2 in the Supplementary Material).

The predicted secondary structure of the CS29544 FliC protein was analogous to homologous flagellins with a highly conserved flagellin N and C termini linked by a hypervariable region (Figure 8A). Analysis of the hydrophobicity scores of the linear amino acid sequence revealed a peak followed by a valley in hydrophobicity (Figure 8B) located in the hypervariable region of the CS29544 FliC. The bacterial filament is composed of several thousand flagellin proteins with the conserved regions stacked laterally. The hypervariable region is externally exposed and able to interact with its environment. We hypothesize that the hypervariable hydrophobic peaks and valleys of neighboring cells interact laterally to promote autoaggregation in CS29544. Further work will be required to test this hypothesis.

The total cellular population of wild-type CS29544 biofilms on PVC was 1.5 × 10⁴ ± 3.0 × 10³ CFU/cm². The total cellular populations of ΔmotAB, ΔfliC, and ΔfliC/cfliC on PVC biofilms were 4.8 × 10³ ± 2.2 × 10³, 2.4 × 10⁴ ± 7.0 × 10³, and 1.0 × 10⁴ ± 6.5 × 10² CFU/cm², respectively. CS29544 biofilm formation on PVC was not mediated by FliC under the tested conditions (P = 0.0928; Figure 9).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.