L. lactis used in this study (Table 1) were grown as standing cultures at 30°C in M17 (Oxoid, Thermo Scientific, Hampshire, UK) broth supplemented with 1% glucose (GM17) or 1% lactose (LM17). E. coli E10 containing pUC19 with an erythromycin resistance gene, pUC19E, was grown in tryptone yeast extract broth (TYB) at 37°C under vigorous shaking and access of oxygen. When required, antibiotics were added to the media: erythromycin (Em) was used at 10 μg/ml; chloramphenicol (Cm)—10 μg/ml; nisin—10 ng/ml. Optical density at 600 nm (OD₆₀₀) was measured using a single cell spectrophotometer (Ultrospec 2000, Pharmacia Biotech, Centerville, USA). Exponentially growing cells were prepared by diluting an overnight culture to an OD₆₀₀ of 0.01 and subsequent incubation until an OD₆₀₀ of 0.45 ± 0.04 was reached. Stationary cells were prepared similarly by growing a culture for 16–18 h.

Cells were harvested by centrifugation at 2,676 g for 3 min at room temperature and the cell pellet was washed 2x with 1 volume of 10 mM phosphate buffer (PB; pH = 6.7) and re-suspended in the same buffer to an optical density OD₆₀₀ of 1. Approximately 2 ml of this cell suspension was filled into the ZetaSizer DST1070 cuvette, which was inserted into ZetaSizer (Nano-ZS, Malvern, Malvern, UK). The electrophoretic mobility of cells was measured at 20°C and automatically re-calculated into the values of zeta potential (mV).

Cell surface hydrophobicity (CSH, %) was measured as described previously (Rosenberg et al., 1980) with the following modifications: 5 ml of cell suspension (OD₆₀₀ = 1) in PB was mixed with 2 ml of either petroleum or hexane (both from Sigma-Aldrich Chemie Gmbh, Munich, Germany) in surfactant-free glass tubes with a surfactant-free stopper. Tubes were vortexed for 2 min and kept still for 15 min at room temperature to allow phase separation to occur. Subsequently, 1 mL of the aqueous phase was transferred to a spectrophotometer cuvette and optical density (OD₆₀₀) of the cell suspension was measured at 600 nm. The surface hydrophobicity was calculated according to the following formula:

Cell Surface Hydrophobicity (CSH, %) = (A₀−A₁)/A₀·100, in which A₀ represents the initial OD₆₀₀ of cell suspension before mixing and A₁ is the OD₆₀₀ of the water phase after mixing with petroleum or hexane and subsequent phase separation.

Emulsion Stability (E24, %) was determined as described earlier (Khopade et al., 2012; Padmapriya, 2012). Initially samples of cells in exponential and stationary growth phases were prepared in the same way as described for CSH with slight modifications. Briefly, 5 ml of cell suspension (OD₆₀₀ = 1) in PB was mixed with 2 ml hydrocarbon, vortexed for 2 min and left standing for 24 h. The E24 index is given as a percentage according to: E24(%) = h₁/h₂*100%, in which h₁ is the height of the emulsified layer and h₂ is the total height of the emulsified layer and liquid column, both in mm.

To 1 ml of cell suspension in either stationary or exponential growth phase prepared as described above (OD₆₀₀ = 1) 1 μL Syto 9 was added (Green Fluorescent Nucleic Acid Stain, Life Technologies, Bleiswijk, The Netherlands) after which the suspension was incubated in the dark for 30 min at room temperature. The cells were washed twice with PB to remove free dye.

Cells and proteins were stained as described above; 100 μl of stained cells (OD₆₀₀ = 1) were mixed with the same volume of 1% protein solution. Subsequently, 1–2 μl of this mixture was placed on a microscope slide, covered with a cover slip and examined at 100-fold magnification using an Olympus BX41 microscope (Olympus Corporation, Tokyo, Japan) with excitation wavelengths of 485 ± 10 and 635 ± 10 nm and emission wavelengths of 530 ± 30 and 660 ± 20 nm, respectively. Images were acquired with a charge-coupled-device camera with identical acquisition settings for all images; exposure to excitation light was for 200 ms for the Syto 9-stained cells and for 2,000 ms for the Nile Blue A-stained proteins. Image overlays were generated using ImageJ version 1.45s (https://imagej.nih.gov/ij).

Gene-trait matching (GTM) was performed using PhenoLink (Bayjanov et al., 2013). Data visualization was done using R (https://cran.r-project.org/bin/windows/base/). The heatmap.plus function using Euclidian distance matrices and average hierarchical clustering and data scaling was used for the generation of heat maps.

Genes targeted for overexpression were PCR amplified using the hot-start KOD polymerase (Novagen, Madison, USA) according to the protocol of the manufacturer with primers listed in Table S1. Amplicons were purified using MSB® Spin PCRapase (Invitek, Gmbh, Berlin, Germany), digested with the restriction enzymes ScaI and XbaI (Fermentas GmbH, St. Leonn-Rot, Germany) and ligated into plasmid pNZ8150 (Table 1) digested with the same enzymes using T4 DNA ligase (Invitrogen, Breda, the Netherlands). DNA purification was carried out according to the protocols supplied by the manufacturers with Wizard®SV gel and PCR Clean-Up system (Promega, Leiden, The Netherlands). Ligations were carried out at 16°C for 16–18 h and the ligated product, which was precipitated with 3 M sodium acetate (pH 5.5) and 70% ethanol, was used to transform electrocompetent cells of L. lactis NZ9000 (Holo and Nes, 1989). After electroporation, cells were plated on GM17 agar plates containing 10 μg/ml chloramphenicol and incubated at 30°C for 3 days. Single colonies were isolated and insert DNA in the plasmid was confirmed using colony PCR with the appropriate primers (Table S1A).

Double crossover knock-outs of 4 genes were made in L. lactis MG1363 using pUC19 harboring an erythromycin (Em) resistance gene, pUC19E. Upstream and downstream flanking regions (left flank, LF; right flank, RF) of the target genes were amplified with the primer pairs described in Table S1B. Typically the left amplified flanking regions contained a sequence overlapping with the right flanking region (termed +vlag in the primer name in Table S1B) which allowed to perform a splicing by overlap extension (SOE) PCR (Horton et al., 2013) to generate amplicons which were digested and subsequently ligated into similarly digested pUC19E. The ligation mixture was used to transform L. lactis MG1363 as described previously (Holo and Nes, 1989). Strains that were the result of a single crossover event were selected after plating and incubating the transformation mixture for 2 days at 30°C on GM17 agar plates supplemented with erythromycin. These single cross-over strains were grown for at least 100 generations in GM17 broth without Em to obtain strains that were the results of a double crossover event and that became Em-sensitive. The presence of the correct, clean gene deletions was confirmed by PCR using specific primers (Table S1B).

Protein overexpression was verified with Sodium Dodecyl Sulfate—(10%) Polyacrylamide Gel Electrophoresis (SDS-PAGE) according to NuPAGE technical guide (Invitrogen, Carlsbad, CA) in cell extracts and supernatant fractions of exponentially growing cells (OD₆₀₀ = 0.42–0.45) grown for 5h after addition of 10 ng/ml nisin.

Cell extracts were obtained by collecting the cell pellet from 10 ml culture (2,927 g for 5 min) and protein in supernatants was precipitated using trichloroacetic acid (TCA). The cell pellet was re-suspended in sterile demineralized water to a final OD₆₀₀ of 5 and transferred to microfuge tubes containing 1 g of zirconium beads and cooled on ice for 5 min. This was followed by 3 × 30 s of bead beating with a FastPrep FP100 bead-beater (Qbiogene, Cedex, France) and 1 min off cooling on ice in between the three cycles. After this treatment the tubes were left on ice for 5 min to allow the beads to sink and the supernatant was transferred to sterile Eppendorf tubes and kept at −20°C.

TCA precipitation was conducted by adding 2.5 ml 100% trichloracetic acid to 10 ml of culture supernatant, followed by vigorous mixing and incubation for 30 min on ice. Subsequently the tubes were centrifuged at 2,927 g at 4°C for 30 min, the supernatant was discarded carefully and the pellet was dissolved in 0.5 ml cold acetone. This was followed by centrifugation at the 2,081 g at 4°C for 15 min, acetone was discarded; pellet was dried on air and r-suspended in sterile reverse osmosis water.

As 30 out of 55 strains used in this study were isolated from a dairy environment, we decided to examine the affinity of L. lactis to different milk proteins. For this purpose, we developed a protocol that allowed quantifying the attachment of proteins to microbial cells. Bacterial cells were stained with the fluorescent DNA stain Syto 9, while proteins were colored with the fluorescent dye Nile Blue A. The emission spectra of these two dyes show little overlap and the individually stained particles could thus be distinguished by flow cytometry (Figure 1A). Cells give a high fluorescent emission signal at 530 nm while the proteins emit the highest signal at 680 nm. When Nile Blue A-stained protein binds to a Syto 9-labeled bacterial cell, the resulting particle should show high fluorescence at both wavelengths while this should not be the case when the two do not interact. This basic concept was tested using L. lactis TIFN5 (Figure 1B). We performed two additional experiments to validate the methodology with independent techniques. In the first experiment, Syto-9 labeled cells were mixed with stained proteins. The strain used was predicted to give limited to no binding to the protein based on flow cytometry results. Using the flow cytometer we sorted 10,000 events from the population identified as protein based on the described method. The plating of these proteins showed that ~5% of the sorted events led to the formation of a colony. On the other hand, sorting of events from the population identified as cells resulted in ~95% of the cases leading to growth of a colony. These results were consistent with what was expected with an acceptable error margin of ~5%. In the second validation experiment, cells that were identified by flow cytometry as either binding or not binding to milk proteins were incubated with the protein of interest and subsequently visualized by light microscopy (Figure 2). Clear differences in Nile Blue A fluorescence intensity indicate different levels of protein binding to the cells. The level of protein binding to a cell, as detected by fluorescence microscopy at 680 nm (Figure 2), is in agreement with the protein binding levels observed with flow cytometry, which decreases from strain HP to SK11, KW10 and P7266 respectively (Figure S2, Table S2).

Sodium caseinate and sodium para-caseinate were selected as the proteins to be studied, as they represent the major milk proteins, i.e., the caseins. In sodium para-caseinate, the C-terminal caseinomacropeptide (CMP) of κ-casein is removed by enzymatic hydrolysis, as in cheese-making. Sodium caseinate was used in two forms: either or not pre-treated for 10 min at 90°C. The heat treatment was performed to account for possible heat-induced changes in the proteins, which is a common processing step when manufacturing fermented milk products (Hashizume and Sato, 1988; Lucey and Singh, 1997).

Using this approach, 55 strains of L. lactis were screened for their ability to bind the three milk proteins. The strains bound all proteins to some extent, but clear differences were observed (Table S2, Figure 3, Figures S1–S3). Significant differences in growth phase-dependent protein binding were seen for 12 of the 55 strains (Figure S4). While the ability of L. lactis to bind proteins is strain-specific the results also show that the capacity of binding milk proteins appeared to be larger than 61% with the average between 82 and 97% for strains of dairy origin. These mostly belong to the L. lactis ssp. cremoris and ssp. lactis biovar. diacetylactis. The 10 out of 23 L. lactis strains of plant origin, which belong to the ssp. lactis (indicated as dark and light green in the “origin” and “species” column in Figure 3, respectively), showed either poor (<50%) or no protein binding for both growth phases (Figure 3, Figures S1–S3, Table S2).

Cell surface charge, hydrophobicity, and emulsion stabilizing ability of the individual strains were determined on exponentially growing and stationary cells (Table S2). The results show a considerable phenotypic diversity (Figure 3, Figures S1–S3, S5–S7). The cell surface charge—measured as zeta potential—varied from −3.4 to −42.3 mV between the different strains. CSH is the measure of the extent to which cells suspended in a water phase are attracted to a hydrocarbon phase when both phases are mixed vigorously and left for phase separation to occur. We found that 25 out of the 55 strains have a CSH ranging from 0 to 20% for both growth phases. On the contrary, 9 strains showed >95% hydrophobicity for stationary growth phase, while another 5 strains showed such a high hydrophobicity in the exponential growth phase. In stationary growth phase strains NCDO712 and MG1299 displayed 60–99% hydrophobicity while the plasmid cured derivatives MG1063, MG1261, MG1362. MG1363, MG1365, and NZ9000 showed a hydrophobicity of 5–35% (Figures S5, S8, Table S2). This indicates that hydrophobicity in NCDO712 might be a plasmid-encoded trait.

Hierarchical cluster analysis revealed no correlations between the CSH of the strains as determined with different hydrocarbons and their binding of proteins (Figure 3). Strains in clade 1 show low surface hydrophobicity in the exponential growth phase, but high binding affinity to milk protein is seen. This clade consists mainly of L. lactis ssp cremoris strains of dairy origin. In contrast, strains in clade 2 poorly bind to milk proteins while they have a high CSH. Clade 3-strains show poor protein binding capacity and a low surface hydrophobicity. Clade 2 and 3 consist mainly of L. lactis ssp. lactis strains originating from the plant environment. For cells from stationary growth phase some differences are seen but the overall trends are the same in the clades 1, 3, and 4 (Figure 3B). Together, the results demonstrate that surface hydrophobicity of L. lactis cells and their protein binding capacity are independent parameters. Interestingly, the analysis of the origin of the strain (plant or dairy) and species of L. lactis revealed that the majority of L. lactis ssp. cremoris and L. lactis ssp. lactis biovar. diacetylactis, both of dairy origin, have a high capacity to bind to milk proteins. In contrast, the majority of strains originating from plan material only poorly bind milk proteins (Figure 3).

The stability of the emulsions obtained by mixing the water and hydrocarbon phases varied between the strains from no observed emulsification to total hydrocarbon phase emulsification and stability for at least 24 h. No correlation was found between emulsion stabilizing capacity of the strains and their cell hydrophobicity, between their charge and hydrophobicity, and between their charge/hydrophobicity and capacity to bind proteins.

A comparison of stationary-phase and exponential-phase cells revealed that 12 of the 55 strains show significant growth phase-related differences in their capacity to bind the milk protein samples tested, 3 strains show significant changes in hydrophobicity and 9 of the 55 strains show differences in the E24 measured (Figures S4, S8). The charge of stationary and exponentially growing cells differed significantly for 22 out of 55 strains tested (Figure S7). Taken together, the results indicate that the measured cell surface properties have some growth phase dependency, but strain specific properties are much more determinant. The binding of different milk proteins (regression coefficient r = 0.88–0.93) and cell hydrophobicity measured with different hydrocarbons (r = 0.98) are in relatively good agreement. However, little to no correlation is observed between hydrophobicity, cell surface charge and protein binding. Interestingly, there is a clear overrepresentation of strains of dairy origin in the clusters that show high binding ability of milk proteins which suggests that this trait might be beneficial during evolution in a dairy environment.

A random forest-based genotype-phenotype matching algorithm (Bayjanov et al., 2013) was employed to identify genes potentially involved in cell surface properties. The analysis was run separately for each individual phenotype measured, resulting in candidate lists for genes involved in the individual traits. This resulted initially in 201 candidate genes which were selected based on the highest importance score and gene description. From these 201 candidate genes, 18 were selected for further characterization on the basis of three parameters: (i) the importance score in the individual GTM analyses; (ii) multiple appearances in the GTM analysis for the individual phenotypes; and (iii) the predicted gene function being related to cell surface (Table S3). The following choices were made for the further characterization of these genes. If the presence of a selected candidate gene was found to be associated with a phenotype and this gene is absent in L. lactis MG1363, it was overexpressed in MG1363. Conversely, when gene absence was found to be associated with a phenotype and this gene is present in MG1363, it was knocked-out in MG1363. We selected the longest gene within the orthologous group of the 55 strains for overexpression purposes, to eliminate the risk of working with a truncated protein. Eight of the 18 genes could not be deleted and/or overexpressed, possibly because they are essential or deleterious. Ultimately, 6 of the selected candidate genes were successfully cloned downstream of the nisin-inducible promoter P_nisA in pNZ8150 (Table 1) in L. lactis NZ9000, an L. lactis MG1363 derivative, while 4 of the genes were deleted from the chromosome of L. lactis MG1363 (Table 2). While we cannot exclude that the addition of nisin itself has an effect on surface properties we would like to point out that the over-expression results reported here are in relation to the nisin induced control strain which carried an empty plasmid vector. In addition no significant differences in surface properties are seen between the uninduced L. lactis MG1363 and its nisin induced derivative NZ9000 harboring pNZ8150 (Tables S2, S4, Table 1). Over-expression of proteins of the predicted sizes, and their cellular localization were examined with SDS-PAGE after induction with 10 ng/ml nisin of exponentially growing cells carrying the expression plasmids and incubation for another 5 h (Figure S9). The results showed that all 6 proteins could be successfully overexpressed, as they were detected either intracellularly or in the medium.

The deletion of the gene llmg_1383 decreased the binding of exponentially growing cells of MG1363 to the 3 milk proteins tested by 53 ± 3%. The protein Llmg_1383 of L. lactis MG1363 is annotated as a conjugal transfer protein TraG, which is described to aid in the transfer of DNA during bacterial conjugation. As it is predicted to be involved in membrane pore formation (type IV secretion; Schroder et al., 2002), a role in surface alteration seems plausible.

Similar to llmg_1383, the deletion of the hypothetical gene llmg_1093 also decreased the binding of L. lactis MG1363 to the three milk proteins examined by 60 ± 27% for cells in the exponential growth phase and by 42 ± 37% for cells in the stationary growth phase. Protein Llmg_1093 is a putative secreted protein but its actual function is not known.

As several of the genes studied here harbor sortase dependent LPXTG signals we also deleted the sortase A gene (llmg_1449) from the chromosome of strain L. lactis MG1363. This deletion did not change the cell surface properties of L. lactis MG1363 (Table S4), indicating that SrtA might not be involved directly in cell surface properties. We also tried to obtain a knock-out mutant of the srtC gene but were not able to obtain it after two attempts, indicating a possibly essential role or SrtC for growth.

The cell surface protein precursor B40_0084 is a putative mucus binding protein as it has 4 mucus binding domains (MucBP) (pfam06458) and a LPXTG-motif cell wall anchor domain (TIGR01167). GTM predicted that the presence of this protein leads to a more negative cell surface charge. This prediction was not confirmed, but changes were observed for hydrophobicity. For example, for stationary phase cells where the control strain shows a CSH of 8.8% (±5%), the overexpression of B40_0084 resulted in a CSH of 89% (±7%) (Table S4). Interestingly, while SDS PAGE analysis (Figure S9) verified the overexpression of B40_0084 in exponentially growing cells, this overexpression did not affect the CSH of the cells in this growth phase. Mucus binding proteins are described to be involved in the binding of carbohydrates such as mannose (Pretzer et al., 2005) and they are speculated to be important for probiotic function (Kleerebezem et al., 2010). The effect of the overexpression of B40_0084 only leads to hydrophobicity changes in stationary phase cells, indicating that other surface decoration(s) dominates surface properties in a growth phase-dependent manner.

Another gene identified by gene-trait matching is a putative internalin containing four MucBP domains (pfam06458) and a surface-anchoring domain (COG4932). Internalins were originally described in Listeria monocytogenes as surface proteins that are involved in adhesion to mammalian epithelial cells (Lecuit et al., 1997). For example, protein internalin A (InlA) mediates bacterial adhesion and invasion of epithelial cells in the human intestine through specific interaction with its host cell receptor E-cadherin (Schubert et al., 2002). Here, we overexpressed the ortholog from strain KF282 (KF282_0409). While gene-trait matching associated the presence this gene with lower surface charge, its overexpression in MG1363 increased the CSH, from 8.8% (±5) in the control strain to 49.3% (±21.8). Furthermore, overexpression of a cell wall surface anchor family protein (LLKF_0311) led to the decreased attachment of cells to milk proteins from 98% (±0.7) to about 31% (±47%) in the stationary growth phase and to the increase in CSH from 9 ± 5 to 78 ± 9% in the stationary and exponential growth phases (Table S4) while initially the presence of LLKF_0311 was predicted to effect cells surface hydrophobicity and emulsion stability in exponential phase of growth. (Table S4).

The endo-beta-N-acetylglucosaminidase ypcCD (LLKF_1605) was found to be associated to the binding of cells to milk proteins. Overexpression of this protein did not lead to an alteration in attachment of cells in stationary growth phase to milk proteins, but led to an increase in surface hydrophobicity from about 8.8 ± 5 to 23.7 ± 16.1%.

Overall, the phenotype predicted by gene-trait matching could be confirmed experimentally for 3 out of 10 engineered strains, and for 2 out of 3 strains additional altered surface properties were detected. Overexpression of the selected candidate genes did not influence the protein binding properties. A total of 4 strains showed altered surface properties but not the predicted ones whereas for 3 strains no changes were observed. While not all predictions were correct, the approach did allow identifying targets that are of importance for lactococcal surface properties.

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.