High-throughput screening for texturing Lactococcus strains

Abstract In the food industry, lactic acid bacteria (LAB) are used in dairy fermentations, extending the shelf life by lowering the pH and also affecting taste and texture of the fermented milk. The texture of fermented milk is an important quality parameter, affecting consumer acceptance. Finding LAB providing desired texture of a product is time consuming and laborious when using standard methods for measuring texture, e.g. rheology measurements. Screening of 986 Lactococcus lactis strains resulted in few strains with the ability to enhance texture, demonstrating the necessity of implementation of high-throughput screening methods. A high-throughput screening assay was developed, combining small-scale 96-well microtiter plates and pressure measurements during liquid handling, e.g. aspiration, to find strains that give good texture in fermented milk. Only about 1% of the strains were found to enhance milk texture. Two of the texturing strains belong to L. lactis subsp. lactis, which are the first texturing strains from this subsp. reported. Mining for eps gene clusters responsible for exocellular polysaccharide production was performed, as polysaccharide production can contribute positively to fermented milk texture. Comparative genomics approach revealed four types of texturing L. lactis strains with diverse eps gene clusters.


INTRODUCTION
Conversion of fresh milk to fermented milk by lactic acid bacteria (LAB) is used to extend the shelf life of milk due to acidification, to provide taste as well as texture. The ability of LAB to produce polysaccharides is associated with improvement of the texturing properties and increased health benefits of fermented products (Caggianiello, Kleerebezem and Spano 2016). Polysaccharides from food-grade LAB are non-toxic, biodegradable, environment friendly and act as natural viscosifiers, emulsifiers, stabilisers, binders, gelling agents, coagulants and suspending agents in food industry as well as in cosmetics (Jindal and Singh Khattar 2018). For this reason, screening and selection of polysaccharide-producing LAB is of importance to both academia and industry.
Lactococcus lactis is used to produce numerous fermented dairy products including cheese and mesophilic fermented milk, such as buttermilk and sour cream. Polysaccharide-producing strains are of great interest for these applications, as polysaccharides released into the medium can result in improved texturing properties of buttermilk and sour cream, while capsular polysaccharides can result in improved water-holding capacity and thus improved yields of cheese.
Genes encoding Wzy-dependent exocellular polysaccharide biosynthesis proteins in LAB are typically organised in a cluster with an operon structure. The major genera of LAB used in food processing (Lactococcus, Streptococcus, Lactobacillus, Leuconostoc, Oenococcus and Pediococcus) possess eps gene clusters (Zeidan et al. 2017). Generally, eps gene clusters are highly diverse and their nucleotide sequences are among the most variable sequences in LAB genomes. Also, polysaccharides structures are very diverse as a result of variations in monosaccharide building blocks, the presence of branches, decoration with noncarbohydrate moieties and linkages: for instance, two glucose residues can be joined together in 30 different ways (Laine 1994, Zeidan et al. 2017. The type and size of polysaccharides and their interaction with milk proteins are the determining factors for texture development (Hassan 2008;Mende, Rohm and Jaros 2016;Birch et al. 2017). Even though the potential to synthesise polysaccharides is encoded within the genomes of many LAB, production of polysaccharides and their functional properties need to be evaluated for successful industrial applications.
A great variety of different techniques to identify polysaccharide-producing microorganisms have been devised. Screening strategies can be generally divided into screening for polysaccharide-producing strains or phenotypes associated with these polysaccharides. They include visual inspection of slimy or mucoid colonies on solidified media, observation of viscosity in culture broth (e.g. measuring ropiness with a pipette or by microhaematocrit capillaries), staining methods (aniline blue, ruthenium red, neutral red, Calcofluor white, Congo red, Indian ink, lectins), carbohydrate determination (quantitative HPLC analysis, total sugar determination using phenol-sulphuric acid method), cell sedimentation in semi-liquid agar medium, measurement of the change in the electrical conductivity in the growth medium during polysaccharide production (impedance microbiology), precipitation with different alcohols as a common detection and isolation method, electron microscopy and FITC-dextran exclusion assay for visualisation of the capsule production (Llull et al. 2005;Hathaway et al. 2012;Rühmann, Schmid and Sieber 2015;Zeidan et al. 2017;Bancalari et al. 2018).
The texture of fermented milk is dependent on both the bacteria used for fermentation and process parameters. Polysaccharide-producing bacteria can positively influence product characteristics such as texture and sensory properties (Mende, Rohm and Jaros 2016). Rheometer and texture analyser are typically used to assess texturing properties of milk gels, such as shear stress. Sensory textural attributes are often correlated with the results from instrumental tests, e.g. shear stress is related to viscosity and perceived mouth thickness (Folkenberg et al. 2006). Standard texture measurements require relatively large sample volumes, e.g. 100 ml. Although these methods of measurement are accurate and reproducible, they are highly demanding in respect of the time required per sample, technical skills and precision. The state of the art equipment for measuring viscosity takes about 20 min per sample, and only one sample at a time can be tested. The washing and changing step between each sample makes it difficult to be used in a high-throughput screening method. There is a need for high-throughput screening methods that may be used for relevant food matrices such as acidified milk.
The aim of this work was to develop a method for highthroughput texture screening of fermented milk samples and screen a collection of 986 Lactococcus strains for texturing capabilities. Moreover, we wanted to examine if any of the texturing strains found had novel eps gene clusters, compared to what has previously been described in the literature.

High-throughput screening for texturing strains
A total of 986 L. lactis strains originating from the Chr. Hansen culture collection were screened for ability to enhance texture in fermented milk. The strains were inoculated in 96 lowwell microtiter plates in 200 μl M17 broth (Terzaghi and Sandine 1975) containing 1% glucose and 1% lactose as C-source and incubated overnight at 30 • C. A volume of 10 μl was transferred to 990 μl B-milk containing pH colour indicator ±0.2% yeast extract in 96 deep-well plates. B-milk was prepared by reconstituting low fat skim milk powder to a level of dry matter of 9.5% and pasteurised at 99 • C for 30 min, followed by cooling to 30 • C. The pH colour indicator milk was prepared in the following way: 50 mg bromocresol purple salt and 50 mg bromocresol green salt (both from Sigma Aldrich, St. Louis, Missouri, United States) were dissolved in a final volume of 40 ml dH 2 O, pH was adjusted to 7.0 with NaOH and the final volume was adjusted to 50 ml with dH 2 O. The pH indicator was sterile filtered (0.2 μm) and 5 ml was added to 95 ml B-milk. The inoculated pH colour indicator milk samples were incubated for 18-20 h on top of flat-bed scanners (HP ScanJet G4010) with temperature-controlled hoods set at 30 • C ( Fig. 1). pH-dependent changes in colour were recorded every 6 min, using pH Multiscan software (v.5.1, HNH Consult Aps, 9530 Støvring, Denmark). After 18-h static incubation at 30 • C, most samples in the plate had a pH of 4.3-4.5, where the fresh milk was converted to fermented milk gel. The plates were kept at 4 • C overnight, and TADM (Total Aspiration Dispense Monitoring) pressure curves were obtained by aspirating the samples using Hamilton liquid robot.
A Hamilton MicroLab Star liquid handling device (Hamilton, Bonaduz, Switzerland) was used to collect pressure versus time data using TADM software of the Hamilton (Camenisch 2001). Aspiration pressure curves were used to identify samples with elevated texture. A volume of 500 μl was aspirated (350 μl/s) using wide-bore tips (Hamilton Robotics). Pressure versus time data (TADM) were expressed as a single number by either recording pressure at a particular time point (e.g. 1 s) or as TADM curve area.
Three different tools for carbohydrate active enzyme annotation in dbCAN2 (http://cys.bios.niu.edu/dbCAN2) were combined to classify glucosyltransferases: HMMER search against the db-CAN HMM (hidden Markov model) database, DIAMOND search against the CAZy pre-annotated CAZyme sequence database and Hotpep search against the conserved CAZyme short peptide database (Zhang, Yohe and Huang 2018).

Rheology measurements
Shear stress data were obtained by inoculating strains in semifat milk (1.5% fat) enriched with 2% skim milk powder. Milk was heated at 90 • C for 20 min and cooled down to the inoculation temperature, prior to inoculation with 1% overnight microbial culture. The inoculation took place for 7-20 h at 30 • C in 200-ml scale until pH = 4.55, as monitored by pH electrodes, followed by cooling to 4 • C and storage for 5 days at 4 • C. After storage, the fermented milk was manually stirred gently by means of a stick fitted with a bored disc until homogeneity of the sample. Shear stress of the samples was assessed on a rheometer (Anton Paar Physica Rheometer with ASC, Automatic Sample Changer, Anton Paar R GmbH, Austria). For the data analysis, the shear stress at shear rate 300 s −1 was chosen. Rheology measurements were made on duplicate biological samples.

Genome data used for eps gene cluster mining
Mining for eps gene clusters was performed using BLAST analysis of the conserved part of the lactococcal eps gene clusters, against the proprietary genomes and genomes available on the NCBI website. Proprietary genomes were obtained by the purification of total DNA from a culture grown overnight at 30 • C in M17 medium containing 1% lactose and 1% glucose using DNeasy Blood and Tissue Kit (Qiagen). The DNA quality was checked using gel electrophoresis, the DNA concentration was estimated using Nanodrop 2000 spectrophotometer and 15 μg DNA (approximately 150 ng/μl) was used for sequencing at BGI (HongKong, China) using Illumina HiSeq equipment with pairend channel module with 2 × 100 bp read length and an insert size of 500 bp. Assembly of raw data into contigs was performed using CLC Genomics workbench 7.0 software.
Easyfig software (version 2.1) was used for comparative eps gene cluster visualisation (Sullivan, Petty and Beatson 2011). In order to compare the most related eps gene clusters between each other in Easyfig, the relatedness of the eps gene clusters on nucleotide level was assessed using 'Create Tree' tool within 'Alignments and Trees' set of tools of the CLC Main Workbench 7 software (tree construction method: neighbour joining; nucleotide distance measure: Jukes-Cantor; bootstrapping was performed using 200 replicates).
Eps gene cluster sequences obtained from proprietary strains were deposited in GenBank under the following accession Aspiration pressure values at 1 s (Pa, mean and its standard error, two to eight biological replicates in one to two independent experiments) measured by pipetting using Hamilton liquid handling unit plotted against shear stress (Pa, mean and its standard error, two to four biological replicates in one to two independent experiments) at shear rate 300 s −1 values measured using rheometer for selected milk gel samples obtained by fermenting milk using L. lactis strains as in Fig. 1
To confirm acidification and texturing abilities of selected strains in a larger scale, the strains were inoculated in 200 ml milk until pH = 4.5, followed by rheometer measurements. The acidification of milk in 200-ml scale was followed using pH electrodes; the acidification curves resembled those obtained in 1ml scale for the strains under investigation (data not shown).
TADM data obtained from 1-ml scale milk acidification were plotted against shear stress values for the same strains from 200-ml scale milk acidification experiment, where the correlation of R 2 = 0.82 between TADM data and shear stress measurements was observed (Fig. 2). Shear stress measurements were used to differentiate between texturing and non-texturing strains. Strains resulting in shear stress values below 50 Pa were considered non-texturing, while those above 50 Pa texturing, based on sensory analysis of food-grade fermented milk samples (data not shown). Time to pH 4.5 was between 9.0 and 13.0 h in the case of non-texturing, and 12-20.0 h in the case of texturing strains. Most of the texturing strains belong to L. lactis subsp. cremoris. To the best of our knowledge, Lll3 is the first texturing strain belonging to L. lactis subsp. lactis reported (Poulsen, Oeregaard and Derkx 2015). Lll3 and Lll8 are the only texturing strains belonging to the subsp. lactis reported so far.
Using the 96 channel pipetting head of the Hamilton robot, 96 samples were examined in a few seconds, while this would take ∼32 h non-stop using a conventional rheometer. The highthroughput method for texture screening in fermented milk can be used to screen thousands of samples within a few hours.

Comparative genomics of polysaccharide biosynthesis in L. lactis
Since an enhanced texture is associated with the production of polysaccharides, mining for eps gene clusters was performed. The eps gene clusters are generally chromosomal in L. lactis subsp. lactis but can reside on a plasmid in L. lactis subsp. cremoris (Table 1).
A large diversity of genes encoding the synthesis of exocellular polysaccharides via the Wzy-dependent pathway were found in the genomes of L. lactis (Fig. 3). Here, we used the nomenclature suggested by Zeidan et al. (2017). The conserved genes in the beginning of the eps gene cluster were denominated epsRXCDB, and those at the end, epsL and lytR, while the polymerase was named wzy, and the flippase, wzx (Fig. 3). The organisation of the genes in the L. lactis eps clusters is similar to other LAB, but there are some distinguishing features. LytR, which is likely involved in the attachment of polysaccharides to the cell wall in LAB (Zeidan et al. 2017), is the first gene of the cluster in most LAB (usually denominated epsA or cpsA) and in Streptococcus pneumoniae, while in L. lactis it is situated at the end of the eps gene clusters (Fig. 3). The first gene of the lactococcal eps gene cluster, epsR, is a regulator belonging to XRE family of proteins, which is not present in the eps gene clusters of other LAB (Zeidan et al. 2017), indicating that L. lactis has a different way or an additional mechanism of polysaccharide biosynthesis regulation compared to other LAB.
Genes located at the 5' end of the eps gene cluster epsRXCDB, which are involved in the modulation and assembly machinery of polysaccharide biosynthesis, as well as epsL and lytR at the 3' end, displayed the highest level of conservation. The genes of the variable part, including polymerase wzy, flippase wzx and glucosyltransferases (GT) or other polymer-modifying enzymes, were rarely similar between the strains (Fig. 3), in agreement with what is observed in other organisms (Bentley et al. 2006;Zeidan et al. 2017). Based on CAZy (database of Carbohydrate-Active enZYmes) classification, one third of GT found in the eps gene clusters of the selected strains belong to GT2, one third to GT4 and one third to other CAZy groups (Fig. 3).
Among the non-texturing lactococcal strains from Fig. 2, Llc7 and Lll9 did not contain an eps gene cluster. Strains Llc1, Lll1, Lll2, Lll4, Lll5, Lll6 and Lll7 all contained an eps gene cluster. Three (Lll1, Lll2, Lll5) out of seven eps gene cluster positive but non-texturing strains contained presumably non-functional eps gene clusters. Lll1 had neither a putative polymerase wzy nor a flippase wzx; Lll2 was lacking both a polymerase wzy and GT; Lll5 had no polymerase wzy (Fig. 3). The four remaining non-texturing strains contained presumably functional eps gene clusters. However, it is not known if all the genes are functional. Even if strains can produce polysaccharides, they may not lead to texture in fermented milk. Because the presence of eps genes per se does not indicate whether a strain will contribute with texture in a particular food matrix, it is of importance to have a screening method that allows to detect strains with the desired texturing phenotype. This is achieved with the TADM screening method.
Based on eps gene cluster similarities, the eight texturing strains found in this study seem to fall into four different groups. Llc2, Llc3, Llc4, Llc5 and Lll8 all resemble NIZO B40, whereas strains Lll3, Llc6 and Ropy352 all have unique eps gene clusters (Fig. 3). All eps gene clusters have high similarity within the conserved regions (epsRXCDBE, epsL and lytR), but the remaining part of the eps gene clusters, including wzy (polymerase), wzx (flippase) and the GT genes were indeed variable, both in terms of sequence and in terms of number of genes present (Fig. 3). The eps gene clusters of the strains from the four groups (NIZO B40 vs Lll3 vs Llc6 vs Ropy352) were 61%-78% identical, based on the nucleotide sequences of the complete eps gene clusters, while their identity in the variable part was between 48 and 59%. Lll3 has three GT, while the NIZO B40-like strains have four GT and Ropy352 and Llc6 have five GT (Fig. 3). We speculate this may lead to polysaccharides with different repeating units. The common denominator for the texturing strains is that they all contain the genes required for the polysaccharide production, e.g. epsCDBEwzy-wzx and GT (Zeidan et al. 2017). Several texturing strains from the subsp. cremoris have been reported, e.g. NIZO B40, SMQ-461, Ropy352, JFR1 (van Kranenburg et al. 1997;Dabour and LaPointe 2005;Knoshaug, Ahlgren and Trempy 2007;Ayala-Hernández et al. 2008). We report the first texturing strains from the subsp. lactis, Lll3 and Lll8; the species identity of both strains was confirmed using 16S. Lll3 has a unique eps gene cluster including 14 open reading frames covering 13 kb, which contains three novel GT. All three predicted GT gene products of Lll3 showed low amino acid similarity (up to 51% identity) with known GT. The three GT together with a putative nucleotide sugar dehydrogenase are potentially involved in sequential building of the repeating unit, although their specific functions and therefore order of action have not been demonstrated. Lll8 seems to have acquired the eps gene cluster plasmid typically found in the subsp. cremoris such as NIZO B40 strain, but also in several proprietary strains from the subsp. cremoris, where the eps gene clusters differ from each other by several nucleotides.
More research is needed to link eps genes directly with functional properties of fermented milk, such as texture. A study of the structure, molecular weight and amount of polysaccharides resulting from the eps gene clusters is necessary to establish such a link. A combination of two activities, linking polysaccharide gene clusters with polysaccharide structures, e.g. as it was done with S. pneumoniae capsular polysaccharides (Aanensen et al. 2007), and linking polysaccharide structures with their functionality in the desired matrix, e.g. milk, might reveal the direct link between polysaccharide genes and functionality of the polysaccharides. This might enable prediction of the texturing properties of the strains based on their genome sequences.

CONCLUSION
The present work describes a high-throughput screening method for texturing strains, which requires only small sample volume, e.g. 500 μl, and is significantly faster compared to the standard texture measurement by a rheometer. The method is useful when it comes to screening for microorganisms that are suitable for obtaining specific texturing properties of the milk gel. It also provides a tool for determining the impact of milk additives on their ability to impact shear stress of milk gels. The TADM-based high-throughput method allowed us to rapidly screen for novel texturing lactococcal strains suitable for use in preparation of mesophilic food products.
Comparative genomics of polysaccharide gene clusters of publicly available and proprietary texturing and non-texturing L. lactis strains revealed four different groups of texturing strains, based on their eps gene clusters. The two strains Lll3 and Lll8  Zeidan et al. (2017). These include modulatory genes (yellow; phosphoregulatory module epsCDB), polysaccharide assembly machinery genes (green; initiation epsE, polymerisation wzy, export/flippase wzx and attachment lytR), genes encoding GT (orange; glucosyl transferases) necessary for the assembly of the repeating units, and genes encoding non-housekeeping functions (pink) required for the synthesis of activated sugar precursors and modification of the sugar residues. The functions of the three genes typically only present in the lactococcal eps gene clusters, epsR, epsX and epsL, remain to be elucidated. Eps gene cluster regions with 65%-100% identity are indicated with grey connection bars. Mobile genetic elements are marked with stars. Genes with unknown functions or functions that might not be related to the polysaccharide production, e.g. mobile genetic elements, are in light grey. Glucosyltransferases were marked with GT group numbers according to CAZy (database of Carbohydrate-Active enZYmes) classification, e.g. 2 means GT2 group. Abbreviations: GT, glycosyltransferase; wzy, polymerase; wzx, flippase; NDP-sugar, nucleotide diphospho-sugar.
are the first L. lactis subsp. lactis reported to enhance texture in fermented milk. Strain Lll3 has a novel eps gene cluster, whereas strain Lll8 has an eps gene cluster similar to that in NIZO B40. Texturing strains from the four groups are presumably producing different types of polysaccharides, as they have variable GT genes along with different wzy and wzx genes.