Genomics accelerates food innovation

Published: 09-07-2014

Gene-trait matching enables NIZO food research to predict functional properties of starter and probiotic cultures and to optimize fermentation conditions.

Fermented foods constitute a large part of the human diet. Examples are products like cheese, yoghurt, bread, wine, beer and soy sauce. In order to obtain specific product characteristics, microorganisms such as bacteria, yeasts and molds are added to milk, fruits, cereals, soy and other food substrates. Well-known examples of bacteria used in this way are Lactococcus lactis, Lactobacillus plantarum and Streptococcus thermophilus. Enzymes of these microorganisms, particularly amylases, proteases, and lipases, hydrolyze polysaccharides, proteins, and lipids into nontoxic products with flavors, aromas, and textures that consumers find appealing. Modern genomics technologies offer unprecedented options for optimization of fermentation conditions. NIZO food research, an independent contract research company working on tasty, healthy, sustainable, affordable and safe foods, is at the forefront of this development.


Predicting functional properties
Genomics has matured into a powerful technique for analyzing microbial activity. Next-generation sequencing techniques have made it possible to determine the genetics of microorganisms quickly and relatively cheaply. Much information about the genetic code of industrially employed microorganisms is currently becoming available.

The datasets obtained through “omics” technologies offer interesting strategies for the delivery of tasty, safe and healthy food products. NIZO food research has developed various analytical and predictive bioinformatic tools that can help food companies optimize their fermentation processes. One of these techniques is referred to as genotype-phenotype-matching analysis, or gene-trait matching. Gene-trait matching makes it possible to predict the behavior and functionalities of a microorganism in a given fermentation, based on the microorganism’s genetic code. “Application of our tools can optimize fermentation processes in various ways, for example with the respect to the growth of a bacterial strain on certain sugars, for instance, or related to production of specific compounds such as flavors and vitamins,” said Dr Sacha van Hijum, principal scientist bioinformatics at NIZO food research and group leader of the Radboud umc bacterial (meta) genomics group. Gene-trait matching may also lead to identification of factors contributing to quality and shelf-life survival of the food product in a knowledgebased manner. If research data on the robustness/resistance of spoilage strains to cleaning protocols (high temperatures, types of detergents) can be associated with specific genes responsible for this resistance, this could lead to new ways of combating these spoilage organisms. Also, if spoilage of a specific product could be linked to one or several genes (functionalities), and an anti-microbial treatment can be devised based on this knowledge, adjusting the product composition in a certain way might reduce the chance of spoilage considerably.

Transcriptome-trait matching (TTM, i.e. associating gene-expression profiles to bacterial traits) is a logical and useful variant of gene-trait matching, since some complex phenotypes (traits) correspond to differences in the level of gene expression, rather than to the presences or absence of (a set of) genes. Another strategy is to use gene-metabolite matching (GMM) which involves coupling the presence of metabolites to the presence of specific microbial genes.

Gene-trait matching: the principle
The principle of gene-trait matching is depicted in the figure below. In this example eight bacterial strains are pheno-typically tested for their ability to grow (1= yes, 0= no) on sucrose. The genetic content describes the presence (1) or absence (0) of several genes in these bacterial strains. In this example, the presence of gene 3 is associated with growth on sucrose. These results indicate that if another bacterial strain (other than these eight strains) has gene 3 it will grow on sucrose.

In this example, the relationship between growth on sucrose and the presence of gene 3 is easily discovered visually or by using statistical tests. However, since most studies identifying gene-trait matches are carried out with more than eight strains (in order to provide appropriate statistical power, the number of bacterial strains should preferably be in the order of dozens), and as each characteristic (functionality) of a microorganism can be explained by the combined presence of multiple genes in its genome sequence, retrieving these types of conditional relations can be complex. Huge datasets with genotypic and phenotypic information of a large number of strains may be involved. Yet another complicating factor is that gene-trait matching is not only about the presence or absence of genes. Much smaller differences such as a few mutations in a gene sequence –for example Single Nucleotide Polymorphisms (SNPs)– may lead to different gene products and possibly different functionalities of strains as well. Additionally, a starter culture may consist of multiple strains that are all involved in the fermentation process. Hence, bioinformatics is an essential discipline in this type of work. Predictive modeling starts with the genomic information of a certain microorganism. Then, large-scale integration with genomic information in publicly accessible and in-house databases, as well as large-scale integration with information on genetic content and microbial characteristics in the scientific literature (i.e. text mining) occurs. In the end, this results in predictions for industrially relevant characteristics of microorganisms. One of the bioinformatics tools developed by Van Hijum and co-workers is a web-based iterative gene-selection procedure called Phenolink1. This tool can handle large and diverse data sets and is robust to noise (i.e. aberrations that occur frequently in data sets). It links phenotypes to “omics” data and finds relationships between specific (high throughput-based) measurements and strain phenotypes (traits). With Phenolink and other bioinformatics techniques and approaches, in silico prediction for fermentation optimization has become a reality.

“In order to properly carry out all this work, it is indispensable to have in-depth knowledge of molecular biology and microbial physiology as well as bioinformatics,” which Van Hijum emphasizes as a strength of NIZO food research. He continues to explain: “The genetrait matching associations are selected on the basis of their statistical significance during the predictive modeling process. However, some associations might make more biological sense than others.”


Gene-trait matching: powerful examples
While for some bacterial traits additional work needs to be carried out to validate gene-trait associations in the lab, Van Hijum illustrates the power of gene-trait matching and the benefits that it already brings for today’s food industry.

Survival upon spray drying
Industry may even benefit from gene-trait matching before the actual fermentation process starts. Robustness towards heat and oxidative stress, for example, may differ considerably between individual strains. A NIZO food research study sponsored by the Kluyver Center for Genomics of Industrial Fermentations showed that robust strains of Lactococcus lactis survive heat or oxidative stress conditions up to 4 log units better than the sensitive strains2. When the survival upon spray drying was determined for these Lactococcus strains, it was clearly demonstrated that the high and low levels of survival upon spray drying correlated clearly with the combined robustness under heat and oxidative stress. Robust strains showed a 200 times better survival upon spray drying than the least-tolerant ones. This study linked the presence or absence of certain genes in robust and sensitive strains with survival after spray drying. Hence, a genetic fingerprint for this type of robustness could be assigned, which enables prediction of spray drying survival for other strains of the same species based on their genetic profile.



Selection for cheese starter cultures
Optimizing cheese quality requires in-depth knowledge of the formation of amino acid derived flavors by starter cultures. In a gene-trait matching study conducted by NIZO food research and Top Institute Food and Nutrition partners3 – involving 18 strains of 18 different Lactic Acid Bacteria species – linked the flavor compounds that were formed to the presence or absence of specific genes. This finding paves the way for a targeted production of natural flavor substances during cheese fermentation. It also allows for faster selection of starter cultures with novel flavor-forming characteristics.

Analysis of 38 Lactococcus lactis strains
The power of gene-trait matching was clearly demonstrated in a recent study involving gene occurrence and phenotype data of 38 Lactococcus lactis strains of dairy and plant origin4. These 38 strains had been assessed in 207 previous phenotyping experiments. Experimental variables included various substrates such as sugars, polysaccharides and milk. The study also assessed the effects of nisin, salt and various enzymes on Lactococcus lactis growth and made use of antibiotic resistance and metal resistance data. For gene-trait matching, the presence or absence of 4,026 ortholog groups (genes originating from the same ancestor – which may have the same function in different strains) was used as genotype data. Using a Phenolink, a total of 1,388 gene-phenotype relationsships was found. Some of these relationships were already known, for example the importance of arabinose utilization genes only for Lactococcus lactis of plant origin. However, the researchers also identified a gene cluster related to growth on melibiose, a plant disaccharide. Several novel gene-phenotype relationships could be described, for instance, genes related to arsenite resistance or arginine metabolism. The study results are publicly available and contain many leads into Lactococcus specieswide genotype-phenotype relationships that are to be further analyzed and experimentally validated. This could result in refinement of gene functions and more adequate predictions of phenotypes of new strains.

Limiting food spoilage
Microbial food spoilage occurs through random contamination with microorganisms from the environment. Lactobacilli are common spoilage bacteria in sauces, soups, condiments and spreads, where repeated use can easily lead to contamination. In collaboration with Unilever R & D, NIZO food research tested 121 Lactobacillus strains of six common Lactobacillus species for their phenotypes related to growth on 21 different carbon sources (food media) and 27 preservation conditions5. Bioinformatics tools developed at Radboud umc and NIZO food research established the matches between genes and traits. By linking the genome of Lactobacillus species and strains with phenotypic parameters, NIZO food research was able to identify the genes responsible for growth in specific carbon sources and under specific treatment conditions. “Differences between strains within one species might be even larger than those between species,” says Van Hijum discussing the results of this study. “Hence, diversity at the genotypic level, rather than the species type, determines whether a lactic acid bacterial strain is a food spoiler. “ He adds: “With this type of genotypic information, risks can be better assessed and preservation boundaries for specific food varieties can be determined more accurately.”

Health-promoting properties and safety assessment
“Predictive modeling might become a powerful instrument for companies and even regulating authorities in delivering supporting evidence about the presence or absence of certain characteristics in probiotic bacteria,” Van Hijum explains. If the genome sequence of a strain is available, there is a relatively cheap and quick way to determine whether antibiotic-resistant and virulenceassociated genes are present. With respect to healthpromoting properties, one could predict production of mucosal adhesion protein, essential nutrients and vitamins, exopolysaccharides or antimicrobial proteins, for example. Likewise, it is possible to predict the presence of marker genes for specific immunomodulatory properties.

Current research
Van Hijum concludes by saying that NIZO food research further strengthens its expertise on genomics through various partnerships. He points to the recently launched GENOBOX project (, in which NIZO food research collaborates intensively with the bacterial (meta)genomics group at the Centre for Molecular and Biomolecular Informatics (CMBI, Radboud umc, Nijmegen) and with four SMEs to create a toolbox in which genomic analysis can be used for prediction of functional properties of bacteria. The project involves the study of both starter cultures and probiotic strains. It is supported by funding from the European Union under the Seventh Framework Programme.

Contact details
Sacha van Hijum, PhD
NIZO food research
Telephone: +31 318 659 570

1. Bayjanov et al, BMC Genomics 2012, 13:170
2. Dijkstra et al, Appl Environ Microbiol 2014: 80(2): 603
3. Bayjanov et al, BMC Microbiology 2013, 13: 68
4. Ten Bruggencate, S. Case on and Wiersma et al, manuscript in preparation
5. Wels, M. Latest news at , March 13, 2014