Work

Integration of molecular tools with technological evaluation to monitor the performance of starter cultures used for yogurt production

Year: 2023
Type: dissertation
Abstract: The combination of strains in a starter culture should promote efficient mutual growth between S. thermophilus and L. bulgaricus without the loss of metabolites. Therefore, screening mixtures of isolates for metabolic compatibility can explain balanced growth, gel formation and aroma production in yogurt. This study explored the metabolic compatibility of starter mixes using multivariate analysis methods to integrate multiple phenotypes, metagenomic single nucleotide variants (SNVs), volatile compounds, sensory attributes and rheological parameters. Hierarchal clustering of 72 isolates (36 of S. thermophilus and 36 of L. bulgaricus), representing three starter communities, was grouped by the similarity of phenotypic profiles: four clusters for S. thermophilus and five clusters for L. bulgaricus. Phenotypic characterization of isolates showed that diversification may have occurred over repeated culturing of yogurt starter mixtures, leading to deviations in fermentation performance. All starter mixtures were classified as compatible or incompatible based on the proportion of acidification profiles and competition between isolates in a mixture. A tuned model on 100 mL lab-scale mixtures selected 11 out of 19 volatile compounds and time to pH 4.6 as significant features that correctly classified 78.9% of their corresponding compatibility class. Nineteen SNVs of S. thermophilus and L. bulgaricus of starter mixtures were significantly enriched in compatible (11 SNVs) or incompatible (8 SNVs) mixtures. The compatible SNVs showed significant positive correlations with acetaldehyde, fruity aroma and gelation time, which were negatively correlated with dimethyl sulfoxide, sulphur flavour and broken texture (p-value < 0.05). A tuned model trained on 100 mL lab-scale fermentations tested against nine compatible and three incompatible 4 L bench products had 100% correct predictions for compatible and 6% correct predictions for incompatible mixtures. The one incompatible mixture that was predicted to be compatible indicated that strain mixtures may give compatible products due to strain compensation. This can be a basis for predicting the extent of incompatibility, producing off-flavour products. Machine learning can drive decision-making with tuned and validated models to predict starter culture performance in terms of improving flavour profiles. Multivariate statistical analysis can select genomic and metabolic markers of interest to understand the metabolic balances of consumer-preferred characteristics of yogurt.
Language: English
Subfield: Food Science
Open Access status: green
OpenAlex ID: W7201918019
Cites: 194
Cited by: 1
Related to: 10