Dairy Science & Technology · Postgraduate & Research · also Advances in Dairy Microbiology, Molecular Dairy Microbiology
Genomics of dairy organisms, microbiome science, phage biology and advanced detection, as examined in ICAR AICE subject 36.
In the exams: ICAR AICE subject 36 (all units); ASRB NET Dairy Microbiology
Dairy microbiology has been transformed in the past twenty-five years by genome sequencing. The complete genomes of starter and probiotic bacteria explain their metabolism, their adaptation to milk and their defences against phages; CRISPR–Cas systems, first shown to work as adaptive immunity in a yoghurt bacterium, are now used to type and improve starters; amplicon and shotgun sequencing reveal the microbial communities of raw milk, cheese and the human gut; and whole-genome sequencing has replaced older typing methods in outbreak investigation. At the same time, classic questions remain central: how bacteria survive stress and injury, how they form biofilms on plant surfaces, how resistance genes spread, how bacteriocins can protect foods, and how growth and death can be predicted from the conditions of a product.
This manual covers ten topics at doctoral and research level: the genomics of lactic acid bacteria; CRISPR systems and phage resistance; the microbiomes of raw milk, cheese and the gut; metagenomics and amplicon sequencing as methods; molecular typing for outbreak investigation, from pulsed-field gel electrophoresis to whole-genome sequencing; stress responses and injured cells; quorum sensing and biofilm biology; antimicrobial resistance genes; bacteriocins, their structure and use; and predictive microbiology.
Doctoral fellowship and research-service examinations in dairy microbiology expect candidates to explain mechanisms, interpret sequencing and typing data, choose methods, and apply models to food safety and quality. The manual presents established knowledge with its quantitative tools and notes where methods are still developing or interpretations debated, for example the existence of a resident milk microbiome inside the healthy udder.
The first complete genome of a lactic acid bacterium, Lactococcus lactis subsp. lactis IL1403, was published in 2001, and the genomes of hundreds of strains of every important dairy species have followed. Lactic acid bacteria have small genomes, typically about 1.8 to 3.3 million base pairs, and low guanine-plus-cytosine contents, mostly about 35 to 50%, consistent with their place in the low-G + C Gram-positive phylum. Species living in many habitats, such as Lactiplantibacillus plantarum, have the larger genomes and wide metabolic repertoires; species adapted to milk have smaller ones.
Adaptation to milk has shaped dairy genomes by gene loss. Streptococcus thermophilus and Lactobacillus delbrueckii subsp. bulgaricus, the yoghurt bacteria, carry many pseudogenes, genes inactivated by mutation, particularly for carbohydrate metabolism and for virulence-related functions present in their pathogenic relatives; this reductive evolution reflects life in a rich, stable medium where lactose is the main sugar and amino acids come from casein. S. thermophilus, though closely related to pathogenic streptococci, has lost or inactivated the virulence genes that make them dangerous, which is consistent with its long history of safe use.
Plasmids carry many technologically important traits in lactococci: lactose transport and metabolism through the phosphotransferase system and the tagatose pathway, the cell-envelope proteinase PrtP that starts casein breakdown, citrate permease in diacetyl-producing strains, phage-resistance systems and some bacteriocins. Their loss or transfer explains much of the variability of starter performance.
Sequencing many strains of a species reveals its pan-genome, the total set of genes found in any strain, divided into a core genome shared by all strains and an accessory genome present in some. Dairy lactococci and lactobacilli have large accessory genomes, reflecting horizontal gene transfer by plasmids, transposons, insertion sequences and prophages. Strain-level differences in the accessory genome explain differences in acidification rate, proteolysis, flavour, exopolysaccharide production, phage resistance and stress tolerance, and are now used to select and match strains for cultures.
Genomics has clarified the cooperation of the yoghurt bacteria. S. thermophilus grows first, lowering oxygen and producing formate, folate and carbon dioxide that stimulate L. delbrueckii subsp. bulgaricus; the lactobacillus, which has a strong cell-wall proteinase, releases peptides and amino acids from casein that the streptococcus, often with weaker proteolysis, needs. Each partner's genome shows the gaps the other fills. The result is faster acidification and more flavour, especially acetaldehyde, than either produces alone.
Genome-scale metabolic models, reconstructed from genome annotations and refined by experiments, predict growth requirements, metabolic fluxes and products, and help design media and processes. Transcriptomics and proteomics show which genes are expressed during growth in milk, during cheese ripening and under stress.
Genome sequences are now central to the safety assessment of food cultures. Regulators and expert bodies use genomic data to confirm the identity of a strain, to search for genes of concern, such as acquired antimicrobial resistance genes, toxin and virulence genes and genes for biogenic amine production, and to show that the strain belongs to a species with a history of safe use. The European Food Safety Authority's qualified presumption of safety (QPS) approach, for example, lists species considered safe subject to qualifications, and whole-genome sequencing of strains used in the food chain is required in several evaluation procedures. Similar principles apply in other jurisdictions' assessments of microbial cultures.
Genomics also guides strain improvement without recombinant DNA. Natural variants with desired traits, such as reduced post-acidification, increased exopolysaccharide production, improved texture or phage resistance, can be selected and their genetic basis confirmed by sequencing; genes for traits can be moved by conjugation; and adaptive laboratory evolution under selective conditions can be followed by resequencing to identify the mutations involved. For example, lactose-negative or galactose-positive variants of S. thermophilus have been selected to reduce post-acidification or to remove residual galactose, which matters in some cheeses.
Average nucleotide identity calculated from whole genomes, with values above about 95 to 96% indicating the same species, underlies modern taxonomy, including the 2020 division of the genus Lactobacillus into 25 genera.
CRISPR (clustered regularly interspaced short palindromic repeats) loci consist of short repeated sequences separated by unique spacers, preceded by a leader sequence and accompanied by CRISPR-associated (cas) genes. In 2007, Barrangou and colleagues showed, in Streptococcus thermophilus, that challenging the bacterium with phages produced resistant mutants that had acquired new spacers matching the phage genome, and that removing or adding spacers removed or conferred resistance. This established CRISPR–Cas as an adaptive immune system of prokaryotes.
Immunity works in three stages. In adaptation, a fragment of invading DNA, the protospacer, is captured and inserted as a new spacer at the leader end of the array; selection of protospacers depends on a short protospacer-adjacent motif (PAM) beside them in the invading DNA. In expression, the array is transcribed and processed into CRISPR RNAs (crRNAs), each carrying one spacer. In interference, crRNAs guide Cas nucleases to complementary sequences in invading phage DNA or RNA, which are cleaved; the PAM requirement prevents the system from attacking its own array.
CRISPR–Cas systems are grouped into two classes. Class 1 systems (types I, III and IV) use complexes of several Cas proteins; class 2 systems (types II, V and VI) use a single large effector protein, such as Cas9 in type II, Cas12 in type V and Cas13, which targets RNA, in type VI. S. thermophilus carries type II-A systems, called CRISPR1 and CRISPR3, which are the most active in phage resistance, and type III-A systems. Cas9 from bacteria, adapted as a programmable DNA-cutting tool, became the basis of CRISPR genome editing.
Phages counter CRISPR immunity. Point mutations or deletions in the protospacer or PAM let phages escape a particular spacer, and phage populations quickly select such escape mutants; bacteria in turn acquire further spacers. Some phages encode anti-CRISPR proteins that inhibit Cas proteins. Bacteriophage-insensitive mutants (BIMs) of S. thermophilus carrying several new spacers against the same phage are more robust, because a phage must mutate at several sites to escape.
Dairy bacteria have many other defences, often carried on plasmids in lactococci. Adsorption inhibition prevents phages binding, for example by changing or masking the cell-wall polysaccharides that lactococcal phages of the 936 group recognise. Superinfection exclusion, often encoded by prophages, blocks DNA injection. Restriction–modification systems cut unmethylated foreign DNA while the host's own DNA is protected by methylation; they are classified as types I to IV. Abortive infection systems allow phage DNA to enter but stop phage development, killing the infected cell and so protecting the population, a form of altruistic suicide; over twenty Abi systems are known in lactococci. Genome mining has recently revealed many further defence systems, with names such as BREX, DISARM and CBASS, showing that bacterial immunity is far more diverse than once thought.
Phage receptors and host specificity are now understood at molecular level for major dairy phage groups, explaining why some strains are attacked and others are not, and guiding the choice of strains for culture rotations.
CRISPR loci provide a powerful typing tool. Because new spacers are added at the leader end, the spacer content of an array records the strain's history and differs between closely related strains, so CRISPR spacer typing can identify and track starter strains, detect strain substitution in cultures and products, and serve as a molecular barcode. Culture companies have used it to authenticate strains and to monitor their use.
The industrial use of CRISPR immunity rests on selecting natural BIMs. A starter strain is challenged with phages isolated from dairy plants; survivors carrying new spacers against those phages are selected, characterised and confirmed to retain their technological properties. Because spacer acquisition is a natural process, such strains are not genetically modified organisms under most regulations, and strains with enhanced resistance have been commercialised, especially for S. thermophilus, where phages are the main cause of fermentation failure in yoghurt and some cheeses.
Genome editing with Cas9 or base editors has been demonstrated in lactic acid bacteria in the laboratory, for gene deletion, gene insertion and metabolic engineering. Its use in food cultures is constrained by the regulatory status of edited organisms, which differs between jurisdictions, and by consumer acceptance. CRISPR-based detection methods, which use Cas12 or Cas13 activity triggered by a target sequence to release a fluorescent signal, are being developed for rapid detection of pathogens and adulteration in dairy products.
Milk from a healthy udder was long thought to be sterile as it leaves the alveoli, with all bacteria entering from the teat canal, teat skin and environment. Culture-independent sequencing has detected bacterial DNA in aseptically collected milk, and some researchers propose a resident udder microbiome and even an entero-mammary pathway transferring gut bacteria to the gland; others attribute such signals to contamination of low-biomass samples, a well-recognised problem. The question remains open.
What is clear is that bulk raw milk contains a diverse and variable community. Typical members include Gram-negative psychrotrophs such as Pseudomonas and Acinetobacter; lactic acid bacteria such as Lactococcus, Streptococcus, Enterococcus, Leuconostoc and lactobacilli; staphylococci and corynebacteria from skin; spore-formers such as Bacillus, Paenibacillus and Clostridium from soil, feed and bedding; yeasts and moulds; and occasionally pathogens. Composition varies with season, housing, bedding, feed, milking hygiene, teat preparation, equipment cleaning and farm, and it shifts during cold storage, when psychrotrophs, especially pseudomonads, come to dominate within days.
These communities matter for quality and for products. Psychrotrophs produce heat-stable enzymes; spore-formers survive heat treatment; and the lactic acid bacteria and other organisms of raw milk give raw-milk cheeses their diversity of flavour, which is lost when milk is pasteurised or when farm hygiene greatly reduces microbial load.
Cheese contains two broad groups of bacteria inside its body. Starter lactic acid bacteria (SLAB), added deliberately, produce acid during manufacture, reach high numbers and then decline during ripening, releasing intracellular enzymes that contribute to flavour. Non-starter lactic acid bacteria (NSLAB), present at low numbers in the milk or from the plant environment, are mostly facultatively heterofermentative species such as Lacticaseibacillus paracasei, Lacticaseibacillus casei, Lactiplantibacillus plantarum and Lacticaseibacillus rhamnosus; they survive the harsh conditions of ripened cheese, using energy sources such as lactate, citrate, peptides, amino acids and nucleotides from lysed starter cells, and grow over weeks to months to reach about 10⁷ to 10⁸ cfu per gram in cheeses such as cheddar types, influencing flavour and sometimes causing defects. Adjunct cultures of selected NSLAB are added to control this.
Cheese rinds host distinct, often complex surface communities. Smear-ripened cheeses carry yeasts such as Debaryomyces hansenii that deacidify the surface, followed by coryneform bacteria such as Brevibacterium, Corynebacterium and Glutamicibacter, coagulase-negative staphylococci and salt-tolerant Gram-negative bacteria such as Psychrobacter and Halomonas. Bloomy rinds carry Penicillium camemberti and Geotrichum candidum, and natural rinds carry diverse moulds. Studies of cheese rinds from many countries found that communities assemble reproducibly according to rind type and conditions, making them model systems for microbial ecology.
Traditional cheeses made with natural starters, such as whey cultures carried over from the previous day's production, have complex, stable communities that give regional character; understanding them helps preserve traditional products while managing safety.
The human gut microbiome, the community of trillions of micro-organisms in the intestine, influences digestion, immunity and metabolism. Diet is a major influence, and dairy foods affect it in several ways. Fermented dairy foods deliver live bacteria, mostly transiently: yoghurt bacteria and most probiotic strains are detected in faeces during consumption but disappear within days or weeks after it stops, and few colonise long-term. They may nonetheless affect the resident community and host while passing through.
Lactose reaching the colon in lactase non-persistent people is fermented by bacteria, increasing bifidobacteria and lactic acid bacteria in some studies, which may contribute to adaptation and improved tolerance with regular intake. Milk oligosaccharides are central in infancy: human milk oligosaccharides are not digested by the infant but selectively feed bifidobacteria, particularly Bifidobacterium longum subsp. infantis, which carries gene clusters for their uptake and breakdown; this explains the dominance of bifidobacteria in breastfed infants and the interest in adding human-identical oligosaccharides to infant formula.
Human studies of fermented foods have shown, for example, that a diet high in fermented foods, including yoghurt and kefir, increased gut microbiome diversity and lowered markers of inflammation over ten weeks in a randomised study, while a high-fibre diet had different effects. Such findings are promising but come from small trials; links between specific microbiome changes and health outcomes remain largely associative, and many commercial claims go beyond the evidence.
Amplicon sequencing profiles microbial communities by amplifying and sequencing a marker gene from all organisms in a sample. For bacteria and archaea the marker is usually one or two hypervariable regions of the 16S rRNA gene, such as V3–V4 or V4, sequenced on short-read platforms; for fungi it is the internal transcribed spacer (ITS) region or parts of the 18S or 28S rRNA genes. The reads are processed to remove errors and chimeras and grouped either into operational taxonomic units (OTUs), traditionally clusters of sequences at 97% identity, or, in current practice, into amplicon sequence variants (ASVs), exact sequences inferred by denoising algorithms, which are reproducible between studies. They are then assigned taxonomy by comparison with reference databases.
Amplicon methods are cheap and sensitive but have important limitations. Short 16S fragments usually resolve bacteria only to genus, not species or strain; closely related dairy species such as Lactococcus lactis subspecies or the members of the Lactobacillus casei group may be indistinguishable. Primer choice biases which taxa are detected; PCR amplifies some templates more than others; the number of 16S gene copies per genome varies between species, distorting abundances; and DNA extraction efficiency differs between Gram-positive and Gram-negative cells and spores. Full-length 16S sequencing with long-read technologies improves resolution.
Low-biomass samples such as aseptically collected milk are especially vulnerable to contamination from reagents and laboratory environments, the so-called kitome, which can dominate the results. Negative extraction and PCR controls, mock communities of known composition, and careful interpretation are essential.
Shotgun metagenomics sequences all the DNA in a sample without targeting a marker gene. It identifies organisms across all domains, bacteria, archaea, fungi and viruses including phages, often to species and sometimes strain level, and reveals the functional genes present, such as those for lactose metabolism, proteolysis, bacteriocin production, antimicrobial resistance and virulence. Reads can be assembled into contigs and binned into metagenome-assembled genomes (MAGs), draft genomes of uncultured organisms. Its costs are higher than amplicon sequencing, and samples with abundant host or food DNA, such as milk with somatic cells, need deeper sequencing or host-DNA depletion.
DNA-based methods detect dead as well as live cells, and dead bacteria are abundant in heated products and in cheese. Treating samples with propidium monoazide or similar dyes, which enter damaged cells and bind their DNA so that it cannot be amplified, enriches for intact cells. Metatranscriptomics, sequencing RNA, shows which genes are being expressed and so which organisms are active, for example during cheese ripening; metaproteomics and metabolomics show the proteins and metabolites produced. Combining these layers links community composition to function.
Long-read sequencing, with nanopore and single-molecule real-time technologies, generates reads of thousands of bases, improving assembly and strain resolution and enabling rapid, portable sequencing for diagnostics. Culture remains complementary: isolating organisms is necessary to confirm viability, test properties and obtain strains for use.
Community data are analysed for diversity and composition. Alpha diversity describes diversity within a sample: richness, the number of taxa observed, estimated with indices such as Chao1 that correct for unseen taxa; and indices combining richness and evenness, such as Shannon and Simpson. Beta diversity compares samples, using distance or dissimilarity measures such as Bray–Curtis, based on abundances, Jaccard, based on presence and absence, and UniFrac, which incorporates phylogenetic relationships. Ordinations such as principal coordinates analysis display sample relationships, and permutational multivariate analysis of variance (PERMANOVA) tests whether groups differ in composition.
Sequencing data are compositional: each sample gives relative proportions summing to one, not absolute counts, because sequencing depth is arbitrary. An increase in one taxon's proportion may reflect a fall in others rather than its own growth. Analyses therefore use compositional methods, such as centred log-ratio transformations, or combine sequencing with absolute quantification by quantitative PCR, flow cytometry or spike-in standards. Differences in sequencing depth between samples are handled by rarefaction or by statistical models designed for count data.
Interpretation requires caution. Statistical associations between taxa and conditions do not establish causation; databases contain misannotations; and low-abundance signals may be artefacts. Reporting standards that require raw data deposition, description of controls, bioinformatic parameters and software versions make results reproducible. In dairy research, these methods have mapped microbial sources in farms and plants, followed cheese ripening, tracked contamination routes of spoilage organisms, and characterised traditional starters.
Investigating a foodborne outbreak requires determining whether cases are linked to each other and to a food source. Subtyping isolates of the pathogen, from patients, foods and processing environments, below the species level provides this evidence when combined with epidemiological data. A good typing method has high discriminatory power, the probability that two unrelated isolates are distinguished, often measured by Simpson's index of diversity; high reproducibility within and between laboratories; typeability of all isolates; and epidemiological concordance, grouping isolates known to be related.
Phenotypic methods came first: serotyping, which classifies isolates by surface antigens, such as the O and H antigens of Salmonella and Escherichia coli and the serotypes 1/2a, 1/2b, 4b and others of Listeria monocytogenes; phage typing; and antimicrobial susceptibility patterns. Their discrimination is limited, because many unrelated isolates share serotypes.
Pulsed-field gel electrophoresis (PFGE) became the reference method from the 1990s. Genomic DNA, embedded in agarose plugs to avoid shearing, is cut with a rare-cutting restriction enzyme, such as XbaI for Salmonella and E. coli or AscI and ApaI for Listeria, into a few large fragments, which are separated by periodically reversing the electric field. The banding patterns are compared, and national and international surveillance networks such as PulseNet shared standardised protocols and databases of patterns, allowing clusters of cases across regions to be detected. PFGE is laborious, its patterns can be indistinguishable for unrelated isolates of common clones, and small genetic changes can alter patterns unpredictably.
Multilocus sequence typing (MLST) sequences internal fragments of usually seven housekeeping genes, assigns each unique sequence an allele number, and combines the alleles into a sequence type (ST); related STs form clonal complexes (CCs). Its results are unambiguous, portable between laboratories and suitable for population studies. For Listeria monocytogenes, MLST revealed that certain clonal complexes, such as CC1, CC2, CC4 and CC6, are hypervirulent and over-represented among clinical cases, while others, such as CC9 and CC121, are more often found in foods and processing environments and may be adapted to persistence there. MLST is too coarse, however, to separate isolates within a common clone during an outbreak. Multiple-locus variable-number tandem repeat analysis (MLVA), counting repeats at several loci, offered higher discrimination for some species.
Whole-genome sequencing (WGS) has now replaced PFGE in many public health laboratories. Two main analytical approaches are used. Gene-by-gene methods extend MLST to hundreds or thousands of loci: core-genome MLST (cgMLST) uses the genes shared by nearly all members of a species, such as a scheme of about 1,700 loci for Listeria, and whole-genome MLST adds accessory genes. Single-nucleotide polymorphism (SNP) analysis maps reads to a reference genome and counts the nucleotide differences between isolates. Isolates differing by only a few alleles or SNPs are likely to share a recent common source, while large differences exclude a link.
WGS also predicts serotype, antimicrobial resistance genes, virulence genes and plasmids from the same data, replacing several separate tests.
No universal threshold defines an outbreak. Commonly cited guidance treats Listeria isolates within about seven to ten cgMLST alleles, or a small number of SNPs, as possibly related, but the expected diversity depends on the species, the time over which isolates were collected, the mutation rate and the structure of the source: a processing plant with a persistent strain colonising drains for years can produce isolates that differ by several alleles yet share a source. Genomic data must therefore be interpreted together with epidemiological, traceback and environmental evidence.
Dairy products have been linked to many investigations resolved or strengthened by WGS: listeriosis from soft and semi-soft cheeses, including those made from pasteurised milk contaminated after heat treatment; listeriosis from ice cream and milkshakes; Shiga toxin-producing E. coli infections from raw-milk cheeses; salmonellosis from cheeses and milk powders; and Cronobacter sakazakii infections linked to powdered infant formula. WGS has also shown that Listeria persisting in a plant for years can cause sporadic cases spread over long periods that earlier methods would not have linked.
For dairy companies, the implications are practical. Isolates from environmental monitoring can be sequenced to distinguish persistent resident strains from transient introductions, to locate harbourage sites and to verify the success of sanitation; and public health authorities increasingly compare isolates from food businesses with clinical databases, so a company's environmental isolates may link it to past cases. Strong environmental monitoring, root-cause analysis and corrective action, rather than reliance on end-product testing, are the response.
Micro-organisms in dairy processing and products meet heat, cold, acid, salt, drying, oxidation and starvation, and they respond with protective programmes. The heat shock response induces chaperones such as DnaK and GroEL, which refold or prevent aggregation of damaged proteins, and proteases such as Clp that remove damaged proteins; in Gram-positive bacteria it is controlled by repressors such as HrcA and CtsR. The cold shock response induces cold shock proteins that act as RNA chaperones and adjusts membrane fatty acids to maintain fluidity, allowing psychrotrophs such as Listeria and Pseudomonas to grow in refrigerated milk.
Acid tolerance involves several systems. The F₀F₁-ATPase pumps protons out of the cell; the glutamate decarboxylase system of Listeria and some lactic acid bacteria consumes a proton when glutamate is decarboxylated to γ-aminobutyrate and exchanges it for external glutamate; the arginine deiminase pathway produces ammonia that neutralises acid. Exposure to mild acid induces an acid tolerance response that protects against otherwise lethal acidity. Osmotic stress from salt in cheese and brines is countered by accumulating compatible solutes, such as glycine betaine and carnitine, through dedicated transporters; Listeria's ability to do so explains its survival in brines and high-salt products. Oxidative stress is countered by catalase, superoxide dismutase and, in lactic acid bacteria, which lack catalase, by NADH oxidases and peroxidases.
General stress regulators coordinate many responses: the alternative sigma factor σB in Listeria, Bacillus and staphylococci, and RpoS in Gram-negative bacteria. Exposure to one stress often protects against others, a phenomenon called cross-protection, so mild sublethal treatments can make pathogens more resistant to later processing steps.
Treatments that do not kill all cells often injure many of the survivors. Sublethally injured cells have damaged membranes, walls, ribosomes or nucleic acids; they cannot grow under conditions that healthy cells tolerate, especially on selective media containing bile salts, salts, antibiotics or dyes, but they can repair the damage and recover in a favourable, non-selective environment. Heat, freezing, drying, acid, sanitisers, high pressure and other non-thermal treatments all cause injury.
Injury has important consequences for testing. If a sample of pasteurised or acidified product containing injured pathogens is plated directly onto a selective medium, many cells fail to form colonies and the count is underestimated, sometimes by orders of magnitude, or the pathogen is not detected at all. Yet in the food, the injured cells may repair during storage and grow. Standard detection methods therefore include resuscitation steps: non-selective or mildly selective pre-enrichment, such as buffered peptone water for Salmonella, before selective enrichment; and enumeration methods may use overlay techniques, in which the sample is first plated on a non-selective medium and overlaid with selective medium after a repair period. The extent of injury is estimated as the difference between counts on non-selective and selective media.
The viable but non-culturable (VBNC) state is a related condition in which cells remain metabolically active and intact but cannot form colonies on any routine medium, sometimes recovering under particular conditions. Its significance for dairy pathogens is debated, but it means culture-negative results do not always prove absence of live cells.
Stress biology matters for both beneficial and harmful bacteria. Starter and probiotic cultures must survive freezing, freeze-drying, storage, and, for probiotics, passage through stomach acid and bile. Manufacturers improve survival by harvesting cells at the right growth phase, often early stationary phase, by mild pre-adaptation to heat, acid or osmotic stress that induces protective responses, by cryoprotectants such as skim milk solids, sugars like trehalose and sucrose, and polyols during freezing and drying, and by encapsulation in protective matrices. Viability during shelf life of probiotic yoghurts and powders depends on oxygen, acidity, water activity and temperature.
For pathogens, the same responses can undermine safety. Listeria adapted to cold, acid and salt in a plant can survive cheese making and brining better than laboratory strains; acid-adapted Escherichia coli O157 survives better in fermented products; and heat-shocked Salmonella can be more heat-resistant. Process validation should therefore use strains and preparation conditions relevant to the real hazard, including stress-adapted cells, and challenge tests follow guidance on inoculum preparation.
Spores have their own dormancy and resistance: some spores in a population, called superdormant, fail to germinate under conditions that trigger most others, so treatments designed to germinate spores before killing them, as in double heat treatments, leave survivors. Persister cells, a small fraction of a vegetative population tolerant to antibiotics or disinfectants without genetic resistance, are a further source of survival after sanitation.
Quorum sensing is cell-to-cell communication by which bacteria coordinate gene expression according to population density. Cells produce and release signalling molecules, autoinducers; as the population grows, the concentration rises until it exceeds a threshold, at which receptors activate or repress target genes across the population. Gram-negative bacteria commonly use N-acyl-homoserine lactones (AHLs), made by LuxI-type synthases and detected by LuxR-type receptor regulators; Gram-positive bacteria commonly use modified oligopeptides, autoinducing peptides, detected by two-component signal transduction systems, as in the agr system of Staphylococcus aureus, which regulates virulence factors including toxins. Autoinducer-2, a furanosyl borate diester made by the LuxS enzyme, is produced by many species and has been proposed as an interspecies signal, though its role is debated.
Quorum sensing controls traits relevant to dairy. In Pseudomonas and other Gram-negative spoilage bacteria, it regulates production of extracellular enzymes, such as proteases and lipases, and biofilm formation, linking cell density in cold-stored raw milk to the onset of enzyme production. In S. aureus, the agr system regulates toxin production, relevant to enterotoxin formation in milk and cheese.
The bacteriocin nisin is itself a quorum-sensing pheromone: extracellular nisin is sensed by the two-component system NisK and NisR in the producing Lactococcus lactis, which induces transcription of the nisin biosynthesis genes. This autoregulation has been harnessed in the nisin-controlled gene expression system, a widely used tool for inducible protein production in lactic acid bacteria. Quorum quenching, the interference with signals by enzymes that degrade AHLs or by signal analogues, is studied as a way to control spoilage and biofilms.
A biofilm is a community of micro-organisms attached to a surface and embedded in a self-produced matrix of extracellular polymeric substances (EPS): polysaccharides, proteins, extracellular DNA and lipids. Biofilm formation proceeds through stages. A conditioning film of adsorbed milk proteins and other organic molecules forms on a clean surface within seconds to minutes; cells attach reversibly through weak forces, then irreversibly through adhesins and EPS; they divide into microcolonies; the biofilm matures into a three-dimensional structure with water channels; and cells or clumps disperse to colonise new sites.
Cells in biofilms are far more resistant to cleaning and disinfection than the same cells suspended in liquid, by factors of tens to a thousand or more. The reasons include limited penetration of chemicals through the matrix, reaction of oxidising sanitisers with matrix material, slow growth and altered physiology of cells within the biofilm, persister cells, and protection by other species in mixed communities. Biofilms also release cells continuously, so a small biofilm can contaminate large volumes of product.
In dairy plants, biofilms form wherever milk residues, moisture and time coincide: on plate heat exchangers, especially in regeneration and cooling sections; on gaskets, valves and dead legs; in raw-milk silos and tankers; on membranes of ultrafiltration and reverse osmosis plants; and in floor drains, conveyors and other environmental sites, where Listeria monocytogenes can persist for years. Thermophilic spore-formers such as Geobacillus stearothermophilus and Anoxybacillus flavithermus form biofilms in evaporators and in the hot sections of milk powder plants, growing at 55 to 65 °C and releasing spores and cells into the product; their numbers rise with run time and limit the length of production runs.
Biofilm control relies first on prevention through hygienic design and effective cleaning. Equipment designed without crevices, dead legs and rough surfaces, built of smooth stainless steel with drainable geometry, limits attachment and allows cleaning solutions to reach every surface. Cleaning-in-place programmes combine a pre-rinse, an alkaline detergent wash that removes proteins and fats, an acid wash that removes mineral deposits, and sanitising, at controlled concentrations, temperatures, contact times and flow velocities high enough to give turbulent shear at the walls. Run lengths of evaporators and powder plants are limited to stop thermophile biofilms from reaching high numbers, and cleaning frequency is set from monitoring.
Additional measures target established biofilms. Enzymatic cleaners containing proteases, polysaccharidases or DNases degrade matrix components; peracetic acid and other peroxygen sanitisers penetrate biofilms better than some alternatives; and physical methods such as ultrasound or high-pressure water assist removal in some applications. Research explores phages and phage enzymes that degrade matrices, quorum-sensing inhibitors, antimicrobial or anti-adhesive surface coatings, and competitive exclusion by harmless biofilm-forming bacteria, though few of these are yet routine.
Verification shows whether control works. Environmental monitoring programmes sample food-contact and non-contact surfaces for indicator organisms and for Listeria, using sponges and swabs, with zones defined by proximity to the product; adenosine triphosphate bioluminescence gives rapid readings of residual organic soil; test coupons placed in pipelines or tanks are removed and examined by culture or microscopy, including confocal laser scanning microscopy with fluorescent stains; and isolates are typed to determine whether they represent persistent strains. Positive findings trigger investigation, intensified cleaning and, where necessary, equipment redesign.
Antimicrobial resistance (AMR) is the ability of a micro-organism to survive or grow in the presence of an antimicrobial agent that would normally inhibit or kill it. Resistance mechanisms fall into a few classes: enzymatic inactivation of the drug, such as the hydrolysis of β-lactams by β-lactamases including extended-spectrum β-lactamases and carbapenemases; modification of the target, such as the altered penicillin-binding protein PBP2a encoded by mecA in methicillin-resistant Staphylococcus aureus or the methylation of ribosomal RNA by erm genes conferring macrolide resistance; ribosomal protection, as by the tet(M) and tet(O) proteins for tetracyclines; active efflux of the drug by membrane pumps; and reduced permeability.
Resistance may be intrinsic, a property of a whole species resulting from its normal structure, or acquired, gained by mutation or by horizontal transfer of resistance genes. Acquired genes are often carried on mobile genetic elements, plasmids, transposons, integrative and conjugative elements and gene cassettes in integrons, which can move between strains and species by conjugation, transformation and transduction. Acquired, transferable resistance is the main concern, because it can spread to pathogens.
Many lactic acid bacteria are intrinsically resistant to some antimicrobials. Most lactobacilli of the Lacticaseibacillus and Lactiplantibacillus groups, Leuconostoc and Pediococcus are intrinsically resistant to vancomycin because their peptidoglycan precursors end in D-alanine–D-lactate rather than D-alanine–D-alanine, to which vancomycin binds poorly. This resistance is chromosomal and not transferable and is not considered a hazard, whereas the transferable van genes of enterococci are.
Cultures used in foods should not carry acquired, transferable resistance genes, because fermented foods deliver very large numbers of live bacteria to the gut, where gene transfer to other bacteria is possible. Genes such as tet(M), tet(S), tet(W), erm(B) and others have been found in some strains of lactic acid bacteria and bifidobacteria from foods and probiotics. Regulatory guidance, such as that of the European Food Safety Authority, assesses strains by comparing their minimum inhibitory concentrations (MICs) for a panel of antimicrobials with microbiological cut-off values for the species; strains above the cut-off are examined, now largely by whole-genome sequencing, to determine whether resistance is intrinsic or due to acquired genes, and strains with acquired transferable genes are considered unsuitable.
Raw milk and the dairy farm environment are reservoirs of resistant bacteria. Use of antimicrobials to treat and prevent mastitis and other diseases selects for resistance; studies have found extended-spectrum β-lactamase-producing Escherichia coli, livestock-associated methicillin-resistant S. aureus and resistant enterococci and streptococci in milk, on farms and in calves fed waste milk containing antibiotic residues. Pasteurisation kills these bacteria, but their DNA and resistance genes may persist, and raw-milk products can carry live resistant bacteria to consumers.
Antimicrobial residues in milk are a separate issue: they are controlled by withdrawal periods and by maximum residue limits set by Codex and national authorities, and tested at milk reception, because residues are a food-safety concern and inhibit starter cultures.
Phenotypic testing measures susceptibility directly. Broth microdilution determines the MIC, the lowest concentration that prevents visible growth, using standardised methods and media; lactic acid bacteria need special media because standard ones do not support them. Disc diffusion and gradient strips are simpler alternatives. Results are interpreted against clinical breakpoints for pathogens or epidemiological cut-off values (ECOFFs), which separate wild-type populations from those with acquired resistance.
Genotypic methods detect resistance genes. PCR targets specific genes; microarrays detect many at once; and whole-genome sequencing of isolates, analysed with curated databases of resistance genes and mutations, identifies the complete known resistome of a strain and its mobile elements. Shotgun metagenomics surveys resistance genes in whole communities, such as farm environments, raw milk and cheese. Genotype does not always predict phenotype, because genes may be silent or novel mechanisms unknown, so both approaches are used.
AMR is a One Health problem linking human, animal and environmental health. International action plans of the World Health Organization, the Food and Agriculture Organization and the World Organisation for Animal Health promote prudent use of antimicrobials in animals, surveillance of use and resistance in the food chain, infection prevention, and alternatives to antimicrobials. For the dairy sector, these translate into mastitis prevention through hygiene and vaccination, selective dry-cow therapy rather than blanket treatment, responsible veterinary prescribing, safe disposal of waste milk, screening of cultures for acquired resistance genes, and participation in national surveillance.
Bacteriocins are ribosomally synthesised antimicrobial peptides or proteins produced by bacteria, usually active against closely related species. Those of lactic acid bacteria are of particular interest because the producers have a long history of safe use in foods. Classification schemes have been revised several times; a widely used version distinguishes the following groups.
Class I bacteriocins, the lantibiotics and related modified peptides, undergo extensive post-translational modification: serine and threonine residues are dehydrated and linked to cysteines to form the thioether amino acids lanthionine and methyllanthionine, creating ring structures. Nisin, produced by Lactococcus lactis, is the best-known example, a 34-amino-acid peptide; lacticin 3147 is a two-peptide lantibiotic. Class II bacteriocins are small, heat-stable, unmodified peptides. Class IIa, the pediocin-like bacteriocins, share a conserved N-terminal motif, YGNGV, and are strongly active against Listeria; pediocin PA-1 is the classic example. Class IIb bacteriocins need two different peptides to act; class IIc are circular peptides; and class IId groups other unmodified peptides. Larger, heat-labile antimicrobial proteins, sometimes called bacteriolysins and formerly class III, such as helveticin J, act differently and are often considered separately.
Producer strains protect themselves with immunity proteins and dedicated transporters, encoded with the bacteriocin structural gene in operons on plasmids, transposons or the chromosome; the nisin genes, for example, lie on a conjugative transposon.
Most bacteriocins of lactic acid bacteria act on the cytoplasmic membrane of sensitive cells. Nisin has a dual mode of action: it binds lipid II, the membrane-bound carrier of peptidoglycan building blocks, which blocks cell wall synthesis, and it uses lipid II as a docking molecule to form pores in the membrane, causing rapid efflux of ions and small metabolites, collapse of the proton-motive force and cell death. The specific lipid II target explains its high potency at nanomolar concentrations. Pediocin-like class IIa bacteriocins use the mannose phosphotransferase system as a receptor in the target membrane and form pores, which explains their specificity for Listeria and some other Gram-positive bacteria.
Bacteriocins of lactic acid bacteria are generally active against Gram-positive bacteria, including Listeria, Staphylococcus aureus, Bacillus and Clostridium, and nisin also inhibits the outgrowth of germinating spores of Bacillus and Clostridium. Gram-negative bacteria are normally protected by their outer membrane, which prevents access to the cytoplasmic membrane, but become sensitive when the outer membrane is disrupted, for example by chelating agents such as EDTA, by high pressure or by sublethal heat. Yeasts and moulds are not affected; natamycin, used on cheese surfaces against moulds and yeasts, is a polyene antifungal produced by Streptomyces, not a bacteriocin.
Resistance to bacteriocins can arise in target bacteria through changes in membrane and cell wall composition, alterations in receptors such as the mannose phosphotransferase system, and enzymes that degrade the peptides, such as nisin resistance protein. Using bacteriocins as one of several hurdles reduces the risk of resistant populations emerging.
Nisin is the most widely approved bacteriocin as a food preservative, permitted in many countries and in Codex standards for products including processed cheese, cheese spreads and some other dairy products, at levels set by national rules. Its classic use is in processed cheese and spreads, where it prevents spoilage by spores of Clostridium species that survive the melting process, such as C. tyrobutyricum and C. sporogenes, which can cause gas and off-flavours, and contributes to the control of C. botulinum in formulated products. It is also used in some desserts, flavoured milks and ricotta-type cheeses, often to extend shelf life against spore-formers and Listeria.
There are three main ways of using bacteriocins in dairy foods. First, they can be added as purified or semi-purified preparations, such as commercial nisin concentrates or fermentates, which are cultured media containing bacteriocins and other metabolites, used where permitted. Second, bacteriocin-producing starter or adjunct cultures can produce the bacteriocin in situ during fermentation; nisin-producing lactococci, or lacticin 3147-producing strains, have been studied to control Listeria and clostridia in cheese, though nisin producers may also inhibit other starter bacteria. Third, protective cultures, selected strains that inhibit pathogens or spoilage organisms without changing the product's character, often through bacteriocins as well as competition and acid, are added to fresh cheeses, yoghurts and other products.
Limitations include binding of bacteriocins to fat and proteins, which reduces their activity in full-fat products; degradation by proteases; reduced activity at neutral pH for some, such as nisin, which is more soluble and stable in acid; narrow spectra; regulatory restrictions; and the need to keep good hygiene as the primary control. Bioengineering of nisin and other lantibiotics to improve activity and stability is an active research field.
Predictive microbiology describes the responses of micro-organisms to their environment with mathematical models, so that growth, survival or death can be predicted for conditions not directly tested. Models are classified as primary, secondary and tertiary. Primary models describe the change in numbers with time under constant conditions. Growth curves, plotted as the logarithm of counts against time, are characterised by the lag time λ, the maximum specific growth rate μmax and the maximum population density, and are fitted with sigmoid models. The modified Gompertz model, introduced to food microbiology by Zwietering and colleagues in 1990, and the Baranyi and Roberts model of 1994, which describes lag as the time needed for cells to synthesise a critical substance and is widely used, are the commonest; the logistic model and Buchanan's three-phase linear model are simpler alternatives.
Inactivation is also modelled by primary models. The classical log-linear model assumes first-order death and is characterised by the D value. Many survival curves are not linear, showing shoulders, where little death occurs at first, or tails, where a resistant fraction survives. The Weibull model, log(N/N₀) = −(t/δ)^p, describes such curves with a scale parameter δ, the time to the first decimal reduction, and a shape parameter p, which is less than one for tailing and greater than one for shouldered curves. Geeraerd's model includes both shoulders and tails. Choosing an appropriate model matters, because extrapolating a log-linear model fitted to a tailing curve overestimates lethality at long times.
Secondary models describe how the parameters of primary models, such as μmax and λ, depend on environmental factors: temperature, pH, water activity, organic acids, preservatives, carbon dioxide and others. The square-root model of Ratkowsky and colleagues, published in 1982, relates growth rate to temperature below the optimum: √μ = b(T − Tmin), where Tmin is a notional minimum growth temperature, found by extrapolation, and b a constant. A worked example: for an organism with Tmin = −2 °C and b = 0.03 per °C per h^0.5, √μ at 6 °C is 0.03 × 8 = 0.24, so μ = 0.0576 per hour and the doubling time ln 2/μ is about 12 hours; at 4 °C, √μ = 0.18, μ = 0.0324 per hour, and the doubling time is about 21 hours. This shows how strongly small temperature changes in the cold chain affect growth.
Extended square-root models add terms for pH, water activity and other factors, and cardinal parameter models express each factor's effect in terms of its minimum, optimum and maximum values for growth, multiplying the effects of independent factors according to the gamma concept of Zwietering. Polynomial models fitted by response surface methods are also used but are less mechanistic.
Growth/no-growth boundary models, usually logistic regression of observed growth or no growth against combinations of factors, predict the conditions under which a pathogen cannot grow at all, important for products such as cheeses and fermented milks designed to prevent the growth of Listeria. Their predictions are probabilities rather than certainties.
Tertiary models are software tools and databases that combine primary and secondary models with user interfaces. ComBase, an international database of tens of thousands of microbial growth and survival curves with associated predictive tools, and other freely available programs, such as the Pathogen Modeling Program and various national tools, allow users to predict pathogen and spoilage organism behaviour in their products. Commercial and research platforms add stochastic simulations that account for variability between strains, products and storage conditions, giving distributions of outcomes rather than single values.
Models must be validated against independent data, ideally from the real product. Ross proposed two widely used indices: the bias factor, Bf = 10^(mean of log(predicted/observed)), which shows whether a model on average predicts faster (Bf above one) or slower (below one) growth than observed; and the accuracy factor, Af = 10^(mean of |log(predicted/observed)|), which measures average deviation regardless of direction. A model for growth rate with Bf between about 0.9 and 1.05 is often considered good for pathogens, with fail-safe predictions slightly faster than observed preferred. Models built in broth often predict faster growth than in real foods, because foods contain structures, competing flora and inhibitors not represented in broth.
Predictive models are used to set shelf lives, to assess the effect of temperature abuse in distribution, to design product formulations that prevent pathogen growth, to support HACCP decisions on critical limits, and as the exposure assessment step of quantitative microbial risk assessment. In regulatory frameworks that limit Listeria monocytogenes in ready-to-eat foods, such as limits applying throughout shelf life in some jurisdictions, manufacturers use predictive models together with challenge tests and durability studies to show compliance.
Read the key primary papers: the first lactococcal genome, the 2007 CRISPR phage-resistance study, the Baranyi–Roberts and Ratkowsky models, and a major cheese-rind ecology study, and explain each in your own words. Practise analysing a small amplicon dataset with a standard pipeline, including negative controls, and fit primary and secondary growth models to real data. Learn to calculate injury, D and Weibull parameters, and bias and accuracy factors, because examinations and research both need them, and keep a log of every analysis you run.
A culture scientist or food safety microbiologist five years into the job selects phage-resistant strains and designs rotations, screens cultures for acquired resistance genes, investigates biofilms and persistent Listeria in plants with environmental monitoring and sequencing, validates detection methods for injured cells, designs challenge tests and uses predictive models to set shelf lives and assess temperature abuse. Typical decisions concern which isolates to sequence, whether a positive environmental sample represents a resident strain, and whether a formulation prevents Listeria growth.
A head of microbiology, culture research director or senior regulator with 25 years' experience sets strategy: investment in sequencing and bioinformatics, phage-control and culture development programmes, environmental monitoring policy, responses to outbreak investigations that link plant isolates to clinical cases, and the company's or agency's position on genome-edited cultures and novel antimicrobials. They interpret complex genomic evidence for management and courts and ensure that models and methods are validated before decisions rest on them. They also decide when the evidence justifies a recall or a public statement.
A professor or research leader advances the field: phage–host interactions and new defence systems, functional ecology of cheese and milk microbiomes with multi-omics, mechanisms of persistence of pathogens in plants, stress biology of probiotics, and stochastic predictive models for risk assessment. They train students in rigorous methods, controls and reproducible bioinformatics, and contribute to international standards, culture safety guidance and surveillance networks. Their teaching links classical culture-based microbiology with genomics so that graduates can use both, and their reviews of grant proposals and manuscripts shape what the field accepts as evidence.
Detailed bioinformatic pipelines, parameter choices and statistical models for microbiome data are described in principle only. Outbreak relatedness thresholds, regulatory status of genome-edited cultures, bacteriocin permitted levels and culture safety rules differ between agencies and change; only general principles and examples are given. Viral (non-phage) and protozoal hazards, mycotoxigenic fungi genomics and the immunology of probiotics are not covered. Evidence on a resident udder microbiome and on many gut microbiome effects of dairy is still developing.
Original practice questions written to the syllabus and pattern of the exams named in each question's examTags. They are not past papers and do not reproduce any board's questions.
These notes are free to read here, with no account.
What the Laisha app adds:
Free for ten days, then a subscription. We would rather write that here than have you find it later.