Random amplified polymorphic DNA polymerase chain reaction uses short random primers to amplify multiple random DNA segments which, once visualized on an agarose gel, give unique patterns (Williams et al., 1990). Analysis of these fingerprints allows differentiation of species and strains by comparing profiles of various known strains (Ronimus et al., 1997, 2003). This method has been applied to colonies obtained from dairy powders in New Zealand to identify Geobacillus stearothermophilus, Anoxybacillus flavithermus, Bacillus licheniformis, and B. subtilis as the main contaminants of WMPs and SMPs, as well as buttermilk and goat milk powders (Ronimus et al., 2003). It has also been applied to whole and SMPs in Uruguay, correctly identifying the presence of B. licheniformis, B. megaterium, B. pumilus, A. flavithermus, and B. subtilis (Reginensi et al., 2011). Indeed, using this approach, G. stearothermophilus, A. flavithermus, and B. licheniformis have been identified as the dominant species in whole and SMPs from multiple countries including; Poland, Germany, Switzerland, France, Portugal, Netherlands, Great Britain, Ireland, Canada, USA, Mexico, Chile, Brazil, South Africa, Thailand, Australia, and New Zealand. B. subtilis, B. circulans, Ureibacillus thermosphaericus, B. coagulans, and B. pumilus have also being identified, albeit in lower quantities (Ruckert et al., 2004). A common feature of the RAPD-PCR approach is the highlighting of the 3–4 most dominant species. However, species of lower abundance might be the most interesting in terms of food security and spoilage. One exceptional study described the use of RAPD-PCR, and revealed a more in depth array of species, in Chinese dairy powders (Table 1) (Sadiq et al., 2016). Apart from identifying previously unreported species, other details worth noting are that B. licheniformis, G. stearothermophilus, and A. flavithermus were again established as being present in high abundance while, importantly, B. cereus group species were also identified. This observation obviously has implications for food safety (Sadiq et al., 2016). Although informative, analysis of the gel bands in RAPD PCR is very subjective allowing errors in classification and bias. Furthermore, the method requires time-consuming and laborious preparation of reference strains and there may also be variability between gels with the same samples, thus large-scale analysis would be difficult.

Housekeeping genes are genes that are essential for the functions of the cell and viability of the organism, and thus typically contain highly conserved regions (Gil et al., 2004; Eisenberg and Levanon, 2013). Genes that contain such highly conserved regions at either end of a more variable region are particularly useful for strain identification purposes as the conserved regions can be targeted using degenerate primers to facilitate PCR amplification and sequencing of the variable region (Case et al., 2007). Identification of genera present is facilitated by comparison with databases of corresponding variable region sequences of known origin (Case et al., 2007). Many genes have been utilized for classification of species in fluid milk in the form of molecular typing (Durak et al., 2006). Other typing methods have been described for milk powder isolates of Geobacillus spp. and B. licheniformis based on variable number tandem repeat analysis (Seale et al., 2012; Dhakal et al., 2013). The 16S rRNA gene is ubiquitous among bacteria, and contains multiple conserved and variable regions making it extremely useful, in general, for taxonomic classification. However, 16S rRNA gene sequencing cannot differentiate between closely related species or subtypes and other housekeeping genes such as gyrB or rpoB have been utilized to do so (Durak et al., 2006; Case et al., 2007). Recently, both the rpoB and 16S rRNA genes have been used to characterize the contaminating psychrophilic, mesophilic, and thermophilic spore populations isolated from sweet whey, WPC, non-fat dry milk and acid whey powders. At least 14 different species were identified, with B. licheniformis, Geobacillus spp., and Anoxybacillus spp. being the most abundant (Miller et al., 2015). These methods have the potential to allow identification and monitoring of persistent species and subtypes throughout dairy powder processing plants (Seale et al., 2012; Dhakal et al., 2013). Although not currently employed in sequencing dairy powder isolates, cpn60 (Durak et al., 2006; Schellenberg et al., 2011), pycA, ccpA (Liu et al., 2015), and groEL (Chang et al., 2003) have all been used to varying success in the sequencing of isolates from fluid milk (Durak et al., 2006), vaginal (Schellenberg et al., 2016) and marine (Liu et al., 2013) populations and remain as potential targets for future application to study dairy powder-associated microbes.

Pyroprinting utilizes sequencing by synthesis on multiple copy polymorphic loci simultaneously. The sequence reads are digitalized and can be compared using Pearsons correlation distance matrix to identify strains (Black et al., 2014). This method has been developed and utilized for source tracking, i.e., tracing sources of microbial contamination in end products or, more specifically, of endospore-forming bacilli in raw milk through to dairy powders. Presumptive species identified in powder included G. thermoleovorans, A. flavithermus, B. licheniformis, B. pumilus, and B. amyloliquefaciens (VanderKelen et al., 2016). These results correlate well with previous studies on raw milk and powders using the Sanger sequencing approach (Durak et al., 2006; Miller et al., 2015).

All of the above tests allow identification of the most abundant culturable species identified in dairy powders. However, they are limited by an initial requirement for culturing and, unless these methods are modified for identification of species directly from dairy powders, they are not suitable for the identification of non-culturable species or species of lower abundance which can be out competed when culturing, unless selective media is employed. Ultimately, while promising, these methods when compared to culture-independent sequencing (see Next Generation Sequencing for the Identification of Dairy Powder Contaminants) are labor intensive and time consuming.

Polymerase Chain Reaction (PCR)-based assays have been developed for the detection of B. cereus toxin genes. Taqman quantitative PCR (qPCR) assay of a single component of the hemolysin toxin gene in B. cereus has been developed (Cattani et al., 2016), amplifying the sequence corresponding to one component of one tripartite toxin. It has been reported that the Taqman probe is specific for B. cereus strains that contain this gene, however, not all B. cereus strains contain the hemolysin gene (Cui et al., 2016). This assay reportedly does not give false positives with related species, such as other members of the B. cereus sensu lato including B. thuringiensis and B. mycoides. However, this assay could lead to false negatives. The assay may fail to detect other species that have the toxin genes, or other strains of B. cereus that do not have this particular toxin, but may be pathogenic due to the presence of other toxins. This assay also gives accurate quantification of viable B. cereus by comparison to standard curves. Multiplex endpoint PCR of toxin genes has also been performed to identify B. cereus in dairy samples. These assays included primers to amplify single components of B. cereus enterotoxin genes, i.e., those encoding Nhe, CytK, and Hbl (Zhang et al., 2016) as well as enterotoxin FM (EntFM) and emetic toxin Ces (Forghani et al., 2015). However, the specificity of these assays was only tested using B. cereus and non-Bacillus species. Multiplex PCR of multiple components of B. cereus toxin genes has also been performed on single bacterial colonies isolated from dairy products and environments (Wehrle et al., 2009). This approach allows detection of all components needed to produce viable enterotoxins, and thus lessening the chance of false readings compared to other assays that only identify one toxin gene component. Multiplex endpoint PCR assays have also been developed for hygiene indicator species, G. stearothermophilus and A. flavithermus isolated from dairy powders. These assays rely in the species specific conserved regions of ITS 16S-23S rRNA region and the rpoB gene (Pennacchia et al., 2014). Further validation of these assays could lead to their use on DNA isolated directly from dairy powders. Finally, droplet digital PCR (ddPCR) allows precise, absolute quantification of a target DNA sequence. The DNA is encapsulated into many water in oil emulsion droplets and a PCR performed on each (Pinheiro et al., 2012). This culture-independent method has recently been used to detect B. cereus in fluid milk and can provide absolute quantification without need for comparison to standard curves. In this instance ddPCR was implemented using primers that target the gyrB gene of B. cereus sensu lato and the assay was found to have a lower detection limit than traditional qPCR (Porcellato et al., 2016), which is ideal for dairy powders that have low levels of contamination.

The assays described above also have the potential to be employed in the form of biosensors. Indeed, biosensors are already being developed for detection of a toxin gene found in B. cereus in milk and powder (Izadi et al., 2016). These biosensors are DNA based pencil graphite electrode (PGE) biosensors, in which a nhe toxin gene primer is immobilized on gold nanoparticles. Positive results are measured by an increase in charge resistance on the biosensor from the hybridization of the target DNA to nhe toxin sequence.

Although these methods do not give a complete view of the microbial composition in a dairy powder, they are useful as a test for key spoilage and pathogenic bacteria, including producers of harmful toxins. It is important to note that B. cereus sensu lato toxin genes are not specific to any one species of the group, nor is one toxin found in all B. cereus (Liu et al., 2015; Cui et al., 2016; Zhu K. et al., 2016). However, targeting toxins allows detection of all possible pathogenic species. Singleplex assays that target one component of one toxin may be prone to false negatives (Cui et al., 2016), i.e., producers of other toxin types being overlooked, thus underestimating the number of pathogenic B. cereus cells in a sample. Multiplex assays targeting many toxins, are more robust and can be beneficial for the food industry as they are a good indicator of potential food pathogens. Targeting all components of a toxin system may be required to confirm if there is a true potential for toxin production. Furthermore, while the genes for toxins may be present, it is unclear from these assays whether any active proteins are functionally expressed. The alternative use of a non-toxin gene for identification of B. cereus (gyrB) does not distinguish between members of B. cereus sensu lato, nor does it identify if the species identified are capable of being pathogenic. Overall the detection of toxin and species specific genes are a good indicator of potential pathogenic and other species of interest being present. Although issues remain, future improvement and development should result in the full potential of these approaches being realized.

In the last decade, considerable advances have meant that next generation DNA sequencing platforms have surpassed traditional Sanger sequencing platforms in terms of speed and potential applications. Their initially extremely short sequencing read lengths are less of a concern as sequencing lengths of Illumina and Ion platforms have increased (Quail et al., 2012) and new, even longer read, platforms have been developed by PacBio and Oxford Nanopore (Quail et al., 2012; Madoui et al., 2015). The advantages and disadvantages of the various sequencing platforms have been previously reviewed elsewhere (Goodwin et al., 2016). Regardless, research laboratories now have a much greater choice when determining which sequencing technology to use, though it should be noted that results generated using different methods, technologies or bioinformatics pipelines are not always consistent (Clooney et al., 2016). Whole genome shotgun sequencing is the process by where the whole genome of a single colony is sequenced. The DNA is extracted and sheared it into small pieces, before sequencing of these pieces and the use of computer software to assemble these sequences reads back together. This process can be applied to metagenomics, the term used to denote all of the genomic information from an entire community of different cells, for example the contaminants in dairy powders (Sharpton, 2014). The application of metagenomic techniques to the analysis of dairy products presents exciting opportunities. Metagenomic sequencing eliminates the need to culture, thus reducing bias, and allows the identification of species that are difficult to, or cannot be, cultured in the laboratory. Metagenomic sequencing has been applied to single gene products, such as the aforementioned 16S rRNA gene that can differentiate between all bacteria present to the genus level, while the spo0A gene has been targeted to specifically identify spore-forming Firmicutes in mixed populations. A whole metagenome ‘shotgun,’ i.e., untargeted, approach has also been attempted and comparison of 16S amplicon sequencing, spo0A amplicon sequencing and metagenomic shotgun sequencing performed for the identification of Firmicutes in metagenomic samples (Filippidou et al., 2015). Each method has advantages and disadvantages. Amplicon sequencing is more cost effective, high throughput and rapid but often only gives accurate classification to genus level, and may over-estimate microbial diversity in the sample (Acinas et al., 2004; Poretsky et al., 2014). In contrast, shotgun sequencing is more expensive, less samples can be analyzed at one time, but it gives the opportunity to accurately classify to species level provided there are accurate reference databases to compare sequence reads to Sharpton (2014). Shotgun sequencing also reduces the bias of amplicon sequencing that can arise due to need for an initial PCR amplification and, where relevant, variable gene copy numbers (Sharpton, 2014; Brooks et al., 2015). The other advantage of shotgun metagenomic approaches is that additional information regarding other genes of interest within the microbial community can be generated. Such genes include toxin genes (Steffen et al., 2012; Leonard et al., 2015), sporulation genes (Filippidou et al., 2015), non-ribosomal peptide synthase (NRPS) gene clusters (Schirmer et al., 2005), antibiotic resistance genes (Bengtsson-Palme et al., 2014), and phage genes (Dutilh et al., 2014), all of which may be interesting from a food safety point of view. The sequencing reads from this approach can be difficult to analyze as they can be biased toward genomes of higher abundance. This is a particular issue when studying samples from specific human and animal microbiomes where there is a considerable amount of DNA from host cells present (Feehery et al., 2013). It is important to note that, due to the high sensitivity of shotgun metagenomic sequencing, care needs to be taken to ensure the absence of contaminating cells or DNA from other environments (Salter et al., 2014; Glassing et al., 2016).

Regardless of the sequencing approach taken, bioinformatic expertise is needed to analyze sequencing data and compare sequence reads to databases. Databases and bioinformatics software are updated continuously and newer, more accessible programs are constantly being developed (Vincent and Charette, 2015), including more targeted programs and databases specifically for food microbes (Vangay et al., 2013; Parente et al., 2016).

Both amplicon and shotgun metagenomic sequencing reveal the relative abundance of bacteria in a sample. Furthermore, the quantification of total bacterial load can be achieved by coupling these techniques with qPCR or ddPCR analysis, (Porcellato et al., 2016).

While the benefits of next generation sequencing in determining the safety and quality of dairy powders provide cause for optimism, there are several hurdles. Culture-independent DNA analyses rely on one’s ability to extract all genomic DNA directly from the substrate for analysis. Extracting DNA from dairy powder can be difficult, especially from spore-forming bacteria. Although, many studies have endeavored to optimize methods for the extraction of DNA from spores that have been spiked into food, success has been varied (Wielinga et al., 2011; Mertens et al., 2014). Furthermore, the bacterial load is likely to be lower in dried dairy powders than other environmental samples in which this sort of analysis has been previously performed, such as the gut (Gill et al., 2006), soil (Fierer et al., 2012), and fermented food (Jung et al., 2011). Low DNA concentration can be overcome through use of whole genome amplification kits (Yokouchi et al., 2006; Binga et al., 2008). Although expensive, these provide for culture independent non-targeted analysis of all bacteria present in dairy powders even if present at low cell numbers. However, these kits are notoriously susceptible to contamination (de Bourcy et al., 2014) and, ideally, ultra clean laboratory environments are needed for their use (Weinmaier et al., 2015).

Isolation of DNA solely from spore-formers. There may be instances where there is a specific desire to specifically focus on the sequencing of DNA from the spore-forming community within a powder sample. Isolation of DNA solely from spores/spore-forming bacteria is a challenge. One possible method would be to perform standard spore pasteurization at 80°C for 12 min (see Spore Count Methods) (Frank and Yousef, 2004; Watterson et al., 2014) or other forms of targeted vegetative cell lysis (Wunderlin et al., 2016). However, free DNA could still be present in the samples from the lysed vegetative cells. Elimination of this signal could be performed using an intercalating dye (described below). Post-heat treatment, subsequent culture-based enrichment could be employed prior to DNA extraction (Frank and Yousef, 2004; Watterson et al., 2014) but, as described with respect to the culture-based approaches, this has the potential to lead to bias.

Sequencing-based approaches can also be adapted to specifically focus on spore-formers by, for example targeting of the spo0A gene for amplicon sequencing, or through focusing specifically on this gene from within shotgun sequence data. However, yet again, the need to ensure optimal DNA extraction and the removal of DNA from dead cells is a key consideration. A less conventional way of overcoming such challenges could involve the isolation of spores from dairy powder using density gradient centrifugation (Tamir and Gilvarg, 1966).

As noted above, free DNA from lysed vegetative cells can be present in samples following heat-treatments. Elimination of this signal could be performed using an intercalating dye. The use of intercalating dyes is especially relevant in the case of amplicon metagenomic sequencing where PCR amplification is performed (Rudi et al., 2005). This has been performed utilizing the dyes propidium monoazide (PMA) or ethidium monoazide bromide (EMA) to bind free DNA in the samples (Rudi et al., 2005; Forghani et al., 2015; Cattani et al., 2016; Zhang et al., 2016). Further testing and optimization would be needed to determine if its results are as promising for dairy powder samples with mixed populations. There are contradicting studies with regard to whether EMA or PMA is best for particular applications (Seinige et al., 2014; Wu et al., 2015). Very few studies have compared EMA and PMA in mixed populations, though EMA was reported to be favorable at penetrating heat damaged bacterial cells in fish fillets (Lee and Levin, 2009). EMA has been known to penetrate some live bacteria (Nocker et al., 2006; Seinige et al., 2014) whereas PMA has been seen not to penetrate all dead cells (Cattani et al., 2016). The concentrations of EMA used has seen a decrease in recent years (possibly to circumvent the penetration of live cells) and, so, while early studies used 100 μg/ml (Rudi et al., 2005; Nocker and Camper, 2006), more recent studies used 8–10 μg/ml (Seinige et al., 2014; Wu et al., 2015). Alternatives, including the use of platinum (Soejima et al., 2016) to bind extracellular DNA, appear promising as they have been reported to be more selective at differentiating live/dead E. coli and C. sakazakii than PMA in water and milk. Ultimately, optimization needs to take place to develop the system that is best suited to the low microbial load of mixed populations present in powdered dairy products. It should also be noted that these approaches are not effective when performing metagenomic shotgun sequencing, as there is no amplification step to eliminate the dye-bound DNA.

Currently culture-independent, population-based, analysis is relatively expensive and, thus, further developments are needed to increase its relevance to the food industry. It is, however, becoming more accessible as a test method for companies to strategically analyze processing pipelines and end products, allowing development of targeted treatments and intervention strategies against persistent or troublesome microorganisms. To provide thorough and reproducible analysis of dairy powders in this fashion, it will be particularly important to arrive at a consensus regarding the standardized sample preparation, use of specific sequencing platforms and analysis methodologies to facilitate comparison across multiple investigations (Clooney et al., 2016).