Shelf-life storage temperature has a considerably larger effect than high-temperature, short-time pasteurization temperature on the growth of spore-forming bacteria in fluid milk

In the absence of postpasteurization contamination, psychrotolerant, aerobic spore-forming bacteria that survive high-temperature, short-time (HTST) pas-teurization, limit the ability to achieve HTST extended shelf-life milk. Therefore, the goal of the current study was to evaluate bacterial outgrowth in milk pasteurized at different temperatures (75, 85, or 90°C, each for 20 s) and subsequently stored at 3, 6.5, or 10°C. An initial ANOVA of bacterial concentrations over 14 d of storage revealed a highly significant effect of storage temperatures, but no significant effect of HTST. At d 14, average bacterial counts for milk stored at 3, 6.5, and 10°C were 1.82, 3.55, and 6.86 log 10 cfu/mL, respectively. Time to reach 1,000,000 cfu/mL (a bacterial concentration where consumers begin to notice microbially induced sensory defects in fluid milk) was estimated to be 68, 27, and 10 d for milk stored at 3, 6.5, and 10°C, respectively. Out of 95 isolates characterized with rpoB allelic typing, 6 unique genera, 15 unique species, and 44 unique rpoB allelic types were represented. The most common genera identified were Paenibacillus , Bacillus , and Lysinibacillus . Nonmetric multidimensional scaling identified that Bacillus was significantly associated with 3 and 10°C, whereas Paenibacillus was consistently found across all storage temperatures. Overall, our data show that storage temperature has a substantially larger effect on fluid milk shelf life than HTST and suggests that abuse temperatures (e.g., storage at 10°C) allow for growth of Bacillus species (including Bacillus cereus genomospecies) that do not grow at lower temperatures. This indicates that stringent control of storage and distribution temperatures is critical for producing extended shelf-life HTST milk, particularly concerning new distribution pathways for HTST pasteurized milk (e.g., electronic commerce), and when enhanced control of spores in raw milk is not feasible.


INTRODUCTION
Production of fluid milk with extended shelf life (ESL) and improved quality over shelf life (including ability to maintain quality during mild or even extended temperature abuse) remains a key challenge for the dairy industry (Boor et al., 2017;Reichler et al., 2018;Sadhu, 2018).Key drivers for this include consumer concerns about sustainability (e.g., UN Sustainable Development Goals; UN General Assembly, 2015) and food spoilage, new distribution pathways that may require ESL products (e.g., electronic commerce), as well as transportation related challenges (e.g., disruptions in the trucking industry in some countries, which may lead to a need for less frequent deliveries, requiring products to have a longer shelf life).Although in many countries, this need for ESL products is addressed through production of UHT milk, there continues to be a need for production and distribution of HTST fluid milk with longer shelf life, especially in the United States where consumers prefer the sensory attributes of HTST milk over those of UHT milk (Horner et al., 1980;Chapman et al., 2001).
Although the FDA Pasteurized Milk Ordinance (PMO) has a set minimum for HTST pasteurization of 72°C for 15 s, it is well known that commercial processing plants pasteurize well above the minimum requirements (Fromm and Boor, 2004;Martin et al., 2011).Some published studies have reported on the potential of different strategies for shelf-life extension of HTST fluid milk, such as improved temperature control throughout shelf life (Buehler et al., 2018) or that pasteurization at lower temperatures (i.e., closer to the minimum HTST temperature of 72°C compared with above 80°C) may actually prolong shelf life (Ranieri et al., 2009;Martin et al., 2012).Most of these studies, however, represented small and highly controlled pilot studies with defined raw milk quality (e.g., raw milk from a single farm; Ranieri et al., 2009) or modeling studies with limited validation (Buehler et al., 2018; Shelf-life storage temperature has a considerably larger effect than high-temperature, short-time pasteurization temperature on the growth of spore-forming bacteria in fluid milk Ziyaina et al., 2018).To provide further information on the ability of different strategies (i.e., improved temperature control, different HTST temperatures) to extend shelf life of fluid milk produced under more industry relevant conditions, we used a commercial raw milk supply with substantial raw milk quality variation to assess the effect of (1) HTST pasteurization temperatures and (2) storage temperatures on the growth of psychrotolerant, aerobic spore-forming bacteria.These bacteria were characterized using rpoB PCR and sequencing, as this has been proven to be a useful tool for identifying spore-formers of concern (Ivy et al., 2012;Gaballa et al., 2021).In addition, we also measured pH as a measure of acid formation and performed particle analysis as an indicator of coagulation, as physiochemical properties can be additional indicators of fluid milk spoilage (Spoilage of Milk and Milk Products, 2016).This study was a full factorial design, utilizing different classical microbiological and molecular methods, as well as prediction tools.

Sample Collection, HTST Pasteurization, and Storage
On 4 separate occasions (4 biological replicates, hereby referred to as trials) between November 2018 and April 2019, approximately 70 L (18 gallons) of raw bovine milk were collected from storage tanks at a processing plant in Texas.Milk was collected into 3.8 L (1 gallon) sterile Nalgene bottles (Thermo Fisher Scientific) and shipped overnight, on ice, to Cornell University (Ithaca, NY).Upon receipt, raw milk was stored at 4°C and held for up to 24 h before processing.All 70 L of raw milk was then commingled in a 76-L sanitized stainless steel kettle, a 500-mL raw milk sample was then collected for microbiological analyses, and then the commingled milk was pasteurized at 75, 85, or 90°C for 20 s at a flow rate of 1 L/min and subsequently cooled to 10°C in a small-scale HTST pasteurizer (Mi-croThermics Model 25-DH) in the Food Processing and Development Laboratory (Cornell University, Ithaca, NY).For each trial and for each of the 3 pasteurization temperatures, a total of approximately 5 L of pasteurized milk was aseptically collected and distributed into 250-mL sterile, screw-capped Pyrex (Corning, Inc.) bottles (each filled with 200 mL of pasteurized milk).For all trials, these samples were held at 1 of 3 temperatures (3, 6.5, or 10°C) until at least d 42, 35, and 15, respectively.Incubator temperatures were checked daily to ensure temperatures were ± 0.5°C throughout shelf life.Pasteurized milk samples held at 3°C were evaluated throughout shelf life on d 0, 7, 14, 21, 28, 35, and 42.Pasteurized milk samples held at 6.5°C were evaluated throughout shelf life on d 0, 7, 14, 21, 28, and 35.Samples held at 10°C were evaluated on d 0, 3, 5, 7, 9, 11, 13, and 15.Although samples stored at each of the 3 storage temperatures were tested for all 4 trials, data for trial 2 samples stored at 3°C were removed due to an incubator failure.Additionally, data for trial 1 samples stored at 10°C were removed because samples were tested only on d 0, 7, 14, 21, 28, and 35; this testing scheme missed data in the lag and exponential growth phases, hence why samples were tested on d 3, 5, 7, 9, 11, 13, and 15 in trials 2, 3, and 4. In summary, data were available for 3, 4, and 3 trials for samples stored at 3, 6.5, and 10°C, respectively.In addition to the shelf-life samples, a 60-mL pasteurized milk sample, representing the initial day (d 0), was collected for each trial from each of the 3 pasteurization temperatures for analyses.

Microbiological Analyses of Raw Milk and Pasteurized Fluid Milk Samples
For each trial, raw milk was assessed for (1) total aerobic plate count (APC), performed on standard methods agar and incubated at 32°C for 48 h; (2) laboratory pasteurization count (LPC), which involves heat treatment of a raw milk sample at 63°C for 30 min before performing an APC, and (3) preliminary incubation count (PIC), which involves incubating a raw milk sample at 21°C for 18 h before performing an APC; the method details described in Standard Methods for the Evaluation of Dairy Products (Frank and Yousef, 2004;Laird et al., 2004) were followed except that all APC were performed in duplicate by using a spiral plater (Neu-tec Eddy Jet 2) to plate 50 μL of milk onto standard methods agar.The APC were incubated for 48 h at 32°C and subsequently enumerated.Raw milk was also assessed for (4) mesophilic aerobic spore count (MSC), and (5) psychrotolerant aerobic spore count (PSC) as previously described (Frank and Yousef, 2004).Briefly, for the MSC and PSC, raw milk was heat treated at 80°C for 12 min, followed by pour plating 1 mL of heat-treated milk on each of 10 separate standard method agar plates for each MSC and PSC (yielding a limit of detection of 1 spore/10 mL).The MSC plates were incubated at 32°C for 48 h before enumeration, whereas enumeration of colonies on PSC plates was performed after 10 and 21 d of incubation at 6°C.All enumeration was performed using an automatic colony counter (IUL, S.A.).
Pasteurized milk samples were tested for total APC (as described above) and total gram-negative bacteria count until the end of the evaluated shelf life.Total gram-negative bacteria counts were determined by spi-ral plating 50 μL of sample onto crystal violet tetrazolium agar (CVTA) in duplicate, followed by incubation at 21°C for 48 h and subsequent enumeration of only red colonies.As CVTA is a selective agar for gramnegative bacteria, and psychrotolerant gram-negative bacteria, including Pseudomonas, have been associated with postpasteurization contamination, the presence of gram-negative bacteria on CVTA, was used as an indicator of postpasteurization contamination (Alles et al., 2018).Atypical growth on CVTA, defined as nonred colonies, is typically small, light-to-dark purple or translucent-to-white colonies, was noted as present or absent, but these colonies were not included in the CVTA count.

Physiochemical Analysis
In addition to microbiological analyses, pH and particle-size analysis measurements were performed to assess physiochemical properties of the milk samples over shelf life.These tests were performed at the same time points in shelf life as the microbiological tests (as detailed above).Milk pH was evaluated according to (Hooi et al., 2004), using a combination electrode (Model Beckman Coulter pHi 470) to measure the pH of duplicate 60-mL aliquots of each sample.Particlesize analysis was performed using a Nanoparticle Size Analyzer equipped with a Peltier temperature control system (Brookhaven Instruments Corp.); the resulting measurements allowed us to assess the beginning of protein aggregation, as an indicator of sample coagulation.Duplicate measurements were conducted at a fixed 90° angle and a wavelength of 658 nm.Before particlesize analysis, samples were removed from storage and kept on crushed ice until analysis was performed.Milk samples were diluted in ambient-temperature deionized (DI) water to a concentration of 0.5 to 1% to ensure the sample measurement was within the measurable kilocounts per second for the particle analyzer.Samples were equilibrated to 25°C within the particle analyzer before measurements began, and light scattering intensity fluctuation measurements were taken for 4 min at 8 different 30 s intervals; the Stokes-Einstein equation was used for calculating the equivalent spherical particle sizes from the particle diffusion coefficients.Similar methods using particle analysis for assessing physiochemical properties of milk during shelf life were used by Yang et al. (2021).

Bacterial Isolation and Preservation
For the 217 of 218 pasteurized milk samples that yielded at least 1 colony on standard methods agar (on a given day of shelf life), the corresponding agar plates used for enumeration were visually examined for unique colony morphologies.A colony representative of each distinct morphology was chosen for isolation and a subset of the isolates were later used for characterization.Unique atypical colonies on CVTA, observed on 21% of CVTA plates, were also isolated to confirm that these colonies were not gram-negative bacteria that would be representative of postpasteurization contamination (PPC).Additionally, unique colonies with morphologies consistent with spore-forming bacteria were isolated from LPC plates to allow for the comparison of these isolates to spore-forming bacteria subtypes isolated after HTST pasteurization (i.e., d 0 samples).Typically for all isolation, 1 to 5 colonies per sample plate were selected and streaked for purity on brain heart infusion (BHI) agar.Purified isolates were grown overnight in BHI broth, and 850 μL of the overnight culture was mixed with 150 μL of glycerol and subsequently frozen at −80°C.

rpoB Sequencing and Allelic Type Assignment
Identification of Bacillus spp.and Paenibacillus spp.isolates was performed by PCR amplification and subsequent sequencing of a 632-nucleotide fragment of the rpoB gene, as described previously by Huck et al. (2007).Sequence data obtained were compared with an rpoB sequence database (Gaballa et al., 2021), which is available through Food Microbe Tracker (https: / / www .foodmicrobetracker.net/), to assign an rpoB allelic type (AT; e.g., AT 15) to a given isolate; different AT are assigned to gene sequences that differ from each other by 1 or more nucleotides.Isolates with differing AT were considered to represent different subtypes, even though we appreciate that this approach has limited discriminatory power compared with other molecular methods (e.g., whole genome sequencing).The AT data, along with rpoB phylogenies (if needed), were also used to assign putative species names to a given isolate, as previously described (Gaballa et al., 2021).Most isolates allowed for successful PCR amplification and subsequent sequencing with the standard rpoB primers previously described as well as subsequent classification to genus and species by using the existing rpoB sequence database and phylogenetic analyses (Gaballa et al., 2021).However, a single isolate did not amplify with rpoB PCR and was thus characterized by sequencing the 3′ end of the 16S rDNA, as described previously (Huck et al., 2007).The partial 3′ 16S rDNA sequence obtained was used for a similarity search against the Ribosomal Database Project Seqmatch tool (Center for Microbial Ecology, Michigan State University, East Lansing, MI).If the top 10 matches returned by the search with Seqmatch represented the same genus, isolates were assigned to this genus.

Data Curation and Statistical Analysis
All raw data collected were saved in Excel (Microsoft Excel for Microsoft 365 MSO;Version 2112 Build 16.0.14729.20254;64-bit;Microsoft Corp.).The majority of data curation was performed in OpenRefine (version 3.4.1,https: / / openrefine .org/).Data manipulation was performed using the rstatix package (Kassambara, 2021) and the dplyr package (Wickham et al., 2021).Figures were created with ggplot2 (Wikcham, 2016).Descriptive statistics, ANOVA, Tukey honestly significant difference, and general linear models were performed with Base-R (R Core Team, 2022).The R package biogrowth was then used to estimate the time to 20,000 and 1,000,000 cfu/mL using the "time_to_ logcount" function which calculates the time based on linear interpolation of a simulated growth curve (Garre et al., 2021).Levene's test was performed using the car package (Fox and Weisberg, 2019).Linear mixed effects models were created using the lme4 package (Bates et al., 2015).Kruskal-Wallis rank sum tests and Dunn's tests for pairwise multiple comparisons of the ranked data were performed using rstatix.Nonmetric multidimensional scaling (NMDS) was performed to determine if bacterial populations between HTST temperatures or storage temperatures were different.For the NMDS performed, an initial presence-absence distance matrix of rpoB AT was constructed and duplicates of trial, storage temperature, HTST temperature, and rpoB AT were removed.To reach convergence, singletons were removed by first removing all instances where a given combination of trial-storage temperature-HTST temperature included only a single AT; second, removing all remaining AT that only represented a single isolate; and third, again removing all instances where a given combination of trial-storage temperature and HTST temperature that included only a single AT.The NMDS scores were plotted using ggplot2.Then, analy-sis of similarities (ANOSIM, an ANOVA-like test for dissimilarity matrices) using the Bray-Curtis dissimilarity measure and 9,999 permutations was performed to determine whether 1 or multiple AT, species, or genera were significantly associated with a given storage temperature, HTST temperature, or trial.Both NMDS and ANOSIM were performed using the vegan package (Oksanen et al., 2020).For combinations of taxonomic classifications (AT, species, genus) and predictor variables (trial, HTST temperature, storage temperature) that we found to have P < 0.05 from the ANOSIM results, we performed a multiple-level pattern analysis to further identify significant differences.Multiple-level pattern analysis was performed using the indicspecies package (De Caceres and Legendre, 2009).All analyses were performed in RStudio (version 1.2.5033,RStudio Team).All raw data files and code used for this paper are available in the Supplemental Files (https: / / github .com/FSL -MQIP/ ESL _HTST).
We performed an initial formal statistical analysis using a 3-way ANOVA to assess the effect of HTST tem-perature, storage temperature, day, and the interaction of storage temperature and day on the net log 10 growth (defined as log 10 cfu/mL at a given day minus log 10 cfu/ mL at d 0; these analyses included only d 0, 7, and 14, as these were the only sampling days for which data were available for all 3 storage temperatures.For milk stored at 10°C, d 14 data were calculated as the mean of the d 13 and 15 net log 10 growth data as d 14 data were not collected.Based on this ANOVA, we observed a difference of net log 10 cfu between storage temperatures (P < 0.001), days (P < 0.001), and the interaction of storage temperature and day (P < 0.001).We did not observe a difference between net log 10 growth between different HTST temperatures (P = 0.54).From Tukey tests, we observed that (1) for d 14, net log 10 growth was different between samples stored at 3 and 6.5°C (P < 0.001); and (2) for d 7 and 14, net log 10 growth for samples stored at both 3 and 6.5°C was different than the net log 10 growth of samples stored at 10°C (all P < 0.001; Figure 2).We did not observe a difference of net log 10 growth between samples stored at 3 and 6.5°C on d 7 (P = 0.91).
Subsequently, we performed 3 separate 2-way ANOVAs, 1 for each storage temperature, to assess the effect of HTST temperature and day on net log 10 growth (Supplemental Table 1, https://github.com/FSL-MQIP/ESL_HTST).As the initial ANOVAs with the data for all days of shelf life did not meet the assumptions of normality, analyses were performed with only those days in shelf life that showed mean increased bacterial concentrations over d 0, which included (1) d 21, 28, 35, and 42 for storage at 3°C; (2) d 14, 21, 28, and 35 for storage at 6.5°C; and (3) d 5, 7, 9, 11, 13, and 15 for storage at 10°C.We did observe a difference of net log 10 growth due to day of shelf life for samples stored at 6.5 and 10°C (both P < 0.001).We did not observe a difference between days for samples stored at 3°C (P = 0.059); however, this does not imply that there was no significant growth at this temperature but reflects the fact that data for the initial 3 d of shelf life were excluded in the analyses to ensure normality of the data (as detailed above).For storage at 6.5°C, we observed a difference of net log 10 growth between HTST temperatures (P < 0.05), and a Tukey's test revealed that net log 10 growth for milk pasteurized at an HTST temperature of 85°C (4.25 net log 10 growth) was different compared with milk pasteurized at 75°C (3.15 log 10 net growth); milk pasteurized at 90°C showed net log 10 growth of 3.52 cfu; net log 10 growth here represents the average across all days included in the data set used for the statistical analyses (i.e., d 14, 21, 28, and 35).
To estimate the time to important milk quality thresholds (i.e., 20,000 cfu/mL, the FDA PMO limit, and 1,000,000 cfu/mL, the bacterial level where consumers tend to begin to notice microbially induced sensory defects (Carey et al., 2005)), we used microbial count data for each HTST and storage temperature combination (e.g., 3°C/75°C) to fit 3 primary growth models (Buchanan, Baranyi, and modified Gompertz; Zwietering et al., 1990).We used the best-fitting model, based on BIC values, to estimate initial concentration (N 0 ), lag (λ), maximum growth rate (μ max ), and maximum concentration (N max ; see Supplemental Table 2, https://github.com/FSL-MQIP/ESL_HTST).We estimated samples to reach 20,000 cfu/mL by d 45, 19, and 7, and 1,000,000 cfu/mL by d 68, 27, and 10 for storage temperatures of 3, 6.5, and 10°C, respectively (Table 2).For milk stored at 3 and 6.5°C, we estimated that milk pasteurized at 75°C will take longer to reach microbial thresholds of 20,000 and 1,000,000 cfu/mL compared with milk pasteurized at 85 and 90°C.For example, for milk stored at 6.5°C, we estimated that milk pasteurized at 75°C would reach 1,000,000 cfu/ mL by d 32, whereas milk pasteurized at 85 and 90°C was estimated to reach this threshold at 24 and 25 d, respectively.

Effect of HTST and Storage Temperature on Physicochemical Spoilage Associated Parameters (pH and Particle Size)
In addition to assessing bacterial growth across shelf life, we also performed pH measurements and particlesize analyses on samples collected over shelf life as alternative chemical and physical proxies for spoilage.Overall, we only observed pH decreases and particle size increases on shelf-life days subsequent to microbial numbers reaching >6.5 log 10 cfu/mL.We performed an initial formal statistical analysis using a 3-way ANOVA to assess the effect of HTST temperature, storage tem- perature, day, and the interaction of storage temperature and day on pH; these analyses included only d 0, 7, and 14, as these were the only sampling days for which data were available for all 3 storage temperatures.For milk stored at 10°C, d 14 data were calculated as the mean of the d 13 and 15 pH data, as d 14 data were not collected.Using this ANOVA, we did not observe a difference of pH due to HTST temperature (P = 0.67), storage temperature (P = 0.070), day (P = 0.073), or the interaction of storage temperature and day (P = 0.80).Regardless of pasteurization temperature, we did not observe milk stored at 3°C drop below pH 6.5 through the last day of monitored shelf life (i.e., d 42; Figure 3a).By comparison, we observed the mean pH dropped below 6.5 for milk stored at both 6.5°C and 10°C on d 35 and 15, respectively.Given that the assumptions of ANOVA were not met for pH at the individual storage temperatures, we performed statistical analyses using the Kruskal-Wallis test, a 1-way ANOVA equivalent.We computed pairwise comparisons using Dunn's tests with Bonferroni correction.Although we observed no clear patterns of overall pH differences between milk pasteurized at different HTST temperatures, using Kruskal-Wallis and subsequent Dunn's tests, we revealed differences between (1) 75 and 90°C on d 28 for samples stored at 3°C (P < 0.01; with mean pH values of 6.68 and 6.80 for 75 and 90°C, respectively); and (2) 85 and 90°C on d 14 for samples stored at 6.5°C (P < 0.01; with mean pH values of 6.77 and 6.69 for 85 and 90°C, respectively).
Regarding particle size, we observed the mean log 10 particle size (nm) at d 0 (across all storage and pasteurization temperature) to be 2.50 log 10 nm.For milk stored at 3 and 6.5°C, we observed that the maximum particle size over the full storage time (42 and 35 d, respectively) never exceeded 2.71 log 10 nm, even though general patterns suggested a slight increase in particle size over storage time (Figure 3b).For milk stored at 10°C, we observed that mean particle size exceeded 3.00 log 10 nm in 5 of 72 samples.These 5 samples surpassed 3.00 log 10 nm on d 11 of shelf life or later.Visible observation of coagulation (by a single observer) classified all 5 samples with particle size >3.00 log 10 nm as visibly coagulated; conversely, no visible coagulation was observed in any samples with particle size <3.00 log 10 nm.We performed initial formal statistical analysis using a 3-way ANOVA to assess the effect of HTST temperature, storage temperature, day, and the interaction of storage temperature and day on particle size; these analyses included only d 0, 7, and 14 as these were the only sampling days for which data were available for all 3 storage temperatures.For milk stored at 10°C, d 14 data were calculated as the mean of the d 13 and 15 log 10 particle size data as d 14 particle size data were not collected.With this ANOVA, we observed a difference in particle size between days (P < 0.01) and storage temperatures (P < 0.01); however, we did not observe a difference in particle size due to HTST temperature (P = 0.075).The Tukey tests that we performed revealed that particle size for samples stored at 3 and 6.5°C were not observed to be different from each other (P = 0.85) but were observed to be different from samples stored at 10°C (both P < 0.05).
As the data for separate particle-size analysis ANO-VAs to determine the effect of HTST temperature and day on particle size for each storage temperature (which included the data for all days of shelf life) did not meet the assumptions of normality, we performed analyses with the same subset of days used in the individual storage temperature ANOVAs for assessing bacterial concentration as mentioned above.For samples stored at 3°C or 10°C, we did not observe particle size to be different at the level of HTST temperature (P = 0.21, P = 0.49, respectively, or day P = 0.072, P = 0.16, respectively).For samples stored at 6.5°C, we observed a difference in particle size because of day (P < 0.001), but not HTST temperature (P = 0.43).
Physicochemical data paired with rpoB AT data allowed us to initially assess whether specific sporeformer subtypes are associated with samples that show reduced pH.Importantly for trials 1 and 4, we found that samples stored at 6.5°C showed different pH values for shelf-life samples pasteurized at different temperatures.For trial 1 samples stored at 6.5°C, we found that on d 28 and 35, samples pasteurized at 85°C showed substantially lower pH as compared with corresponding Computed as the mean from the predicted values from each pasteurization temperature.For example, for the mean time to 20,000 cfu/mL for milk stored at 3°C was calculated as (48 + 41 + 45)/3 = 45.
samples pasteurized at 75 and 90°C.For trial 4 samples stored at 6.5°C, we found that on d 28 and 35, samples pasteurized at 75 and 85°C showed substantially lower pH as compared with corresponding samples pasteurized at 90°C (Table 3).The rpoB data show that 5 Paenibacillus peoriae (representing 5 different AT) were isolated from the 3 shelf-life samples with pH values <6.6, whereas the matching shelf-life samples with pH values >6.6 yielded different genera and species (e.g., different Bacillus species, including Bacillus licheniformis, Bacillus paralicheniformis, as well as different Paenibacillus species, such as Paenibacillus odorifer) but did not yield P. peoriae.We performed similar comparisons for particle size (Table 4); however, only a single set of shelf-life samples was available where there were noticeable differences in particle size; for trial 3 samples stored at 10°C, we observed that on d 13 and 15, samples pasteurized at 75°C had particle sizes >3.00 log 10 nm and visible coagulation while corresponding samples pasteurized at 85 and 90°C had particle sizes <3.00 log 10 nm and did not have visible coagulation.Although we observed 5 subtypes to be unique to the samples with higher particle size (e.g., Bacillus mosaicus AT 410 and AT 785), the fact that only 1 sample set was available for comparison prevented us from developing a well-supported hypothesis as to whether specific spore-former subtypes may be more likely to cause coagulation (e.g., due to production of proteolytic enzymes).

Characterization of Bacteria Isolated from Milk Samples Pasteurized at Different Temperatures and Stored at Different Temperatures
Overall, we obtained 155 isolates from standard methods agar (aerobic plate counts of pasteurized milk To test the hypothesis that different types of organisms are selected for based on initial pasteurization temperature as well as shelf-life storage temperature, we used subset (2) described above, which represents isolates that were obtained from the 2 shelf-life samples from the days that showed the highest bacterial counts for each combination of pasteurization temperature, storage temperature, and trial (e.g., for trial 4, isolates were selected from d 13 and 15 for samples that were HTST pasteurized at 75°C and stored at 10°C).If a given sample included multiple isolates with the same AT (e.g., 2 B. licheniformis AT 1 isolates from d 13 of trial 4, for the sample that was HTST pasteurized at 75°C and stored at 10°C), we only included 1 isolate in this isolate set; we used this approach to prevent overrepresentation of isolates of the same subtype.This selection procedure yielded a total of 96 isolates from the original 111 category (2) isolates, and we deemed them to represent the "predominant" organisms for the shelf-life samples that showed the greatest bacterial concentration.Only 1 of these isolates (classified as Micrococcus spp.) was characterized by 16S sequencing, whereas 95 of these 96 isolates were characterized by rpoB sequencing.These 95 isolates represented the genera Paenibacillus (n = 52), Bacillus (n = 35), Lysinibacillus (n = 6), Brevibacillus (n = 1), and Streptococcus (n = 1) (Table 5); the 3 most common species among these isolates were Paenibacillus odorifer (n = 23), Paenibacillus cf.peoriae (n = 21), B. licheniformis (n = 11), with B. licheniformis AT 1 representing the most frequently isolated AT for each of the 3 HTST temperatures (n = 10).By contrast, different AT were predominant for different storage temperatures; AT 1 (B.licheniformis, n = 8), AT 15 (Paenibacillus odorifer, n = 4), and AT 239 (Paenibacillus cf.peoriae, n = 6) were the most frequently isolated AT obtained from samples stored at 3, 6.5, and 10°C, respectively.Overall, Bacillus spp.represented 52% (15/29), 11% (2/18), and 38% (18/48) of isolates that we obtained from samples stored at 3, 6.5, and 10°C, respectively, whereas Paenibacillus spp.represented 41% (12/29), 89% (16/18), and 50% (24/48) of isolates that we obtained from samples stored at 3, 6.5, and 10°C, respectively.Interestingly, we found 3 genomospecies representing the B. cereus group only in shelf-life samples stored at 10°C, including B. mosaicus (n = 6, 3 unique AT), B. cereus sensu stricto (s.s.; n = 8, 4 unique AT), and B. toyonensis (n = 3, 2 unique AT), these 3 genomospecies represented 35% (17/48) of the isolates obtained from 10°C samples.As these genomospecies were not isolated from shelf-life samples stored at 3 or 6.5°C, our data suggest that these organisms show limited or no ability to effectively grow at 3 and 6.5°C.
Based on these initial observations, we performed NMDS analysis using the 96-representative shelf-life sample isolates detailed above to formally determine if microbial communities were different between storage temperatures or HTST temperatures.The initial presence-absence distance matrix constructed contained 89 unique combinations of trial, storage temperature, HTST temperature, and AT and the ordination based on this matrix did not reach convergence after 100 iterations.Removing singletons yielded 58 unique combinations of trial, storage temperature, HTST temperature, and AT, and the ordination run with this matrix reached convergence after 44 iterations.The NMDS data scores were then plotted, showing that populations with the same storage temperature clustered closely to each other, while there was no apparent clustering of populations with the same HTST temperatures (Figure 4).After performing ANOSIM at the level of AT, we observed a difference based on storage temperature (P < 0.01; Supplemental Table 3, https://github.com/FSL-MQIP/ESL_HTST).We did not observe a difference for level of trial (P = 0.25) or HTST temperature (P = 0.48).At the level of species, we observed a difference due to storage temperature (P < 0.001).We did not observe a difference for level of HTST temperature (P = 0.55) or trial (P = 0.089).At the level of genus, we also observed a difference due to storage temperature (P < 0.001), but also due to trial (P < 0.05); however, we did not observe a difference for level of HTST temperature (P = 0.84).
With multiple-level pattern analyses, we did not identify any genera to be associated with 1 or more trials.However, we identified multiple associations of AT, species, and genera with storage temperatures, including (1) at the level of AT, AT 1 (B.licheniformis) and storage temperature 3°C (P < 0.01); (2) at the level of species, B. licheniformis and 3°C (P < 0.001), Bacillus safensis and 3°C (P < 0.01), Paenibacillus cf.peoriae at both 6.5 and 10°C (P < 0.05 for both temperatures), and Bacillus cereus s.s.(P < 0.05) at 10°C; and (3) at the level of genus, Lysinibacillus and 10°C (P < 0.05) and Bacillus at both 3 and 10°C (P < 0.001 for both temperatures).

Comparison of Raw Milk Isolates Obtained by LPC and Isolates Obtained After Pasteurization at 75, 85, and 90°C
Initially, we collected isolates with spore-former morphologies from LPC, a commonly used raw milk test.To compare the ability of LPC to allow for the isolation of the predominant spore-formers found in milk immediately after pasteurization, we used rpoB AT data to compare the spore-former subtypes obtained from LPC and directly after milk was pasteurized at 75, 85, or 90°C.Overall, we observed that AT obtained from LPC were typically also found in milk pasteurized with at least 1 of the 3 temperatures used in this study; notably in trials 1, 2, and 4, we found B. licheniformis AT 1 in LPC as well as milk pasteurized at all 3 HTST temperatures (Table 6).In addition, we obtained some AT only from LPC, but not in any pasteurized milk samples from corresponding trials (e.g., B. safensis AT 140).Similarly, we found some rpoB AT that were found only in pasteurized samples, but not from LPCs (e.g., B. paralicheniformis AT 215).In addition, we observed that Paenibacillus AT found toward the end of shelf life, were neither isolated from LPC nor from milk immediately after HTST pasteurization.

Paenibacillus Growth on CVTA
As part the experiments performed here, all samples that we collected over shelf life were also plated on CVTA, a selective medium often used for the enumeration of gram-negative bacteria.Although we did not observe any samples exhibiting typical growth on CVTA (i.e., red-colored colonies), we observed several samples with atypical growth on CVTA, which is defined as colonies without red color (Frank and Yousef, 2004).Atypical growth was more common in samples stored at 6.5 and 10°C (12/77 and 33/78 total samples with atypical growth, respectively), as compared with samples stored at 3°C (1/63 total samples with atypical growth).To identify and further characterize the bacteria responsible for this atypical growth, we selected a single colony from a total of 4 samples (n = 1 and n = 3 stored at 6.5 and 10°C respectively) for characterization by Gram staining; for all colonies that we characterized with this approach, Gram staining showed small rods that stained either violet or pink, typically with similar proportions of cells stained with these 2 colors; this phenotype designated as "gram-variable," is a characteristic previously described for Paenibacillus, which s.s.denotes sensu stricto.s.l.denotes sensu lato.cf.denotes conferre: isolate matching 16S rDNA sequence (i.e., 97.0%-98.0%sequence similarity) of more than 1 type strain from family Bacillaceae.sp.denotes species designated to an isolate for which genus, but no species, could be assigned.
Table 5 (Continued).Genera, species, and allelic type frequency of the 96 representative isolates selected from aerobic plate count media from the 2 shelf-life samples with the highest aerobic plate counts for each unique combination of trial-HTST temperature-storage temperature can stain gram-negative despite having the structure of a gram-positive organism (Priest, 2015).The rpoB sequencing of 13 isolates that represented atypical growth on CVTA further confirmed our Gram stain findings; 11/13 isolates were identified as Paenibacillus peoriae or cf.peoriae (with a total of 8 unique AT; see Table 7).We identified 1 isolate as Paenibacillus jamilae, while we also identified 1 isolate as being from the genus Brevundimonas, a gram-negative organism.However, the fact that we did not identify Brevundimonas or any other gram-negative organism from APC plates or CVTA suggests that this isolate was likely a plate contaminant.

DISCUSSION
Reducing fluid milk spoilage due to psychrotolerant, aerobic spore-forming bacteria remains a major challenge, therefore, we used commingled raw milk representing multiple farms to assess the effect of different HTST pasteurization temperatures and different shelf-life storage temperatures on reducing fluid milk spoilage due to outgrowth of psychrotolerant spores.Importantly, the work we performed, described here, used a commercial raw milk supply with raw milk quality variation, which is expected in commercial systems.This allowed us to identify relatively easily implementable opportunities for HTST fluid milk shelflife extension, which should help the dairy industry better capture existing and emerging market opportunities (e.g., electronic commerce, where ESL may be of particular value; reduced shipping frequency, which can help address emerging transportation and trucking challenges).Overall, our data suggest that lower storage temperatures significantly increase fluid milk shelf life, and use of lower HTST pasteurization (e.g., around 75°C) can achieve a similar shelf life for fluid milk as compared with milk pasteurized at higher pasteurization temperatures (e.g., 85-90°C).In addition, we  found that potentially pathogenic spore-formers in the Bacillus cereus group may grow in fluid milk exposed to temperature abuse, further emphasizing the importance of maintaining stringent temperature control, including in emerging distribution channels.Finally, our data support that spore-formers classified into the genus Paenibacillus can appear atypical on CVTA, a medium typically used to detect gram-negative spoilage organisms.This highlights that atypical growth on CVTA should not be interpreted as an indicator of PPC.

Lower Storage Temperatures Significantly Increase Fluid Milk Shelf Life
Our data exhibit that milk stored at 3°C can achieve substantially longer shelf life as compared with 6.5 and 10°C, as supported by an estimated time to reach 1,000,000 cfu/mL (a threshold commonly used to indicate the likely occurrence of microbially induced sensory defects).We estimated that the 1,000,000 cfu/ mL threshold would be reached by 68 d for milk stored at 3°C, as compared with 27 and 10 d for milk stored at 6.5 and 10°C.The positive effect of storage at a low temperature (i.e., 3°C) was also supported by our findings that the milk stored at these temperatures did not show any reduction in pH or increase in particle size (which were found for milk stored at 10°C).Our results are consistent with previous findings regarding the effect of storage temperature on bacterial growth in fluid milk, including 1 study that reported a shelf life of 30, 24, and 12 d for milk stored at 5, 7, and 10°C with shelf life defined as time to 5.0 log 10 cfu/ mL (Ziyaina et al., 2018).Another study reported that lowering storage temperature by as little as 2°C will lower the relative rate of spoilage (Rowe, 1993).Lowering storage temperature has also been reported to delay chemical changes that result in flavor defects in fluid milk (Santos et al., 2003).
Subtyping and characterization data that we obtained here also revealed substantial differences in the type of spoilage organisms that were predominantly associated with different temperatures.More specifically, we found that Paenibacillus was the predominant genus isolated from milk stored at 6.5°C, whereas for milk stored at 3 and 10°C, Bacillus and Paenibacillus both represented more than 35% of isolates with neither genus having a large majority.Our findings at 3°C most likely represent residual spore populations that survived processing but showed little to no growth during storage.The distribution of Bacillus and Paenibacillus observed here is similar to previous studies that characterized raw milk spore-former populations; for example, (Masiello et al., 2014) found that Bacillus represented 71.4% (317/444) of isolates and Paenibacillus represented 26.4% (117/444) of isolates from spore-pasteurized bulk tank raw milk samples.The same study also reported that 32.9% (146/444) of isolates were Bacillus licheniformis, consistent with our findings that B. licheniformis also made-up a substantial proportion of the total number of isolates.The finding of Paenibacillus being predominant at 6.5°C is consistent with several studies that at this temperature and below, Paenibacillus spp.are the predominant genera after 14 to 21 d of storage (Huck et al., 2007;Ranieri and Boor, 2009).Our findings at 10°C most likely represent the fact that both Paenibacillus and Bacillus grow at these temperatures, which is consistent with (Trmčić et al., 2015) where it was reported that the majority of Paenibacillus spp.grew at 6 and 21°C, while the majority of Bacillus spp.did not grow at 6°C but did grow at 21°C, suggesting that some Bacillus spp.will grow at temperatures between 6 and 21°C.Our findings are also consistent with other studies where both Bacillus and Paenibacillus are found at the beginning of shelf life, whereas Paenibacillus is predominantly found at the end of shelf life (Fromm and Boor, 2004;Ranieri and Boor, 2009).
Future studies may use a metataxonomic approach to better understand population changes in fluid milk, however, the signal from dead cells may be a challenge, particularly for milk tested soon after pasteurization or for milk where there has been no bacterial growth, where the signal from dead cells is still likely to be detectable and substantial.Additionally, in matched samples (i.e., same trial, day, and storage temperature), Paenibacillus cf.peoriae was isolated from samples with lower pH, consistent with the fact that Paenibacillus peoriae has been reported to produce acid by fermenting lactose (Heyndrickx et al., 1996); another study reported that other Paenibacillus spp.(e.g., Paenibacillus glucanolyticus) also ferment lactose (Berge et al., 2002).Additionally, our physiochemical data revealed that visibly coagulated milk had particle sizes >3.00 log 10 nm, which is supported by (O'Connell and Fox, 2000) where all casein micelles in milk heated to 140°C were initially <1,000 nm, but just before and following the onset of coagulation, a large proportion of casein micelles were >1,000 nm.This reveals that particle-size measurements of fluid milk across shelf life can determine if a sample is coagulated or not without introducing bias (i.e., observer bias during visible assessment for coagulation).Future research should focus on comparing pH and particle size data along with the presence of different Paenibacillus species to understand physiochemical changes in fluid milk during shelf life.
Importantly, for the experiments discussed here, we held milk at a constant temperature over the whole duration of shelf life, which generally would not be feasible for commercial distribution pathways where fluid milk is likely stored at different temperatures at the processing facility, during distribution, at retail, and at consumer homes.This is also supported by a survey of US home refrigerators (n = 939), which reported a mean refrigeration temperature of 4°C (39.2°F) with a standard deviation of 2.66°C (4.78°F; FDA, 1999), suggesting substantial variation in home storage temperatures, which is expected to significantly affect the shelf life of fluid milk, particularly if stored at consumer homes for extended times.Hence, future studies should use the data generated here to develop predictive models that can assess the growth of spoilage organisms in fluid milk exposed to different storage temperatures; this could be a modification of initial models for microbial fluid milk spoilage (Buehler et al., 2018;Lau et al., 2022) and also could use previously reported approaches to modeling microbial growth under varying temperatures (Zwietering et al., 1994).Future modeling studies, as well as experimental validation studies of models may also consider modeling for assessing sensory defects (Hayes et al., 2002) or additional proxies for spoilage (e.g., degradation of casein as a percentage of total protein; Santos et al., 2003).However, despite the fact that shelf-life data were evaluated at a single static temperature, it is important to note that our data reported here may still be practically valid as this will allow industry to assess the possibility of producing and distributing HTST fluid milk with >42 d of shelf life, even though this would require stringent end-toend temperature control.Stringent temperature control could be feasible for vertically integrated supply chains, particularly if coupled with temperature indicators that identify milk that was exposed to temperatures above the threshold needed for this type of extreme shelf life.

Use of Lower HTST Pasteurization May Assist with Achieving ESL
Overall, our data also supported that milk pasteurized at higher temperatures is likely to have a shorter shelf life, at least at lower storage temperatures (3 and 6.5°C); whereas with statistical analyses, we only observed effects of pasteurization temperature on microbial growth at 6.5°C, numerical data also indicated ESL for milk pasteurized at 75°C and stored at a temperature of 3°C.Our data are consistent with previous findings that there is a counter-intuitive relationship between HTST temperature and microbial growth, with higher temperatures leading to greater microbial growth over shelf life (Ranieri et al., 2009;Martin et al., 2012).More specifically, in a highly controlled study where milk from a single farm was pasteurized at 1 of 4 temperatures (72.9, 77.2, 79.9, or 85.2°C) for 25 s, Ranieri et al. (2009) found a significant effect of pasteurization temperature on microbial counts over time; for example on d 21, the mean APC for milk processed at 72.9°C was 1.31 log 10 cfu lower as compared with milk pasteurized at 85.2°C where the mean APC on d 21 was 7.10 log 10 cfu/mL.By comparison, we found that milk stored at 6.5°C and pasteurized at 75°C exhibited significantly lower bacterial counts than milk pasteurized at 85°C (1.1 log 10 cfu mean difference).The fact that we did not see an effect of pasteurization temperature on microbial counts for milk stored at 10°C is noteworthy because it suggests that further research is needed on the impact of pasteurization temperature on fluid milk spoilage if product is stored at more extreme temperature abuse conditions (where different spore-formers can grow, as compared with milk stored at 6.5°C).

Potentially Pathogenic Spore-formers in the Bacillus cereus Group May Grow in Fluid Milk Exposed to Temperature Abuse (i.e., 10°C and above)
One important finding of our study was that we obtained several isolates classified into the B. cereus group from milk stored at 10°C, but we did not obtain any isolates classified into this group from milk stored at 3 or 6.5°C.More specifically, the B. cereus group isolates that we obtained from milk stored at 10°C represented 3 of the genomospecies of the B. cereus group (Carroll et al., 2020), including B. cereus s.s., B. mosaicus, and B. toyonensis.Our findings are consistent with a previous study that reported B. cereus growth to levels of >10 6 cfu/mL in milk stored at 10°C (Eneroth et al., 2001).Although the fact that we did not isolate B. cereus genomospecies from milk stored at 6.5°C suggests that the B. cereus group organisms present in the milk used here are not capable of growing in fluid milk at lower temperatures, it is also possible that B. cereus group isolates can grow at this temperature but were outcompeted by other spore-formers (e.g., Paenibacillus) in our experiments.Importantly, previous studies have also reported that B. wiedmannii, which also belongs to the B. cereus genomospecies B. mosaicus (Carroll et al., 2020), can grow at temperatures as low as 5°C in BHI broth (Miller et al., 2016).
As Bacillus cereus has previously been isolated from milk products (Stenfors Arnesen et al., 2008;Bennett et al., 2013), this emphasizes the importance of a properly managed cold chain.Further risk assessments that characterize the likelihood of human exposure and illness due to B. cereus presence and growth in HTST milk may also be valuable and should be informed by improved time-temperature data for HTST milk along the supply chain (including novel distribution pathways such as electronic commerce) and improved data on Bacillus cereus growth in milk focusing on growth of different strains and subtypes at temperatures between 5 and 10°C, and determining the minimum temperature for growth and cereulide toxin production in fluid milk for different Bacillus cereus group strains.

Atypical Growth of Spore-formers on CVTA May Lead to Misidentification of Fluid Milk Quality Issues
We characterized several bacterial isolates that represented atypical colonies on the gram-negative selective medium CVTA, that has been used previously to identify fluid milk with PPC (Alles et al., 2018;Reichler et al., 2021), and we identified the majority of these isolates as Paenibacillus peoriae and cf.peoriae.This indicates that gram-positive aerobic spore-formers could be mischaracterized as gram-negative contamination, particularly by inexperienced laboratory analysts that would not recognize atypical growth patterns associated with spore-formers.This mischaracterization could lead to improperly targeted, thus ineffective intervention strategies aimed at improving quality.A recently described method for detecting gram-negative bacteria in fluid milk using Petrifilms (Rojas et al., 2020), may provide the industry with a way to enumerate gramnegative bacteria with less chance of misclassification while maintaining convenience.Overall, our findings support the continued need for development of better laboratory methods that can help support the control of spore-formers in the HTST fluid milk supply chain.This issue is further illustrated by our preliminary findings that LPC, a commonly used raw milk test, does not appear to be suitable for sensitive detection and isolation of the specific psychrotolerant spore-formers that have been shown here and elsewhere to be responsible for fluid milk spoilage.

Figure 1 .
Figure 1.Scatter plots of aerobic plate count (APC) data across shelf life for milk stored at each storage temperature [3°C (n = 3), 6.5°C (n = 4), or 10°C (n = 3)] and pasteurized at each HTST temperature (75, 85, or 90°C).Storage temperatures are separated into 3 panels with labels on the right side, and HTST temperatures are separated by color.Individual points in the figure represent bacterial counts of samples from individual trials.The function geom_smooth from ggplot2 was used to fit the points with a regression line using the LOESS (locally estimated scatterplot smoothing) method, where the shaded area in colors corresponding to the HTST temperatures represents the 95% CI of the fitted line.
Figure 2. Mean net log 10 growth of HTST temperatures (75, 85, or 90°C) for each storage temperature [3°C (n = 3), 6.5°C (n = 4), or 10°C (n = 3)] on d 7 and 14.Error bars represent 1 standard deviation from the mean.***P < 0.001 based on Tukey posthoc tests performed following an ANOVA that found storage temperature to be significant (as described in the Results section).

Table 1 .
Lott et al.: STORAGE AND HTST TEMPERATURE EFFECT ON MILK Mean bacterial concentration for microbiological tests of raw milk samples (n = 4) collected from a processing facility in Texas before pasteurization

Table 2 .
Lott et al.: STORAGE AND HTST TEMPERATURE EFFECT ON MILK Average predicted time (d), based on primary growth models, to 20,000 and 1,000,000 cfu/mL for milk stored at 3, 6.5, or 10°C and pasteurized at 75, 85, or 90°C

Table 3 .
Lott et al.: STORAGE AND HTST TEMPERATURE EFFECT ON MILKThe rpoB allelic types of spore-forming bacteria isolated from fluid milk shelf-life samples with low pH values (i.e., <6.6) and matched fluid milk shelf-life samples with no pH drop, differing only by pasteurization temperatures 5Sample was measured on d 28, but not on d 35 due to laboratory error.

Table 5 .
Lott et al.: STORAGE AND HTST TEMPERATURE EFFECT ON MILKLott et al.: STORAGE AND HTST TEMPERATURE EFFECT ON MILK Genera, species, and allelic type frequency of the 96 representative isolates selected from aerobic plate count media from the 2 shelf-life samples with the highest aerobic plate counts for each unique combination of trial-HTST temperature-storage temperature Lott et al.: STORAGE AND HTST TEMPERATURE EFFECT ON MILK

Table 6 .
Genus, species, and rpoB allelic types of spore-forming bacteria isolated from laboratory-pasteurized (63°C for 30 min) raw milk and milk plated immediately after pasteurization at 75, 85, or 90°C 1 .denotes sensu stricto; s.l.denotes sensu lato; sp.denotes species designated to an isolate for which genus, but no species, could be assigned; Bacillus sp. 2 mojavensis halotolerans designates an AT that has 2 rpoB type strains that are closest relatives: B. mojavensis and B. halotolerans.

Table 7 .
Lott et al.: STORAGE AND HTST TEMPERATURE EFFECT ON MILK Genera, species, and allelic type frequency of atypical isolates selected from crystal violet tetrazolium agar (CVTA) and successfully characterized by rpoB sequencing Conferre: isolate matching 16S rDNA sequence (i.e., 97.0%-98.0%sequence similarity) of more than 1 type strain from family Bacillaceae.sp.denotes species designated to an isolate for which genus, but no species, could be assigned.
2No atypical isolates were characterized from CVTA for samples stored at 3°C.