APPLICATION OF RAPID DIAGNOSTICS AND NEXT-GENERATION 

AMPLICON SEQUENCING TO ADDRESS PRODUCT QUALITY IN THE FOOD 

INDUSTRY 

 

 

 

 

 

 

 

A Dissertation 

Presented to the Faculty of the Graduate School 

of Cornell University 

In Partial Fulfillment of the Requirements for the Degree of 

Doctor of Philosophy 

 

 

 

 

 

by 

Jonathan Harold Sogin 

May 2023 



 

© 2023 Jonathan Harold Sogin



 

APPLICATION OF RAPID DIAGNOSTICS AND NEXT-GENERATION 

AMPLICON SEQUENCING TO ADDRESS PRODUCT QUALITY IN THE FOOD 

INDUSTRY 

 

Jonathan Harold Sogin, Ph.D. 

Cornell University 2023 

 

The cost of foodborne illness and food loss to the United States (US) economy is 

approximately $300 billion per year. This number monetarily quantifies the almost 10 

million individuals who contract a foodborne illness and 141 trillion kcal of disposed 

food, enough to feed half the US population, each year. Despite the US having one of 

the safest and highest quality food supplies in the world, these figures illustrate the 

need to improve upon food safety and quality within the US supply chain and globally. 

Increasing consumer desires for minimally processed and clean label foods contributes 

to the challenge of addressing this need, as proven processing methods and food 

additives are being substituted or otherwise avoided/removed. To improve food safety 

and quality, food producers are seeking novel processing methods and additives to 

eliminate or kill foodborne pathogens and prevent the growth of spoilage 

microorganisms. Additional strategies to accomplish these goals include 

improvements to hygiene management within food processing facilities, and 

optimization of fermentation conditions or cultures as determined by an improved 

understanding of the microbial communities involved in commercial food 

fermentations. This dissertation presents examples of these strategies within the 



 

context of tools usage, namely adenosine triphosphate (ATP) monitoring of a food 

processing environment and next-generation amplicon sequencing of commercial 

tempeh and kombucha. The following studies demonstrate that the application of these 

tools is useful within the context of improving food product quality.  

 

 

 
 



 

BIOGRAPHICAL SKETCH 

 Jonathan Harold Sogin was born in Minneapolis, MN to Emily Elizabeth Duke 

and Daniel Aaron Sogin, and grew up just across the Mississippi River in St. Paul, 

MN. Jonathan received his Bachelor of Science degree in Food Science from the 

University of Wisconsin–Madison in 2018, where he developed a keen interest in 

Food Microbiology. He began his studies at Cornell in the fall of 2018 and has since 

worked on a variety of projects related to fermentation, food quality, and food safety. 

He received his Master of Science degree from Cornell University in the spring of 

2021 during the completion of this Ph.D. 

 



iv 

To my family, friends, and cats, who supported me through thick and thin, 

and 

To all the individuals who support the operations of the University, which allowed me 
to learn and do work throughout my studies at Cornell. 

 



v 

ACKNOWLEDGMENTS 

“It takes a village to raise a child Ph.D.”  

 I can say without a doubt I would not have finished this degree without the 

patience, support, and encouragement of my parents, Emily Duke and Daniel Sogin; 

sisters, Hallie and Jennifer Sogin; partner, Caitlyn Berk; friends, Caitlin Carmody and 

Amanyi Richardson; and advisor, Randy Worobo. Each of these individuals, and 

collectively the core of my ‘village’, has celebrated with me in good times and been a 

refuge for me in tough times. 

 Although it’s somewhat cliché to say that my Ph.D. progressed atypically, I 

feel justified in this assessment. First came COVID, and then came ACL 

reconstruction surgery with a side of complications which ultimately landed me on a 

continuous self-administered intravenous infusion of 12 g ampicillin daily for 30 days 

– for reference, that’s a boatload of ampicillin. Luckily, I have no lasting 

complications from either of these impositions, but in those moments and the many 

months that followed, I considered tapping out of this degree many many times. 

Luckily, I had a good village. 

 In addition to the core of my village, there are many people at Cornell who 

helped me along the way, including Linda Cote and Peter Schweitzer at the Cornell 

Institute of Biotechnology; the other members of my special committee Patrick 

Gibney and Daniel Buckley; and the other members of the Worobo lab, who were 

almost always in good spirits and created an environment that was supportive and 

collaborative. Each of these individuals were genuinely nice and helpful, even when I 



vi 

didn’t know what I was doing – and never made me feel bad about it. A special 

shoutout goes to Dwayne Bershaw, as I had an incredibly fun time as his teaching 

assistant for the beers and distilled spirits lectures and labs, and would otherwise not 

have spent any time in the Cornell Food Science Department’s teaching winery. 

 I’d also like to acknowledge the staff at Cornell, whose names I largely do not 

know, but whose work keeps the place running. These are the individuals that keep the 

lights on and the water flowing, fix things when they’re broken, salt the sidewalks in 

the winters, and do the many other things that keep Cornell operational, even through 

a pandemic. As much as I would not have reached this point without my core village, I 

would not have reached this point without the hard and often unappreciated work of 

these individuals. 

 Finally, I’d like to thank Randy for his unconditional academic support and 

appreciation for personal well-being. There are a lot of things that I could thank him 

for, but I will sum it up with this: he is a nice person, and he cares about people. I am 

extremely grateful for his advising over the past five years and hope that if I ever find 

myself in a similar position that I can pass along the same selflessness and support he 

has shown me.   



vii 

TABLE OF CONTENTS 

BIOGRAPHICAL SKETCH III 

ACKNOWLEDGMENTS V 

TABLE OF CONTENTS VII 

LIST OF FIGURES VIII 

LIST OF TABLES X 

CHAPTER 1 1 

A BRIEF CONTEXTUALIZATION OF FOOD SAFETY AND QUALITY 
MANAGEMENT WITHIN THE UNITED STATES ECONOMY AND 
CONSUMER PREFERENCES FOR MINIMALLY PROCESSED FOODS  

REFERENCES 7 

CHAPTER 2 10 

IMPLEMENTATION OF ATP AND MICROBIAL INDICATOR TESTING 
FOR HYGIENE MONITORING IN A TOFU PRODUCTION FACILITY 
IMPROVES PRODUCT QUALITY AND HYGIENIC CONDITIONS OF 
FOOD CONTACT SURFACES: A CASE STUDY  

REFERENCES 37 

CHAPTER 3 41 

NEXT-GENERATION AMPLICON SEQUENCING ANALYSIS OF 
COMMERCIAL TEMPEH DURING FERMENTATION  

REFERENCES 73 

APPENDIX 78 

CHAPTER 4 82 

MICROBIOLOGICAL AND PHYSIOCHEMICAL ANALYSIS OF 
COMMERCIAL KOMBUCHA PRODUCTS SOLD IN US RETAIL MARKETS 
REVEALS BRAND-DISTINCT PRODUCT PROFILES AND FURTHER 
INDICATES THE ROLE OF LACTIC ACID BACTERIA IN COMMERCIAL 
KOMBUCHA PRODUCTION  

REFERENCES 119 

APPENDIX 125 

CHAPTER 5 130 

CONCLUDING THOUGHTS AND FUTURE DIRECTIONS  

REFERENCES 133 

 



viii 

LIST OF FIGURES 

Figure 2-1. Conceptual overview of process to implement hygiene management 17 

Figure 2-2. ATP swab failure rate over time 23 

Figure 2-3. ATP and microbial swab failure rates by zone 25 

Figure 2-4. Microbial loads of products before pasteurization 29 

Figure 2-5. Microbial loads of products after pasteurization 29 

 

Figure 3-1. Bacterial Family Relative Abundance 51 

Figure 3-2. Bacterial alpha diversity comparisons 52 

Figure 3-3. Bacterial Bray-Curtis NMDS ordination 54 

Figure 3-4. Bacterial species PERMANOVA coefficients for product type 56 

Figure 3-5. Core bacterial species and differential abundance analysis 57 

Figure 3-6. Fungal Family Relative Abundance 61 

Figure 3-7. Fungal alpha diversity comparisons 62 

Figure 3-8. Fungal Bray-Curtis NMDS ordination 63 

Figure 3-9. Fungal species PERMANOVA coefficients for product type 64 

Figure 3-10. Core fungal species and differential abundance analysis 65 

 

Figure 4-1. Chemical properties of kombucha samples 93 

Figure 4-2. Chemical Mahalanobis NMDS ordination 94 

Figure 4-3. Microbial loads of Kombucha samples plated on various media 95 

Figure 4-4. Bacterial genera relative abundances 99 

Figure 4-5. Bacterial Bray-Curtis NMDS ordination 100 



ix 

Figure 4-6. Bacterial species PERMANOVA coefficients lactic acid content 102 

Figure 4-7. Correlation between lactic acid and Lactobacillales relative abundance 103 

Figure 4-8. Differentially abundant bacterial species between kombucha brands 105 

Figure 4-9. Fungal species relative abundances 107 

Figure 4-10. Fungal Bray-Curtis NMDS ordination 108 

Figure 4-11. Differentially abundant fungal species between kombucha brands 110 

Figure 4-12. Bacterial and fungal species cooccurrence network in kombucha 113 

Appendix Figure 4-13. Correlation between ethanol content and relative sugar content 
versus the label 125 

Appendix Figure 4-14. Correlation between ethanol content and pH 126 

Appendix Figure 4-15. Fungal species relative abundance when using forward and 
reverse reads 126 

Appendix Figure 4-16. Results of MaAsLin2 analysis of bacterial species associations 
with chemical and microbial load variables 127 

Appendix Figure 4-17. Results of MaAsLin2 analysis of fungal species associations 
with chemical and microbial load variables 128 

Appendix Figure 4-18. Differentially abundant bacterial species between kombucha 
samples due to the inclusion of ginger 129 

 

  



x 

LIST OF TABLES 

Table 2-1. Swab failure rate by site 27 

 

Table 3-1. Significantly differentially abundance core bacterial species 59 

Table 3-2. Significantly differentially abundance core fungal species 67 

Appendix Table 3-3. 16S primers used for Illumina sequencing 1st step PCR 78 

Appendix Table 3-4. ITS primers used for Illumina sequencing 1st step PCR 79 

Appendix Table 3-5. Alpha diversity metrics significance table 80 

 

 
 



1 

CHAPTER 1 

A BRIEF CONTEXTUALIZATION OF FOOD SAFETY AND QUALITY 
MANAGEMENT WITHIN THE UNITED STATES ECONOMY AND 

CONSUMER PREFERENCES FOR MINIMALLY PROCESSED FOODS 

The Economics of Foodborne Illness and Food Loss 

 Agriculture, food, and related industries contributed $1.25 trillion and 21.1 

million jobs to the United States (US) economy in 2021, representing 5.4% of gross 

domestic product and 11% of the US workforce (1). Despite the scale of these 

industries and the long distances foods travel from farm to fork, which averages 

approximately 1020 miles (1640 km) (2), the US has one of the highest quality and 

safest food supplies in the world (3rd in Quality and Safety, 2022, Global Food 

Security Index) (3). The two countries with food supplies that ranked higher than the 

US for quality and safety in 2022 were Canada and Denmark, first and second (3), 

both which have populations at least 85% lower than the US (4). Food safety 

regulations that are reinforced by strong food safety culture (5) are primary 

contributors to the US’ success. 

 Federal legislation has broadly required the safe production of foods in the US 

since congress enacted the Pure Food and Drug Act and Federal Meat Inspection Act 

in 1906 (6). Since those initial acts, many amendments and additional laws have been 

enacted, most recently with the passage of the Food Safety Modernization Act 

(FSMA), which was signed into law by President Barak Obama on January 4th, 2011, 

(7). The two major regulatory bodies responsible for food safety enforcement in the 

US are the Food Safety and Inspection Service (FSIS) and the Food and Drug 



2 

Administration (FDA). The FSIS regulates products containing meat, poultry, and 

eggs, which constitutes 10-20% of the US food supply, whereas the FDA regulates 

everything else, which constitutes 80-90% of the US food supply (8). 

 Despite the US’ relatively safe food supply, a 2011 study estimated that 9.4 

million illnesses, 56 thousand hospitalizations, and 1,400 deaths occur every year due 

to food-acquired bacterial, viral, and parasitic microorganisms (9). Most illnesses are 

attributed to viruses (5.5 million vs 3.6 million bacterial), whereas most 

hospitalizations and deaths are attributed to bacteria (9). The most common food 

source of bacterial pathogens is land animals, and the most common source of viral 

pathogens is plants (10). Monetary loss per illness ranges from $208 to $7.0 million, 

and the total annual cost of all foodborne illnesses is $36-56 billion (2015 estimates) 

(11, 12). Furthermore, the cost to a food producer implicated in selling products that 

caused a foodborne illness outbreak can reach hundreds of millions of dollars (13). 

 Quality, in addition to safety, is critically important to food producers and 

consumers. If foods are deemed too poor quality to sell, which can occur for a variety 

of reasons, they are disposed of by landfilling, composting, anaerobic digestion, 

incineration, or other means (14). The disposal (loss) of food at the retail and 

consumer levels was 31% in the US in 2010, which equated to 67 million tons of food 

worth $162 billion (15). The energy and nutrient contents of food loss are substantial, 

representing 2% of the annual US energy consumption (16) and 141 trillion kcal 

(enough to feed 194 million people for a year at 2000 kcal/day) in 2010 (15). The top 

three categories of foods lost in 2010 were meat, poultry, and fish (30%), vegetables 

(19%), and dairy (17%) (15). 



3 

 The combined economic loss due to foodborne illnesses and food loss is 

almost $300 billion every year (based on est. ~$56 b due to illnesses in 2015, and 

$162 b due to food loss in 2010, adjusted for inflation with consumer price index) (11, 

12, 15). As the above data illustrates, individual companies, industries, and the 

government should be economically motivated to improve food safety and quality. 

Consumer Desires for Healthy and Minimally Processed Foods 

  Food processing is broadly defined as the act of preparing or transforming a 

raw agricultural commodity for human consumption, which may improve food safety, 

shelf-life, availability, convenience, affordability, and nutrition (17, 18). It is not a 

uniquely human trait, as evidenced by the storage of nuts by grey squirrels (19) or the 

use of stone tools by Burmese long-tailed macaques to process marine prey (20), but 

specifically cooking food is unique to humans and may be biologically obligatory 

(21). Agriculture and food processing were fundamental to the growth of the world’s 

population and civilizations and will remain fundamental to sustain adequate nutrition 

to a growing world population (22). 

 Counterproductive to improving food safety and quality, consumers are 

increasingly interested in purchasing ‘clean label’ or minimally processed foods (23). 

There is no regulatory definition of clean label, nor is there a consensus as to what 

clean label means (24). Clean label products are often characterized by co-occurring 

themes of sustainability, nutrition equity and justice, and health and wellness (23–25), 

but a commonality of most clean label products is an ingredients label with 

components that an average consumer can understand and do not sound ‘chemical-ly’. 

 Reducing or eliminating chemically-sounding functional ingredients can 



4 

worsen sensory qualities of products, but those negative changes may be overlooked 

by consumers because the product is clean label (26). This same logic cannot be 

applied to food safety due to federal regulations and because consumers highly value 

food safety. Food producers struggle to substitute known and effective food 

antimicrobials such as sodium benzoate and sodium nitrite to prevent the growth of 

pathogenic and spoilage causing microorganisms (27) but may attempt to do so with 

naturally occurring antimicrobials (28). 

 Related to consumer perceptions of minimally processed foods and food 

producers’ need to reduce or eliminate the presence of microorganisms that may cause 

disease or spoilage, food producers are increasingly utilizing a variety of non-thermal 

processing methods. Such methods include high pressure processing, sterile filtration, 

ultraviolet light treatment, use of protective cultures, and others (29, 30). These 

methods are varyingly effective depending on the food matrix and target 

microorganisms but can provide improved nutritional and sensory characteristics 

compared to thermal processing (30). Although consumers may be averse from buying 

products explicitly marketed as being produced using such technologies because they 

are unfamiliar with them (30), they may be more likely to buy such products if they 

are marketed as ‘non-thermally treated’, with the implication that they are more 

minimally processed and fresher. 

Tools Used to Investigate the Microbiological Landscapes of Food and Food 
Processing Facilities 

 Maintaining the sterility of food processing environments is impossible. 

Microorganisms can enter food processing facilities through three broad channels: raw 



5 

materials, personnel, and the surrounding facility environment. Each of these channels 

is a likely source of pathogenic and spoilage-causing microorganisms. As most 

processing facilities contain an abundance of food residues and moisture, 

microorganisms will grow and may form resident communities via biofilm formation 

in drains, hard to clean portions of equipment, and building crevasses. While typically 

avoided for most food products, such biofilms may be an integral requirement for 

complex fermented foods such as cheese (31). It is the job of facility personnel to 

manage the relative risks of product contamination and ensure that products will not 

contain pathogenic microorganisms or other microorganisms that may cause 

premature product spoilage. 

 Facility personnel accomplish this through the implementation of the hazard 

analysis and critical control point system (HACCP), and a variety of supporting 

programs including standard sanitation operating procedures (SSOPs), environmental 

monitoring, raw materials sourcing specifications, and others. Depending on what type 

of product is being manufactured, a microbial load reduction may be required to 

remove any pertinent pathogens of concern. Although food producers are not 

mandated to control spoilage microorganisms by FSIS or FDA, some of the methods 

to reduce or remove pathogens may also reduce or remove relevant spoilage 

microorganisms. Other times, additional processing treatments or food additives may 

be required to prevent product spoilage. As previously mentioned, it is more difficult 

to do this for clean label or minimally processed products. 

 A variety of tools are used by food manufacturers to track the microorganisms 

that are present in food production facilities and foods. These include biomarker tools, 



6 

culture-based tools, and molecular sequencing-based tools. This dissertation presents 

examples managing and investigating the quality of foods, specifically tofu, tempeh, 

and kombucha using a biomarker tool, adenosine triphosphate luminescence (ATP) 

testing, and a molecular sequencing-based tool, next-generation amplicon sequencing. 

 The choice of tools a producer uses is dictated by a slew of factors which 

includes the nature of the facility, the product, the processing steps, and the desired 

information yielded by the test. Some tests such as ATP testing are used for 

monitoring, whereas next-generation amplicon sequencing may be used to address 

specific quality attributes that cannot be solved through cheaper and easier means like 

culture-based analyses or single-strain polymerase chain reaction (PCR) followed by 

Sanger Sequencing. The benefits of using ATP testing and next-generation amplicon 

sequencing are discussed throughout this dissertation. 

 

  



7 

REFERENCES 

1.  Kassel K, Lanigan T, Martin A, Michael-Midkiff J, Russell D, Ruth T, Sanguinett C, 
Smits J. 2023. Selected Charts from Ag and Food Statistics: Charting the Essentials, 
February 2023. Economic Research Service 111. 

2.  Weber CL, Matthews HS. 2008. Food-miles and the relative climate impacts of food 
choices in the United States. Environ Sci Technol 42:3508–3513. 

3.   2022. Global Food Security Index 2022. Economist Impact. 

4.   2023. Country Comparisons - Population. The CIA World Factbook. 
https://www.cia.gov/the-world-factbook/field/population/country-comparison. 
Retrieved 4 April 2023. 

5.  Powell DA, Jacob CJ, Chapman BJ. 2011. Enhancing food safety culture to reduce 
rates of foodborne illness. Food Control 22:817–822. 

6.  Barkan ID. 1984. Industry Invites Regulation: The Passage of the Pure Food and Drug 
Act of 1906. Am J Public Health 75:18–26. 

7.  Strauss DM. 2011. An Analysis of the FDA Food Safety Modernization Act: 
Protection for Consumers and Boon for Business. Drug Law Journal 66:353–376. 

8.  Johnson R. 2016. The Federal Food Safety System: A Primer. Congressional Research 
Service RS22600. 

9.  Scallan E, Hoekstra RM, Angulo FJ, Tauxe R V., Widdowson M-A, Roy SL, Jones 
JL, Griffin PM. 2011. Foodborne Illness Acquired in the United States—Major 
Pathogens. Emerg Infect Dis 17:7–15. 

10.  Painter JA, Hoekstra RM, Ayers T, Tauxe R V., Braden CR, Angulo FJ, Griffin PM. 
2013. Attribution of Foodborne Illnesses, Hospitalizations, and Deaths to Food 
Commodities by using Outbreak Data, United States, 1998–2008. Emerg Infect Dis 
19:407–415. 

11.  Minor T, Lasher A, Klontz K, Brown B, Nardinelli C, Zorn D. 2015. The Per Case and 
Total Annual Costs of Foodborne Illness in the United States. Risk Analysis 35:1125–
1139. 

12.  Scharff RL. 2015. State Estimates for the Annual Cost of Foodborne Illness. J Food 
Prot 78:1064–1071. 

13.  Hussain M, Dawson C. 2013. Economic Impact of Food Safety Outbreaks on Food 
Businesses. Foods 2:585–589. 



8 

14.  Badgett A, Milbrandt A. 2021. Food waste disposal and utilization in the United 
States: A spatial cost benefit analysis. J Clean Prod 314:128057. 

15.  Buzby JC, Wells HF, Hyman J. 2014. The Estimated Amount, Value, and Calories of 
Postharvest Food Losses at the Retail and Consumer Levels in the United States. 
Economic Research Service 121. 

16.  Cuéllar AD, Webber ME. 2010. Wasted Food, Wasted Energy: The Embedded Energy 
in Food Waste in the United States. Environ Sci Technol 44:6464–6469. 

17.   Get the Facts: Food Processing. Institute of Food Technologists. https://www.ift.org/-
/media/policy-advocacy/ift-comments/efsa/ift-food-processing-toolkit.pdf. Retrieved 3 
April 2023. 

18.  van Boekel M, Fogliano V, Pellegrini N, Stanton C, Scholz G, Lalljie S, Somoza V, 
Knorr D, Jasti PR, Eisenbrand G. 2010. A review on the beneficial aspects of food 
processing. Mol Nutr Food Res 54:1215–1247. 

19.  Hopewell LJ, Leaver LA. 2008. Evidence of Social Influences on Cache-Making by 
Grey Squirrels (Sciurus carolinensis). Ethology 114:1061–1068. 

20.  Gumert MD, Malaivijitnond S. 2012. Marine prey processed with stone tools by 
burmese long-tailed macaques (Macaca fascicularis aurea) in intertidal habitats. Am J 
Phys Anthropol 149:447–457. 

21.  Wrangham R, Conklin-Brittain N. 2003. ‘Cooking as a biological trait.’ Comp 
Biochem Physiol A Mol Integr Physiol 136:35–46. 

22.  Floros JD, Newsome R, Fisher W, Barbosa-Cánovas G V., Chen H, Dunne CP, 
German JB, Hall RL, Heldman DR, Karwe M V., Knabel SJ, Labuza TP, Lund DB, 
Newell-McGloughlin M, Robinson JL, Sebranek JG, Shewfelt RL, Tracy WF, Weaver 
CM, Ziegler GR. 2010. Feeding the World Today and Tomorrow: The Importance of 
Food Science and Technology. Compr Rev Food Sci Food Saf 9:572–599. 

23.  Dornblaser L. 2020. Clean Label: Shifting consumer perceptions. Mintel. 

24.  Asioli D, Aschemann-Witzel J, Caputo V, Vecchio R, Annunziata A, Næs T, Varela 
P. 2017. Making sense of the “clean label” trends: A review of consumer food choice 
behavior and discussion of industry implications. Food Research International 99:58–
71. 

25.  Bartelme MZ, Mattucci S, Zegler J, Beckett A, Koyenikan ayisha, Hong TH, Faulkner 
D, Li D, Henry C, Gilsogamo AP. 2022. 2023 Global Food & Drink Trends. Mintel. 

26.  Maruyama S, Lim J, Streletskaya NA. 2021. Clean Label Trade-Offs: A Case Study of 
Plain Yogurt. Front Nutr 8:704473. 



9 

27.  Erickson MC, Doyle MP. 2017. The Challenges of Eliminating or Substituting 
Antimicrobial Preservatives in Foods. Annu Rev Food Sci Technol 8:371–390. 

28.  Juneja VK, Dwivedi HP, Yan X. 2012. Novel Natural Food Antimicrobials. Annu Rev 
Food Sci Technol 3:381–403. 

29.  Jadhav HB, Annapure US, Deshmukh RR. 2021. Non-thermal Technologies for Food 
Processing. Front Nutr 8:657090. 

30.  dos Santos Rocha C, Magnani M, de Paiva Anciens Ramos GL, Bezerril FF, Freitas 
MQ, Cruz AG, Pimentel TC. 2022. Emerging technologies in food processing: 
impacts on sensory characteristics and consumer perception. Curr Opin Food Sci 
47:100892. 

31.  Bokulich NA, Mills DA. 2013. Facility-Specific “House” Microbiome Drives 
Microbial Landscapes of Artisan Cheesemaking Plants. Appl Environ Microbiol 
79:5214–5223. 

 

  



10 

CHAPTER 2 

IMPLEMENTATION OF ATP AND MICROBIAL INDICATOR TESTING 
FOR HYGIENE MONITORING IN A TOFU PRODUCTION FACILITY 
IMPROVES PRODUCT QUALITY AND HYGIENIC CONDITIONS OF 

FOOD CONTACT SURFACES: A CASE STUDY 

ABSTRACT 

 Rapid ATP testing and microbiological enumeration are two common methods 

to monitor the effectiveness of cleaning and sanitation in the food industry. In this 

study, ATP testing and microbiological enumeration were implemented at a tofu 

production facility with the goal of improving cleaning practices and overall plant 

hygiene. Results from ATP monitoring were used to target areas of the production 

environment needing additional cleaning. ATP results were verified by 

microbiological enumeration of aerobic microorganisms, lactic acid bacteria, and 

yeasts and molds. Products from the production line were enumerated for the same 

microorganisms to determine if there was an impact on product quality. After the 

implementation of ATP monitoring and targeted cleaning, there was a statistically 

lower proportion of swabs that failed to meet established sanitary requirements for 

ATP, aerobic microorganisms, and lactic acid bacteria (p < 0.05), but not for yeasts 

and molds. ATP swabs and microbiological enumeration agreed on site hygiene 75.1% 

(72.3-77.7%, 95% CI) of the time. Product data indicated that unpasteurized finished 

products contained a statistically lower microbial load of the three groups of 

organisms following implementation of targeted cleaning (p < 0.05).  

  



11 

IMPORTANCE 

 Cleaning and sanitation are critical to maintaining safe and high-quality food 

production. Monitoring these activities is important to ensure proper execution of 

procedure and to assure compliance with regulatory guidelines. The results from 

monitoring activities can direct targeted cleaning of areas with higher risk of 

contamination from foodstuffs and microorganisms. The results of this study show 

that ATP monitoring and microbiological enumeration are useful tools to verify and 

improve the efficacy of cleaning and sanitation practices, which can have a positive 

impact on both plant hygiene and product quality. However, testing regimes and 

critical parameters will vary based on the product and facility. 

 

 

 

 

 

 

 

 

This study was published in Applied and Environmental Microbiology: 

Sogin JH, Lopez-Velasco G, Yordem B, Lingle CK, David JM, Çobo M, Worobo 
RW. 2021. Implementation of ATP and Microbial Indicator Testing for Hygiene 
Monitoring in a Tofu Production Facility Improves Product Quality and Hygienic 
Conditions of Food Contact Surfaces: a Case Study. Appl Environ Microbiol 
87:e02278. https://doi.org/10.1128/AEM.02278-20. 
  



12 

INTRODUCTION 

 Many factors contribute to the microbial ecology of a food processing plant 

including the product, processing steps, and scale of the operation. However, all food 

processing facilities regardless of product or size must maintain a sanitary processing 

environment (9 CFR § 416, 21 CFR §117) (1, 2). Companies execute Sanitary 

Standard Operating Procedures (SSOPs) to reduce and control for the infiltration, 

movement, and growth of microorganisms in a plant (e.g. use of personal protective 

equipment, water treatment, air filtering, pest exclusion), but these programs do not 

and cannot completely prevent microorganisms from entering the processing 

environment. A major component of SSOPs involves cleaning and sanitizing the 

processing line and surrounding areas. Both the Food & Drug Administration and 

United States Department of Agriculture require that food-contact and non-food-

contact surfaces be cleaned and sanitized as frequently as necessary to prevent 

contamination of products (9 CFR § 416.4, 21 CFR §117.35) (3, 4). These rules exist 

to prevent product contamination with hazards – “any biological, chemical (including 

radiological), or physical agent that has the potential to cause illness or injury” (21 

CFR §117.3, see 9 CFR § 417.1 for equivalent USDA definition) (5, 6). However, 

cleaning and sanitation (as well as other SSOPs) also control for spoilage 

microorganisms. 

 Cleaning and sanitation involve removal of material build-up from surfaces 

and subsequent application of a substance to reduce target microorganisms to an 

acceptable level. Cleaning should specifically remove proteins, carbohydrates, FOG 

(fats, oils, and grease), minerals, and water (7). These substances are growth substrates 



13 

for various pathogenic and non-pathogenic microorganisms and quench the 

effectiveness of sanitizers by serving as off-target substrates for sanitizing compounds. 

Thus, sanitation is only effective if adequate cleaning precedes it. Cleaning and 

sanitation occur before and after production runs by a qualified and trained team; 

efforts are prioritized based on production zones, distinguished by likelihood to 

contaminate food. Zone 1 corresponds to food-contact surfaces (highest risk of 

introducing contamination), Zone 2 corresponds to non-food-contact surfaces near 

Zone 1, Zone 3 corresponds to more distant non-food-contact surfaces than Zone 2, 

and Zone 4 corresponds to surfaces outside the production room (lowest risk of 

introducing contamination) (8). An individual on the cleaning and sanitation team 

utilizes visual inspection and timers to monitor progress and determine the endpoint of 

cleaning and sanitation activities. However, visual inspection has limited 

effectiveness, as food residues and/or microorganisms may be present even on a 

visually clean surface (9). Visual inspection should be an expectation, but companies 

use other verification methodologies to verify the efficacy of cleaning and sanitation.  

 Hygiene monitoring is the regular, systematic, and site-specific testing of a 

processing plant for an attribute relevant to the processing environment to verify the 

efficacy of a sanitation program or overall cleanliness of the plant (10). Hygiene 

monitoring differs from environmental monitoring in that it does not identify specific 

organisms, namely environmental pathogens, but rather detects broad groups of 

microorganisms and/or food residues relevant to the processing environment, thus 

enabling the identification of sites that can harbor or support the growth of 

microorganisms due to the presence of food residues. Despite its non-specific nature, 



14 

hygiene monitoring is an important activity that supports the effectiveness of both 

food safety and food quality programs. Two common hygiene monitoring methods are 

culture-based quantification assays and rapid indicator-testing for biological markers. 

 Culture-based enumeration provides quantification of microorganisms at a site, 

but is biased due to sampling procedure, processing, and growth conditions (e.g., 

media composition, growth temperature, growth time) (11). The use of different 

growth media for hygiene monitoring selects for various groups of microorganisms, 

but not individual taxonomic clades. Unlike environmental monitoring, hygiene 

monitoring seldom integrates sequencing tools (i.e., Sanger sequencing and next-

generation platforms) for regular use; nonetheless, and beneficially, culture-based 

enumeration allows for the isolation and identification of suspected spoilage and/or 

residential microorganisms. Because culture-based methods rely on the growth of 

microorganisms, results are obtained after a minimum of 24-72 hours, depending on 

the group of microorganisms in question (12, 13). Thus, enumeration-based 

monitoring tools lead to strictly reactive, but not real-time, solutions to deviations 

from cleaning and sanitation, delaying implementation of corrective actions when 

improper cleaning has been performed. Further, enumeration requires use of a 

dedicated laboratory space, trained personnel to process samples, and the purchase of 

several additional laboratory materials (e.g., growth media, incubators, pipettes), so 

only companies able to regularly pay for and manage or outsource these activities can 

effectively utilize culture-based monitoring tools.  

 Conversely, the use of rapid tests can be used to detect for the presence of 

biological markers resulting from microbial metabolism or food. Rapid tests are easy 



15 

to conduct, fast, portable, and can direct real-time improvements to cleaning and 

sanitation regimes. Unlike culture-based monitoring assays, rapid tools do not require 

a dedicated laboratory space nor extensively trained personnel to conduct the tests. 

One of the most common rapid methodologies tests for the presence of adenosine 

triphosphate (ATP), which is a molecule produced by all living cells, and detected via 

an enzymatic assay utilizing a luciferin/luciferase complex for light production (14). 

Light production can be measured as relative light units (RLU), thus converting ATP 

levels at individual sites to numerical values. The amount of light produced is directly 

proportional to the level of ATP, and therefore used to assess cleanliness of a site. 

Compared to culture-based methods, ATP detection has no sensitivity for specific 

groups of microbes (spoilage and/or pathogenic) because the structure of ATP is 

identical in all cells; ATP in food processing environments is derived from 

microorganisms, food residues, and other organic matter (15). Therefore, it can help 

identify niche sites where food is not efficiently removed during cleaning and 

sanitation (15). ATP detection is a more rapid (providing results in minutes) and 

accessible tool to a greater range of food production facilities for cleaning verification 

and hygiene monitoring compared to culture-based enumeration. However, ATP 

monitoring, like culture-based methods, is susceptible to sampling biases (e.g. 

swabbing pattern, swabbing area, surface characteristics), degradation of ATP or 

interference with the assay due to cleaning and sanitation compounds, and inability to 

effectively detect spores (16). 

 Even though ATP testing is usually performed after every cleaning and 

sanitation operation, microbial quantification should be performed periodically, in 



16 

addition to ATP testing, to verify that sanitation is effective. The use of both microbial 

and ATP monitoring can provide a robust set of data that verifies the efficacy of 

SSOPs (9). To effectively verify and improve cleaning and sanitation processes, 

hygiene monitoring programs need to address the frequency of testing, the location of 

test sites, and actionable limits for tests. These considerations are product and process 

specific but require a systematic framework to implement. Selecting sampling sites 

may require mapping the complete facility and production process, dividing the 

facility into zones based on microbiological risk to the product, and completing an 

assessment of the most appropriate test sites (8). Test sites should be selected after 

conducting an appropriate risk analysis to understand the risks associated with sites 

given the processing stage, proximity to food, potential for cross-contamination, ease 

of cleaning, and condition of the surface being tested (17). A conceptual overview of 

the process to implement hygiene monitoring is presented in Figure 2-1. 

 



17 

 
Figure 2-1. Conceptual overview of process to implement hygiene management 
 In this study, targeted cleaning directed by site-specific ATP bioluminescence 

detection was implemented as a measure to improve environmental cleanliness of a 

tofu production facility. This process was chosen for study because tofu production 

applies few hurdles to control for microbial growth and is therefore particularly 

sensitive to spoilage (18). Thus, it was hypothesized that targeted improvements to the 

cleanliness of the processing environment, monitored by ATP luminescence detection, 

would improve the microbiological quality of products and the processing 

environment. This study was conducted over three phases (Figure 2-1): establishment 

of a baseline hygiene level of the plant and products without targeted cleaning (Phase 

1), implementation of targeted cleaning practices directed by ATP results from Phase 

1 while maintaining extensive ATP testing to verify efficacy (Phase 2), and 

maintenance of cleaning and sanitation practices with reduced ATP testing (Phase 3). 



18 

ATP testing was complemented by culture-based testing of environmental and product 

samples for three groups of target microorganisms: total aerobic microorganisms, 

yeasts and mold, and lactic acid bacteria, selected because they are common measures 

of environmental cleanliness, and because they often cause food spoilage (19–21).  

MATERIALS AND METHODS 

Facility and Process Overview 

 This study was conducted in a medium-sized facility that produces only soy-

based products. In brief, the tofu produced at this facility is made from coagulated 

soymilk extracted from hydrated soybeans; extraction occurs at approximately 88℃ 

(190° F). After coagulation, the curd is pressed, cut, water-cooled, and vacuum 

packaged for retail or institutional use (pre-pasteurization). After packaging, products 

are pasteurized in package, water-cooled, and refrigerated for cold chain distribution 

(post-pasteurization); the shelf life of products is declared as 60 days. The facility in 

which this study was conducted was chosen due to a professional relationship between 

the authors and facility management.  

Study Design and Implementation 

 The goal of this study was to investigate the impact of targeted cleaning on the 

microbial quality of the environment and finished products. Microbiological 

environmental and product testing occurred in two phases for analysis: pre-

intervention (Phase 1) and post-intervention (Phase 3) of targeted cleaning activities. 

Prior to the start of study, 30 sites (21 Zone 1, 9 Zone 2) were identified by the authors 

and facility management for ATP monitoring and microbiological enumeration. These 



19 

sites were chosen based on phase of production, and relative cleaning and sanitation 

difficulty (i.e., sites deemed harder to clean were favored for testing over others); the 

sites and zone designations are listed in Table 2-1. ATP monitoring occurred over 

three phases; Phase 1 (baseline assessment): verification of cleaning and sanitation 

procedures utilizing extensive ATP testing (30 sites targeted per day), Phase 2: post-

implementation of targeted cleaning, maintaining extensive ATP testing (30 sites 

targeted per day) and Phase 3: post-implementation with maintenance of cleaning and 

sanitation practices and reduced ATP testing (18 randomized sites targeted per day). 

Reduction and randomization of sites for ATP testing conducted in Phase 3 were 

performed utilizing the 3M™ Clean-Trace™ Data Management Software (v 1.3.0.0) 

with the randomization function. 

 To establish a baseline hygiene level of the facility and products, the pre-

intervention phase of the study occurred over 3 weeks following the facility’s normal 

cleaning and sanitation program. After this, an adjustment period of 6 weeks was 

allowed for the cleaning and sanitation crew to adopt targeted cleaning practices 

indicated by Phase 1 data. Targeted cleaning practices incorporated results from ATP 

and microbiological testing, described below, into the cleaning and sanitation 

program. Management informed the crew which sites consistently had high levels 

(RLU/swab > 500) of ATP during the pre-intervention phase, and the cleaning and 

sanitation crew subsequently targeted those sites for enhanced cleaning. Enhanced 

cleaning included increased time spent cleaning portions of the line associated with 

the pre-identified sites, and some disassembly of equipment to access hard-to-clean 

areas. Sanitation proceeded as usual after cleaning. Following the adjustment period, 



20 

the post-intervention phase of study occurred over 16 weeks to determine the impact 

of targeted cleaning. 

Biochemical and Microbiological Testing 

 During Phases 1, 2, and 3, ATP was quantified at the 30 predetermined sites 

using 3M™ Clean-Trace™ Surface ATP Swabs (3M Company, St. Paul. MN) and the 

3M™ Clean-Trace™ Hygiene Monitoring System LM1 Luminometer (v 1.1.0.0) (3M 

Company, St. Paul MN) immediately following cleaning and sanitation according to 

manufacturer instructions. During Phases 1 and 3, yeasts and molds, lactic acid 

bacteria, and aerobic microorganisms were quantified from a single complimentary 

3M™ Quick Swab (3M Company, St. Paul. MN) taken adjacent ( < 15 cm or 6”) to 

the area swabbed for ATP. Samples were collected by swabbing in two directions: 

horizontally and vertically an area of approximately 100 cm2. Both the ATP and 

microbiological swabs were taken by a trained member of the cleaning and sanitation 

crew; thus, the swabbing portion of this study represents ‘real-world’ execution. 

Microbiological swabs were kept refrigerated ( < 4℃) and processed within 1-6 days 

(dictated by transportation time, shift, and day of the week). Swabs were serially 

diluted in Butterfield’s buffer (3M, St. Paul, MN) and plated onto 3M™ Petrifilm™ 

Plates according to manufacturer instructions: Rapid Yeast and Mold Count Plates 

(RYM; 3 days at 25℃), Lactic Acid Bacteria Count Plates (LAB; 2 days at 30℃), and 

Rapid Aerobic Count Plates (RAC; 1 day at 35℃). 

 Two packaged products from each production lot, one pre-pasteurized and the 

other post-pasteurized, were taken directly from the production line during Phases 1 

and 3. These products were kept refrigerated ( < 4℃) and microbiologically 



21 

characterized within 1-6 days. In brief, one 25 g subsample of tofu was aseptically 

removed from each packaged product and stomached in a sterile filter bag with 225 

mL 0.1% peptone water (Becton Dickinson, Franklin Lakes, NJ); stomaching occurred 

at 200 RPM for 90 s. Samples were serially diluted with Butterfield’s buffer and 

plated onto RYM, LAB, and RAC plates according to manufacturer instructions (see 

above). 

Data and Statistical Analyses 

 Quantitative data collected from ATP and microbiological swabs was 

transformed to binary pass/fail values with the following cutoffs, i.e. the minimum 

sanitary requirements (specific to this study): ATP – RLU/swab (100 cm2) > 500, 

RYM – log10 CFU/swab > 1.30 (20 CFU), LAB – log10 CFU/swab > 2.30 (200 CFU), 

and RAC – log10 CFU/swab > 2.30. Measurements that exceeded these cutoffs failed 

the test, indicating that the site was not adequately cleaned and sanitized for 

processing, i.e., these swabs failed to meet the minimum sanitary requirements. These 

values were chosen based on manufacturer recommendations, in coordination with 

plant management, and based on the authors’ experience. They are in alignment with 

previously established levels (22, 23), but it is important to note that present day 

hygiene monitoring emphasizes risk-based decision making and thus these cutoffs will 

vary depending on the product and process (24). Quantitative data obtained from 

microbiological analysis of finished products was not transformed except that for 

values less than or greater than the limit of detection (based on the chosen dilutions 

and countable ranges for each of the utilized media) were set to the limit of detection. 

 Data analysis was conducted in R (v4.0.2) (25) using R Studio (v1.3.1073) 



22 

(26) with the following packages: readxl (v1.3.1) (27), dplyr (v1.0.2) (28), ggpubr 

(v0.4.0) (29), tidyr (v1.1.1) (30), and kableExtra (v1.2.1) (31). Trending data were 

visualized via locally estimated scatterplot smoothing (LOESS) with the following 

parameters: span = 0.75, degree = 2, and confidence interval = 95%. All statistical 

tests were conducted as two-sided tests with α = 0.05. Binary data were analyzed 

using the binomial distribution to obtain 95% confidence intervals for groups; groups 

were compared using Fisher’s exact test. Quantitative product data were compared 

using the nonparametric Mann-Whitney U test due to the skewed nature of the data. 

Agreement between the results from ATP swabs and microbiological swabs was 

analyzed using the sampling distribution to obtain 95% confidence intervals for 

groups. 

Data Availability 

 The data and code used to draw conclusions in this study are deposited in a 

zenodo repository under doi: 10.5281/zenodo.4287499 (32). 

RESULTS 

Environmental Quality 

 ATP, yeasts and molds, lactic acid bacteria, and aerobic microorganisms were 

quantified from swabs at 30 predetermined sites (21 Zone 1, 9 Zone 2), then 

transformed to binary pass/fail results based on predetermined cutoffs specific to each 

measure to determine the impact of targeted cleaning on the hygiene of the processing 

environment. After excluding data due to excessive sample processing time ( > 6 days) 

a total of 5,196 measurements were retained across all phases of the study.  



23 

 Over the course of study, the proportion of sites failing to meet the minimum 

sanitary requirement day to day, based on ATP swabs, was highest during Phase 1 and 

then steadily decreased during Phase 2 before leveling off in Phase 3 (Figure 2-2); 

this indicated that targeted cleaning was improving the cleanliness of the facility. 

 
Figure 2-2. ATP swab failure rate over time  
Trend lines are locally fitted polynomial regressions computed via the LOESS 
method, grouped by zone. Vertical lines correspond to the separation of the three 
phases utilized in this study (Phase 1: pre-intervention – 30 sites targeted per day (21 
Z1, 9 Z2), Phase 2: post-intervention – 30 sites targeted per day (21 Z1, 9 Z2), Phase 
3: post-intervention – 18 randomized sites targeted per day (12 Z1, 6 Z2). 

 When aggregated by zone, the results show that targeted cleaning significantly 

lowered the proportion of swabs that failed to meet the minimum sanitary 

requirements between Phases 1 and 3 for lactic acid bacteria and aerobic 

microorganisms in both Zones 1 and 2, but did not significantly change the proportion 

of swabs that failed to meet the minimum sanitary requirement for yeasts and molds in 

either Zones 1 or 2 (p < 0.05, Fisher’s exact test) (Figure 2-3). The reduction between 



24 

Phases 1 and 3 was larger for aerobic microorganisms (21.8% – Z1, 26.8% – Z2) 

compared to lactic acid bacteria (9.7% – Z1, 14.1% – Z2), and was reflected in the 

significantly lower proportion of ATP swabs that failed to meet the minimum sanitary 

requirements in both Zones 1 and 2 (p < 0.05, Fisher’s exact test) (Figure 2-3). The 

reduction between Phases 1 and 3 was largest for ATP swabs among all the tests 

(26.5% – Z1, 51.0% – Z2). Finally, there was a significantly higher proportion of 

swabs in Zone 2 that failed to meet the minimum sanitary requirements compared to 

Zone 1 for only two groups: ATP swabs in Phase 1 and lactic acid bacteria swabs in 

Phase 3 (though the difference between Zones 1 and 2 for lactic acid bacteria was 

small – 2.4%). 

  



25 

 
Figure 2-3. ATP and microbial swab failure rates by zone 
Proportion of swabs, aggregated by zone, failing to meet the minimum sanitary 
requirements based on the measurement of adenosine triphosphate (ATP), yeasts and 
molds (RYM), lactic acid bacteria (LAB), and aerobic microorganisms (RAC) during 
Phases 1 and 3. Error bars represent the 95% confidence interval for each group based 
on the binomial distribution. Asterisks correspond to a significant difference between 
phases for a given zone (p < 0.001, Fisher’s exact test). 

 When aggregated by site, as opposed to zone, targeted cleaning caused a 

decrease in the proportion of swabs failing to meet the minimum sanitary requirements 

between Phases 1 and 3 for the majority of sites (Table 2-1). Though some sites 

exhibited an increase in the proportion of failing swabs, these increases were not 

significant (p > 0.05, Fisher’s exact test). There was a significant decrease in the 

proportion of swabs that failed to meet the minimum sanitary requirement for two sites 

when quantifying lactic acid bacteria (1 – Z1, 1 – Z2) and nine sites when quantifying 

aerobic microorganisms (6 – Z1, 3 – Z2); targeted cleaning did not result in a 



26 

significant decrease in the proportion of swabs that failed to meet the minimum 

sanitary requirement for any sites when measuring yeasts and molds (p < 0.05, 

Fisher’s exact test). Similar to when data were aggregated by zone, the significant 

decrease in the proportion of swabs failing to meet the minimum sanitary requirements 

across specific sites for lactic acid bacteria and aerobic microorganisms was reflected 

in 14 sites that showed a significant decrease in the proportion of swabs failing to meet 

the minimum sanitary requirement when measuring ATP (9 – Z1, 5 – Z2). The 

reduction in the proportion of failing swabs for all metrics between the pre- and post-

intervention phases of study across all sites is presented in Table 2-1.  

  



27 

Table 2-1. Swab failure rate by site 
Reduction in the proportion of swabs failing to meet the minimum sanitary 
requirements across all sites and measurements between Phases 1 and 3 

Zone Site 
ATPa RYMb LABc RACd 

Red.e Sig.f Red. Sig. Red. Sig. Red. Sig. 

1 

01. Soybean Hopper 
Corner 

49.6% *** 24.9% ns 79.1% *** -15.8% ns 

02. Auger shaft- 
flexicon 

52.9% *** -7.9% ns -5.5% ns 12.6% ns 

03. Slurry tank inside 0.0% ns 0.0% ns 7.7% ns 15.4% ns 
04. Bulk (Soymilk) 
tank inside 

23.1% ns 0.0% ns 7.7% ns 23.1% ns 

05. Bulk (Soymilk) 
tank agitator 

92.3% *** -7.9% ns 15.4% ns 20.4% * 

06. Roller extractor 
shaft 

30.8% * 3.1% ns 23.1% ns 24.1% ns 

07. Roller extractor 
roller 

38.5% * 9.1% ns 9.1% ns 24.5% ns 

08. Bucket inside 40.9% * 1.8% ns -5.9% ns 23.1% ns 
09. Bucket agitator 23.1% ns -15.9% ns 0.0% ns 38.5% * 
10. Bucket turbulent 
stick 

56.2% *** -2.6% ns 7.7% ns 38.5% *** 

11. Curd holding 
tank/ Curd transfer 
barrel 

47.9% *** -3.0% ns 5.1% ns 12.8% ns 

12. Conveyor belt 
white mat/ Auto 
press belt 

20.3% ns -0.2% ns 0.0% ns 15.4% ns 

13. Conveyor green 
plastic side belt 

7.7% ns 1.8% ns 23.1% ns 30.8% * 

14. Chain conveyor/ 
Transfer conveyor 

49.1% ** 1.8% ns 0.0% ns 15.4% ns 

15. Chilling tank 
smooth surface/ 
Conveyor tank 

0.0% ns -11.8% ns 0.0% ns 7.7% ns 

16. Chilling tank 
inside corner 

0.0% ns 0.0% ns 0.0% ns 0.0% ns 

17. Chilling tank 
conveyor 

23.1% ns -12.3% ns 0.0% ns 15.4% ns 

18. Chilling tank 
roller shaft 

0.0% ns -7.7% ns 23.1% ns 38.5% * 

19. Chilling tank 
roller sprocket 

16.7% ns 16.8% ns 23.1% ns 38.5% * 

20. Overflow tofu 
tank inside/ Rolling 
tanks 

25.8% ns -3.4% ns 7.7% ns -3.4% ns 

21. Overflow tofu 
tank corner 

7.7% ns -7.1% ns 0.0% ns 0.0% ns 

table continues next page   

  



28 

Zone Site 
ATPa RYMb LABc RACd 

Red.e Sig.f Red. Sig. Red. Sig. Red. Sig. 

2 

50. MV4 HMI 
Screen 

85.6% *** 10.1% ns 7.7% ns 4.9% ns 

51. MV4 HMI 
Screen control 
button and E-stop 

-5.5% ns 10.0% ns 15.4% ns 10.0% ns 

52. MV4 film 
Rollers 

89.2% *** -0.6% ns 7.7% ns 20.3% ns 

53. Rolling rack 40.5% * 3.1% ns 7.7% ns 23.1% ns 
54. Rolling rack 
trays 

-11.8% ns 2.1% ns 7.7% ns 10.8% ns 

55. MV side rail 23.8% ns 1.4% ns 23.1% ns 26.0% ns 
56. Chiller tank 
outside/ Conveyor 
tank outside 

9.7% ns -11.1% ns 7.7% ns 30.8% * 

57. Waterpack 
control panel 
buttons 

76.5% *** -10.8% ns 28.1% * 51.1% *** 

58. Waterpack 
upper guide rails 
prior to sealer 

38.5% ** 8.7% ns 23.1% ns 61.5% *** 

a Adenosine triphosphate, b yeasts and mold, c lactic acid bacteria, and d 

aerobic microorganisms 
e Reduction calculated as Pfail, P1 – Pfail, P3.  
f Significance according to Fisher’s exact test: ns – not significant, * – p < 
0.05, ** – p < 0.01, *** – p < 0.001. 
 

Microbiological Product Quality 

 Tofu products taken from the line during production were sampled for yeasts 

and molds, lactic acid bacteria, and aerobic microorganisms to determine the impact of 

targeted cleaning on the quality of finished products. Among pre-pasteurized products, 

the mean rank log10 CFU/g of post-intervention products (n = 68) was significantly 

lower than pre-intervention products (n = 19) for yeasts and mold, lactic acid bacteria, 

and aerobic microorganisms (p < 0.05, Mann-Whitney U test) (Figure 2-4). Among 

post-pasteurized products, the mean rank log10 CFU/g of post-intervention products (n 

= 69) was not significantly different than pre-intervention products (n = 19) for yeasts 

and mold, lactic acid bacteria, and aerobic microorganisms (p > 0.05, Mann-Whitney 

U test) (Figure 2-5). 



29 

 
Figure 2-4. Microbial loads of products before pasteurization 
Microbial load of yeasts and mold (RYM), lactic acid bacteria (LAB), and aerobic 
microorganisms (RAC) in packaged tofu products sampled pre-pasteurization. Red 
lines correspond to the limits of detection for each group of interest. Numbers 
correspond to the p value between Phases 1 and 3. 

 
Figure 2-5. Microbial loads of products after pasteurization 
Microbial load of yeasts and mold (RYM), lactic acid bacteria (LAB), and aerobic 
microorganisms (RAC) in packaged tofu products sampled post-pasteurization. Red 
lines correspond to the limits of detection for each group of interest. Numbers 
correspond to the p value between Phases 1 and 3. 

 

  



30 

ATP and Microbiological Swab Agreement 

 Because microbiological swabs were taken immediately adjacent to ATP 

swabs, it was possible to analyze post-hoc the agreement between ATP swabs and 

microbiological swabs at those sites. In total, 960 samples over the pre- and post-

intervention phases of study were quantified for both ATP and viable microorganisms. 

Data were transformed to binary pass/fail results based on the same predetermined 

cutoffs as previously mentioned, but rather than analyze each microbiological group 

individually (i.e., yeasts and molds, lactic acid bacteria, and aerobic microorganisms) 

a binary transformation was made such that if any of the microbiological 

measurements exceeded their respective cutoffs, then the site failed to meet the 

minimum sanitary requirement. For a site to pass the minimum microbiological 

sanitary requirements, every microbiological group had to be below its individual 

predetermined cutoff. ATP and microbiological swab results agreed for 75.1% (72.3-

77.7%) of samples (95% C.I., sampling distribution). ATP swab results failed the 

minimum sanitary requirements but microbiological swabs passed for 11.7% (9.8-

13.9%) of samples (95% C.I., sampling distribution), and ATP swabs passed the 

minimum sanitary requirements but microbiological swabs failed for 13.2% (11.2-

15.5%) of samples (data not shown). 

DISCUSSION 

 Cleaning and sanitation procedures are critical in the food industry and are the 

first line of defense to prevent contamination of food products from the production 

environment. They are required to prevent the presence and proliferation of 



31 

pathogenic and spoilage microorganisms. In this study, cleaning and sanitation 

operations were monitored in a soy-based manufacturing production facility over a 

period of three weeks utilizing both ATP bioluminescence and microbial indicators to 

assess the cleaning efficacy. After this time, targeted cleaning of specific sites that 

showed the highest rate of cleaning and sanitation failures was implemented; the effect 

of targeted cleaning was monitored with the same verification methods. 

Environmental Quality 

 The results from this study indicate that targeted cleaning directed by ATP 

monitoring may improve the environmental hygiene of food processing facilities 

(Figure 2-2); this was verified by microbiological tests. A significant decrease in the 

proportion of swabs failing to meet the minimum sanitary requirements for lactic acid 

bacteria and aerobic microorganisms indicates that the targeted cleaning applied after 

Phase 1 had a positive effect on the facility’s hygiene. In contrast, the hygiene 

measure for yeasts and molds remained unchanged with targeted cleaning efforts. 

However, it is important to consider that equipment surfaces may not be the only 

sources of yeast and molds. Other sources of yeasts and molds include air, raw 

materials, and packaging, which would likely be unaffected by improvements to 

cleaning and sanitation (19).  

 There was minimal difference between the microbial indicator failure rates 

between Zones 1 and 2 over the course of the study. However, there was a significant 

difference between failure rates of Zone 1 and Zone 2 sites for ATP during Phase 1 

that was absent in Phase 3. Though the data do not suggest that Zone 2 surfaces were 

less hygienic than Zone 1 with regards to microorganisms, the insignificant difference 



32 

for ATP between Zone 1 and Zone 2 sites during Phase 3 suggests the plant was 

overall cleaner. In this case, cleaner Zone 2 surfaces corresponded with improved 

plant hygiene, which was reflected in the improvement of hygiene measures 

associated with aerobic microorganisms and lactic acid bacteria. Such an effect 

protects against the establishment of spoilage and likely pathogenic microorganism 

populations (33). 

 This study also showed that ATP monitoring and the use of microbiological 

indicators may result in more effective equipment surface cleaning. When comparing 

individual sites (Table 2-1), targeted cleaning was most effective in specific portions 

of the manufacturing line. A total of 14 sites for ATP and 9 sites for aerobic 

microorganisms showed a significant reduction in the proportion of failing swabs 

between Phases 1 and 3. Some sites exhibited an increase in the proportion of failing 

swabs, but these increases were not significant. Taken together, these data indicate that 

targeted cleaning may only improve hygiene for specific sites and that other factors 

aside from cleaning (e.g., equipment geometry, temperature of product at the 

processing step, etc.) may have a greater effect on the hygienic quality of some sites. 

For this plant, ATP monitoring and the use of total aerobic count best reflected 

(greatest change) the effect of targeted cleaning; these could be selected as methods 

for routine verification of cleaning and sanitation operations in this facility. These two 

measures may not apply to all products and facilities. It is important for facilities to 

choose metrics that are sensitive to changes in the plant environment and can easily 

detect deviations to cleaning and sanitation. Setting critical parameters for monitoring 

ATP as well as microbiological criteria on equipment surfaces is highly dependent on 



33 

the manufacturing site, design and state of the equipment, product, process, and 

cleaning process. Relying on data that can be trended over time is often helpful to 

establish an appropriate baseline. Baseline testing for ATP and microbiological 

parameters should be periodically reviewed and reassessed to verify that cleaning and 

sanitation operation procedures remain effective.  

Microbiological Product Quality 

 In this study, there was an improvement in the microbial load of yeasts and 

molds, lactic acid bacteria, and aerobic microorganisms for pre-pasteurized products 

(Figure 2-4), but not post-pasteurized products after implementation of targeted 

cleaning (Figure 2-5). It is important to note that for this process, the product 

undergoes two thermal processing steps. Soymilk requires an extraction that occurs at 

88℃, which significantly reduces the microbial load associated with the raw 

materials; after this process, coagulated product is pressed, cut, and packaged before it 

is pasteurized in-package. Because the microbial load is reduced in soymilk during 

extraction, the microbial load of products determined in this study in pre-pasteurized 

products is primarily associated with post-process contamination, including contact 

with equipment surfaces. The microbial load of post-pasteurized products was not 

significantly different, which is expected as the packaged product is heat treated again, 

killing vegetative cells. The results from this study indicate that targeted cleaning 

monitored by ATP bioluminescence and microbial indicators could improve 

microbiological product quality for products that do not undergo an in-package 

pasteurization step (e.g., fresh fruits and vegetables).  

  



34 

Hygiene monitoring with ATP bioluminescence 

 ATP bioluminescence was utilized as a tool for hygiene monitoring during this 

study. Prior to implementation of targeted cleaning, site selection was done to identify 

areas in the production environment that were most likely to pose challenges during 

cleaning and sanitation. Site selection should comprehensively cover the production 

environment such that results obtained during monitoring activities reflect the 

cleanliness of the area (34). ATP bioluminescence monitoring accompanied by 

microbial evaluation demonstrated the effectiveness of targeted cleaning. Significant 

differences were determined between the proportion of swabs failing to meet the 

minimum sanitary requirements set for both ATP and microbiological parameters for 

specific sites (Table 2-1). In cases where ATP did not agree with the microbiological 

result, it is important to consider sampling area (which although adjacent could display 

differences) as well as the microbial load required for ATP detection (approximately 

103-104 CFU) (35). 

 Overall, the ATP data over time showed a decrease in the proportion of failing 

swabs as Phases 2 and 3 were implemented (Figure 2-2). Evaluation of ATP swabs as 

a tool to identify microbial contamination showed that 75.1% (72.3-77.7%) of ATP 

swabs reflected the microbiological status of test sites. This correlation is likely due to 

inadequate sanitation resulting from poor cleaning or adequate sanitation resulting 

from proper cleaning. Although ATP generally does not directly correlate to the 

number of microorganisms on a given surface (16, 36) it can be used as a rapid tool to 

assess equipment cleanliness (22). ATP sources are not exclusive to microbial ATP; 

food residues, which cannot be detected by microbiological tests, can also account for 



35 

failing ATP swabs. In this study, this occurred when the microbial counts passed the 

minimum sanitary requirements, but the ATP levels did not – 11.7% (9.8-13.9%). The 

ATP from non-bacterial sources can evidence potential niches where food residues 

may accumulate over time and thus enable microbial growth. The need to conduct 

both biochemical and microbiological tests is especially highlighted where the 

microbiological results did not meet the minimum sanitary requirements, but ATP did 

– 13.2% (11.2-15.5%). In these cases, spores may cause the discrepancy because they 

do not produce ATP while in spore form (37, 38), but can still enter packages and 

induce spoilage after germination. 

 ATP is considered a method to rapidly verify cleaning while microbial tests 

provide results to verify the sanitation status; both tests can provide a more holistic 

assessment of the effectiveness of cleaning and sanitation operations. ATP swabs 

alone can provide rapid and robust daily verification for cleaning and sanitation 

operations of a facility. In this study, ATP swabs taken after cleaning and sanitation 

would have either correctly verified the microbiological hygiene status or elicited 

additional cleaning due to food residues in 86.8% (84.5-88.8%) of cases. 

Microbiological testing could supplement this on a less frequent basis (e.g., weekly) to 

ensure continued efficacy of cleaning and sanitation. 

CONCLUSION 

 This study showed that the microbiological quality of products improved 

following targeted cleaning implemented to improve the hygiene of the production 

environment. The use of ATP bioluminescence and microbial indicators seem to be 



36 

effective tools to monitor cleaning and sanitation operations and to direct the efforts of 

the cleaning and sanitation crew. Further, these results indicate that both biochemical 

and microbiological tests should be used to monitor hygiene, as they are 

complementary in efficiently assessing the cleaning and sanitary status of the 

manufacturing environment and processed products. This study serves as a framework 

for companies to implement hygiene monitoring in their own facilities, but it is 

important to note that different products and plants may require different tests and/or 

critical limits to the tests used in this study. 

  



37 

REFERENCES 

1.  USDA FSIS. 2020. Sanitation, 9 CFR § 416. Code of Federal Regulations Regulatory 
Requirements Under the Federal Meat Inspection Act and the Poultry Products 
Inspection Act. 

2.  DHHS FDA. 2020. Current Good Manufacturing Practice, Hazard Analysis, and Risk-
Based Preventive Controls for Human Consumption, 21 CFR § 117. Code of Federal 
Regulations Food for Human Consumption. 

3.  USDA FSIS. 2020. Sanitation: Sanitary Operations, 9 CFR § 416.4. Code of Federal 
Regulations Regulatory Requirements Under the Federal Meat Inspection Act and the 
Poultry Products Inspection Act. 

4.  DHHS FDA. 2020. Current Good Manufacturing Practice, Hazard Analysis, and Risk-
Based Preventive Controls for Human Consumption: Sanitary Operations, 21 CFR § 
117.35. Code of Federal Regulations Food for Human Consumption. 

5.  DHHS FDA. 2020. Current Good Manufacturing Practice, Hazard Analysis, and Risk-
Based Preventive Controls for Human Consumption: Definitions, 21 CFR § 117.3. 
Code of Federal Regulations Food for Human Consumption. 

6.  USDA FSIS. 2020. Hazard Analysis and Critical Control Point (HACCP) Systems: 
Definitions, 9 CFR § 417.1. Code of Federal Regulations Regulatory Requirements 
Under the Federal Meat Inspection Act and the Poultry Products Inspection Act. 

7.  Keener L. 2005. Improving cleaning-out-of-place (COP), p. 445–467. In Lelieveld, 
HLM, Mostert, MA, Holah, J (eds.), Handbook of Hygiene Control in the Food 
Industry. 

8.  Gombas D, Bierschwale S, Blackman S, Butts JN, Carter D, Coles C, Crawford WM, 
Denault-Bryce P, Eisenberg BA, Estrada Jr M, Ewell H, Foster S, Hardin M, Hau H, 
Kerr J, Mills B, Owens EM, Parker C, Petran RL, Prince G, Raede J, Roberson M, 
Shergill G, Snyder K, Stoltenberg SK, Suslow T, Zomorodi B. 2013. Guidance on 
Environmental Monitoring and Control of Listeria for the Fresh Produce Industry. 
United Fresh Produce Association. 

9.  Moore G, Griffith C. 2002. A comparison of traditional and recently developed 
methods for monitoring surface hygiene within the food industry: An industry trial. Int 
J Environ Health Res 12:317–329. 

10.  Holah JT. 1992. Industrial Monitoring: Hygiene in Food Processing, p. 645–659. In 
Melo, LF, Bott, TR, Fletcher, M, Capdeville, B (eds.), Biofilms — Science and 
Technology. Springer Netherlands, Dordrecht. 



38 

11.  Nivens DE, Co BM, Franklin MJ. 2009. Sampling and quantification of biofilms in 
food processing and other environments, p. 539–568. In Biofilms in the Food and 
Beverage Industries. Elsevier. 

12.  Maturin L, Peeler JT. 2001. BAM Chapter 3: Aerobic Plate Count. Bacteriological 
Analytical Manual (BAM). https://www.fda.gov/food/laboratory-methods-food/bam-
chapter-3-aerobic-plate-count. 

13.  Tournas V, Stack ME, Mislivec PB, Koch H a, Bandler R. 2001. BAM Chapter 18: 
Yeasts, Molds and Mycotoxins. FDA Bacteriological Analytical Manual. 
https://www.fda.gov/food/laboratory-methods-food/bam-chapter-18-yeasts-molds-
and-mycotoxins. 

14.  Hawronskyj JM, Holah J. 1997. ATP: A universal hygiene monitor. Trends Food Sci 
Technol 8:79–84. 

15.  Mildenhall KB, Rankin SA. 2020. Implications of Adenylate Metabolism in Hygiene 
Assessment: A Review. J Food Prot 83:1619–1631. 

16.  Shama G, Malik DJ. 2013. The uses and abuses of rapid bioluminescence-based ATP 
assays. Int J Hyg Environ Health 216:115–125. 

17.  Roberts L, Lang G, Yordem B. ATP and Protein-based Hygiene 
MonitoringEnvironmental Monitoring Handbook for the Food and Beverage 
Industries, 1st ed. 3M. 

18.  Rahman MS. 2015. Hurdle Technology in Food Preservation, p. 17–33. In Rahman, 
MS, Siddiqui, MW (eds.), Minimally Processed Foods. Springer International 
Publishing. 

19.  Snyder AB, Worobo RW. 2018. Fungal Spoilage in Food Processing. J Food Prot 
81:1035–1040. 

20.  Clavero R. 2010. Solving Microbial Spoilage Problems in Processed Foods, p. 63–78. 
In Kornacki, JL (ed.), Principles of Microbiological Troubleshooting in the Industrial 
Food Processing Environment. Springer New York, New York, NY. 

21.  Lorenzo JM, Munekata PE, Dominguez R, Pateiro M, Saraiva JA, Franco D. 2018. 
Main Groups of Microorganisms of Relevance for Food Safety and Stability, p. 53–
107. In Barba, FJ, Sant’Ana, AS, Orlien, V, Koubaa, M (eds.), Innovative 
Technologies for Food Preservation. Elsevier. 

22.  Cunningham AE, Rajagopal R, Lauer J, Allwood P. 2011. Assessment of Hygienic 
Quality of Surfaces in Retail Food Service Establishments Based on Microbial Counts 
and Real-Time Detection of ATP. J Food Prot 74:686–690. 

23.  Griffith C. 2005. Improving surface sampling and detection of contamination, p. 588–



39 

618. In Handbook of hygiene control in the food industry. Woodhead Publishing 
Limited. 

24.  Møretrø T, Langsrud S. 2017. Residential Bacteria on Surfaces in the Food Industry 
and Their Implications for Food Safety and Quality. Compr Rev Food Sci Food Saf 
16:1022–1041. 

25.  R Core Team. 2020. R: A language and environment for statistical computing. 4.0.2. R 
Foundation for Statistical Computing, Vienna, Austria. 

26.  RStudio Team. 2020. RStudio: Integrated Development Environment for R. 1.3.1073. 
RStudio, Inc., Boston, MA. 

27.  Wickham H, Bryan J. 2019. readxl: Read Excel Files. 1.3.1. 

28.  Wickham H, François R, Henry L, Müller K. 2020. dplyr: A Grammar of Data 
Manipulation. 1.0.2. 

29.  Kassambara A. 2020. ggpubr: “ggplot2” Based Publication Ready Plots. 0.4.0. 

30.  Wickham H, Henry L. 2020. tidyr: Tidy Messy Data. 1.1.1. 

31.  Zhu H, Travison T, Tsai T, Beasley W, Xie Y, Yu G, Laurent S, Shepard R, Sidi Y, 
Salzer B, Gui G, Fan Y, Murdoch D. 2020. kableExtra. 1.2.1. 

32.  Sogin JH, Velasco Lopez G, Yordem B, Lingle CK, David JM, Cobo M, Worobo RW. 
2020. Tofu Case Study Statistical Analyses. 1. Zenodo 
https://doi.org/10.5281/zenodo.4287499. 

33.  Hammons SR, Stasiewicz MJ, Roof S, Oliver HF. 2015. Aerobic Plate Counts and 
ATP Levels Correlate with Listeria monocytogenes Detection in Retail Delis. J Food 
Prot 78:825–830. 

34.  Simmons CK, Wiedmann M. 2018. Identification and classification of sampling sites 
for pathogen environmental monitoring programs for Listeria monocytogenes: Results 
from an expert elicitation. Food Microbiol 75:2–17. 

35.  Ukuku DO, Pilizota V, Sapers GM. 2001. Bioluminescence ATP Assay for Estimating 
Total Plate Counts of Surface Microflora of Whole Cantaloupe and Determining 
Efficacy of Washing Treatments†. J Food Prot 64:813–819. 

36.  Osimani A, Garofalo C, Clementi F, Tavoletti S, Aquilanti L. 2014. Bioluminescence 
ATP Monitoring for the Routine Assessment of Food Contact Surface Cleanliness in a 
University Canteen. Int J Environ Res Public Health 11:10824–10837. 

37.  Setlow P, Kornberg A. 1970. Biochemical Studies of Bacterial Sporulation and 
Germination. XXII. Energy metabolism in early stages of germination of Bacillus 



40 

megaterium sporesThe Journal of Biological Chemistry. 

38.  Korza G, Setlow B, Rao L, Li Q, Setlow P. 2016. Changes in Bacillus Spore Small 
Molecules, rRNA, Germination, and Outgrowth after Extended Sublethal Exposure to 
Various Temperatures: Evidence that Protein Synthesis Is Not Essential for Spore 
Germination. J Bacteriol 198:3254–3264. 

 

  



41 

CHAPTER 3 

NEXT-GENERATION AMPLICON SEQUENCING ANALYSIS OF 
COMMERCIAL TEMPEH DURING FERMENTATION  

ABSTRACT 

 Tempeh is a mold-fermented food product native to Indonesia. Rhizopus 

microsporus is typically used as an inoculum for cooked soybeans, but production 

processes introduce other bacteria and other fungi which form a microbial community. 

This study investigated the influence of product type (i.e., raw materials composition) 

between soybean and multigrain tempeh and time point in fermentation on the 

bacterial and fungal communities of commercially produced tempeh. Next-generation 

amplicon sequencing of the bacterial 16S rRNA gene V3-V4 regions (16S) and fungal 

internal transcribed spacer region 2 (ITS2) were used to investigate community 

differences. For both bacterial and fungal communities, multigrain tempeh exhibited 

overall higher alpha diversity than soybean tempeh. The results of PERMANOVA and 

ZicoSeq differential abundance analyses indicated that product type was significantly 

(p < 0.05) associated with differences in bacterial communities, whereas time point 

was significantly associated with differences in fungal communities. Core microbiome 

analysis showed that Leuconostoc, Enterococcus, Lactococcus, and Kocuria species 

were the core bacterial species, and that Rhizopus microsporus and Trichosporon 

asteroides were the core fungal species. This study reports tempeh community 

information over the course of fermentation and provides a direct comparison of the 

associated communities due to ingredient composition. The discussion suggests 

further study on the influence of Acinetobacter and Trichosporon species on sensory 



42 

characteristics of tempeh. 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
This study is being prepared for publication and is currently under review: 

Sogin JH, & Worobo RW. 2023. Next-Generation Amplicon Sequencing Analysis of 
Commercial Tempeh During Fermentation. In review. 



43 

INTRODUCTION 

 Tempeh is a plant-based protein source originating from ethnic Javanese 

people in Indonesia (1). It is mainly made from cooked soybeans, but other grains 

including broad beans, peas, and chickpeas can be used too (2). The typical steps of 

tempeh production are raw soybean soaking, dehulling, washing, boiling, draining, 

cooling, inoculation with Rhizopus starter culture, packaging and incubation, but many 

variations exist due to processing scale (3, 4). The most noticeable role of 

fermentation in tempeh production is the physical aggregation of individual soybeans 

into a cohesive mass (‘soybean cake’). Rhizopus molds cause this transformation 

through the breakdown of soy constituents and growth of hyphae that fill vacant space 

between beans, thus forming a mycelium (5, 6). Traditionally produced tempeh in 

Indonesia used to contain three Rhizopus species: R. arrhizus, R. delmar, and R. 

microsporus, but widespread commercial starter use has led to the loss of R. arrhizus 

and R. delmar (7). 

 The nature of tempeh production makes it extremely difficult to maintain 

single-strain fermentations outside a laboratory like other fermented foods such as 

sauerkraut and koji (8, 9). Soybeans and food production environments harbor 

microorganisms that can grow during soybean soaking and will subsequently grow in 

the tempeh fermentation (2, 10, 11). Although soybeans are cooked during production, 

thereby killing many of the organisms that may have originated from the raw 

materials, food processing environments harbor resident microbial communities that 

will persist even with frequent cleaning and sanitation (12), especially in a fermented 

food facility. Furthermore, mycelium development throughout the soybean cake 



44 

during fermentation requires oxygen (among other factors), which is accomplished via 

a high surface area to volume approach, leaving portions of the cake exposed to the 

environment, unlike other fermentations such as beer, wine, or vegetable products 

which are often fermented in closed vessels. 

 A handful of previous studies have conducted community analyses of tempeh. 

One amplicon sequencing study identified differences in bacterial communities 

between two tempeh producers in Indonesia, finding the two majorly abundant genera 

in the products to be Enterococcus and Lactobacillus (13). Another amplicon 

sequencing study investigated the bacterial and fungal communities of over-fermented 

tempeh; some of the predominant bacterial genera were Chryseobacterium, 

Lactococcus, Lactobacillus, Streptococcus, Acetobacter, and Klebsiella, and some of 

the predominant fungal genera were Rhizopus and Trichosporon (14). One shotgun 

metagenomic study comparing the same tempeh products compared by Radita et al. 

(2018), found the predominant bacterial genera to be Novosphingobium and 

Enterobacter, with lower relative abundances of Leuconostoc, Enterococcus, and 

Lactobacillus observed (15). 

 The aforementioned studies analyzed finished or past-finished tempeh. We 

were interested in how the bacterial and fungal communities of tempeh changed over 

the course of fermentation. Furthermore, two of the studies (13, 15), compared 

products made by different manufacturers, indicating general differences between 

different tempeh, but no comparisons were made between products due to specific 

processing factors. Some studies have compared the microbiology of tempeh soak 

water between different grains (2) or due to different wraps (11), but neither used 



45 

next-generation sequencing to directly compare the bacterial and fungal communities 

of tempeh due to grain composition. Therefore, the purpose of this study was to 

characterize the bacterial and fungal communities of soybean and multigrain tempeh 

produced in a commercial facility throughout fermentation. 

METHODS 

Tempeh Production & Sample Acquisition 

 Tempeh was acquired directly from a commercial manufacturer of various soy-

based products; other products including tofu were produced near the tempeh. Two 

product types were sampled: soy and multigrain. Soy tempeh was made using only 

soybeans, whereas multigrain tempeh was made from soybeans, brown rice, millet, 

kasha, and quinoa. Sampling was conducted by an employee of the manufacturer at 

predetermined time points during production. Sampling occurred over the course of 

several weeks to obtain samples from multiple production runs/product lots. Samples 

were ultimately taken from four soy production runs (n = 24) and two multigrain runs 

(n = 12). Samples were taken before inoculation (‘pre-inoculation’), after inoculation 

(‘post-inoculation’), during fermentation (‘early’ ~5 h, ‘middle’ 10-15 h, ‘end’ ~20 h), 

and after product packaging (‘packaged’). Samples were defined as a single 8-ounce 

(227 g) product unit except for bean samples (pre-inoculation and post-inoculation), 

which were defined as an approximately 500 g portion of the bulk rehydrated, cooked, 

and drained beans. Samples were immediately frozen in bags, stored at -20℃ at the 

plant, and eventually transported to Cornell University for further processing. All 

samples were processed within 90 days of acquisition. 



46 

DNA Extraction 

 Frozen tempeh samples were defrosted under refrigeration (2℃) overnight 

(~12 h) prior to DNA extraction. For each sample, the product was aseptically opened 

and a 10 g portion from the center of the tempeh block was placed in a stomaching 

bag. 90 mL PBS was added to the stomaching bag, which was then hand massaged to 

break up the tempeh. The bag was then stomached at 230 rpm for 1 min. 1.8 mL of the 

stomached tempeh mixture was centrifuged at 13,000 rcf for 1 min. The supernatant 

was decanted, and DNA was extracted from the resulting pellet using the Qiagen 

Powerfood DNA extraction kit (Qiagen, Germantown, MD, USA). An ‘extraction 

negative control’ was subject to the same procedure using 100 mL PBS instead of 10 g 

tempeh + 90 mL PBS. 

Next-Generation Amplicon Sequencing 

 A two-step amplicon sequencing protocol was followed to amplify and pool 

marker gene fragments for bacteria and fungi from the tempeh DNA samples prior to 

short-read next-generation sequencing. 

First Step 

 The 16S rRNA V3-V4 region (16S) from bacteria and internal transcribed 

spacer region 2 from fungi (ITS2) were amplified in separate reactions. The locus-

specific portions of the 16S primers were F-5’-CCTACGGGNGGCWGCAG-3’ and 

R-5’-GACTACHVGGGTATCTAATCC-3’, colloquially known as Bakt_341F and 

Bakt_805R (16, 17); the locus-specific portions of the ITS primers were F-5’-

AACTTTYRRCAAYGGATCWCT-3’ and R-5’-

AGCCTCCGCTTATTGATATGCTTAART-3’, colloquially known as 5.8S-Fun and 



47 

ITS4-Fun (18). Both 16S and ITS primers were designed to include the required 

Illumina indexing overhangs (19), a 0-4 bp heterogeneity spacer (20), a 2 bp linker 

(21), and the appropriate locus-specific primer region. The use of a heterogeneity 

spacer resulted in there being five variations of each primer, which were equimolarly 

mixed prior to PCR amplification; a complete list of the primers used for 16S rRNA 

and ITS2 amplifications can be found in Appendix Tables 3-3 & 3-4. 

 Samples were amplified using Invitrogen Platinum II Taq Hot-Start Green 

PCR Master Mix (2X) in 25 µL reactions (Invitrogen, Waltham, MA, USA). Each 

reaction contained 12.5 µL 2X master mix, 2.5 µL 25 mM supplemental magnesium 

(as MgCl2), 1 uL 10 uM forward primer mixture, 1 uL 10 uM reverse primer mixture, 

3 µL GC enhancer (included with master mix), and 5 µL DNA template. PCR 

conditions for both 16S and ITS amplifications were the same: 1 cycle 94 ℃ for 5 

min; 25 or 35 cycles 94 ℃ for 15 sec, 50 ℃ for 15 sec, 68 ℃ for 15 sec; 1 cycle 68 ℃ 

for 2.5 min. Successful amplification was verified via gel electrophoresis and EtBr 

staining. An attempt was made to amplify all samples with 25 amplification cycles, 

but several samples required 35 amplification cycles to yield a visible band of EtBr-

stained PCR product following gel electrophoresis. In general, tempeh sampled at the 

pre-inoculation and post-inoculation time points required 35 cycles, whereas those 

sampled at later time points only required 25 cycles. ‘PCR negative controls’ using 

nuclease-free water instead of DNA template were subject to the same procedure for 

every PCR condition. Additionally, the ‘extraction negative control’ was subject to the 

same procedure for every PCR condition. A ‘PCR negative control’ and ‘extraction 

negative control’ were included for every PCR condition. 



48 

Second Step 

 The amplified samples were submitted to the Cornell Biotechnology Resource 

Center for quantification, indexing, multiplexing, cleanup, quality control, and 

Illumina MiSeq sequencing (Illumina, San Diego, CA, USA). Unique dual indexing 

was applied to the samples to prevent index hopping (22). Sequencing proceeded 

according to manufacturer specifications using a MiSeq 2×300 bp V3 sequencing kit. 

Data Processing & Analysis 

 Raw reads were processed in Qiime2 (23) using Cutadapt (24) to trim adapter 

sequences and DADA2 (25) to merge, denoise, and bin reads into amplicon sequence 

variants (ASVs). Sequences were taxonomically classified using the Qiime2 naïve-

bayes feature classifier (26) trained on amplicon specific regions of 16S sequences 

from the Silva version 138 99% sequence identity OTU data release (27, 28) 

processed with RESCRIPt (29), and on full length ITS sequences from the expert 

curated OTU thresholds UNITE 9.0 database (30). Data analysis was conducted using 

R (31) within RStudio (32) using renv (33). Processed read data were then transferred 

to R using Phyloseq (34), decontaminated using Decontam (35), ‘denoised’ (rare 

sequence removal) with PERFect (36), and analyzed with vegan (37), microbiome 

(38), and GUniFrac (39) R packages; GUniFrac was used for differential abundance 

testing with ZicoSeq (40). When data were agglomerated to higher taxonomic levels, 

taxonomically unresolved OTUs at the desired agglomeration level were kept separate. 

Other packages were used for data visualization: ggpubr (41), ggtext (42), and 

ggVennDiagram (43). The full code used to process raw reads and analyze data is 

deposited and accessible at 



49 

https://github.com/jsogin574/Sogin_and_Worobo_Tempeh_2023. 

Data Availability 

 Sample metadata and raw sequencing data were deposited to the US National 

Center for Biotechnology Information (NCBI) under BioProject number 

PRJNA941282 and Sequence Read Archives SRR23815527- SRR23815527606. 

Analysis codes are available at 

https://github.com/jsogin574/Sogin_and_Worobo_Tempeh_2023. 

RESULTS 

Next-Generation Sequencing and Sequence Processing 

 Samples were sequenced on an Illumina MiSeq with a 2×300 bp V3 

sequencing kit (Illumina, Madison, WI, USA). 11 M raw reads were obtained from 

Illumina sequencing (5.3 M 16S, 5.8 M ITS); 7.8 M reads were retained (3.2 M 16S, 

4.6 M ITS) after raw sequence processing. 5.8 M reads were retained (1.6 M 16S, 4.2 

M ITS) after filtering, contaminant removal with Decontam, and exclusion of negative 

controls. Chloroplast reads were a large fraction of those removed from 16S data (1.0 

M). At this point, samples from pre-inoculation and post-inoculation time points were 

removed for statistical analyses. Because fermentation had not begun, these samples 

had inherently low biomass compared to other samples, which resulted in low read 

counts, a larger number of contaminant sequences present, and many different OTUs 

compared to samples during fermentation (early, middle, end, packaged). Had they not 

been excluded, the differences in read depth and large number of zeroes in the data 

would have drastically reduced the sensitivity and specificity of analyses. Unless 



50 

explicitly noted, the pre-inoculation and post-inoculation samples were not included in 

any analyses. The remaining samples were ‘denoised’ with PERFect to remove rare 

and uninformative OTUs. 

Bacterial Community Analyses 

Bacterial Family Relative Abundance 

 Filtered and decontaminated data were agglomerated to family taxonomic 

level, converted to proportions, and plotted in Figure 3-1. Because no statistical 

comparisons were being made, pre-inoculation and post-inoculation time point 

samples were included. Taxa with mean relative abundance less than 0.025 across all 

samples were collapsed into an ‘other’ category. For early, middle, end, and packaged 

samples, most of the community was represented by seven families: 

Leuconostocaceae, Moraxellaceae, Streptococcaceae, Staphylococcaceae, 

Micrococcaceae, Bacillaceae, and Enterococcaceae. The multigrain tempeh seemed 

to exhibit more variability and had larger ‘other’ portions than the soy tempeh. The 

pre-inoculation and post-inoculation samples were poorly represented by these seven 

families and exhibited very large ‘other’ portions compared to later time points. 

However, many of the families that appeared dominant in the other samples were 

observed in the pre-inoculation and post-inoculation samples, notably 

Leuconostocaceae, Moraxellaceae, and Streptococcaceae. 



51 

 
Figure 3-1. Bacterial Family Relative Abundance 
Relative abundances of bacterial families in individual samples grouped by product 
and time point. Only families present at mean relative abundance greater than 0.025 
across all samples were plotted; other families were collapsed into ‘other’. Plotted 
from family-agglomerated proportion data. 

Bacterial Alpha Diversity 

 Several alpha diversity metrics were calculated for samples from denoised and 

rarefied data: Shannon-Wiener diversity (44, 45), Inverse Simpson diversity (46), 50% 

Coverage diversity, Observed richness, Chao1 richness (47), Pielou evenness (48), 

Relative dominance, Core Abundance dominance, and Low Abundance rarity. A 

description of these metrics can be found in Appendix Table 3-5. Samples were then 

grouped by product, time point, and a combined product*time point factor. Figure 3-2 

shows a boxplot comparison of the various metrics with samples grouped by product. 



52 

For simplicity, only the product comparisons were plotted. Multigrain tempeh 

exhibited generally higher diversity, richness, and evenness, whereas soy tempeh 

exhibited higher dominance; both exhibited similar rarity, although this measure is 

highly impacted by denoising with PERFect. 

 
Figure 3-2. Bacterial alpha diversity comparisons 
Boxplot comparisons between products of calculated bacterial alpha diversity metrics. 
Alpha diversity metrics were calculated for every sample from rarefied OTU-level 
count data. 

 A Kruskal-Wallis test was used to determine whether any of the alpha diversity 

metrics differed overall between group levels. Significant differences (p < 0.05) were 

observed between products for Shannon-Wiener, Inverse Simpson, 50% Coverage, 

and Core Abundance metrics. None of the metrics significantly differed (p < 0.05) 

when samples were grouped by time point. Only Shannon-Wiener and Core 



53 

Abundance metrics were significantly different (p < 0.05) when grouped by the 

product*time point interaction factor. Pairwise comparisons were not made due to the 

limited number of samples for each time point; the product grouping is inherently 

paired. Multiple comparisons were controlled for by the false discovery rate method. 

See Appendix Table 3-5 for a complete listing of significance values from the various 

comparisons.  

Bacterial Beta Diversity Analysis 

 A distance matrix was computed using the Bray-Curtis (49) method between 

samples on denoised species-agglomerated proportion data. Nonmetric 

multidimensional scaling (NMDS) was used to ordinate the samples in 2D, as seen in 

Figure 3-3. The ordination was grouped in two ways based on product and time point 

levels. The product ordination shows the presence of two separate clusters for soy and 

multigrain tempeh. The time point ordination shows the presence of three separate 

clusters for middle, end, and packaged time points, and a completely dispersed cluster 

from early time points that spans the other time points. Interestingly, the end and 

packaged time points are more compact compared to the early and middle time points. 



54 

 
Figure 3-3. Bacterial Bray-Curtis NMDS ordination 
Nonmetric multidimensional scaling ordination of bacterial Bray-Curtis dissimilarities 
grouped by product and time point. Bray-Curtis dissimilarities were calculated 
between all samples from species-agglomerated proportion data. 

 Statistical analysis was then conducted using the distance matrix to determine 

whether product, time point, and their interaction product*time point were 

significantly associated with the observed sample differences. To determine the 

marginal effect of each of these factors, a distance-based redundancy analysis was 

conducted (dbRDA) and an analysis of variance (ANOVA) was used to interpret the 

result. The ANOVA showed that product (p < 0.01) and time point (p < 0.05) 

significantly influenced the dbRDA but that their interaction did not (p = 0.072). A 

permutational multivariate analysis of variance was then conducted (PERMANOVA) 

(50) to determine which species were most associated with the observed community 



55 

dissimilarities. Two PERMANOVA models were run, one analyzing product while 

controlling for time point, and the other vice versa. The interaction of product*time 

point was not included due to the result of the dbRDA. The two PERMANOVA 

results agreed with the dbRDA analysis, showing that both product (p < 0.01) and time 

point (p < 0.05) were significantly associated with differences in the observed 

community dissimilarities. The top 10 species coefficients contributing to product 

differences were plotted in Figure 3-4. A Leuconostoc sp. and Weissella sp. were 

most strongly associated with soy tempeh compared to multigrain tempeh, whereas 

Enterococcus italicus and Acinetobacter tandoii were most strongly associated with 

multigrain tempeh compared to soy tempeh. Pairwise comparisons of species 

coefficients between time points were not made. PERMANOVA analysis can result in 

false positives due to different sample dispersions between levels; a multivariate 

homogeneity of groups dispersions (PERMDISP2) (51) test was conducted to 

determine whether dispersions significantly differed between group levels. ANOVA 

test of the PERMDISP2 result showed that dispersions did not significantly differ (p > 

0.05) between product levels, but dispersions did significantly differ between time 

point levels (p = 0.0325). 

  



56 

 

 
Figure 3-4. Bacterial species PERMANOVA coefficients for product type 
Top ten (absolute value) bacterial species coefficients contributing to community 
dissimilarity due to product, as determined by PERMANOVA analysis. Positive 
values indicate the species was observed more frequently in soy tempeh, whereas 
negative values indicate the species was observed more frequently in multigrain 
tempeh. UR preceding the species name indicates the OTU was taxonomically 
unresolved; the four-character hash is unique to the specific OTU. Bray-Curtis 
dissimilarities were calculated between all samples from species-agglomerated 
proportion data. 

Bacterial Core Species 

 To further investigate community differences, core species were identified 

between product and time point levels using denoised species-agglomerated 

proportion data. Core species were defined as species present at a relative abundance 

greater than 0.005 in greater than 66% of samples within group levels. To better 

understand core species, the mean abundance of core species within group levels were 

calculated. The number, identification, and mean abundance of core species was 

visualized in Figure 3-5; mean abundances were negative natural log transformed to 

emphasize differences in abundance between group levels. 



57 

 
Figure 3-5. Core bacterial species and differential abundance analysis 
Core bacterial species grouped by product and time point. Venn diagrams indicate the 
number of shared core species between levels of each group. Heat maps show negative 
natural log transformations of mean proportion data for core species within levels of 
each group; blue indicates high mean abundance and yellow indicates low mean 
abundance. Species name coloring corresponds to which group levels species are core 
to, and match Venn diagram region label coloring. UR preceding the species name 
indicates the OTU was taxonomically unresolved; the four-character hash is unique to 
the specific OTU. Core species were defined as those species present at a relative 
abundance greater than 0.005 in greater than 66% of samples within levels of each 
group. Species that were not identified as core to any group level were collapsed into 
‘other’. Analysis was done with species-agglomerated proportion data. 

 Seven species were identified as core to both soy and multigrain tempeh; 

multigrain tempeh had six unique core species, and soy tempeh had two unique core 

species. Eight species were identified as core to early, middle, end, and packaged time 

points; early and middle time points shared one unique core species, and middle and 

packaged time points each had one unique core species. Many of the core species 



58 

identified between products were also identified as core between time points. These 

species included three Leuconostoc spp., Enterococcus italicus, two Lactococcus spp., 

and Kocuria rhizophila. Species identified as uniquely core to one group level were 

generally more abundant than those same species in other levels. Interestingly, core 

species identified in different time point levels generally appeared to become less 

abundant at later stages of the fermentation. 

Bacterial Differentially Abundant Species 

 Differential abundance (DA) analysis was conducted using ZicoSeq on 

denoised species-agglomerated count data. Species abundance was compared within 

product and time point groups separately while controlling for the other factor; no 

pairwise comparisons were made between different time points. Only species present 

in at least 33% of samples within a group and with mean relative abundance greater 

than 0.0005 were included. Multiple comparisons were controlled by the false 

discovery rate method. Core species that were significantly different (p < 0.05) 

between soy and multigrain tempeh are listed in Table 3-1. The four most 

significantly differentially abundant core species between soy and multigrain tempeh 

were all Acinetobacter spp. (p ≤ 0.001). Non-core species were also identified as 

differentially abundant. No species were found to be significantly differentially 

abundant between time point levels. 

  



59 

Table 3-1. Significantly differentially abundance core bacterial species 
Significantly differentially abundant (p < 0.05) core bacterial speciesa,b between Product 
levels identified using ZicoSeq. 
   Soy Multigrain 
Species p padj

c μ σ/ μ μ σ/ μ 
Acinetobacter tandoii 1.0 × 10-4 1.7 × 10-5 5.4 × 10-4 2.2 5.0 × 10-2 0.86 
URd Acinetobacter 5b17 1.0 × 10-4 1.7 × 10-5 8.3 × 10-4 1.2 3.1 × 10-2 0.66 
UR Acinetobacter 3099 1.0 × 10-4 1.7 × 10-5 2.9 × 10-4 2.7 4.2 × 10-2 0.72 
Acinetobacter bereziniae 1.0 × 10-4 2.0 × 10-5 1.8 × 10-4 1.4 1.5 × 10-2 0.75 
UR Weissella d395 3.0 × 10-4 3.6 × 10-4 9.4 × 10-2 0.88 5.7 × 10-3 0.97 
UR Weissella ac2f 5.0 × 10-4 8.7 × 10-4 1.1 × 10-2 0.94 5.3 × 10-4 1.5 
UR Kurthia cfe3 6.0 × 10-4 1.6 × 10-3 1.1 × 10-3 1.8 1.2 × 10-2 0.91 
Enterococcus italicus 1.5 × 10-3 1.8 × 10-3 5.4 × 10-2 0.44 1.1 × 10-1 0.41 
UR Enhydrobacter 006f 3.1 × 10-3 5.2 × 10-3 1.8 × 10-3 3.8 8.2 × 10-3 0.76 
Lactococcus lactis 1.1 × 10-2 1.9 × 10-2 7.2 × 10-2 0.59 3.1 × 10-2 0.47 
a Species-agglomerated count data. 
b Core bacteri