Point-of-use colorimetric detection of Escherichia coli in food matrices with DNAzyme crosslinked hydrogels

TL;DR

A new sensor detects E. coli in food using a color change, enabling easy on-site monitoring to prevent foodborne illness.

Contribution

A DNAzyme-crosslinked hydrogel sensor for point-of-use detection of E. coli in various food matrices is developed.

Findings

The sensor can detect E. coli at 105 CFU mL-1 in milk and food samples.

The platform works with produce, leafy greens, and ready-to-eat foods.

The method is equipment-free and visually detectable.

Abstract

Due to the significant healthcare burden associated with foodborne illness, developing platforms suitable for the on-site detection of food pathogens is of critical importance to public health. Low-cost, equipment-free approaches are desired to allow for point-of-use contamination monitoring along the food supply chain. Here, we demonstrate the compatibility of an Escherichia coli responsive colorimetric DNAzyme-crosslinked hydrogel sensor with a wide range of food products. Sensor functionality involves an E. coli detecting DNAzyme-substrate complex that cleaves the hydrogel crosslinking in the presence of the target bacteria, resulting in a release of gold nanoparticles that is visible to the naked eye. Naked-eye detection of E. coli at concentrations of 105 CFU mL-1 has been shown in milk as well as samples extracted from produce, leafy greens, and ready-to-eat foods such as rotisserie chickens. The functionality, simplicity, and versatility of this sensing platform may improve the feasibility of frequent pathogen monitoring in the food production pipeline, with the potential to mitigate future outbreaks of foodborne illness.

Figures

Microscale and macroscale schematics of the sensing platform.A Microscale functionality of the sensing platform. Phage infection of *E. coli* bacteria and the resulting cell lysis increases the release of target protein. This target protein triggers DNAzyme cleavage and breakdown of the polyacrylamide hydrogel crosslinking, releasing the trapped gold nanoparticles. B Macroscale depiction of the hydrogel sensing platform. An intact gel is visible in the absence of *E. coli*, while a contaminated sample results in a visibly degraded gel. Schematics created using Biorender.

Naked-eye visible bacterial detection and sensor performance with various food products.A Categories of foods that can be contaminated with pathogenic *E. coli*, and their known avenues of contamination. B Optical images of *E. coli* detection in milk. C Detection performance in milk at 10^5^ CFU mL^-1^ *E. coli*. Sample size *n* = 4 at each concentration. D Optical images of *E. coli* detection in RTE chicken purge. E Detection performance in RTE chicken purge at 10^5^ CFU mL^-1^ *E. coli*. Sample size *n* = 4 at each concentration. F Optical images of *E. coli* detection in RTE eggs. G Detec

Sensor performance and workflow with leafy greens.A Optical images of *E. coli* detection for lettuce. B Detection performance for lettuce at 10^5^ CFU mL^-1^ *E. coli*. Sample size *n* = 4 at each concentration. C Optical images of *E. coli* detection for salad mix. D Detection performance for salad mix at 10^3^ and 10^5^ CFU mL^-1^ *E. coli*. Sample size *n* = 4 at each concentration. E Schematic of the process to extract and test a liquid sample from contaminated leafy greens. F *E. coli* colony counts extracted from the spiked samples of leafy greens after stomaching, plotted on a log 10 s

Full text

Introduction

Foodborne pathogens continue to be major causes of illness, in North America and around the world. The World Health Organization has estimated that foodborne hazards resulted in 600 million illnesses in 2010, and 420 000 deaths^1^. This includes contamination with pathogenic strains of Escherichia coli, such as E. coli O157. Pathogenic E. coli strains have been known to contaminate various categories of food including meat, dairy, and produce, such as leafy greens, sometimes resulting in extensive recalls^2–5^. While food recalls are often only instituted after reports of illness, this retroactive detection can result in a delayed response and thus more widespread infection. In light of this, regular monitoring at various stages of the food supply chain is key for the detection of pathogenic bacteria before contaminated products are consumed.

Conventional methods of pathogen detection in food include bacterial culturing, as well as strategies such as polymerase chain reaction (PCR). However, these approaches typically require equipment, user training, and a laboratory environment, limiting their suitability for on-site testing applications. This renders them impractical for some stages of the food supply chain, such as at the retail or consumer level.

To address the general need for on-site bacterial testing, numerous pathogen sensing platforms have been developed. This can include immunoassays, enzymatic assays, nucleic acid-based methods including loop-mediated isothermal amplification, and electrochemical detection methods, such as impedance flow cytometry^6–8^. However, not all of these on-site platforms are compatible with food testing. The complex nature of food matrices may interfere with some biosensors through non-specific binding, sensor fouling, or other mechanisms^8,9^. In particular, samples extracted from food contain various salts, proteins, fats, polysaccharides, and other particulate matter^10,11^. These can interfere with sensing platforms in multiple ways, such as through non-specific binding between these matrix components and sensor biorecognition elements, suppression of sensor optical, fluorescent, or enzymatic signals by the matrix, or low recovery of pathogens bound to fats or proteins^12^. Enzymes in food samples, such as endogenous nucleases may contribute to degradation of certain nucleic acid-based sensor types^13^. The high prevalence of non-target bacteria in many food samples may also pose a challenge for sensors that lack bacterial species specificity. Variable pH of food samples has additionally been reported as an issue for some platforms^12^, as well as sample viscosity^13^. For these reasons, many sensing approaches for food require pre-processing steps to separate bacteria from other components of sample, increasing platform complexity^10,14^.

There have, however, been many colorimetric food pathogen sensing platforms reported in the literature despite these challenges. This includes lateral flow immunoassays, such as enzyme-linked immunosorbent assays which use antibodies to detect pathogenic bacteria in food samples^15^. This type of platform is typically used as one-time test of an extracted sample, often less suitable for continuous monitoring, such as in-package sensing. Bacterial capture and separation with coated magnetic particles has also been a popular strategy for target amplification, often paired with colorimetric readout^14^. Loop mediated isothermal amplification (LAMP) assays have been used to colorimetrically detect bacterial DNA in food samples, often in the form of microfluidic chips, but require a nucleic acid extraction step to function^16^. Colorimetric sensing platforms for on-site food pathogen detection are extensive, and have been thoroughly reviewed elsewhere^17–19^.

Our group has previously reported a hydrogel biosensor for the on-site colorimetric detection of E. coli^20^. The hydrogel is made up of polyacrylamide, and is crosslinked together with a type of catalytic nucleic acid called DNAzymes. RNA-cleaving DNAzymes have a substrate strand and an enzymatic strand, which cleaves the substrate in response to a specific target^21^. For this sensing platform, a DNAzyme responsive to a target protein from E. coli was chosen (Table S1)^22^. This target protein has been referred to in previous works as EC1^22–24^, and is known to be expressed by multiple E. coli strains^20^. Upon encountering the target protein, the DNAzyme crosslinking holding together the polyacrylamide hydrogel breaks down as the substrate strand is cleaved, and the gel degrades. This allows for the highly sensitive and specific detection of E. coli in an equipment-free colorimetric platform. In our previous work, however, the effectiveness of the sensing platform was only demonstrated for water and urine samples.

Herein, we demonstrate the compatibility of our hydrogel E. coli sensor with a wide range of food samples. The sensing platform performed colorimetric, naked-eye detection with samples from contaminated milk, produce, and ready-to-eat (RTE) food products in 18 h. Detection has also been demonstrated with a known vector of E. coli, contaminated leafy greens. This widespread applicability to food products at risk of E. coli contamination increases the real-world relevance of our biosensing platform for infection prevention. To our knowledge, this is the first report of using a DNAzyme-crosslinked hydrogel sensor for food pathogen detection.

Results and Discussion

Design and Functionality of the Biosensing Platform

The basic design of our sensing platform has been described in our previous publication^20^. It consists of polyacrylamide chains in which the acrylamide monomers have been co-polymerized with acrydite-modified oligonucleotides during free-radical polymerization. This results in the oligonucleotide strands, which are an E. coli responsive DNAzyme and substrate pair, being grafted to the polymer backbone. Complementary binding between the DNAzyme and substrate oligonucleotides crosslinks the polymer together to form a hydrogel, and bovine serum albumin (BSA) coated gold nanoparticles (AuNPs) are entrapped in the polymer matrix to achieve a visible red color.

To function, the hydrogel sensor is combined with buffer, growth media, a potentially E. coli contaminated liquid sample, and T7 bacteriophage. This species of bacteriophage targets E. coli^25,26^, inducing cell lysis in the infected bacteria and thus increasing the amount of target protein available to the sensor. If E. coli is not present in the sample, the gel will remain intact. However, if E. coli is present, the target protein will interact with the DNAzyme-substrate complex (Fig. 1A). This induces cleavage of the oligonucleotide crosslinking holding the gel together, and thus the degradation of the gel matrix. The difference between an intact and degraded gel can be seen on a macroscale with the naked eye (Fig. 1B). The visual result has also been measured with ImageJ to produce the graphs seen in this publication.Fig. 1. Microscale and macroscale schematics of the sensing platform.A Microscale functionality of the sensing platform. Phage infection of E. coli bacteria and the resulting cell lysis increases the release of target protein. This target protein triggers DNAzyme cleavage and breakdown of the polyacrylamide hydrogel crosslinking, releasing the trapped gold nanoparticles. B Macroscale depiction of the hydrogel sensing platform. An intact gel is visible in the absence of E. coli, while a contaminated sample results in a visibly degraded gel. Schematics created using Biorender.

Sensor Functionality with a Range of Food Samples

As previously mentioned, E. coli contamination can occur in a wide variety of food products including dairy products, produce, meat, and RTE foods (Fig. 2A). The E. coli O157 strain is particularly notable, causing severe illness and with an infectious dose estimated to be as low as under 100 colony forming units (CFU)^27^. Due to this hazard the USDA has a zero-tolerance policy for E. coli O157:H7 in ground beef, employing culture pre-enrichment for highly sensitive detection^28^. E. coli is ubiquitous in the digestive system of cattle, with some strains being enteropathogenic to humans^2^. Thus, fecal contamination of dairy processing equipment can result in the presence of E. coli in unpasteurized or improperly pasteurized milk. Contamination of produce with pathogenic E. coli is also a concern, especially as vegetables are frequently eaten raw. This is often due to contaminated soil, irrigation water, or manure that has been inadequately composted^3,29^. Leafy greens are a particular concern, and have been estimated by some to be responsible for around half of reported foodborne disease outbreaks in developed countries^30^. While raw meat, in particular beef, is known to be a vector for pathogenic E. coli, the bacteria is typically killed during the cooking process^4,31^. However, bacteria may remain if the cooking temperature is not adequate. Undercooked raw ingredients such as these are one pathway for E. coli contamination of RTE food products, in addition to improperly sterilized equipment and food preparation areas.Fig. 2. Naked-eye visible bacterial detection and sensor performance with various food products.A Categories of foods that can be contaminated with pathogenic E. coli, and their known avenues of contamination. B Optical images of E. coli detection in milk. C Detection performance in milk at 10^5^ CFU mL^-1^ E. coli. Sample size n = 4 at each concentration. D Optical images of E. coli detection in RTE chicken purge. E Detection performance in RTE chicken purge at 10^5^ CFU mL^-1^ E. coli. Sample size n = 4 at each concentration. F Optical images of E. coli detection in RTE eggs. G Detection performance in RTE eggs at 10^5^ CFU mL^-1^ E. coli. Sample size n = 5 at each concentration. H Optical images of E. coli detection in carrots. I Detection performance in carrots at 10^5^ CFU mL^-1^ E. coli. Sample size n = 4 at each concentration. All bacterial concentrations are given as CFU mL^-1^. Brightness values on each graph have been normalized to the mean of their respective controls. All asterisks represent significant differences at corresponding significance levels, and error bars represent standard error of the mean. Schematics created using Biorender.

To address these varied applications, the sensing platform has been tested with several foods. First, milk was spiked to a concentration of 10^5^ CFU mL^-1^ E. coli and tested with the sensing platform as described in the methods section. The first iteration of the test was unsuccessful, as the viscosity of the solution was too high for sensor functionality (Fig. S1). To mediate this, the spiked milk was diluted as described in the methods before being added to the sensor. Here, E. coli was successfully detected after 18 h at 37 °C. The signal can be interpreted optically with the naked eye (Fig. 2B) and has also been measured with ImageJ to show statistical significance (Fig. 2C). In a supplementary limit of detection test (Fig. S2), the platform detected contaminated milk spiked with E. coli on the order of 10^4^ CFU mL^-1^ in the original spiked milk sample, corresponding to a concentration on the order of 10^2^ CFU mL^-1^ in the diluted milk sample being added to the sensor.

Next, E. coli detection in RTE foods was assessed, including rotisserie chicken and hard boiled eggs. Chicken purge from a grocery store RTE rotisserie chicken was spiked with 10^5^ CFU mL^-1^ E. coli, which was then detected with our sensing platform both optically (Fig. 2D) and graphically (Fig. 2E). Similar to the milk experiment, dilution of the spiked chicken purge was necessary due to viscosity challenges. The RTE eggs were individually packaged suspended in liquid, which was extracted and spiked with E. coli. Successful detection at 10^5^ CFU mL^-1^ was also achieved for these samples (Fig. 2F, G). Since RTE foods are intended for consumption without further washing or cooking steps, detection of bacterial contamination in these types of products can be a last line of defense before pathogen exposure.

To assess the functionality of the sensing platform with produce, the liquid from a bag of baby-cut carrots was extracted and spiked to a concentration of 10^5^ CFU mL^-1^ E. coli. Like with the previously described food samples, this was successfully detected visually and when measured (Fig. 2H, I). Carrots have been known to experience E. coli contamination, including during a notable 2024 recall across North America that sickened 48 with E. coli O121:H19^32^. Thus, the sensor compatibility with carrots has a strong relevance for future real-world applications.

In summary, the demonstrated functionality of our sensing platform with a range of food products is critical for its practical use. The versatility of the platform may make it suitable for use at the general retail and consumer level, where using a different type of E. coli sensor for each category of food would likely not be feasible or practical.

Sensor Functionality with Leafy Greens

While several food products of interest have been discussed, E. coli contamination of leafy greens specifically is a notable public health issue^33^. In light of this, the compatibility of the sensor with this type of food product was tested using multiple approaches for sample extraction.

In many grocery stores, the produce section is frequently misted with water to maintain optimal humidity. The resulting droplets of water from the surface of a head of iceberg lettuce were collected, and spiked with 10^5^ CFU mL^-1^ E. coli. The contaminated samples were successfully identified by the sensing platform, as shown visually (Fig. 3A) and on the graph (Fig. 3B). However, extracting a liquid sample from lettuce in such a way may not be practical for real-world applications. Therefore, we subsequently explored other methods of extracting liquid samples from lettuce for bacterial detection.Fig. 3. Sensor performance and workflow with leafy greens.A Optical images of E. coli detection for lettuce. B Detection performance for lettuce at 10^5^ CFU mL^-1^ E. coli. Sample size n = 4 at each concentration. C Optical images of E. coli detection for salad mix. D Detection performance for salad mix at 10^3^ and 10^5^ CFU mL^-1^ E. coli. Sample size n = 4 at each concentration. E Schematic of the process to extract and test a liquid sample from contaminated leafy greens. F E. coli colony counts extracted from the spiked samples of leafy greens after stomaching, plotted on a log 10 scale. Each bar is a separate stomaching test, with n = 3 technical replicates for each. G Optical images of E. coli detection in the stomached leafy greens samples. H Detection performance for stomached leafy greens at initial spiked concentrations of 10^5^ and 10^7^ CFU g^-1^ E. coli. Sample size n = 4 at each concentration. Brightness values on each graph have been normalized to the mean of their respective controls. All asterisks represent significant differences at corresponding significance levels, and error bars represent standard error of the mean. Schematics created using Biorender.

As it is customary for leafy greens to be washed before being used in salads, wash water is another potential avenue for acquiring a liquid sample from such products. A purchased salad mix containing romaine lettuce, kale, spinach, and other greens was rinsed using a salad spinner, with the wash water being collected. This wash water was subsequently spiked with 10^3^ or 10^5^ CFU mL^-1^ E. coli. Both of these contamination concentrations were detected by the platform, visually and when measured (Fig. 3C, D).

Up until this point for the experiments with leafy greens, the extracted surface or wash water was being directly spiked with a given concentration of E. coli. For an actual contaminated product, the bacteria may not necessarily all be extracted into the liquid sample under these conditions. To account for this, another test was completed, this time involving contaminating the food product directly before extracting a liquid sample. This time a stomaching approach was used to break down the product, in order to more accurately simulate the process of sample collection (Fig. 3E).

Leafy greens from the previously mentioned salad mix were contaminated with E. coli at concentrations on the order of 10^5^ and 10^7^ CFU g^-1^. Liquid samples were then extracted using a stomaching approach, as described in the methods. The recovered E. coli concentrations in the liquid samples after stomaching were on the order of 10^4^ and 10^6^ CFU mL^-1^ as determined by selective plating, with the control displaying no E. coli growth (Fig. 3F). These liquid samples were then diluted before being added to the sensing platform, to prevent interference from the large amount of particulates. Detection was successful at both tested concentrations, shown in the optical images and the graph (Fig. 3G, H). It is also notable that during selective plating on MacConkey agar all samples including the control displayed some non-E. coli bacterial growth, as evidenced by the presence of non-lactose fermenting colonies. This is to be expected, as food samples such as these are non-sterile and will contain some background microorganisms (Fig. S3). The ability of the sensing platform to specifically detect E. coli without false positive signals for other bacteria is critical for its application in food testing, which often involves the presence of multiple non-target bacteria. The specificity of this DNAzyme for E. coli has also been previously shown in tests against other bacterial species including Listeria monocytogenes, Pseudomonas aeruginosa, and methicillin-resistant Staphylococcus aureus^20^.

In conclusion, in our previous work we developed a novel, phage-amplified hydrogel biosensing platform for the on-site colorimetric detection of E. coli in water and urine^20^. Now in this work we have significantly expanded the potential real-world applications of the platform, by demonstrating its compatibility with a wide variety of foods that have been vectors for E. coli infection. The successful detection of E. coli contamination from food matrices including milk, RTE chicken purge, and leafy greens supports the possible utility of this platform for on-site contamination monitoring along the food supply chain. With this success in mind, a potential avenue of future exploration is to develop similar platforms for the detection of other foodborne pathogens of interest. DNAzymes for other such bacteria including Salmonella Typhimurium exist^34^, and adapting the DNAzyme crosslinked hydrogel platform is a possible future research direction. Strategies for increasing the platform’s sensitivity and reducing its detection time would likely need to be explored for real-world implementation, potentially involving pre-enrichment of samples. While the current 18-hour detection time is comparable to conventional culture-based methods, modern PCR approaches as well as many novel biosensors reported in the literature operate on significantly faster time scales^15,35–37^. Another possible line of inquiry is methods of incorporating the sensor into food packaging. In-package biosensing to detect foodborne pathogen contamination is an active area of research, offering the benefit of continuous monitoring^38^. In such applications, the extended sample contact also makes a fast detection time less critical. Overall, this study combined with our previous work demonstrates the strong versatility of our sensing platform, including its applicability to on-site clinical testing, water, and food safety monitoring, albeit with limitations in terms of sensitivity and detection time. This wide-ranging functionality, as well as its naked-eye colorimetric signal, makes it a promising candidate for future real-world applications in the food production pipeline if its limitations can be addressed.

Methods

Materials

The following reagents were purchased from Millipore Sigma (Ontario, Canada): Ammonium persulfate (APS), N,N,N′,N′-Tetramethyl ethylenediamine (TEMED), 40% acrylamide monomer solution, gold (III) chloride solution, trisodium citrate dihydrate, bovine serum albumin (BSA), and 1.0 M MgC1_2_ buffer. Bacto^TM^ yeast extract and dehydrated MacConkey agar were purchased from ThermoFisher Scientific (Ontario, Canada). Oligonucleotides with the acrydite modification were custom ordered from Integrated DNA Technologies (IDT) (Iowa, USA). All food products were purchased from local grocery stores in Hamilton (Ontario, Canada).

Bacterial Preparation

E. coli K12 was cultured from a glycerol stock in LB media. Overnight culture was grown in a shaking incubator at 37 °C and 180 RPM. Before use in the experiments, the liquid culture was centrifuged for 15 minutes at 7000 RCF, 4 °C, before being resuspended in ultrapure water.

Milk Sample Preparation

Skim milk was spiked with resuspended E. coli K12 to a final concentration of 10^5^ CFU mL^-1^ (or concentrations ranging from 10^3^ to 10^5^ CFU mL^-1^, for the supplementary experiments shown in Fig. S2). This contaminated milk was then serially diluted in ultrapure water to a 1:100 dilution before being added to the sensing platform.

RTE Chicken Sample Preparation

Liquid from the bottom of RTE rotisserie chicken trays was collected. This chicken purge was spiked with resuspended E. coli K12 to a final concentration of 10^5^ CFU mL^-1^, before being serially diluted in ultrapure water to a 1:100 dilution.

RTE Egg Sample Preparation

Liquid from a package containing a hardboiled RTE egg was collected, and spiked with resuspended E. coli K12 to a final concentration of 10^5^ CFU mL^-1^. This contaminated liquid was added to the sensor without dilution.

Carrot Sample Preparation

Liquid from a package of RTE baby-cut carrots was collected, and spiked with resuspended E. coli K12 to a final concentration of 10^5^ CFU mL^-1^. This contaminated liquid was added to the sensor without dilution.

Leafy Green Sample Preparation

For the iceberg lettuce, mist machines at the grocery store resulted in droplets of water on the surface of the head of lettuce. These droplets were collected, and spiked with resuspended E. coli K12 to a final concentration of 10^5^ CFU mL^-1^. This contaminated liquid was added to the sensor without dilution. For the next leafy greens experiment, 142 g of salad mix containing romaine lettuce, kale, spinach, and other greens was rinsed with 250 mL of tap water in a salad spinner. The resulting wash water was spiked with resuspended E. coli K12 to final concentrations of 10^3^ and 10^5^ CFU mL^-1^. This contaminated liquid was added to the sensor without dilution. For the final experiment, 1.5 g of salad mix was contaminated with 1 mL of 10^5^ or 10^7^ CFU mL^-1^ E. coli K12. 10 mL of ultrapure water was added to each tube of 1.5 g salad mix, and each sample was homogenized using a stomaching approach. The resulting liquid was then serially diluted in ultrapure water to a 1:100 dilution before being added to the sensing platform. pH of the samples was measured with a Mettler Toledo SevenExcellence benchtop pH Meter (Figure S4).

Bacterial Plating

Liquid samples were serially diluted in triplicate and then plated on selective MacConkey agar. Plates were subjected to overnight incubation at 37 °C, followed by colony counting to determine CFU mL^-1^.

Sensor Fabrication

Hydrogel sensors were fabricated using the procedure outlined in our previous publication^20^. Briefly, acrylamide monomers were co-polymerized with each of the two acrydite modified oligonucleotides, through APS and TEMED-induced free radical polymerization at 45 °C. BSA coated gold nanoparticles were added. The solution co-polymerized with the DNAzyme oligonucleotide was then mixed with the solution co-polymerized with the substrate oligonucleotide over a heat block at 95 °C, in small PCR tubes containing 3 μL of each solution. Complementary binding between the oligonucleotides resulted in the formation of gels. The resulting 6 μL gels were kept frozen at -20 °C until ready to be used. 24 h before an experiment, these gels were taken out of the freezer and pipetted into smaller 2 μL gels, to each of which 15 μL of MgCl_2_ buffer was added.

Sensor Use

To each tube containing the hydrogel sensor and buffer, 15 μL of 50 mg mL^-1^ autolyzed yeast extract and 70 μL of the potentially contaminated sample were added. After incubation at 37 °C and 80 RPM for 6 h in a Thermo Scientific MaxQ 6000 shaking incubator, 70 μL of T7 bacteriophage at a concentration of 10^9^ plaque forming units (PFU) mL^-1^ (prepared as described in our previous publication)^20^ was added to each tube. 12 h in the shaking incubator later, after a total incubation time of 18 h, samples were removed for imaging.

Image Interpretation

Tubes containing the gel samples were imaged on the Epson Perfection V850 Pro scanner. ImageJ software was used to measure the mean brightness of a specific, consistently defined area of each tube.

Statistical Analysis

All graphs were created and statistics analyzed in GraphPad Prism. Unpaired two-tailed t-tests with Welch’s correction were used to compare two groups. One-way Brown-Forsythe ANOVA with Dunnett’s multiple comparison test was used to compare three or more groups. Experiment sample sizes are described in their respective figure captions. Asterisks on the graphs represent significance levels according to ns (p > 0.05), ^^ (p ≤ 0.05), ^^ (p ≤ 0.01), ^^ (p ≤ 0.001), and ^****^ (p ≤ 0.0001).

Supplementary information

Supplementary Information