Determination and quantification of the microbial communities and antimicrobial resistance genes on food

Abstract

A method of determining and / or quantifying the microorganism communities and genes present on food, said method including the steps of; mixing a sample of said at least one foodstuff with a protease, introducing a triterpene glycoside or saponin to the mixture, introducing a DNase to the mixture, identifying and / or quantifying microorganisms and / or genes by DNA sequencing at least part of the mixture and subjecting the same to metagenomic analysis.

Description

[0001] Determination and quantification of the microbial communities and antimicrobial resistance genes on food

[0002] The invention to which this application relates is to a method for depleting host DNA from foods of animal or plant origin and determination and / or quantification of the microbial communities and antimicrobial resistance genes present.

[0003] It is estimated that foodborne pathogens are responsible for 2.4 million infections in the United Kingdom each year (Holland and Mahmoudzadeh, 2020) and cost the economy £9.1 billion each year (Daniel et al., 2020). Pathogens which are resistant to antimicrobial drugs compound the problem, as these infections may result in treatment failure (WHO, 2020). Identification of pathogens on food is most commonly achieved through the isolation of individual bacterial species of interest followed by confirmatory biochemical testing, a process which can take up to seven days (Mooijman et al., 2019). In terms of identifying problems and applying interventions, this has several drawbacks; first, the shelf life of foods may be less than the time it takes to confirm the presence of pathogens, and the amount of pathogens on food may increase over time (Meldrum et al., 2014) . Second, this approach will only identify the specific bacteria targeted; if another pathogen is present but which is not the subject of investigation, it will be missed. Third, non-pathogenic bacteria can act as reservoirs of antimicrobial resistance (AMR) genes which can be transferred to pathogenic bacteria (Rolain, 2013), thus contributing to the burden of disease in humans. Culture-independent approaches, where identification of microbes does not depend on culture and isolation, offer an alternative which can overcome these drawbacks. All food products, of both animal and plant origin, will carry bacteria unless specifically sterilised. Foods vary in the number and types of microorganism with which they are colonised (Mira Miralles et ah, 2019) . These microorganisms may be pathogenic (Gruetzke et al., 2019), associated with food spoilage (Cauchie et al., 2020) and may additionally be resistant to antimicrobials (de Souza et al., 2019). Of the various culture-free approaches, 16S sequencing has helped quantify food bacterial diversity (Zhang et al., 2016), whilst multiplex PGR methods have helped identify the presence of specific genes of interest, such as AMR genes (de Paula et al., 2018) . However, 16S sequencing does not target non-bacterial microorganisms, and novel AMR genes are routinely discovered, requiring continually revised databases and PGR methods for their detection. In addition, these methods cannot determine what genes are associated with which microorganisms and vice versa. Bacteria can have multiple copies of the ribosomal operon, and variation between these copies can be greater than between different bacterial species (Johansen et al., 2017; Vetrovsky and Baldrian, 2013); in general 16S classifications below the genus level are not very accurate (Gruetzke et al., 2019) . Metagenomic methods which sequence the DNA of all microorganisms are required to understand fully the breadth and diversity of microorganisms and genes present on food samples, and how they relate to each other.

[0004] Shotgun metagenome sequencing involves sequencing all DNA in a sample. It is difficult to apply to food samples of animal or plant origin as they frequently contain DNA from the original host that vastly outnumbers the DNA from microorganisms. For example, the chicken (Gallus gallus) genome is 1.2 Gb in length (Bellott et ah, 2017), 240-times larger than the 5 Mb Escherichia coli genome (Makino et al., 1999) . DNA directly extracted and sequenced from foods of animal or plant origin contains a small proportion of bacterial DNA that is sufficient for population identity and diversity estimates but not for detecting specific genetic elements, such as AMR and virulence genes (Kobus et al., 2020) . Multiple methods have been developed to enrich microorganism DNA or deplete host DNA to increase the efficiency of shotgun metagenome sequencing (Escobar- Zepeda et al., 2016) .

[0005] It is therefore an aim of the present invention to provide a method of host depletion that allows efficient sequencing of the metagenomes of food samples.

[0006] It is a further aim of the present invention to provide a method that increases the proportion of prokaryotic reads sequenced for foodstuff samples.

[0007] In a first aspect of the invention there is provided a method of determining and / or quantifying the microorganism communities and genes present on food, said method including the steps of; a) Mixing a sample of said at least one foodstuff with a protease b) Introducing a triterpene glycoside or saponin to the mixture c) Introducing a DNase to the mixture d) Identifying and / or quantifying microorganisms and / or genes by DNA sequencing at least part of the mixture and subjecting the same to metagenomic analysis. In one embodiment the sample is a host food sample wherein the DNA depleted.

[0008] Typically the sample and reagents are mixed in a blender bag. Further typically the mixture is stomached for a period of time.

[0009] In one embodiment the mixture includes water or water is added to the mixture. Typically the water is buffered. Further typically the water is buffered peptone water.

[0010] Typically the protease is an endoprotease. Further typically the endoprotease is a neutral metalloendopeptidase.

[0011] In one embodiment the protease has, and / or is from the Enzyme Commission number EC 3.4.24 class.

[0012] In one embodiment the endoprotease is a neutral metalloendopeptidase from bacillus spp.

[0013] In one embodiment the protease is from the Bacillolysin or Thermolysin family. Typically the protease is Neutrase.

[0014] In one embodiment Illumina dye sequencing is used to sequence the host DNA depleted samples. Typically libraries were sequenced on an Illumina NextSeq.

[0015] The person skilled in the art will appreciate that other sequencing techniques, and in particular other second / nxt generation sequencing techniques could be used to identify and / or quantify the microorganisms and / or genes present.

[0016] In one embodiment the Thermo Fisher Scientific Qubit dsDNA HS assay kit is used to measure the DNA concentration of metagenomic extractions.

[0017] In a second aspect of the invention there is provided a method of host DNA depletion to classify and quantity bacteria and / or antimicrobial resistance genes present on human food. Through this method unbiased estimates of the exposure to bacteria and AMR genes through food can be obtained.

[0018] In one embodiment for each food sample, a host DNA depleted sample was prepared by placing a sample of food into a liter blender bag. Typically buffered peptone water (BPW) is added to the bag. Further typically the sample is stomached.

[0019] In one embodiment protease, for example Neutrase is added to a portion of stomached food sample. Typically the mixture is incubated and / or rotated. Further typically the mixture is centrifuged and the supernatant discarded and re-suspended.

[0020] In a preferred embodiment saponin is added to the mixture. Typically the mixture is incubated and / or rotated. Further typically the mixture is vortexed and / or incubated In one embodiment NaCl is added and vortexed. Typically the mixture is centrifuged before the supernatant is discarded and the pellet resuspended.

[0021] In one embodiment HL-SAN buffer and HL-SAN (ArcticZymes, Tromso, Norway) is added to the mixture. Typically the sample is vortexed and incubated. Further typically the sample is rotated and PBS is added to the solution before being centrifuged. The supernatant is discarded and the pellet resuspended.

[0022] Typically for each food sample, a raw preparation (along with host DNA depletion) was prepared alongside the host DNA-depleted preparation from the food sample by centrifuging 10 ml of stomached sample for 30 minutes at 4,000xg. The supernatant was discarded and re-suspended in 800 pl of PBS.

[0023] Further typically DNA was extracted from raw and depleted samples using a Maxwell® RSC PureFood Pathogen Kit (Promega, Madison, Wisconsin, USA) on a Maxwell® RSC 48.

[0024] In a preferred embodiment for DNA sequencing libraries were constructed from depleted metagenomic extractions using Nextera DNA flex library prep kits (Illumina, San Diego, California, USA) .

[0025] Typically libraries were sequenced on an Illumina NextSeq at 1 Gb per metagenome, as 150 bp paired-end reads. In a third aspect of the invention there is provided a method for depleting host DNA from at least one foodstuff sample, said method including the steps of a) Mixing a sample of said at least one foodstuff with a protease; b) Introducing a detergent (triterpene glycoside or saponin) to the mixture; and c) Introducing a nuclease (DNase) to the mixture

[0026] Specific embodiments of the invention are now described with reference to the following figures:

[0027] Figure 1. Proportion of reads that belong to each superkingdom and were unclassified for three chicken samples (A), three pork samples (B), three salmon samples (C), three prawn samples (D), three leafy green samples (E), a Zymo mock community (F), and a blank (G) that were not processed (raw), processed using a Molysis kit and the Neutrase-saponin method.

[0028] Figure 2. Bray-Curtis distance between metagenomes depleted using the Molysis kit or Neutrase-saponin method and the raw sample; and coloured by sample type.

[0029] Figure 3. Proportions of the prokaryotic proportion of food and blank metagenomes made up of different taxa at the genus level for 109 food metagenomes and 14 blanks. Taxa that made up less than 3% of each metagenome were placed into the “Other” category.

[0030] Figure 4. Proportion of reads that belong to each superkingdom and were unclassified for the 109 food metagenomes and 14 blanks.

[0031] Figure 5. Non-metric multidimensional scaling of 109 food metagenomes and 14 blanks based on rarefied taxon classifications and coloured by food commodity (stress = 18.5%). Figure 6. Concentration of pathogens calculated for 109 food metagenomes separated by food commodity.

[0032] Figure 7. Number of different AMR genes identified for 109 food metagenome and coloured by the antimicrobial agent they are resistant to.

[0033] Figure 8. AMR gene dosage per gram of food and portion size for 109 food metagenomes separated by food type.

[0034] Food products carry bacteria unless specifically sterilised. These bacteria can be pathogenic, associated with food spoilage, and may additionally be resistant to antimicrobials. Current methods for detecting bacteria on food rely on culturing for specific bacteria, a time-consuming process, or 16S metabarcoding that can identify different taxa but not their genetic content. Directly sequencing the metagenome of food is inefficient as foods of plant and animal origin have their own host DNA that outnumbers the bacterial DNA present.

[0035] 1. Materials and methods

[0036] 1.1 Neutrase-saponin host depletion method

[0037] For each food sample, a host DNA depleted sample was prepared by placing 100 grams of food into a FBAG-03 filter blender bag (Corning, New York, USA) with 225 ml of buffered peptone water (BPW) and was stomached for 30 seconds at 100 rpm. 180 ml of 1 U / ml Neutrase (Novozyme, Bagsvserd, Denmark) was added to 10 ml of stomached food sample and incubated at 45 °C for 40 minutes in a rotating incubator at 200 rpm. The mixture was centrifuged for 30 minutes at 4,000xg, before the supernatant was discarded and resuspended in 250 pl of phosphate buffered saline (PBS) (Sigma- Aldrich, St Louis, Missouri, USA) . 200 pl of 0.225% saponin (Tokyo Chemical Industry, Tokyo, Japan) in PBS was added and the mixture was incubated for 10 minutes at 20 °C whilst rotating at 1 ,000 rpm. 350 pl of molecular water (Merck, Kenilworth, New Jersey, USA) was added to the mixture, before it was vortexed for 2 seconds, incubated at room temperature for 30 seconds, 12 pl of 5M NaCl (Sigma- Aldrich) was added, and vortexed for a further 2 seconds. The mixture was centrifuged at 8,000xg for 5 minutes, before the supernatant was discarded and the pellet was resuspended in 100 pl of PBS. 100 pl of HL-SAN buffer (5M NaCl, 0.1M MgCL (Thermo Fisher Scientific, Waltham, Massachusetts, USA)) and 10 pl of 25 U / pl HL-SAN (ArcticZymes, Tromso, Norway) were added to the mixture, before it was vortexed for 2 seconds and incubated at 37 °C for 15 minutes whilst rotating at 800 rpm. 800 pl of PBS was added to the solution before it was centrifuged at 6,000xg for 3 minutes, the supernatant was discarded and the pellet was resuspended in 500 pl of PBS. 500 pl of PBS was added to the mixture, before it was centrifuged at 6,000xg for 3 minutes, the supernatant was discarded and the pellet was resuspended in 800 pl of PBS.

[0038] For each food sample, a raw preparation (along with host DNA depletion) was prepared alongside the host DNA-depleted preparation from the food sample by centrifuging 10 ml of stomached sample for 30 minutes at 4,000xg. The supernatant was discarded and resuspended in 800 pl of PBS.

[0039] DNA was extracted from raw and depleted samples using a Maxwell® RSC PureFood Pathogen Kit (Promega, Madison, Wisconsin, USA) on a Maxwell® RSC 48.

[0040] 2.2. Method validation 2.2.1 Food sample comparison

[0041] The Neutrase-saponin method and the commercial Molysis (Molzyme, Bremen, Germany) method were trialled on three chicken, three pork, three leafy green, three prawn and three salmon samples, along with the ZymoBiomics microbial community standard (Zymo Research, Irvine, California USA) . The Molysis method was performed according to the manufacturer’s instructions. For this and each subsequent sampling runs, each DNA depletion method was additionally applied to a blank (comprising 225 ml of BPW) in parallel. Blank samples had undetectable DNA concentrations, therefore were spiked with PhiX174 RF1 (New England Biolabs, Ipswich, Massachusetts, USA) up to a concentration of 0.5 ng / pl prior to sequencing.

[0042] 2.2.2 Temperature comparison

[0043] An in-house mock community was assembled from the bacteria of seven genera: Escherichia (10%), lebsiella (20%), Pseudomonas (10%), S almonella (35%), Staphylococcus (5%) and Vibrio (10%). This community was divided into four groups that were incubated for 72 hours at different temperatures: room temperature (20 °C), refrigeration (2-8 °C), freezer (-20 °C), and long-term freezer (-80 °C) . For each of the incubated in-house mock communities, along with a Zymo mock community, the mock communities were divided into five samples, two of which underwent the Neutrase-saponin method to test for method variability, one underwent only the Neutrase section of this method and another only the saponin section to evaluate the individual performance of these components on the results, and one was unaltered for comparison (raw sample) . 2.2.3 DNA sequencing

[0044] Libraries were constructed from depleted, raw and blank metagenomic extractions using Nextera DNA flex library prep kits (Illumina, San Diego, California, USA) . Libraries were sequenced on an Illumina NextSeq at 1 Gb per metagenome, as 150 bp paired-end reads.

[0045] 2.3 Method application

[0046] 2.3.1 Food sampling

[0047] The Neutrase-saponin method was performed on 154 food samples (34 chicken, 33 leafy greens, 31 pork, 28 prawns and 28 salmon) that were collected on fourteen sampling runs conducted on different days. The food samples were collected from commercial stores around Norfolk in the United Kingdom between 20 / 5 / 2019 and 25 / 11 / 2019 Qanecko et al., 2021) .

[0048] 2.3.2 Metagenome depletion quality control

[0049] The Qubit dsDNA HS assay kit (Thermo Fisher Scientific) was used to measure the DNA concentration of metagenomic extractions. Host DNA of the raw and depleted samples were measured using qPCR methods specific to the TGFB3 gene for chicken samples (Cai et al., 2014) and the 18S gene for other host samples (Hadziavdic et al., 2014); bacterial DNA was quantified using a qPCR method specific for the 16S gene (Liu et al., 2012) (Appendix 1). The host-bacteria depletion difference was calculated from the difference in quantification cycles (Cq) for raw and host depleted samples for the host qPCR and subtracting the equivalent difference using the bacterial qPCR (referred to henceforth as the qPCR difference) . Host DNA depleted metagenomes with DNA concentrations greater than 0.05 ng and host-bacteria depletion differences of three or greater were selected for sequencing.

[0050] To determine if the host DNA depletion method had differential efficiency between food types, depleted DNA concentrations and qPCR differences were modelled using linear regression with metadata on the food samples as the explanatory variables. Metadata included the type of food, geographical origin of food, the type of store from which it was purchased (chain or independent supermarket, or butcher), how it was stored (frozen, chilled or previously frozen), for leafy greens the number of plant species present, and for salmon and prawns how it was produced / harvested (wild caught or aquaculture) . Partial-F tests were used to determine if the potential explanatory variables significantly improved the fit of the model.

[0051] Raw samples were obtained for qPCR comparisons but were not sequenced for these food samples.

[0052] 2.3.3 Method application Illumina sequencing

[0053] Libraries were constructed from depleted and blank metagenomic extractions using Nextera DNA flex library prep kits (Illumina) . Libraries were sequenced on an Illumina NovaSeq 6000 at 8 Gb per metagenome, as 150 bp paired-end reads. Raw sequence data were uploaded to the sequence read archive.

[0054] 2.4 Metagenome analysis

[0055] BBsplit v38.75 (Bushnell, 2014) was used to align paired reads from blank samples to the phiX174 genome (SAMN04281799); reads mapping to phiX174 were removed from further analysis. Paired metagenome reads were processed using fastp vO.19.5 (Chen et al., 2018) to trim reads with a minimum quality value of 20 and default other parameters. Trimmed reads were classified using Centrifuge vl .0.4 (Kim et al., 2016) and the Centrifuge NT database. KMA vl .2.3t (Clausen et al., 2018) with the ResFinder v3.2 database (Zankari et al., 2012) was used to identify AMR genes amongst the trimmed reads. BBsplit v35.85 (Bushnell, 2014) was used to align reads to the ResFinder database to estimate the proportion of reads associated with AMR genes.

[0056] Taxonomic read data from the food metagenomes was analysed using the vegan v2.5.6 (Oksanen et al. , 2019) and phyloseq vl .38.0 (McMurdie and Holmes, 2013) packages of R v3.6.1 (R core team, 2019) . Taxonomic data was rarefied before dissimilarity between food metagenomes were measured using Bray-Curtis (Bray and Curtis, 1957) . Permutational Multivariate Analysis of Variance (PERMANOVA) of the distance matrices was used to determine if any metadata variables were associated with metagenome composition.

[0057] The concentration of different bacterial taxa on food was calculated from the proportion of the metagenome made up of the taxon, the proportion of prokaryotic reads as calculated by Centrifuge and the 16S qPCR results (Appendix 2) .

[0058] To evaluate and quantify the potential human exposure to AMR from food, the AMR gene concentration per gram of food was calculated from the proportion of metagenomes originating from prokaryotes, the proportion of metagenomes made up of AMR genes, and the number of bacteria in the metagenomes as estimated from the 16S qPCR result (Appendix 2) . Linear regression was used to model the concentration of AMR genes in samples with metadata as the explanatory variables as above. The European Food Information Resource (Westenbrink et ah, 2016) was used to determine the portion size of the food samples and calculate the amount of AMR genes per portion size.

[0059] 3. Results

[0060] 3.1 Method development

[0061] Multiple methods were trialled to identify the best method for depleting host DNA from food samples (Appendix 1) . The method that depleted the most host DNA whilst preserving the prokaryotic DNA involved a protease step (Neutrase) to break up loose protein in the stomached sample (Escobar-Zepeda et al., 2016), a lysis step that uses a detergent (Saponin) to lyse plant and animal cells in the sample, and a DNAse (HL-SAN) step to degrade any extracellular DNA present. To test the effectiveness of the method, it was trialled alongside a commercial depletion kit (Molysis) on 15 food samples and a commercial mock community. For the 15 food samples investigated, the Neutrase-saponin method increased the proportion of prokaryotic reads compared to the raw 1.3-40-fold and produced the highest prokaryotic read proportion for nine specimens, whilst the Molysis kit increased the proportion by 0.13- 18-fold and produced the highest proportion for six food specimens (Figure 1) . Both methods resulted in altered bacterial genus proportions compared to the raw sample, but metagenomes from the Molysis method were more distant to raw samples than those using the Neutrase-saponin method (Figure 2) - 3.2 Mock communities

[0062] To investigate potential differential effects of the Neutrase-saponin method on bacteria, different sections of the method were applied to an in-house mock community under different conditions and the commercial (Zymo) mock community (Appendix 1) . Comparisons of the raw samples demonstrated that refrigeration increased the proportion of the community made up of S almonella, whilst freezing at -20’C or -80’C increased the proportion of Pseudomonas, but the Neutrase-saponin method had little effect on the relative proportions of the different bacterial genera sequenced. The Neutrase-saponin method was also very consistent, with similar bacterial genus proportions between repeats. However, the Neutrase-saponin method affected the proportions of bacterial genera sequenced in the commercial mock community.

[0063] 3.3 Application of the Neutrase-saponin method

[0064] The Neutrase-saponin method was applied to 154 food samples, 109 of which were chosen for further sequencing based on the selection criteria of DNA concentration and qPCR results. These consisted of 28 chicken, 24 leafy green, 15 pork, 18 prawn and 24 salmon samples, along with 14 blanks for each sampling run.

[0065] Investigation of food metadata found certain categories that were associated with each other (Supplementary table 1). For chicken, the chicken cut, the presence of bone and the presence of skin were associated, so only the presence of skin was investigated; store type and how the chicken were packaged were associated, so only store type was investigated; and only three of the original 34 chicken samples were frozen, so product presentation was excluded. For pork samples, store type and how the pork was packaged were associated, so only store type was investigated; and in regard to geographical origin, samples were domestic or of unknown location, so location was excluded. For leafy green samples, store type and how they were packaged were associated, so only store type was investigated. For prawn samples, the species of prawn, the presence of their head and shell were associated, so only the presence of the shell was investigated; and only one of the initial 28 prawn samples were from an independent store, so store type was excluded. For salmon samples, the species of salmon, the presence of skin and product presentation were associated, so only the presence of skin was investigated; and only one of the initial 28 salmon samples were from an independent store, so store type was excluded.

[0066] Food type was found to be associated with the depleted DNA concentration (p = 5.22 x 1011) and qPCR difference (p = 1.33 x 1015), but not food sampling run (p = 0.198 and p = 0.673, respectively) (Appendix 3). Further analysis of each food type determined that chicken and pork samples from butchers had higher depleted DNA concentrations than those from chain or independent supermarkets (p = 1.39 x 102and p = 5.00 x 104), prawn samples with shells were associated with higher depleted DNA concentrations (p = 1.89 x 103) and lower qPCR differences (p = 1.44 x 103whilst leafy green samples from independent supermarkets were associated with larger qPCR differences than those from chain supermarkets (p = 3.02 x 103).

[0067] 3.4 Metagenome composition The 109 food metagenomes displayed a large amount of variation in the bacterial taxa present (Figure 3), with the predominant bacterial genus being Pseudomonas. However, a large number of other taxa were identified that varied in prevalence between the food samples.

[0068] The proportion of reads that were prokaryote varied between food metagenomes (Figure 4) . Food type was found to be associated with the proportion of prokaryotic reads (p = <2.20 x 1016), with chicken (mean: 0.64; range: 0.14-0.83), pork (mean: 0.74; range: 0.21 -0.87) and leafy green (mean: 0.74; range: 0.64-0.90) metagenomes containing higher proportions than salmon (mean: 0.50; range: 0.028-0.84) and prawns (mean: 0.19; range: 0.031-0.75). There was no difference between the 14 different sampling runs (p = 0.342) (Appendix 4) . Further analysis of each food type found that chicken from butchers were associated with higher prokaryotic read proportions than those from chain and independent supermarkets (p = 0.0279), pork samples containing bones contained higher prokaryotic proportions than those without bones (p = 0.0184), pork steak samples contained higher prokaryotic proportions than other cuts (p = 0.0277).

[0069] Multidimensional scaling and PERMANOVA analysis demonstrated that food metagenomes clustered by food type (p = 0.001), but not sampling run (p = 0.907) (Figure 5) (Appendix 4) . This analysis was repeated on each food type with metadata variables specific to each food type: chicken metagenomes clustered by the presence of skin (p = 0.007) and chicken from butchers formed a small distinct cluster compared to chicken from chain and independent supermarkets (p = 0.006); and prawn samples formed distinct clusters by whether they were caught or farmed (p = 0.007) and whether they contained shells (p — 0.011); but leafy green, pork and salmon samples did not cluster by the metadata variables analysed.

[0070] 3.5 Pathogens

[0071] The concentration of pathogens (Campylobacter, Escherichia coli, Klebsiella / Kaoultella, Eisteria monocytogenes, S almonella, Vibrio and Yersinia enterocoliticd) were calculated from the taxa read classifications and qPCR results and were compared to the most predominant taxa in the food metagenomes, Pseudomonas (Figure 6) . There was a large amount of variation between the concentrations of bacterial pathogens calculated for different food types. However, these pathogens made up a small proportion of the metagenome: Campylobacter mean = 1.8 x 10 ', Escherichia coli mean = 2.7 x 103, Klebsiella / Kaoultella mean = 2.3 x 103, Eisteria monocytogenes mean = 2.9 x 10 ', S almonella mean = 7.0 x 103, Vibrio mean = 2.1 x 103and Yersinia enterocolitica mean = 5.9 x 103, especially compared to Pseudomonas (mean =0.25) .

[0072] 3.6 AMR gene dosage

[0073] The food samples displayed a large amount of variation in the numbers of different AMR genes comprising their resistome (Figure 7) . Most AMR genes encoded resistance to aminoglycoside, betalactam and tetracycline antimicrobial agents. The food samples also displayed a large amount of variation in AMR gene concentration (Figure 8) . Food type was associated with different AMR gene concentrations (p = 4.80 x 10 "') but not sampling run (p = 0.101) (Appendix 5) . The salmon and prawn samples demonstrated lower AMR gene dosages than the chicken, leafy green or pork samples. Metagenome sampling run was not associated with differences in AMR gene concentration (p = 0.101). Analysis was repeated on each food type with metadata variables specific to each food type, but none were found to significantly contribute to the AMR gene concentrations. The mean AMR gene dosage per food portion was 5.2 x 108for chicken, 3.1 x 108leafy greens, 2.8 x 108for pork, 1.5 x 107for prawns and 5.1 x 107for salmon.

[0074] 4. Discussion

[0075] Metagenomic approaches confer the ability to investigate the microbial communities on foods, allowing the identification of pathogens, spoilage organisms and AMR genes in a shorter time span than culture-based methods. Culture-based methods, unlike metagenomic methods, may also be biased towards specific strains of a bacterium (Conrad et al., 2016) and are likely to require multiple isolates to be cultured and sequenced to capture the true diversity of the bacterial population (Worby et al., 2014) . Moreover, 99% of prokaryotes are unculturable and without the use of culture-free approaches like metagenomics, their identification is not possible (Schloss and Handelsman, 2005) . In this study we examined the microbial communities and AMR genes associated with a range of food products at the closest point in the food chain to the consumer. To increase the resolution with which we can investigate the microbiome, here we describe a method (Neutrase-saponin) to deplete plant and animal DNA from food samples, providing greater sequencing depth of the microorganisms present. The method is effective at preserving the DNA of intact bacteria, but damaged bacteria may be lysed using this method. This was evident with mock communities, where the Neutrase-saponin method had minimal effects on the in-house mock community made up of intact bacteria, but a large effect on the Zymo mock community where preservatives were used to inactivate the constituent bacteria. For this reason, it is important to measure the quantities of bacterial DNA through qPCR prior to and following host DNA depletion, to ensure an unbiased representation of the microorganisms present.

[0076] The Neutrase-saponin method caused a 13-fold increase (mean; range: 1.3-40-fold increase) in prokaryotic DNA from food samples. This is not as much as the saponin-based host depletion method that was previously developed to identify pathogens in sputum samples (median: 352-fold increase) (Charalampous et al., 2019) . However, the food samples had larger prokaryotic DNA concentrations in raw samples (1.1-45%) than the sputum samples (as low as 0.001%), and the Neutrase-saponin method was devised to be applicable to a large range of food types, as opposed to just sputum samples. The postdepletion prokaryotic read proportion should then be considered when determining the effectiveness of the method. Based on the mean prokaryotic read proportions post-depletion, the Neutrase-saponin method was effective for chicken (0.64), pork (0.74) and leafy green (0.74) samples, compared to many salmon (0.50) and most prawn (0.19) samples. However, the method used to identify prokaryotic reads relies on a database of known prokaryotes and eukaryotes, resulting in 0.08-0.74 unclassified read proportion. This was of particular a problem for seafood and could be because seafood are colonised by a large number of bacteria that have not been described, as evidenced by Collins et al. (2021) who investigated the intestinal metagenome of deep-sea fish and found that 97% of the 111 metagenome-assembled genomes were novel using average nucleotide identity. In addition, analysis of the parameters used to choose metagenomes for sequencing, identified that prawn samples with shells had high depleted DNA concentrations and low qPCR differences. This suggests that the prawn shell is interfering with Neutrase-saponin method, leading to little depletion overall. In addition, chicken and pork from butchers was associated with higher depleted DNA concentrations, and leafy greens from independent supermarkets were associated with higher qPCR difference. Postsequencing chicken from butchers were associated with higher proportions of prokaryotic reads, as were pork steaks and pork containing bones. It is unclear exactly why the Neutrase-saponin method worked better for these samples, but it may be because they had larger initial bacterial concentrations. The application of the Neutrase-saponin method to more food metagenome samples will help identify the exact reasons the method works better for some samples than others.

[0077] The food microbial communities sequenced in this study showed a large amount of variation in the types and quantity of microbial taxa present. Pseudomonas was the predominant genus for most of the food samples investigated. These bacteria are ubiquitous in soil, plants and animals, so could have colonised food at multiple stages through the food chain (Crone et al., 2020; Stellato et al., 2017) . They are also associated with food spoilage, and their quantification using food metagenomics could be used to measure the effectiveness of food processing and storage techniques (Stellato et al., 2017) .

[0078] Food metagenomes were shown to cluster by the type of food. For chicken samples, they clustered by the types of stores from with they were collected. In the United Kingdom, chain supermarkets obtain their meats from a relatively small number of abattoirs, with animals sources throughout the country (Broadway, 2002), whereas independent butchers are more likely to obtain their meat from local farms (National Craft Butchers, 2021) . This wider geography sourcing and mixing of animals within abattoirs may contribute to the wider range of chicken metagenome compositions from chain supermarkets compared to those obtained from butchers. For prawn samples, the metagenomes clustered by whether they were obtained from wild- caught or farmed prawns. Cornejo-Granados et al. (2017) found that the microbiota of shrimp was similar to the sediment in their environment, and was different amongst farmed and caught shrimp, possibly explaining the differences in composition between the caught and farmed prawn samples. Prawn and chicken samples also clustered by the presence of shells and skin, respectively, which may host different bacteria. The wider application of this type of approach will provide further insight into the associations of food production and processing methods with the food metagenome.

[0079] Reads associated with pathogens and food spoilage organisms were detected in all food metagenomes analysed, which in turn was used to calculate the dosage of the microbes of interest on foods. It is important to note that this does not necessarily represent a food safety risk; for example, reads associated with Vibrio, a marine- associated bacterium were detected on chicken. These unexpected detections are likely the result of analytic algorithms misclassifying reads of closely related bacteria, and as thus must be interpreted with caution. Parallel application of the Neutrase-saponin method with pathogen culturing is required to validate the method for this analysis. Doing so will also provide validation of the method for the identification of specific genetic elements, such as AMR genes, within specific bacteria.

[0080] Antimicrobial resistance is a global problem (IACG, 2019), and food represents one way in which bacteria or AMR genes can be transmitted between the environment, animals, plants and humans. In this study we found a wide variation in the numbers, types and concentration of AMR genes on food. The mean AMR gene concentration was higher in the chicken, pork and leafy green samples than the prawn or salmon samples, indicating that different food types are likely to expose us to different quantities of AMR genes. Chicken, pork, salmon and prawns are usually cooked before consumption, likely decreasing the AMR dosage to which consumers are exposed. The portion sizes varied from 80 grams for leafy greens to 200 grams for pork, but this variation was small compared to the variation in AMR gene concentrations, resulting in similar distributions of AMR dosages for food types compared to AMR gene concentration (102-108AMR genes / gram) . The high AMR gene dosage of leafy greens may be because they contain a larger surface area compared to the other food types, providing more area for bacteria to colonise. The evaluation of other ready-to-eat foods, in addition to the leafy greens examined here, will provide a different perspective on the risks of food as a source of AMR.

[0081] The AMR concentrations calculated for food (mean 1.84 x 106genes / gram) are much lower than that previously calculated from pig faeces (2.48 x 109genes / gram) (Pollock et al., 2020), dewatered sludge from municipal wastewater treatment plants (2.44 xl O9genes / gram) (Gao et al., 2012) and manure (1.14 x 108genes / gram) (Macedo et al., 2020) . However, these concentrations are not directly comparable. People are exposed to food more frequently than these other sources of AMR, and in many cases the foods will be cooked before consumptions and therefore the AMR gene concentration will likely decrease dramatically. In addition, these other sources were calculated using AMR gene-specific qPCR and different AMR gene targets. The food metagenome selection method may have also selected for food metagenomes with more bacteria and therefore higher AMR gene concentrations. Standardised methods are required for calculating the AMR gene concentrations from different sources so they can be compared with each other.

[0082] 5. Conclusions

[0083] In this study we developed a host DNA depletion method for food that allows efficient sequencing of food sample metagenomes, increasing the proportion of prokaryotic reads sequenced 13-fold (median) compared to raw samples. The method worked best on chicken, pork and leafy green samples which had high mean prokaryotic read proportions post-depletion (0.64, 0.74 and 0.74, respectively), some salmon samples (0.50), but few prawn samples (0.19) . We demonstrated its applicability for identifying the different bacteria and AMR genes present and calculating the AMR dosage of food. Different food types were shown to have different AMR gene dosages and metagenome compositions, and for chicken the metagenomes were influenced by the type of store from which they were obtained, and for prawns how they were produced / harvested. The method described in this study was useful at identifying the predominant bacteria on food and their associated AMR genes, but further work is required to use metagenomics to identify specific food pathogens.

Claims

Claims1. A method of determining and / or quantifying the microorganism communities and genes present on food, said method including the steps of; a) Mixing a sample of said at least one foodstuff with a protease b) Introducing a triterpene glycoside or saponin to the mixture c) Introducing a DNase to the mixture d) Identifying and / or quantifying microorganisms and / or genes by DNA sequencing at least part of the mixture and subjecting the same to metagenomic analysis.

2. A method according to claim 1 wherein the sample and reagents are mixed in a blender bag and the mixture is stomached for a period of time.

3. A method according to claim 1 or 2 wherein the protease is an endoprotease.

4. A method according to claim 3 wherein the endoprotease is a neutral metalloendopeptidase.

5. A method according to claim 4 wherein has, and / or is from the Enzyme Commission number EC 3.4.24 class.

6. A method according to claim 5 wherein the endoprotease is a neutral metalloendopeptidase from bacillus spp.

7. A method according to claims 3-6 wherein the protease is from the Bacillolysin or Thermolysin family.

8. A method according to claims 3-7 wherein the protease is Neutrase.

9. A method according to any preceding claim wherein Illumina dye sequencing is used to sequence the host DNA resultant samples.

10. A method according to any preceding claim wherein the DNA concentration of metagenomic extractions is measured.

11. A method of host DNA depletion to classify and quantity bacteria and / or antimicrobial resistance genes present on food, a host DNA depleted sample is prepared by placing a sample of food into blender bag and buffered water is added to the bag, the sample is stomached with the addition of protease.

12. A method according to claim 11 wherein the mixture is incubated and / or rotated.

13. A method according to claim 11 or 12 wherein a detergent and / or saponin is added to the mixture.

14. A method according to claim 13 wherein DNase is added to the mixture.

15. A method according to claim 13 wherein DNA sequencing libraries are constructed from depleted metagenomic extractions and said libraries are sequenced.

Images