Modelling the effectiveness of surveillance based on metagenomics in detecting, monitoring, and forecasting antimicrobial resistance in livestock production under economic constraints

doi:10.21203/rs.3.rs-1994070/v1

Download PDF

Article

Modelling the effectiveness of surveillance based on metagenomics in detecting, monitoring, and forecasting antimicrobial resistance in livestock production under economic constraints

https://doi.org/10.21203/rs.3.rs-1994070/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 21 Nov, 2023

Read the published version in Scientific Reports →

You are reading this latest preprint version

Current surveillance of AMR is mostly based on testing indicator bacteria using MIC panels. Metagenomics has the potential to identify all known ARGs in complex samples and thereby detect changes in the occurrence earlier. Here, we simulate the results of an AMR surveillance program based on metagenomics in the Danish pig population. We modelled both an increase in the occurrence of ARGs and an introduction of a new ARG in a few farms and subsequent spread to the entire population. To make the simulation realistic, total cost of the surveillance was constrained, and the sampling schedule set at one pool per month with 5, 20, 50, 100, or 200 samples. Our simulations demonstrate that a pool of 50–100 samples and a sequencing depth of 250 million fragments resulted in the shortest time to detection in both scenarios, with a time-delay to detection of change of $>$15 months in all scenarios. Overall, our findings suggest that using metagenomics could improve the monitoring of AMR in the animal population.

The ability to implement targeted control measures in due time against emerging antimicrobial resistance (AMR) depends on an efficient monitoring system of AMR in humans, animals, and the environment^1–6. In Denmark, a surveillance programme for antimicrobial usage and AMR (DANMAP) was established in 1995⁷. Since its establishment, DANMAP has been based on phenotypic-based characterisation of clinical isolates of selected pathogens and selected indicator bacteria from samples in animals, humans, and food using minimum inhibitory concentration (MIC) panels^7,8. Antimicrobial resistance genes (ARGs) can occur in all bacterial species^9,10. Using only a few selected bacteria for detecting the emerging or increased occurrence of AMR might miss the emergence and spread of AMR in the entire resistome ^7,8,11. With the ongoing development of gene sequencing technology, it is now possible to investigate the presence and abundance of ARGs in a sample, using shotgun metagenomic sequencing of DNA fragments extracted from all organisms in the sample and subsequently to map obtained sequences of nucleobases against databases with known sequences of nucleobases for ARGs^11–16. It has been shown that the metagenomics approach is a useful and reliable measurement for AMR in pig production ^17–21. It has also been suggested that the use of metagenomics will increase the likelihood that emerging ARGs are detected earlier since the whole microbiota in the sample is being analysed for AMR^22–24.

However, the likelihood of detecting an emerging or significant quantitative change of an ARG in a population using metagenomics²⁵ depends, in addition to the actual occurrence of the ARG, on the design of the monitoring system. This includes a fraction of the population of interest where samples are collected (sample size), the frequency of sampling, and how much of the physical sample is actually investigated (in metagenomics, this is equivalent to how many extracted DNA fragments are sequenced (sequencing depth)).

For national surveillance of AMR in Denmark and many other countries, a key performance parameter is the lag time between the de novo emergence of a novel ARG in a population or a real change in the occurrence of an ARG already present and when this change will be detected as a significant change in the surveillance program (sensitivity). It is also of practical importance that the program does not result in too many false alarms. Thus, in situations with a stable AMR level, the program should not indicate changes in the occurrence of AMR (specificity). The sensitivity and specificity of a monitoring program are often negatively correlated.

The aim of the presented study was to, given economic restrictions, assess different pooling strategies for detecting emergence and the increase in the prevalence of ARGs in the Danish pig population by simulation. In the simulation, we took into account the stochasticity related to sampling and laboratory procedures and the economic restrictions. Based on the simulated data, we estimated the sensitivity and specificity to compare pooling strategies in the surveillance of ARGs in a population of production animals. In the study, we show that using relatively large pools combined with metagenomics, and it is possible to monitor the whole ARGs in a population. The results of the simulations show that with the use of metagenomics, it is possible to do monitoring of all known ARGs, both detecting changes in endemic and detecting newly introduced ARGs.

In this study, we have measured the occurrence of ARGs as the #CPM, which should be interpreted as how many of the sequenced gene fragments were representing a specific ARG (normalised to 1,000,000). Using the rolling method, we estimate the moving mean of simulated "observed" data, which is close to the truth CPM, as shown in Figs. 1 and 2 (Supplementary Figs. 3 and 4).

The more samples collected and pooled at each sampling point, the closer the sample represents the true occurrence of the ARG (Figs. 1 and 2). The deviation of the observed data from the truth is caused by sampling error and random variation due to sequencing (procedure). These two sources of error are distinct from one another. To compare how these two sources of error contribute to the deviation between the observed data and the truth, for each observed data point in each time series, we decompose the total deviation into deviation due to sampling (Eq. 1) and lab procedure (Eq. 2)

$$Deviatio{n}_{-}Sampling\left[CPM\right]= Sample\left[CPM\right] - Truth\left[CPM\right] \left(1\right)$$

$$Deviati{on}_{-}Lab \left[CPM\right]= Observed\left[CMP\right]-Truth\left[CPM\right]-Deviatio{n}_{-}Sampling\left[CPM\right] \left(2\right)$$

We plotted the estimated deviations due to sampling (deviation_sampling) and due to laboratory procedure(deviation_lab), as shown in Figs. 3 and 4 (Supplementary Figs. 5 and 6).

Figure 5 shows the distribution of time to detect for the 1,001 iterations done in each scenario with different numbers of samples in the pool. For any number of samples per pool, the time to detection varies by several months due to the stochasticity of sampling and laboratory, ranging from a detection within the first few months after increased occurrence to that the increase was detected first after several years. This variation in time to detection for any number of samples per pool was a result of the stochasticity of sampling from a population, and the randomness is which fragments that are sequenced and also partly by variation in the actual sequencing depth, which is not fixed. The variation between the results obtained 5, 20,50,100, and 200 only due to the different numbers of samples and thereby also sequencing depth.

According to Fig. 6, the range of time to detection is between 1-120 months, irrespective of how the available money is allocated between the number of samples collected and pooled and sequencing depth. However, for 50 and 100 samples, the median and 1st and 3rd quartiles of the simulated results indicate that the time to detect was generally lower compared to the other number of samples per pool. For the high-level occurring ARG, a median delay in detecting a change ranges from 33 months using 50 samples per pool to 48 months using 5 samples per pool in the scenario using 60.000€ annually. The results for the scenario based on using 90.0000€ annually were similar (for details, see Supplementary Tables 2 and 3).

Figure 6 (Supplementary Figs. 7 and 8) illustrates the 4 years forecast of the occurrence of an ARG based on the data that was observed in iterations in the period between 2000 and 2020. The size of the uncertainty was increased by the time period that was forecasted, but instead of an eyeball forecast, an analytical approach utilising the data is more valid, robust, and objective. The close approximation of the observed and forecasted data to the true data in Fig. 6 reveals that the performance of the surveillance is a valid representation of the truth and can be used to forecast the occurrence.

Figure 7 shows that the 50- and 100- sample pools (Supplementay Tables 2 and 3) are more likely to detect the gene at an early stage of its spread for 60,000€ and 90,000€, respectively, in scenario 3. For the detection of new ARGs in the population, the transition from not present to present, we used a threshold of greater than or equal to five to detect the novel emerging ARG in the population.

The time-delay for detecting a novel ARG emerging in the population after introduction into one farm had a median of between 19 months (at which time 12 farms were infected by the ARG according to the $SI$ model) and 33 months (84 farms infected by the ARG according to the $SI$ model). Overall, the results indicate that the shortest time to detection is achieved by pooling 100 samples at each time of sampling and using a sequencing depth of about 140–210 million with an annual cost of 90,000€.

In this study, we modelled the utility of metagenomics in livestock production for detecting, monitoring, and forecasting the presence of ARGs. The model was developed on two dimensions: first, the dynamic change in the occurrence of AMR, and second, the monitoring procedure - sampling schedule, laboratory method, interpretation, and statistical analysis of laboratory results. Like all mathematical models, these results must be interpreted in the context of the model assumptions and the quality of available simulated data. Also, the results are based on ARG data from Danish pigs and assumptions about the spread of ARGs between farms in the Danish production system. The earlier indication of a change of AMR in a population, as more effective preventive measures will be (regulating or banning the use of certain antimicrobials) in respect to limiting the occurrence of AMR. However, to the best of our knowledge, this is the first mathematical model developed for estimating the performance of an AMR surveillance program based on metagenomics.

All our simulations suggest a considerable time-delay to detection from either an increased occurrence of an ARG already present in the population (33 to 64 months) or the first detection of a completely novel ARG (19 to 33 months). However, our results also show that even using the same amount of funding; we might "gain" 14 to 31 months in time to detection by using the optimal pooling strategy combined with metagenomics. Also, as suggested by our simulations, the correct pooling strategy might be as effective as simply increasing the overall cost used for surveillance of AMR (in our case, 50%).

The generation of time-series data, in general, also gives the opportunity to do some kind of forecasting - either visually or analytically. In most surveillance programs related to health in animal or human populations, e.g., DANMAP²⁶, the forecast is based on a visual inspection of the time-series data, aggregating the data to the annual occurrence. The use of metagenomics, in combination with a very strict sample schedule where samples are collected every month, will result in a time series about the quantitative amount of all known ARGs per, e.g., month. By estimating the trend in the observed data, the estimated trend can be used for a long-time forecast of AMR, given the trend is continuing. Given the potential of ARGs to transfer between bacteria species and the long time it takes to reduce the occurrence of AMR with preventive measures, forecasting the occurrence of AMR can support the implementation of preventive measures in due diligence, applying precautionary principles. This is especially relevant for AMR; decisions need to be taken before the adverse effects of AMR are observed in the human population.

As more sensitive, we make the algorithm detect changes, as we will have a more significant number of false alarms. The sensitivity of the whole surveillance must, in the end, be balanced against the size of the potential risk and the costs of false alarms, and in the end, that is a decision that needs to be made by risk managers. The parameter "duration" in the DBEST significantly detects change within the time series data. The shorter the "duration," the faster a breakpoint will be identified, but also, a short duration time will increase the likelihood of false alarms.

Compared to surveillance based on phenotypic characterisation of isolated bacteria, the metagenomic method allows for monitoring the occurrence of all resistance genes with known nucleotide sequences in the same sample and laboratory procedure. Another strength is that the data structure of the outcome of metagenomics (#counts) makes it possible to aggregate the counts of each gene into higher levels, to obtain the #CPM of genes coding for resistance against antimicrobial class (beta-lactams, macrolides, aminoglycosides, etc.). Thereby, surveillance of AMR based on metagenomics can reveal changes in the occurrence of AMR in the bacterial population at different levels. One apparent disadvantage of metagenomics is that the method does not indicate if the detected genes are phenotypically expressed in the population of bacteria. However, in the case of surveillance of AMR in populations of production animals, the risk is related to the likelihood that the gene is transferred to human pathogens. One of the deciding factors for the size of this risk is the absolute number of resistance genes in the microbiome. Therefore, in this context, whether the genes of interest are expressed or not in the collected sample is of minor concern.

The design of the study is presented in Fig. 8. Initially, we developed a dynamic mathematical model describing the occurrence and the change in the occurrence of ARG of interest in the population over time. This model describes the "true" occurrence of ARG of interest at any given time.

Secondly, a model mimicking the surveillance program overlaid the model of the "true" occurrence to simulate the results that can be observed in surveillance given the settings of the sampling schedule (frequency of sampling, sample size, and pooling) and the metagenomic procedure (sequencing depth). In each iteration, by taking into account the stochasticity of sampling and metagenomics, a time series of observed data was generated. Each time series was subsequently analysed statistically to identify time points for detection of the start of changes in the occurrence as well as estimating long-term trends. In addition, given the time series, the future occurrence of the ARG of interest in the population was forecasted using forecasting methodology. The results from the statistical analysis were finally compared with the known truth that was given in the dynamic model describing occurrence and the change in the occurrence of an arbitrary ARG in the population over time.

Mathematical notification of model describing "true" occurrence and the change in the occurrence of an arbitrary ARG in the population over time.

True occurrence of an ARG in this study was defined as the occurrence of ARG at the farm (proportion of farms positive), within farm prevalence (proportion of pigs within a positive farm with the ARG), and; the concentration of the ARG in faeces of pigs carrying the gene. The concentration of the ARG(s) of interest in faeces was obtained from a Weibull distribution fitted^27–29 to real-world data of the concentration of ARGs in pig faeces^30,31. We chose the Weibull distribution because this gave the best fit to the data as compared to gamma and lognormal distributions (for details, see Supplementary Fig. 1).

In this study, three different scenarios of true occurrence were modelled, each representing ARGs with different occurrences and changes: 1) an increased occurrence of an ARG that is already widely present in all farms in all pigs (high-level endemic), 2) an increased occurrence of an ARG that is present but at a lower frequency (low-level endemic) and 3) the de novo emergence of a novel ARG (emerging). In all scenarios, we use metagenomic data obtained from samples collected in a previous study conducted on the Danish pig population¹⁵. In scenario 1, we use metagenomic data for tet(M) and in scenario 2 we used metagenomic data for $\beta$-lactam. These two genes are widely spread in the Danish pig population, and therefore both in scenarios 1 and 2, the prevalence of positive farms and animals within farms was set at 100%. Within the first 120 months, we assumed that there was no increase in occurrence. After 120 months, we introduced a 5% annual increase in concentration.

In scenario 3, in the model, a new ARG was introduced into one farm in a population of 4,000 farms, where after the ARG was allowed to spread to other farms over time (emerging). The spread was following a generalised Susceptible-Infected ($SI$) compartmental model³² (for details, see Supplementary Eq. 1). Our model was based on a homogeneously mixing system, with a transmission rate (0.148) resulting in 85% contaminated farms after five years. The spread described in the $\left(SI\right)$ model only considers the spread of ARG between farms. As soon as an ARG is introduced into a farm, it was assumed that the occurrence of pigs carrying the ARG was the same as the occurrence in a farm that has been infected for a longer period (90%). We assumed that the within-pig concentration of the emerging ARG was the same as the concentration of the blaOXA gene (a resistance gene with an extremely low concentration in the Danish pig population). The number of samples per pooled that was simulated was 5, 20, 50, 100, and 200.

Model for simulating results obtained in the surveillance program

In the simulation, we generated results obtained in a hypothetical monitoring program, including the effect of the design of the monitoring - the frequency of sampling and the numbers of samples at each timepoint of sampling. In this study, we assumed that all samples collected at each timepoint were pooled into one single technical sample before sequencing. Depending on the number of DNA fragments that are sequenced (sequencing depth)^33–35 and the concentration of the ARG of interest³⁶, the initial output from the model is the number of DNA fragments that were assigned to the ARG of interest.

Subsequently, we calculated the Counts of ARG divided by the total millions number of fragments sequenced in the sample - Counts Per Million (#CPM)^37,38. Thereby, the outcome in the surveillance of AMR using metagenomics is the #CPM in the population. The #CPM in each faecal sample is the product of whether the sample is collected from a farm having the ARG (0 or 1) $binomial(n=1, prob=farmprev)$, the collected animal within a positive farm carrying the ARG (0 or 1) $binomial(n=1, prob=farmprev)$ and the concentration of the ARG if the animal is carrying the ARG (random draw from the Weibull distribution). The model includes a pooling procedure where the number of collected faecal samples at each timepoint of sampling (assumed to be 5, 20, 50, 100, or 200) are pooled into one technical sample before DNA sequencing. The mean of the CPM in the pooled faecal samples represents the "concentration of the ARG in the technical sample at a given point in time." The simulation was conducted over 240 months, with a sampling frequency of once per month.

In theory, when the design factors frequency of sampling, sample size, or sequencing depth goes infinite, the sensitivity of the monitoring program will become 100%, and the lag time between change of occurrence and detection of change will be zero. To obtain realistic values of the design factors, we developed an equation wherein the price of sampling and sequencing were integrated together with the variables sample size and sequencing depth, summing up to a given total annual cost of a monitoring program³⁹. We developed a cost function (Eq. 3) to estimate how deep we would be able to sequence each pooled technical faecal sample given a sampling of n faecal samples (n = 5, 20, 50, 100, and 200) randomly collected once a month. The cost we calculated from

$$Total price=lab cost* k+lab technician cost+n*price per sample+m*price per fragment \left(3\right)$$

where $k$ is the number of technical samples processed in the lab per month (in this study $k=1$), $n$ and $m$ are the number of samples in pooling and the sequencing depth, respectively (for details, see Supplementary Table 1).

An important determinant of the likelihood of metagenomics in detecting an emerging ARG is how many gene fragments are sequenced (sequencing depth)^40–42. Which fragments that are sequenced in a sample are totally random, whereas the number of sequenced fragments can be pre-defined within a certain range. The mean sequencing depth for each sample pool is determined using $rtruncnorm(n, a, b, mean , sd )$: where $n$ is the number of observations (months); $a$ is lower mean sequencing depth; $b$is upper mean sequencing depth; $mean$ is mean sequencing depth, and $sd$is $(mean-a)/3$. To add randomness to the sequencing depth, we assume that the depth is varying stochastically between $\pm$20% of the mean, following a normal distribution where the range between − 20% and +20 % is equal to 6SD. During simulation, the sequencing depth as truncated at $\pm$3SD⁴³. In the simulation of all scenarios and sub-scenarios, 1,001 iterations were performed using Monte Carlo simulation ⁴⁴⁻⁴⁶, representing 1,001 time series of the observations.

Statistical analyses of the time series obtained in the simulations

The performance of the surveillance program⁴⁷ for detecting ongoing changes in the occurrence of ARGs in the population was done by initially identifying a change in the observed data using breakpoint analysis^48,49 and subsequently calculating the time between the start of the change given by the mathematical function defining the true occurrence of an ARG in the population, and the time point when this change was identified in the time-series of observed data.

Also, by applying the breakpoint analysis throughout the time series of observed data, including the initial period of 10 years with no true change, we also obtained the number of false alarms, which is when the surveillance indicates a change in the occurrence of #CPM where the occurrence is actually not changing.

Breakpoint analysis - DBEST method

The simulated observed data of an ARG in a population using metagenomics were analysed for the presence of changes in occurrence over time using breakpoint analysis. In reality, when the amount of endemic occurrence AMR starts to change, the increase/decrease takes place gradually over months/years and not with a direct shift from one level to another within a few months. In the R package DBEST, statistical methods are applied to analyse time trends and detect changes over time. DBEST was originally developed to both detect and forecast changes in vegetation using remote information⁵⁰. We chose DBEST because of its ability to handle #CPM in the form of continuous data, which is not possible with surveillance package analysis, which only considers whole numbers.

In our study, we used DBEST for time-trend analysis, looking for changes in the #CPM of the ARG(s) of interest over time (Supplementary Fig. 2). DBEST detects and estimates trend changes in two ways (abrupt and non-abrupt) and the timing, magnitude, number, and direction of those changes. In the DBEST method, there are five parameters that must be defined before the analysis. The description of the "change_magnitude" is the size of changes in the data that indicate a potential true change. In this study, we used a 5% increase/decrease (change_magnitude, Table 1) from the #CPM mean of the AMR-gene of interest in the population before an increase occurred. The specified "duration" is the time period after the breakpoint that the change must be present to indicate a true persistent change, given the algorithm's sensitivity to detect changes. This was defined as six months in our study (Table 1).

In accordance with our knowledge of changes in ARGs in animal populations over time and how it was specified in the model, the statistical methods should identify non-abrupt changes over a longer time period. Abrupt changes in a short time interval are not expected, and therefore, the parameters in the algorithm were set, so it only looks for non-abrupt changes. Therefore, the values for "first-level-shift" and "second-level-shift" were set to extremely high values, whereby none of the detected change points were classified as abrupt. This is a technical issue, but it is worth noting. The "alpha" is the statistical significance level value used to test the significance of detected changes. Table 1 shows how these five parameters were implemented in the DBEST.

Table 1

The parameters used for the DBEST method
tet(M) and $\beta$-lactam
Parameter	Value
First-level-shift	*Reference_Mean
Second-level-shift	*Reference_Mean
Duration	6 months
Alpha	0.05
Change_magnitude	5% of *Reference_Mean
*Reference_Mean is the mean of the AMR-gene of interest in the population before an increase takes place.

Estimation of the time-trend

We used a rolling average to calculate the average number using CPM data. We calculated the first six-month rolling average. To transform the CPM data set for sequence data points, we specified the window size and took a new average value. This was applied to determine a new CPM average value (Fig. 1, red line). This was done to remove stochasticity from the CPM and gain a solid understanding of how to recognize past data trends and predict what will happen in the future. We used the R software (version 4.0.2) for modelling and data analysis.

Contributors

OOA, FMA, and HV conceived the study. OOA and HV designed the study. All authors acquired and analysed the data. OOA, FMA and HV interpreted the findings. OOA wrote the first draft of the manuscript. FMA and HV contributed to the writing and subsequent versions of the manuscript. OOA developed the code. All authors critically reviewed this paper and approved the final version. The corresponding author (OOA) had final responsibility for the decision to submit for publication.

Funding: Novo Nordisk Foundation (NNF16OC0021856), European Union's Horizon 2020 Research and Innovation programme under Grant Agreement No. 773830: One Health European Joint Programme and the Danish Government.

Estimation of the time-trend

We used a rolling average to calculate the average number using CPM data. We calculated the first six-month rolling average. To transform the CPM data set for sequence data points, we specified the window size and took a new average value. This was applied to determine a new CPM average value (Fig.1, red line). This was done to remove stochasticity from the CPM and gain a solid understanding of how to recognize past data trends and predict what will happen in the future. We used the R software (version 4.0.2) for modelling and data analysis.

Contributors

Declaration of interests

We declare no competing interests.

Data sharing

The data that support the findings of this study are available and are with OOA and HV. The underlying data, statistical analysis plan, and analysis codes will be made available from the corresponding author.

Acknowledgment

The authors would like to express their appreciation to Dr. Sadegh Jamali of Lund University's Department of Physical Geography and Ecosystem Science for his constructive feedback. We gratefully acknowledge the support of Andrea Bordoni HPC Specialist, Department of Applied Mathematics and Computer Science.

Queenan, K., Häsler, B. & Rushton, J. A One Health approach to antimicrobial resistance surveillance: is there a business case for it? International Journal of Antimicrobial Agents 48, (2016).
Graham, D. W. et al. Complexities in understanding antimicrobial resistance across domesticated animal, human, and environmental systems. Annals of the New York Academy of Sciences vol. 1441 Preprint at https://doi.org/10.1111/nyas.14036 (2019).
Ferri, M., Ranucci, E., Romagnoli, P. & Giaccone, V. Antimicrobial resistance: A global emerging threat to public health systems. Critical Reviews in Food Science and Nutrition 57, (2017).
Hendriksen, R. S. et al. Using Genomics to Track Global Antimicrobial Resistance. Frontiers in Public Health vol. 7 Preprint at https://doi.org/10.3389/fpubh.2019.00242 (2019).
Hammerum, A. M. et al. Danish integrated antimicrobial resistance monitoring and research program. Emerging Infectious Diseases vol. 13 Preprint at https://doi.org/10.3201/eid1311.070421 (2007).
Oniciuc, E. A. et al. The present and future of whole genome sequencing (WGS) and whole metagenome sequencing (WMS) for surveillance of antimicrobial resistant microorganisms and antimicrobial resistance genes across the food chain. Genes vol. 9 Preprint at https://doi.org/10.3390/genes9050268 (2018).
Aarestrup, F. M. et al. Resistance to antimicrobial agents used for animal therapy in pathogenic-, zoonotic- and indicator bacteria isolated from different food animals in Denmark: A baseline study for the Danish Integrated Antimicrobial Resistance Monitorng (DANMAP). APMIS 106, (1998).
Bengtsson-Palme, J., Kristiansson, E. & Larsson, D. G. J. Environmental factors influencing the development and spread of antibiotic resistance. FEMS Microbiology Reviews vol. 42 Preprint at https://doi.org/10.1093/femsre/fux053 (2018).
Sivri, N., Bagcigil, A. F., Celik, B., Azaz, D. & Gungor, Y. The antibiotic resistance genes in Escherichia coli isolates from Narman Landfill (NL) area in Erzurum, Turkey. Fresenius Environmental Bulletin 28, (2019).
De, R. Metagenomics: Aid to combat antimicrobial resistance in diarrhea. Gut Pathogens vol. 11 Preprint at https://doi.org/10.1186/s13099-019-0331-8 (2019).
Munk, P. et al. A sampling and metagenomic sequencing-based methodology for monitoring antimicrobial resistance in swine herds. Journal of Antimicrobial Chemotherapy 72, (2017).
Munk, P. et al. Abundance and diversity of the faecal resistome in slaughter pigs and broilers in nine European countries. Nature Microbiology 3, (2018).
Hendriksen, R. S. et al. Global monitoring of antimicrobial resistance based on metagenomics analyses of urban sewage. Nature Communications 10, (2019).
McArthur, A. G. & Tsang, K. K. Antimicrobial resistance surveillance in the genomic age. Ann N Y Acad Sci 1388, (2017).
Aerts, M. et al. Technical specifications on harmonised monitoring of antimicrobial resistance in zoonotic and indicator bacteria from food-producing animals and food. EFSA Journal 17, (2019).
Simjee, S., McDermott, P., Trott, D. J. & Chuanchuen, R. Present and Future Surveillance of Antimicrobial Resistance in Animals: Principles and Practices. Microbiology Spectrum 6, (2018).
Xiao, Y., Tang, S., Zhou, Y. & Smith, R. Predicting the HIV/AIDS epidemic and measuring the effect of mobility in mainland China. Elsevier (2013).
van Gompel, L. et al. The antimicrobial resistome in relation to antimicrobial use and biosecurity in pig farming, a metagenome-wide association study in nine European countries. Journal of Antimicrobial Chemotherapy 74, (2019).
Andersen, V. D., Jensen, M. S., Munk, P. & Vigre, H. Robustness in quantifying the abundance of antimicrobial resistance genes in pooled faeces samples from batches of slaughter pigs using metagenomics analysis. Journal of Global Antimicrobial Resistance 24, (2021).
Luiken, R. E. C. et al. Associations between antimicrobial use and the faecal resistome on broiler farms from nine European countries. Journal of Antimicrobial Chemotherapy 74, (2019).
Andersen, V. D. et al. The association between measurements of antimicrobial use and resistance in the faeces microbiota of finisher batches. Epidemiology and Infection 145, (2017).
Forbes, J. D., Knox, N. C., Ronholm, J., Pagotto, F. & Reimer, A. Metagenomics: The next culture-independent game changer. Frontiers in Microbiology vol. 8 Preprint at https://doi.org/10.3389/fmicb.2017.01069 (2017).
Amico, P. et al. Incidence and prediction of early antibody-mediated rejection due to non-human leukocyte antigen-antibodies. Transplantation 85, (2008).
Warder, L. M. C. et al. Characterization of the microbiota and resistome of bulk tank milk samples from Prince Edward Island dairy farms. Journal of Dairy Science 104, (2021).
K Ko, K. K., Rei Chng, K. & Nagarajan, N. Metagenomics-enabled microbial surveillance. doi:10.1038/s41564-022-01089-w.
Korsgaard, H. B. et al. DANMAP 2019 - Use of antimicrobial agents and occurrence of antimicrobial resistance in bacteria from food animals, food, and humans in Denmark. Danmap (2020).
Datsiou, K. C. & Overend, M. Weibull parameter estimation and goodness-of-fit for glass strength data. Structural Safety 73, (2018).
Ahmad, A. E. B. A. & Ghazal, M. G. M. Exponentiated additive Weibull distribution. Reliability Engineering and System Safety 193, (2020).
Almalki, S. J. & Nadarajah, S. Modifications of the Weibull distribution: A review. Reliability Engineering and System Safety vol. 124 Preprint at https://doi.org/10.1016/j.ress.2013.11.010 (2014).
Hammerum, A. M. et al. Detection of sul1, sul2 and sul3 in sulphonamide resistant Escherichia coli isolates obtained from healthy humans, pork and pigs in Denmark. International Journal of Food Microbiology 106, (2006).
Gerzova, L. et al. Characterization of antibiotic resistance gene abundance and microbiota composition in feces of organic and conventional pigs from four EU countries. PLoS ONE 10, (2015).
Apenteng, O. O., Arnold, M. E. & Vigre, H. Using stochastic dynamic modelling to estimate the sensitivity of current and alternative surveillance program of Salmonella in conventional broiler production. Scientific Reports 10, (2020).
Øvreås, L., Forney, L., Daae, F. L. & Torsvik, V. Distribution of bacterioplankton in meromictic lake Saelenvannet, as determined by denaturing gradient gel electrophoresis of PCR-amplified gene fragments coding for 16S rRNA. Applied and Environmental Microbiology 63, (1997).
de Bustos, A., Cuadrado, A. & Jouve, N. Sequencing of long stretches of repetitive DNA. Scientific Reports 6, (2016).
Fan, H. C., Blumenfeld, Y. J., Chitkara, U., Hudgins, L. & Quake, S. R. Analysis of the size distributions of fetal and maternal cell-free DNA by paired-end sequencing. Clinical Chemistry 56, (2010).
Luiken, R. E. et al. Determinants for antimicrobial resistance genes in farm dust on 333 poultry and pig farms in nine European countries. Environmental Research 208, (2022).
Bushel, P. R., Ferguson, S. S., Ramaiahgari, S. C., Paules, R. S. & Auerbach, S. S. Comparison of Normalization Methods for Analysis of TempO-Seq Targeted RNA Sequencing Data. Frontiers in Genetics 11, (2020).
Townes, F. W., Hicks, S. C., Aryee, M. J. & Irizarry, R. A. Feature selection and dimension reduction for single-cell RNA-Seq based on a multinomial model. Genome Biology 20, (2019).
Sharma, C. et al. Antimicrobial resistance: Its surveillance, impact, and alternative management strategies in dairy animals. Frontiers in Veterinary Science vol. 4 Preprint at https://doi.org/10.3389/fvets.2017.00237 (2018).
Zaheer, R. et al. Impact of sequencing depth on the characterization of the microbiome and resistome. Scientific Reports 8, (2018).
Sims, D., Sudbery, I., Ilott, N. E., Heger, A. & Ponting, C. P. Sequencing depth and coverage: Key considerations in genomic analyses. Nature Reviews Genetics vol. 15 Preprint at https://doi.org/10.1038/nrg3642 (2014).
Pereira-Marques, J. et al. Impact of host DNA and sequencing depth on the taxonomic resolution of whole metagenome sequencing for microbiome analysis. Frontiers in Microbiology 10, (2019).
Burkardt, J. The Truncated Normal Distribution. Department of Scientific Computing , Florida State University (2014).
Papadopoulos, C. E. & Yeung, H. Uncertainty estimation and Monte Carlo simulation method. Flow Measurement and Instrumentation 12, (2001).
Soetaert, K. & Petzoldt, T. Inverse modelling, sensitivity and monte carlo analysis in R using package FME. garr.tucows.com (2010).
Neal, P. & Terry Huang, C. L. Forward Simulation Markov Chain Monte Carlo with Applications to Stochastic Epidemic Models. Scandinavian Journal of Statistics 42, 378–396 (2015).
Gardy, J. L. & Loman, N. J. Towards a genomics-informed, real-time, global pathogen surveillance system. Nature Reviews Genetics vol. 19 Preprint at https://doi.org/10.1038/nrg.2017.88 (2018).
Sholeh, M. et al. Antimicrobial resistance in Clostridioides (Clostridium) difficile derived from humans: A systematic review and meta-analysis. Antimicrobial Resistance and Infection Control vol. 9 Preprint at https://doi.org/10.1186/s13756-020-00815-5 (2020).
Mouton, J. W. et al. The role of pharmacokinetics/pharmacodynamics in setting clinical MIC breakpoints: The EUCAST approach. Clinical Microbiology and Infection 18, (2012).
Jamali, S., Jönsson, P., Eklundh, L., Ardö, J. & Seaquist, J. Detecting changes in vegetation trends using time series segmentation. Remote Sensing of Environment 156, (2015).

No competing interests reported.

Supplementary.docx

Download PDF

Journal Publication

published 21 Nov, 2023

Read the published version in Scientific Reports →

Editorial decision: Major revision
10 Nov, 2022
Reviews received at journal
24 Oct, 2022
Reviews received at journal
23 Oct, 2022
Reviewers agreed at journal
14 Oct, 2022
Reviewers agreed at journal
26 Sep, 2022
Reviewers invited by journal
07 Sep, 2022
Editor assigned by journal
07 Sep, 2022
Editor invited by journal
26 Aug, 2022
Submission checks completed at journal
26 Aug, 2022
First submitted to journal
24 Aug, 2022

You are reading this latest preprint version

Modelling the effectiveness of surveillance based on metagenomics in detecting, monitoring, and forecasting antimicrobial resistance in livestock production under economic constraints

Status:

Journal Publication

Version 1

Abstract

Figures

Introduction

Results

Discussion

Methods

Model for simulating results obtained in the surveillance program

Statistical analyses of the time series obtained in the simulations

Breakpoint analysis - DBEST method

Estimation of the time-trend

Contributors

Declarations

References

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1