Effect of mixing and feed batch sequencing on the prevalence and distribution of African swine fever virus in swine feed

Abstract It is critical to have methods that can detect and mitigate the risk of African swine fever virus (ASFV) in potentially contaminated feed or ingredients bound for the United States. The purpose of this work was to evaluate feed batch sequencing as a mitigation technique for ASFV contamination in a feed mill, and to determine if a feed sampling method could identify ASFV following experimental inoculation. Batches of feed were manufactured in a BSL‐3Ag room at Kansas State University's Biosafety Research Institute in Manhattan, Kansas. First, the pilot feed manufacturing system mixed, conveyed, and discharged an ASFV‐free diet. Next, a diet was manufactured using the same equipment, but contained feed inoculated with ASFV for final concentration of 5.6 × 104 TCID50/g. Then, four subsequent ASFV‐free batches of feed were manufactured. After discharging each batch into a collection container, 10 samples were collected in a double ‘X’ pattern. Samples were analysed using a qPCR assay for ASFV p72 gene then the cycle threshold (Ct) and Log10 genomic copy number (CN)/g of feed were determined. The qPCR Ct values (p < .0001) and the Log10 genomic CN/g (p < .0001) content of feed samples were impacted based on the batch of feed. Feed samples obtained after manufacturing the ASFV‐contaminated diet contained the greatest amounts of ASFV p72 DNA across all criteria (p < .05). Quantity of ASFV p72 DNA decreased sequentially as additional batches of feed were manufactured, but was still detectable after batch sequence 4. This subsampling method was able to identify ASFV genetic material in feed samples using p72 qPCR. In summary, sequencing batches of feed decreases concentration of ASFV contamination in feed, but does not eliminate it. Bulk ingredients can be accurately evaluated for ASFV contamination by collecting 10 subsamples using the sampling method described herein. Future research is needed to evaluate if different mitigation techniques can reduce ASFV feed contamination.


INTRODUCTION
The porcine epidemic diarrhoea virus (PEDV) outbreak of 2013-2014 was the first major disease outbreak to suggest a potential link between contaminated feed and pathogen transmission in pigs (Scott et al., 2016). This hypothesis was never unequivocally proven, but the concept of applying biosecurity practices to the United States swine industry feed manufacturing and delivery systems became heavily emphasized. Research has continued to demonstrate that the risk of feed-based virus transmission extends beyond PEDV and could include viruses such as African swine fever virus (ASFV), foot-and-mouth disease virus (FMDV), or classical swine fever virus (CSFV) (Dee at al., 2018;Stoian et al., 2020). Improved biosecurity practices in the feed industry became particularly important in 2018, when a number of historically ASFV-free countries in Southeast Asia began to report ASFV cases (Gaudreault et al., 2020). The United States maintains trade relationships with a number of countries that are now in ASFVendemic regions, leading to concerns that ASFV may enter the United States through the feed supply chain or other avenues. There is no active surveillance for ASFV in feed or ingredients imported from ASFV-endemic regions, nor is there a validated protocol to sample or analyse for ASFV in a feed or ingredient matrix (USDA-APHIS-VS, 2019). It has been hypothesized that the same methods which demonstrated appropriate sensitivity and specificity for PEDV detection in feed may be applicable to ASFV, but this has not yet been tested. Furthermore, it has been suggested that mitigation measures common in PEDV, such as feed batch sequencing to reduce viral concentration, may be equally effective against ASFV. However, this has also never been evaluated. Therefore, the objectives of this study were to (1) determine if a common sampling strategy could consistently detect ASFV in feed, and (2) evaluate if feed batch sequencing could serve as a potential mitigation technique for ASFV contamination during feed manufacturing.

General
The study was conducted at the Biosecurity Research Institute (BRI)

Inoculation
To prepare the inoculum, 8.5 mL of pooled blood treated with ethylenediaminetetraacetic acid (EDTA) from ASFV-infected pigs was mixed in RPMI media to prepare 530 mL of virus inoculum at a final concentration of 2.7 × 10 6 TCID 50 /mL of ASFV genotype II virus (strain Armenia 2007).

Manufacture and sampling
Feed was manufactured as described by Schumacher et al. (2019). The feed manufacturing system was first primed with an ASFV-free batch of feed, which was subsequently followed by a second batch of feed that was contaminated with ASFV. Four additional batches of ASFV-free feed were subsequently mixed and discharged through the same equipment without any cleaning or disinfection occurring between batches. For this study, a corn and soybean-meal-based diet with a composition normally fed to gestating sows was manu- × 10 4 TCID 50 /g, which was then mixed, conveyed, and discharged using the same equipment and procedures as previously described for the negative control.
c. Sequences 1-4 (Batches 3, 4, 5, and 6)-Manufacture of subsequent batches of feed: Following the discharge of the ASFV-contaminated batch of feed, the same process of mixing, conveying, and discharging 25 kg batches of feed was repeated four additional times using ASFV-free feed.
After a batch of feed was discharged, 10 feed samples were collected just as previously described by Jones et al. (2020). Briefly, the 10 samples were taken from the feed that had been discharged in a biohazard tote through two 'X' patterns. To achieve this pattern, the biohazard tote was divided into two halves and in each half, two imaginary diagonal lines were drawn from corner to corner to make an 'X' pattern. Samples were taken from the corners of each half, along with a sample from the middle where the two imaginary diagonal lines crossed. The 10 samples were not mixed together, but were analysed in separate PCR reactions. This sampling technique resulted in a grand total of 60 feed samples for the entirety of the experiment.

Laboratory analysis
Feed samples were tested at a BSL-3+ laboratory in the BRI. Briefly, 10 g of each feed sample was put in a tube, suspended with 35 mL of PBS, and the tube was capped and inverted, and then incubated overnight at 4 • C. Approximately 10 mL of supernatant was recovered, aliquoted into 5 mL cryovials, and stored at −80 • C until processed for qPCR. In preparation for magnetic bead-based DNA extraction, 500 µL

Statistical analysis
Statistical analysis for this study was performed using R programming language (Version 3.6.1 (2019-07-05), R Core Team, R Foundation for Statistical Computing, Vienna, Austria). The experimental unit for this study was the feed sample. Each feed sample had one extraction for the qPCR assay and each extraction was run in duplicate for qPCR analysis with the exception of samples from batch 2 in which each feed sample had two extractions for the qPCR assay; both extractions were run in duplicate for qPCR analysis as an initial assessment to evaluate the variability present within the extraction and amplification procedures.
Response values for the ASFV p72 gene were analysed using a linear mixed model fit using the lme function in the nlme package, using a normal distribution with the fixed effect as batch, with a random effect of sample to indicate the appropriate level of experimental replication given the duplicate qPCR analysis of feed samples. Results of Ct and genomic CN/g are reported as least squares means ± standard error of the mean. Samples not containing detectable ASFV DNA were assigned a value of 45 because this was the highest number of cycles the qPCR assay performed before concluding a sample did not have detectable ASFV DNA. Genomic CN/g data were Log 10 transformed prior to data analysis to satisfy the assumption of normality. All statistical models were evaluated using visual assessment of studentized residuals and models accounting for heterogeneous residual variance were used when appropriate. A Tukey multiple comparison adjustment was incorporated when appropriate. Results were considered significant at p ≤ .05 and marginally significant between p > .05 and p ≤ .10.

RESULTS AND DISCUSSION
Outbreaks of PEDV in North America were the first events with a potential link between contaminated feed and transmission of disease to pigs (Scott et al., 2016). Since then, veterinarians, producers, Positive 0/10 10/10 10/10 9/10 9/10 7/10 Suspect 0/10 0/10 0/10 1/10 1/10 3/10 Non-detected 10/10 0/10 0/10 0/10 0/10 0/10 † Swine gestation feed was inoculated with African swine fever virus (ASFV) at 5.6 × 10 4 TCID 50 /gram inoculated feed (positive) following an initial priming of the feed manufacturing equipment with ASFV-free feed (negative). Four subsequent batches of feed were manufactured (sequence 1 to 4) and were initially free of ASFV. Ten feed samples were collected from each subsequent batch of feed and analysed using an ASFV p72-specific qPCR assay with each sample analysed in duplicate. Samples were considered qPCR positive if 2 of 2 qPCR reactions had detectable ASFV DNA, suspect if 1 of 2 qPCR reactions had detectable ASFV DNA, and non-detected if 0 of 2 qPCR reactions had detectable ASFV DNA. Log 10 genomic copies/g ¶ 0.0 4.7 ± 0.08 a 3.6 ± 0.09 b 3.1 ± 0.23 b,c 3.1 ± 0.23 b,c 2.8 ± 0.23 c † Swine gestation feed was inoculated with African swine fever virus (ASFV) at 5.6 × 10 4 TCID 50 /gram (positive), following an initial priming of the feed manufacturing equipment with ASFV-free feed (negative). Four subsequent ASFV-free batches of feed were manufactured (sequence 1 to 4). Ten feed samples were collected after each batch of feed and were analysed using an ASFV p72-specific qPCR assay with each sample analysed in duplicate for each assay. Statistical analysis includes all treatment groups except for negative control where samples were collected prior to ASFV inoculation. Values for main effect of batch do not include negative batch of feed. § Cycle threshold values for qPCR reactions with no detectable ASFV p72 gene expression were assigned a value of 45 within the statistical analysis. Batch:

Batch of feed
p < .0001. ¶ Log 10 transformed genomic copies for the ASFV p72 gene per g of feed from feed samples. Batch: p < .0001. a,b,c Means within row lacking common superscript differ (p < .05) using Tukey multiple comparison adjustment.
swine diets are manufactured in ASFV-endemic countries (Shurson et al., 2019). While their manufacture is typically in biosecure laboratories, and the ingredients themselves may pose low risk for foreign animal disease transmission, containers carrying these ingredients may become contaminated and thus become a potential source of ASFV entry into the United States. In theory, ingredients could be sampled for ASFV and screened for safety prior to entry into the country, but surveillance of this magnitude has not been implemented, partially due to the lack of validated bulk sampling or extraction methodologies After the ASFV-positive batch of feed was manufactured, all feed samples had detectable ASFV p72 genetic material (Table 1). A limitation of this experiment is the lack of infectivity data associated with the feed samples containing ASFV p72-specific DNA. This research utilized ASFV, a BSL-3 pathogen, and a US select agent; meaning, to get approval to use this virus is a rigorous progress, requiring special laboratories, and intensive training. Validating these feed samples for ASFV infectivity is important, and will be an area of our future research efforts; however, the focus of this study was to determine if feed sequencing was an effective mitigant strategy for ASFVcontaminated feed and if feed sampling techniques could accurately identify ASFV genetic material. The data presented here provides significant value to the global feed and swine industry by establishing the presence of ASFV DNA in feed after first contaminating and then flushing a feed production system with subsequent batches of 'clean' feed, along with the ability to detect ASFV genetic material in the feed which can provide information for urgently needed surveillance programs.
In conclusion, sequencing with four batches of feed after contamination of a feed mill with ASFV can decrease overall ASFV contamination within feed samples, but not eliminate it entirely. In addition, collecting 10 evenly distributed samples using an 'X' pattern collection system allows for the detection of ASFV genetic material under the conditions of the current investigation. The findings of this study highlight the importance of excluding ingredients from ASFV-endemic countries, but also highlights that proper sampling can be an effective tool to detect ASFV contamination. Additional research is necessary to evaluate the combination of mitigation techniques like chemically treating flush diets (similar to what is done with PEDV) on ASFV-contaminated ingredients.