## Introduction

Bovine tuberculosis (bTB), primarily caused by Mycobacterium bovis, is an important animal health and economic problem for the British cattle industry, and an important zoonotic infection globally, particularly where bovine milk, the main source of human infection, is not treated1. M. bovis, is a member of the M. tuberculosis complex (MTC) group of bacteria and predominantly infects cattle, but has a wide host range that includes badgers (Meles meles), other wildlife, camelids, goats, humans and some companion animals. Badgers are reported to act as one of the ‘reservoirs’ of M. bovis bacilli, at least in some areas2,3,4,5. M. tuberculosis, the overwhelmingly most frequent cause of human TB and M. microti, which mainly circulates in field voles (Microtus agrestis)6, are also found in Great Britain.

The incidence of TB in cattle in England and Wales has been rising since the mid-1980s, and the area in which it is considered endemic has gradually expanded to cover most of the south west of England and south Wales, the West Midlands and parts of mid-Wales2. Although usually described as a single spreading epidemic, the regional distributions of different spoligotypes of M. bovis might suggest a series of regional epidemics7. The government’s strategy for achieving ‘officially TB-free (OTF)’ status for England is based on geographic areas of risk, with different surveillance and control measures applied to different areas: the Low Risk Area (LRA), the Edge Area and the High Risk Area (HRA). The Edge Area is a boundary of intermediate TB incidence in cattle, situated between the HRA and LRA and, at the time of this study, comprised eleven counties (or part-counties), where surveillance suggested that the historically low incidence of TB in cattle was increasing.

The control of TB in cattle in many parts of the world is complicated by the involvement of wildlife species, such as deer8 and, in England, Wales, the island of Ireland and increasingly in some other European countries, badgers (Meles meles)2,3. It is unclear whether or not badgers are technically a reservoir host in the HRA (i.e. not only a source of infection to cattle—the ‘target’ host—but able to maintain infection without reintroduction from cattle)2,9, and even less is known about the role of badgers in cattle infection in the Edge Area. Thus the expansion of the cattle epidemic in the Edge Area could be the result of cattle-to-cattle, badger-to-badger, cross species transmission, or, indeed, other even less well-understood transmission routes such as through soil and farm waste10,11,12,13; and the relative contributions of each potential transmission route, or if they differ between regions in the Edge Area, are unknown.

Ongoing long-term studies of badger TB in Gloucestershire (SW England)23 in the HRA, show long term persistence of TB in badger setts, and recent badger surveys around an isolated hotspot of TB in cattle in Cumbria (NW England), suggested that infection of imported origin in cattle was circulating in both badgers and cattle24. While these studies provide information about M. bovis in badgers and cattle in specific locations, there remains a lack of research into the relationship between M. bovis in cattle and badgers in other risk areas, particularly where the epidemic in cattle is spreading. Our study contributes to this knowledge gap.

Here we describe a study of road-killed badgers in northern Edge Area counties of England (Fig. 1; Cheshire, Derbyshire, Nottinghamshire, Leicestershire, Warwickshire and Northamptonshire), carried out in 2016–2017, and compare those findings with reported outbreaks of TB in cattle over the same area and time, collected by routine, government-regulated surveillance.

## Results

Overall, 610 badger carcasses suitable for necropsy and sampling, were collected from the study area (Fig. 2). Of these, 246 (40%) were collected by farmers and farming groups, 219 (36%) by badger or conservation groups and 145 (24%) by project staff, veterinarians, local authorities or other groups. These proportions differed by county (SI 1). The age and sex of the badger carcasses collected, when recorded (> 98% of carcasses), was overall 53% male and 46% female, and the age ratio 76% adult : 24% juvenile, with 13 (3%) of the adults classified as ‘old’ on the basis of pelage and tooth wear.

The availability of badger carcasses varied over time, with a peak in carcass collection in February 2017 (SI 1), and the number collected was capped at 100 badgers per county.

### Mycobacterium tuberculosis complex (MTC) in badgers

Overall, 51/610 (8.3%, 95% CI 6.4–11%) badgers were positive for MTC by culture and IS6110 PCR. The prevalence of infection in males was approximately twice that in females (11% c.f. 5%, chi-square 6.019 p = 0.0142, odds ratio 2.16, relative risk 2.13). The prevalence did not vary significantly with the age of the badger, the time of year or the type of collector (‘farmer’, ‘wildlife’ group or ‘other’).

At post mortem examination, only one badger, which was MTC culture positive, had obvious, widespread lesions of TB, and another had atypical gross lesions from which MTC was isolated. The prevalence of MTC in the badgers varied between counties, such that Cheshire, Leicestershire and Warwickshire had higher prevalence than Derbyshire, Nottinghamshire and Northamptonshire (χ2 = 12.7, p = 0.026).

Spoligotyping of badger MTC revealed a variety of spoligotypes, including those such as SB0129, SB0263, SB0272 and SB0140, frequently found associated with cattle infection in England25 but also a range of novel and rare spoligotypes (Figs. 2 and 3, and Supplementary Information). The most common spoligotype of M. bovis found in badgers (SB0129), was found particularly in the north of the study area, but also sporadically elsewhere. Other major spoligotypes of M. bovis isolated from cattle were found more sporadically in badgers.

### Mycobacterium tuberculosis complex bacteria in cattle

The results of farm-level surveillance testing of cattle for TB over the same period and area as the badger study are shown in Fig. 3. In total, there were 730,654 individual tests undertaken at 5,765 cattle holdings, and reported herd sizes ranged from 1 to 2,900 animals (median 62). There were 5227 positive animals, giving rise to 660 herd ‘breakdowns’ or outbreaks (11% of holdings) over the study period.

In the vast majority (91%) of herd breakdowns, the index case was detected by skin test during routine surveillance testing. When M. bovis was isolated from cattle, almost all (98.3%) of the spoligotypes from cattle were those previously identified in cattle in England. Each had a distinctive geographical range (Fig. 3) but some were also found sporadically in other regions. For example, the predominant spoligotype found in cattle during 2016–2017 was SB0129, particularly in the northern part of the study area, but the same spoligotype was also more sporadically found in other parts of the study area. Spoligotype SB0272 was the predominant spoligotype found in the southern part of the study areas, SB0263 was found sporadically throughout the study area, but mainly in western Cheshire.

## Discussion

Stakeholders collected the badger carcasses and recorded their locations in this study, as described previously22. Although there was variation between counties, the type of stakeholder (farming group, wildlife group or other) was not overall associated with differences in the sex and age of badger carcasses collected or the prevalence of infection with MTC. Furthermore, the sex ratio, and proportions of animals by age were similar to those described in previous studies of live badgers27,28. Thus, while using road-killed badgers might involve biases in terms of space and time, there was no evidence that different stakeholder groups influenced the outcome of this study in terms of badger demographics or infection prevalence. As the density of badgers is not known in the Edge Area, and 100 carcasses were collected from each county (with counties of different areas), it is impossible to determine if heterogeneities in sampling density reflected heterogeneities in badger population density, stakeholder engagement or roads and road usage.

Overall, 51/610 (8.3%, 95% CI 6.4–11%) carcasses were culture positive and confirmed by PCR as MTC, a figure similar to that found for Wales in 2014–201621. Previous studies found that the prevalence of infection in badgers was higher in adults than cubs, and in males than females, although in some cases these differences were not statistically significant16,20,21,29. This study found no significant difference in prevalence by age, although the numbers of, and prevalence in, younger badgers was small; but that the prevalence of infection in males was approximately twice that in females. It has been suggested this difference might reflect the larger home ranges of, and more territorial conflicts between, males compared to females, both of which could increase transmission20,21.

Most infected badgers had no macroscopic lesions suggestive of tuberculosis on post-mortem examination. The proportion of infected badgers with no obvious lesions (96%) was similar to that found previously in Cheshire22, but higher than that found in most studies of badgers in areas where infection is endemic in both cattle and badgers. The majority of positive cultures were from the ‘head and neck’, ‘thorax’ and ‘carcass’ pools of lymph nodes (27%, 27% and 21%, respectively). This is also similar to the findings of22, but contrasts with many previous studies, which found that isolation of M. bovis was most frequently from lung and thoracic lymph nodes30,31. These differences in infection sites and lesions may reflect differences sources of infection, in transmission routes and dose, or in the pathogenicity of the different genotypes of MTC found in the badgers across this study. It might also reflect differences in how carcasses were collected between these and other studies. Although it is often assumed that transmission of M. bovis among badgers is by the respiratory route, and requires lesions, transmission via bites is also thought to be important. It is conceivable that transmission via saliva and respiratory secretions through bites, grooming and other close contact, or, indeed, contaminated soil and bedding, associated with pharyngeal lymphoid infection is more likely, without the need for lesions. Indeed, such secretions might also enable shedding in faeces, with or without enteric infection.

There was variation in the frequency of infected badger carcasses between counties, with Cheshire, Leicestershire and Warwickshire having headline infection rates (10–15%) higher than Derbyshire, Nottinghamshire and Northamptonshire (4–5%). In Cheshire, 15/104 (13.5%, 95% CI 8–21%) badgers were confirmed MTC-infected, a prevalence not significantly different from that found in 201422. There are no recent studies of MTC prevalence in badgers across the rest of the study area with which the results can be compared. Spoligotyping was performed rether than whole genome sequencing to enable comparison with historical cattle data. The most common spoligotype in MTC isolated from badgers was SB0129, particularly in the northern part of the study area where it was also the most frequent spoligotype found in cattle. Elsewhere, a wide range of spoligotypes were found in badgers, whereas in cattle the predominant spoligotypes were those expected if those infections resulted from expansion of regional cattle spoligotypes from neighbouring High Risk Area counties. The majority of cattle outbreaks in Leicestershire during the study were not confirmed by culture and so the spoligotype(s) responsible could not be identified. Subsequent surveillance reports (APHA, 2019), suggest that cattle outbreaks when spoligotyped remain largely of SB0129 with some suggestion of SB0272 expanding across the border from Warwickshire.

More formal analysis of the prevalence of MTC and spoligotypes in both cattle and badgers across the study area found no pronounced evidence of either MTC or SB0129 -specific spatial foci beyond those in eastern Cheshire and the Derbyshire border with Staffordshire, where SB0129 was and remains the predominant spoligotype found in cattle32. The approach used is robust to differing sampling densities of either farms or badgers across space, necessary with the inevitable variation in population density and sampling density. However, the diversity of spoligotypes found in badgers combined with the relatively low numbers of MTC-positive badgers in much of the study area, made further quantitative analysis difficult.

## Materials and methods

All badgers used in this study were found dead. The majority of badgers collected in this study were collected from road-sides and had lesions compatible with vehicle collisions. A small number were found dead in fields or farm buildings, some of which also had lesions compatible with vehicle collisions. The approach used was that developed for a previous study in Cheshire in 201422: stakeholder groups collected carcasses, recorded where they were found, most often using smart phones, and the carcasses were then brought to the veterinary pathology facilities at either the University of Nottingham or University of Liverpool for necropsy and sampling. Stakeholders included farmers, veterinary practices, wildlife groups, local authorities and businesses as well as university staff. All experimental protocols were approved by the Defra and University of Nottingham.

Badgers were collected from six English counties in the northern edge of cattle TB epidemic (Cheshire, Derbyshire, Nottinghamshire, Leicestershire, Warwickshire and Northamptonshire) over a 16 month period from 2016 to 2017.

Where possible badger carcasses were stored in dry, cool areas or refrigerated. With two exceptions, frozen badgers were not included in the study. Carcass collection and necropsy occurred within 72 h of collection, and obviously autolytic or heavily damaged carcasses were not processed due to the likelihood of contamination. Each badger’s sex, weight and age (adult or juvenile, based on teeth, size and pelage) were recorded.

Lymph nodes and visible ‘TB-like’ lesions were sampled for culture. In addition to lung samples, four pools of lymph nodes were sampled: a ‘thoracic pool’ of bronchial and mediastinal lymph nodes, an ‘abdominal pool’ comprising hepatic and mesenteric lymph nodes, a ‘head and neck’ pool of parotid, mandibular, retropharyngeal and cervical lymph nodes, and a ‘carcass pool’ of prescapular, axillary, and superficial inguinal lymph nodes. If possible lesions were observed, each sample of lesion material was processed separately. Pooled tissues were stored at 4 °C for up to 48 h before processing. Individual tissue samples were stored at − 80 °C in case needed for further study.

### Tissue processing

Tissue samples for microbiological culture were processed in a Containment Level 3 (CL3) facility at the University of Nottingham, Sutton Bonington Campus. Samples were processed according to22. Briefly tissue pools or lesions were gently ground with sterile sand and 2 ml phosphate buffer saline (PBS; Dulbecco A, Oxoid). Samples were mixed with an equal amount of 5% oxalic acid (Fisher Scientific) and incubated at room temperature for 10 min to reduce non-mycobacterial contamination. The decontaminated pools (200 µl) were inoculated onto Stonebrink Selective agar supplemented with PACT (BD Diagnostics) and Middlebrook 7H11 agar slopes supplemented with PANTA (BD Diagnostics), and incubated at 37 °C for a minimum of 12 weeks. Cultures were examined approximately weekly for the appearance of colonies suggestive of mycobacteria.

### Characterisation of mycobacteria

Characteristic and possible colonies were picked from slopes after 12 weeks of incubation. Cells were heat killed (80 °C for 30 min) and DNA was extracted by crude heat lysis, in which heat-killed colonies were frozen and heated to 95 °C for 5 min in sterile distilled water (SDW), centrifuged (13,000 × g; 3 min) and the supernatant used as template DNA for subsequent PCR.

Isolates were screened for the MTB signature genetic element, IS6110 by PCR34. For some colonies, the partial 16 s rRNA gene was amplified, sequenced (Source Bioscience, UK) and analysed to determine whether the colonies were part of the MTB complex or not35. DNA from confirmed members of the MTB complex were further interrogated by microarray-based spoligotyping (Alere Technology) according to36.

### Cattle data

Cattle testing data were obtained from APHA’s surveillance programme. At the time of the study, different cattle testing regimens were used in different counties. In HRA and Edge counties, cattle were routinely tested at herd level six-monthly or annually, with additional testing if infection was detected. Furthermore, some holdings had groups of cattle tested due to movements or because of epidemiological links to outbreaks detected elsewhere. Some TB-infected cattle were also detected during routine post-mortem inspection in the abattoir. As described in22, the original data sets from APHA included cattle holding location, the results of each test (including spoligotype if known), the official ‘breakdown identifier’ and start date, and the number of cattle on the holding and tested on each occasion. For the purposes of this study, all test results in the study period were combined for each site, and TB-positive herds were defined as those with at least one animal having tested positive using the ‘single intradermal comparative cervical tuberculin’ (SICCT) test, at least one animal with two consecutive inconclusive skin test results, positive interferon-gamma blood tests, or if lesions which were culture positive were found at the abattoir.

### Statistical analyses

The prevalence was estimated among the sampled badgers both overall and at a county level, as in previous studies19,20 on the assumption that the carcasses collected were representative of the overall population.

To detect evidence for co-occurrence of MTC positive badgers and SB0129 herd breakdowns, a bivariate Bayesian logistic common-components geostatistical model was used37. Observations consisted of a set of geographically located farms and badgers, with infection (breakdown) status recorded as a binary outcome variable (positive or negative), and observation type (farm or badger) as an explanatory variable. The outcome was modelled for each observation, yi, as a Bernoulli random variable with probability pi such that

$$log\, it\left({p}_{i}\right)=\left\{\begin{array}{l}\alpha +\beta +{S}_{c}\left({x}_{i}\right)+{S}_{b}\left({x}_{i}\right)\quad if \, i \, is \, a \, badger\\ \alpha +{S}_{c}\left({x}_{i}\right)+{S}_{f}\left({x}_{i}\right)\quad if\, i \, is \, a \, farm\end{array}\right.$$

where xi is the geographical location of observation i, and $$\alpha$$ and $$\beta$$ represent respectively the intercept and the log odds ratio for disease in badgers versus farms. The model makes use of three spatial Gaussian Processes, $${S}_{c}\left({x}_{i}\right)$$, $${S}_{b}\left({x}_{i}\right)$$, and $${S}_{f}\left({x}_{i}\right)$$, characterised by a covariance function $${\Sigma }_{ijz}$$ between the locations of observations i and j for “common”, “badger”, and “farm” spatial correlation components respectively. In each case, a Matérn correlation structure is assumed with sill and scale parameters $${\sigma }_{z}$$ and $${\phi }_{z}$$.48 Here, $${S}_{c}\left({x}_{i}\right)$$ represents a disease risk surface common to both farms and badgers that allows disease prevalence to vary with space. Then, $${S}_{b}\left({x}_{i}\right)$$ and $${S}_{f}\left({x}_{i}\right)$$ represent respectively the additional spatial risk for badgers and farms respectively.

$${\Sigma }_{ijz}={\sigma }_{z}\left(1+\frac{\surd 3\Vert {x}_{i}-{x}_{j}\Vert }{{\phi }_{z}}\right)exp\left(\frac{-\surd 3\Vert {x}_{i}-{x}_{j}\Vert }{{\phi }_{z}}\right)$$

In a Bayesian context, prior distributions were assigned to the parameters as shown in Table 1.

In contrast with previous work in Cheshire22, the volume of data analysed here (5522 farm and 565 badger locations) precluded a naïve implementation of the three Gaussian processes. Instead, a low-rank approximation was used in which a number of “knot” points were chosen to represent the Gaussian process in 2-dimensional space, and the resulting representation projected onto the observation points. Here, 600 and 200 points were chosen for farms and badgers respectively by 2-dimensional k-means clustering38.

The model was implemented in Python 3.6. using the PyMC3 3.7 package39, with the No-U-Turn Sampler (NUTS) used to draw samples from the joint posterior distribution. The sampler was run for 20,000 samples, discarding the initial 1500 as tuning and burn-in. The convergence of 4 MCMC chains was inspected visually and found to be satisfactory. Traceplots for these runs are shown in Supplementary Fig. 1 together with Gelman-Rubin statistics demonstrating convergence ($$\widehat{R}$$ in all cases).

To detect co-regionalisation of infection in both badgers and farms, the Bayesian predictive distribution of the Gaussian Process was evaluated on a fine grid over the study area38. To summarise each predictive distribution, both the mean posterior value of the Gaussian process, and the probability that the posterior distribution at each grid point was greater or less than zero, were calculated. Since the Gaussian process represented the spatial log odds ratio, this is equivalent to calculating the posterior probability of the spatial odds ratio being greater or less than 1, i.e. Pr(OR > 1) and Pr(OR < 1) respectively, noting that Pr(OR > 1) = 1 – Pr(OR < 1). To present the results, we plot maps of the mean posterior spatial odds ratio for pixels where Pr(OR > 1) > 0.95 or Pr(OR < 1) > 0.95. Thus our maps highlight regions where there was good evidence that the odds ratio is significantly different to 1 (i.e. no spatial effect).