Edinburgh Research Explorer Histological and immunohistochemical features suggesting aetiological differences in lymph node and (muco)cutaneous feline tuberculosis lesions

O bjectives : To identify and describe histological and immunohistochemical criteria that may differentiate between skin and lymph node lesions associated with Mycobacterium ( M. ) bovis and M. microti in a diagnostic pathology setting. M M ethOds : Archived skin and lymph node biopsies of tuberculous lesions were stained with haematoxylin and eosin, Ziehl-Neelsen and Masson’s Trichrome. Immunohistochemistry was performed to detect the expression of calprotectin, CD3 and Pax5. Samples were scored for histological parameters ( i.e. granulomas with central necrosis versus small granulomas without central necrosis, percentage necrosis and/or multinucleated giant cells), number of acid-fast bacilli (bacterial index) and lesion percentage of fibrosis and positive immunohistochemical staining. r esults : Twenty-two samples were examined ( M. bovis n=11, M. microti n=11). When controlling for age, gender and tissue, feline M. bovis -associated lesions more often featured large multi-layered granulomas with central necrosis. Conversely, this presentation was infrequent in feline M. microti associated lesions, where small granulomas without central necrosis predominated. The presence of an outer fibrous capsule was variable in both groups, as was the bacterial index. There were no differences in intralesional expression of immunohistochemical markers. c linical s ignificance : Differences in the histological appearance of skin and lymph node lesions may help to infer feline infection with either M. bovis or M. microti at an earlier stage when investigating these cases, informing clinicians of the potential zoonotic risk. Importantly, cases of tuberculosis can present with numerous acid-fast bacilli. This implies that a high bacterial index does not infer infection with non-zoonotic non-tuberculous mycobacteria.


INTRODUCTION
Mycobacterial infections are increasingly recognised as a substantial cause of morbidity in the domestic cat population, especially in Great Britain (Broughan et al. 2013), where approximately 1% of all feline biopsy submissions show changes suggestive of myco-bacterial disease (Gunn-Moore et al. 2013). The most common presentation of feline mycobacteriosis is raised cutaneous lesions; these may be ulcerated and discharging sinus tracts can be present. Local lymphadenitis is frequently identified, presumptively due to drainage of mycobacteria-laden immune cells, which are predominantly macrophages (Ganbat et al. 2016), from the cuta-neous lesion (Gunn-Moore et al. 2011a). This clinical presentation is recognised in cases of infection with members of the Mycobacterium (M.) tuberculosis-complex (MTBC) as well as non-tuberculous mycobacteria (NTM) such as members of the M. avium-complex (Gunn-Moore 2014). Infection can spread to the lungs, putatively due to haematogenous spread within monocytes and neutrophils (Latimer et al. 1997, Krishnan et al. 2010, resulting in a broncho-interstitial pulmonary pattern (Bennett et al. 2011). Mycobacterial lesions have also been reported in the eyes (Stavinohova et al. 2019), joints (Lalor et al. 2017) and other organs and tissues, resulting in a range of clinical signs (Gunn-Moore 2014).
The MTBC consists of 10 different mycobacterial species (Rodriguez-Campos et al. 2014, Dippenaar et al. 2015, responsible for causing tuberculosis (TB) across a wide taxon of animals. Those that cause TB in the domestic cat are M. bovis and M. microti, the vole bacillus (Gunn-Moore et al. 2011a), both of which are potentially zoonotic (Emmanuel et al. 2007, O'Connor et al. 2019. While M. bovis has been identified as causing disease across multiple species of veterinary interest, reports of M. microti infection in species other than the domestic cat and its maintenance host the field vole (Microtus agrestis) are uncommon, although they are being identified more frequently (Smith et al. 2009, Michelet et al. 2015. Where positive culture results were obtained, over 70% of feline mycobacterial infections in Great Britain were due to MTBC pathogens, fairly evenly distributed between M. microti and M. bovis (approximately 40% and 33%, respectively) (Gunn-Moore et al. 2011a). These appear to localise to specific regions of the UK; M. bovis is more frequent in cats from TB endemic areas, whereas M. microti infection is more frequent in cats from areas with low or null prevalence of bovine TB (i.e. Scotland), (Burthe et al. 2008).
M. bovis and M. microti infections produce identical macroscopic lesions; however, they differ in their zoonotic potential, and for some owners the decision to treat or euthanise their cat can depend on whether M. bovis is demonstrated as the infective organism (O'Halloran & Gunn-Moore 2017). Therefore, the ability to rapidly discern between the two causative agents of feline TB is essential and can inform the potential risk posed to owners as well as veterinary staff (de la Rua-Domenech 2006).
Histopathology is often the first diagnostic test performed when investigating a potential case of mycobacterial disease in cats (Gunn-Moore 2014, O'Halloran & Gunn-Moore 2017), and lesions featuring granulomatous to pyogranulomatous inflammation, dominated by epithelioid macrophages, raise the suspicion index of mycobacterial disease (Gunn-Moore et al. 2011b). Unlike in other species, multinucleated giant cells (MNGCs) are not routinely observed in feline mycobacterial lesions (Kipar et al. 2003). Special stains, such as Ziehl-Neelsen (ZN), can be performed on sections where granulomatous to pyogranulomatous inflammation is identified to confirm the presence of acid-fast bacilli (AFB) morphologically consistent with mycobacteria. However, many feline mycobacterial lesions are paucibacillary, i.e. they have few to no obvious AFB on ZNstaining, so a negative result does not rule out mycobacteriosis (Gunn-Moore et al. 2013). Abundant numbers of AFB, where identified, have been associated with cases of lepromatous feline leprosy, a disease entity caused by non-zoonotic NTM species (Malik et al. 2002, O'Brien et al. 2017.
Subsequent diagnostic testing for feline mycobacterial infections can be challenging. Specialist culture, which is the validated assay, has poor sensitivity (Gunn-Moore et al. 2011a) and it can take at least 3 months to obtain a positive result for slow-growing mycobacterial species, which includes M. microti (Smith et al. 2009). More rapid testing methodologies are available, including molecular-based diagnostics, such as polymerase chain reaction (PCR) assays (Aranaz et al. 1996, Richter et al. 2003; however, they are less sensitive on formalin-fixed tissues compared to fresh samples (Reppas et al. 2013). Recent advancements in molecular biology have improved the ability of tests to discriminate between members of the MTBC, resulting in an increase in the identification of M. microti as a causative agent of TB lesions across a range of species , Michelet et al. 2015, Landolt et al. 2019, Pérez De Val et al. 2019. Despite this, the financial constraints of owners may limit the use of these techniques for the species-level diagnosis of feline mycobacterial infections. The interferongamma (IFNγ) release assay (IGRA) has excellent sensitivity for detecting MTBC infections and is reasonable at differentiating between M. bovis and M. microti; however, financial constraints may still apply, obtaining sufficient volumes of blood may be a limiting factor in some cats, and the test requires careful handling and processing of heparinised blood (Rhodes et al. 2008, Rhodes et al. 2011.
The hallmark of TB is granulomatous inflammation; this term can encompass a range of histopathological presentations, from unorganised macrophage infiltration, to the presence of well-formed granulomas (Shah et al. 2017). A precise definition for what constitutes a granuloma is difficult, but at its simplest, the granuloma can be thought of as "an organised collection of mature mononuclear phagocytic cells" (Adams 1976, Pagán & Ramakrishnan 2018. Tuberculous granulomas typically consist of epithelioid macrophages surrounding a necrotic core, encapsulated by fibroblasts and an outer layer of lymphocytes (Martin et al. 2016). The tuberculous granuloma within lymph nodes has been well described in cattle (Wangoo et al. 2005) and these histological findings have been applied to a number of other livestock (Sanchez et al. 2011, Vallejo et al. 2018) and wildlife species (Canfield et al. 2002, García-Jiménez et al. 2012, García-Jiménez et al. 2013. The granuloma can be characterised further by its cellular composition using immunohistochemistry (IHC). This has been performed extensively in both experimental and natural M. bovis infections of livestock (Pereira-Suárez et al. 2006, Palmer et al. 2007, Sanchez et al. 2011, Canal et al. 2017, Vallejo et al. 2018) and wildlife species (Canfield et al. 2002, García-Jiménez et al. 2012, García-Jiménez et al. 2013, as well as one study describing M. microti lesions in its natural host, the field vole (Kipar et al. 2014). One study has been performed on feline mycobacterial lesions (Kipar et al. 2003); however, this did not explore differences between M. bovis and M. microti infections.

British Small Animal Veterinary Association
The objective of the current study was to describe and compare the histological and immunohistochemical features of feline tuberculous granulomas in (muco)cutaneous and lymph node lesions to identify potential patterns that help differentiate between the two causes of TB in cats. We hypothesise that the nodal/(muco)cutaneous granulomas observed in cats infected with M. microti differ in structural organisation and immune cell populations compared to those in M. bovis-positive cats.

Animals and samples
Ethical approval for this study was granted by the institutional Veterinary Ethical Review Committee (approval no. 79 14).
A database maintained on Microsoft Excel © 2016 (Microsoft Corporation) by an independent researcher was searched by the lead investigator in May 2019 to identify cats where a formalinfixed paraffin-embedded (FFPE) tuberculous (muco)cutaneous or lymph node lesion biopsy had been submitted to the institution by referring veterinary surgeons (RVS) and commercial histopathology laboratories following owner consent, and that had a culture, PCR or IGRA diagnosis of M. bovis or M. microti infection. (Muco)cutaneous biopsies were taken to include those recorded as being "skin," "dermal," "subcutaneous" and "gum". The biopsies had been taken as part of the diagnostic investigation of these cases or at post-mortem examination and were fixed in 10% neutral buffered formalin for a minimum of 24 hours postsampling and processed for histology using standard methods. Anti-mycobacterial therapy had not been given before sampling. Mycobacterial disease was suspected based on histopathological examination and/or ZN-staining, and subsequent speciation was confirmed by external specialist mycobacterial culture (Animal and Plant Health Agency), PCR (Leeds Teaching Hospitals NHS Trust) or IGRA (Biobest Laboratories). An IGRA diagnosis of M. microti was considered definitive if the IGRA result showed a biased response to purified protein derivative (PPD) from M. bovis (PPDB) over PPD from M. avium (PPDA) and no response to the antigenic cocktail of early secretory antigenic target 6 kDa (ESAT-6)/culture filtrate protein 10 kDa (CFP-10) (Rhodes et al. 2011), the cats had lesions exhibiting histological features consistent with mycobacterial infection, and they lived in regions of England that are low-risk areas for bovine TB or in Officially TB Free regions, i.e. Scotland. Conversely, an IGRA diagnosis of M. bovis was considered definitive if the IGRA result was PPDBbiased over PPDA and ESAT-6/CFP-10 was either positive or negative (Rhodes et al. 2011), with histological features consistent with mycobacteriosis, and the cat had been living in an area with bovine TB. The exception to having been living in a "highrisk" or "edge" area for bovine TB, is any cat fed the raw commercial diet that has recently been associated with an outbreak of M. bovis in cats in England and Scotland . Eligible cases were cross-referenced with clinical histories supplied by the RVS to identify additional clinical information, i.e. results of testing for feline leukaemia virus (FeLV) p27 antigen and feline immunodeficiency virus (FIV) anti-p24 antibodies (SNAP FIV/FeLV Combo Test, Idexx). For each cat, a single FFPE tissue block was selected; serial sections were prepared for histopathology and IHC.
Control FFPE tissues for IHC were a submandibular lymph node from a cat euthanased for reasons not related to mycobacterial infection, and a cutaneous lesion from a cat diagnosed with a NTM infection on IGRA with abundant AFB.

Histopathology
Histopathology was performed to assess the morphological features of the mycobacterial lesions (henceforth, the word lesion will always refer to a granulomatous or pyogranulomatous inflammatory infiltrate). Four-micron thick sections were cut and stained with haematoxylin and eosin (H&E), Ziehl-Neelsen (ZN) and Masson's Trichrome (MT). H&E-stained slides were evaluated to describe the cellular composition and granuloma type of the lesions, with a focus on specific parameters.
A lesion was defined as an area of infiltration by macrophages which were mostly epithelioid. The histological presentation of lesions was assessed and compared to previously described classification systems in other species (Wangoo et al. 2005). However, these systems did not adapt to feline tuberculous lesions, as MNGCs and dystrophic mineralisation were not present in any case. For this reason, a novel, description-based classification was devised for this study (see results). Necrosis was defined as areas with loss of cellular and structural detail, with accumulation of eosinophilic and basophilic (karyorrhectic) debris. This presentation is consistent with caseous necrosis (Miller & Zachary 2017).
Slides stained with ZN were examined under standard light microscopy to determine the bacterial index (BI). To do this, the number of individual AFB were counted in 15 randomly chosen high-power fields (hpf ) (×1000 magnification) and an average was calculated (AFB/hpf ). A modified Ridley BI score, used for leprosy in humans (Ridley 1964), was used to grade the number of AFB/hpf on a scale of 0 to 6: 0=no AFB/hpf; 1=0.01 to 0.1; 2=0.11 to 1.0; 3=1.01 to 10; 4=10.01 to 100; 5=100.01 to 1000; 6=>1000 AFB/hpf ). In accordance with the Ridley BI score, those graded 0 to 2 were categorised as low BI and those 3 to 6 as high BI.
Slides stained with MT, to identify collagen, were analysed as described below to obtain the lesion percentage value of fibrosis.

Immunohistochemistry
IHC was performed to detect the expression of calprotectin, CD3 and Pax5. Briefly, sections were mounted on SuperFrost® Plus-coated slides (Thermo Electron Ltd.), dewaxed, rehydrated, rinsed in distilled water and then washed in Tris-buffered saline (TBS) with Tween®20 (TBS-T) (28,360, Thermo Scientific) before antigen-retrieval. All antibodies were diluted in TBS-T and all washes between steps were in TBS-T. For calprotectin, which is expressed by granulocytes, monocytes and recently blood-derived macrophages (Rugtveit et al. 1996), mouse monoclonal anti-human macrophages antibody, clone MAC387 (MCA874G, Bio-Rad) (Kipar et al. 2003) was diluted to a final concentration of 1/800 (1.25 μg/mL) and incubated for 30 minutes at room temperature (RT) follow- ing epitope retrieval using proteinase K (S302080, Dako) for 20 minutes at RT. For CD3, a pan-T-lymphocyte marker, rabbit polyclonal anti-human CD3 antibody (A045201, Dako) (Kipar et al. 2003) was diluted 1/100 and incubated for 30 minutes at RT following epitope retrieval using 0.01 M sodium citrate buffer, pH 6.0 at 121°C for 40 minutes. Heat-induced epitoperetrieval (HIER) was performed using a SES Little Sister 3 autoclave (Eschmann). For Pax5 (a.k.a B-cell-specific activator protein [BSAP]) which is expressed by all stages of B-lymphocytes, although not plasma cells (Barberis et al. 1990), mouse monoclonal anti-human BSAP antibody, clone DAK-Pax5 (M730701) (Agostinelli et al. 2010) was diluted 1/50 (3.14 μg/ mL) and incubated for 30 minutes at RT following HIER as for CD3. After incubation with the primary antibody, non-specific endogenous peroxidase activity was blocked using REAL Peroxidase-Blocking Solution (S202386, Dako) for 10 minutes at RT. Sections were then incubated with goat anti-mouse/antirabbit secondary detection polymer (EnVision™ + Dual Link System-HRP, K406311, Dako) for 45 minutes at RT, followed by visualisation of positive staining with 3,3′-diaminobenzidine tetrahydrochloride (DAB) for 10 minutes at RT. Slides were counterstained with haematoxylin and Scott's Tap Water for 10 seconds each, then dehydrated, cleared in xylene, mounted and a cover-slip was added. Positive controls were feline tissues known to express the antigen of interest (calprotectin: an NTM-infected cutaneous lesion; CD3 and Pax5: a normal lymph node). An isotype control was used to assess non-specific staining for monoclonal antibodies, using an isotype and concentration-matched mouse anti-chicken Bu-1a/b, clone AV20 (MCA5764, Bio-Rad). Negative controls were run with omission of the primary antibody.

Image analysis
All H&E, MT and immunolabelled-slides were scanned with a NanoZoomer-XR scanner, using the NDP.scan Ver.3.2.12 software (Hamamatsu Photonics) and images viewed on NDP.view2 Ver.2.7.52 (Hamamatsu Photonics) to assess the cellular populations and granuloma type of the (pyo)granulomatous infiltrate. Quantification of necrosis and the total lesion area was determined using QuPath Ver.0.1.2 (Bankhead et al. 2017), with each region of interest (ROI) outlined manually using the drawing tools and the ROI file saved. To quantify the amount of fibrosis (in MT-stained sections), and the area of positive immunostaining (in IHC-stained sections), an adaptation of previously reported methods was used, using image deconvolution (Chen et al. 2017). Briefly, the ROI for each tissue was delineated in QuPath and sent to ImageJ (Schneider et al. 2012). The outside of the image was cleared, the resultant file saved and then opened in FIJI (Schindelin et al. 2012). For MT-stained slides, the inbuilt parameters for "Alcian Blue & H" in the colour deconvolution plug-in gave the best separation of the image file into different colour channels. The threshold for positive fibrosis staining was set as the average auto-threshold (default) for each image file; an ImageJ macro was written to quantify the area of positive staining and the total lesion area. For immunolabelled sections the "H DAB" values of the above plug-in were used and the threshold for positive staining set as the auto-threshold value for the positive control slide. ImageJ macros were then written to calculate the positive staining and total lesion areas. Isotype and negative control slides were analysed using this threshold to calculate corrections in the TB slides for any non-specific staining. All image files were visually assessed to determine accuracy of the automated analysis. Macros written for image analysis are available as Supplemental Data 1.

Statistical analysis
Data were analysed using RStudio (RStudio Team 2018). Univariate binomial logistic regression was performed to identify variables that may predict infection with either M. bovis or M. microti. Variables explored were age, gender, tissue, BI category, presence/absence of small granuloma clusters, necrosis, fibrosis, and expression of calprotectin, CD3 and Pax5. The proportion of necrosis, fibrosis and positive immunolabelling was converted to an ordinal scale variable. As age, gender and tissue were possible confounders, they were included as covariates for univariate analysis of BI category, granuloma type, necrosis, fibrosis, calprotectin, CD3 and Pax5. Multivariate logistic regression was subsequently performed on variables found to be significant on univariate analysis at P < 0.20, as well as the potential confounders age, gender and tissue. Variables were explored for collinearity and removed from the multivariate model where appropriate. Statistical significance for multivariate regression was set at P < 0.05.

Study population characteristics
Seventeen cats with M. bovis infection and 16 with M. microti infection with FFPE tissues available were identified from the database and assessed for eligibility. Six M. bovis-positive cats and two with M. microti were excluded as there was insufficient tissue in the FFPE block for further evaluation. Three cats diagnosed with M. microti on IGRA were identified as coming from highrisk areas for bovine TB and were subsequently excluded. Therefore, a total of 22 (M. bovis n=11, M. microti n=11) archived FFPE (muco)cutaneous and lymph node biopsies of tuberculous lesions were selected; they came from throughout England and Scotland. Further details are available in Table 1

Histopathological classification
Summarised results of histological and IHC analysis are presented in Table 2. Results of univariate analysis are reported in Table 3.
The lesions were granulomatous or pyogranulomatous, with macrophages as the predominant cell type in all cases, which were mostly epithelioid. Neutrophils were often present and were either dispersed throughout the lesion or accumulated in areas of necrosis. Lymphocytes were less frequently identified and were often located towards the periphery of lesions. Multinucleated giant cells were not identified in any sample.
A proportion of lesions of both aetiologies (10/22) were similar in appearance to typical tuberculous granulomas (Martin et al. 2016, Pagán & Ramakrishnan 2018, and these were termed "organised" granulomas ( Fig 1A, B). "Organised" lesions had singular or multifocal zones of central caseous necrosis, surrounded by a layer of macrophages and epithelioid macrophages; neutrophils were also a common feature within this cell layer. Lymphocytes and plasma cells, where present, were often located towards the periphery of the lesions, and/or less often distributed throughout the lesion (where they occasionally clustered). There was inter-individual variation in lesion encapsulation, and an outer layer of concentrically arranged spindle-shaped cells admixed with eosinophilic fibrillary material (fibrous capsule), which ranged from a thick capsule surrounding the entire granuloma to no encapsulation at all. Where no obvious capsule was present, the inflammatory infiltrate extended into surrounding tissue.
The remaining lesions (12/22) had minimal to absent central necrosis. Clusters of epithelioid and non-epithelioid macrophages were present, with interspersed neutrophils and variable numbers of lymphocytes. These lesions were termed "atypical" granulomas (Fig 1C, D). Importantly, "atypical" lesions presented with well defined, mostly round, small epithelioid macrophage clusters (100 μm-1000 μm), often divided by thin fibrous septa into smaller clusters, with very rare and mild central necrosis. The presence of a well-defined outer fibrous capsule was variable, as for "organised" lesions.
Both granuloma types effaced and expanded pre-existing tissue and presented with frequent areas of collagen degeneration with increased intercellular spacing (oedema), with or without inflammation, as well as variable demarcation.
Eight of 11 M. bovis-associated lesions were classified as "organised," compared with two of 11 M. microti-associated lesions; the remaining three and nine lesions were deemed "atypical" for each aetiology, respectively. There was a statistically significant difference between aetiology and granuloma type in univariate logistic regression [P=0.024, OR=0.08 (0.01 to 0.73)], with "organised" lesions being more likely to arise from infection with M. bovis.

Necrosis and fibrosis
Necrosis, as defined previously, was identified in 11 of the 22 cases examined, with larger necrotic areas in lesions with "organised" granulomas (these lesions had central necrosis, as described above). Most "atypical" tuberculous granulomas did not have any central necrosis, but clusters of necrotic neutrophils were identified in case 20 (Fig 1E). The median percentage area of necrosis    AFB Acid-fast bacilli, hpf High power field, BI Bacterial index, H High, L Low Granuloma type was classified as either "organised" or "atypical." "Organised" granulomas consisted of a central necrotic core surrounded bey a cellular layer composed of epithelioid macrophages, macrophages and neutrophils. "Atypical" granulomas lacked a central necrotic core and consisted of tight aggregates of epithelioid macrophages, macrophages and neutrophils. In both granuloma types, lymphocytes were predominantly located at the periphery of the granuloma, and there was variable deposition of collagen fibres forming an external capsule in "organised" granulomas, or thin fibrous septa in "atypical" granulomas. "% Necrosis" and "% Fibrosis" is the area of each histological feature relative to the total area of the lesion, as determined on Haematoxylin & Eosin-, and Masson's Trichrome-stained slides. "% Calprotectin," "% CD3" and "% Pax5" is the area of positive staining for each marker relative to the total area of the lesion. Positive staining for calprotectin indicates the presence of granulocytes and recently blood-derived monocytes, CD3 for T-lymphocytes and Pax5 for B-lymphocytes In terms of fibrosis, there appeared to be qualitative differences in the distribution of collagen fibres within feline TB lesions; M. bovis lesions showed thicker bands of collagen fibres surrounding larger "organised" granulomas, whereas M. microti lesions displayed thinner fibrous septa dividing smaller "atypical" granulomas ( Fig 1F). However, the total proportion of the lesion occupied by collagen fibres did not differ between aetiologies in univariate analysis (P > 0.213).
Acid-fast bacilli counts and bacterial index category AFB were identified in the ZN-stained section on examination of 15 hpf in 91% (20/22) of the samples. For the remaining two samples, the entire slide was examined to look for AFB; a single ZN-positive organism morphologically consistent with mycobacteria was identified in case 16, whereas no AFB were identified in case 13. Fifty-five percent of specimens scored as low BI (12/22); there was no statistical difference between the number of M. bovis and M. microti lesions scoring as low BI (P=0.088, OR=0.14 [0.01 to 1.35]). These data underline the variability in the number of organisms seen in feline tuberculosis lesions (M. bovis median: 8.87, range: 0.07 to 63.40; M. microti median: 0.27, range: 0.00 to 172.93). Case 12 scored BI 5, displaying vast numbers of AFB (Fig 2), often with an S-shaped morphology, and there were x1000 microscope fields from five other cats (cases 1, 2, 6, 9 and 21) with over 100 AFB (data not shown). ZN-positive organisms were found both intra-and extracellularly. Extracellular AFB were often located within areas of necrosis for both pathogens.

Calprotectin immunohistochemistry
Calprotectin-positive cells, i.e. granulocytes, monocytes and recently blood-derived macrophages, with strong granular to diffuse cytoplasmic staining were present in all samples; for each case calprotectin had more percentage-lesion positive immunolabelling than CD3 and Pax5. In lesions with large areas of necrosis, calprotectin-positive cells were common at the cell-necrosis interface, throughout the necrosis and within the regions of (pyo) granulomatous inflammation (Fig 3A, B). Cells morphologically consistent with macrophages and neutrophils were positive for calprotectin, while epithelioid macrophages were negative. Necrotic regions showed variable amounts of positive staining which was not associated with intact cells or tissue; where appropriate, these areas were removed from image analysis when quantifying the total area of cell-specific calprotectin-positive immunolabelling. Differences in calprotectin expression did not infer infection between the pathogens in univariate analysis (P > 0.294).

CD3 immunohistochemistry
Cells with moderate to strong membranous staining for CD3 (i.e. T-lymphocytes) were the second most common cell population identified. They were located peripherally and scattered throughout all lesions (Fig 3C). The percentage-lesion positive staining for CD3 was not statistically significant different between the pathogens in univariate analysis (P > 0.402).

Pax5 immunohistochemistry
Strong, diffuse nuclear staining for Pax5 was seen in cells towards the periphery of lesions. These B-cells were sometimes observed in clusters (Fig 3D). The total area of positive staining for Pax5 did not exceed 1% in any section, and there was comparatively greater staining of CD3 than Pax5 for each case (

DISCUSSION
This study described and compared feline tuberculous lesions, using histology and IHC to identify potential features that could infer infection with either M. bovis or M. microti histologically at an early stage of the diagnostic investigation. Among the companion animal population, the domestic cat appears to be much more susceptible to mycobacterial infections than other species, such as dogs (Broughan et al. 2013), and when outbreaks of disease do occur they can be devastating . The role of companion animal species in the epidemiology of bovine TB in the UK is controversial; it has been shown that cats can become infected with the same M. bovis spoligotypes and genotypes as cattle (Monies et al. 2000, Gunn-Moore et al. 2011a, and endemic M. microti infection may provide a degree of protection against M. bovis infection (Smith et al. 2009). Others argue that cats most likely represent spill-over hosts for mycobacterial infections (Broughan et al. 2013). However, the high prevalence of TB in cats, particularly M. bovis infection (Gunn-Moore et al. 2011a), poses a real zoonotic risk to owners and the veterinary staff who handle and care for these animals (O'Connor et al. 2019). Both M. bovis and M. microti are capable of infecting cats and causing clinical disease. Differentiating between these two pathogens can be complicated, although the geographical location can suggest which is most likely (Gunn-Moore et al. 2011a). The reference standard diagnostic test is specialist mycobacterial culture, however, its sensitivity is only approximately 50%, and while more rapid diagnostic tests are available, they are non-validated and also have their limitations (Gunn-Moore 2014). The need for more commercially available and financially viable diagnostics for mycobacterial disease in cats is therefore imperative.
The median age of cats infected with M. bovis in the current study was higher than previously reported, whereas the age of M. microti-infected cats was similar (Gunn-Moore et al. 2011a). Neutered male cats with outdoor access were the largest demographic in the current study. Additionally, 19 of 22 cats were domestic short-or long-haired, consistent with previous findings (Gunn-Moore et al. 2011a). The single cat not reported to have outdoor access had been fed a commercially available raw food diet, that was previously implicated as the likely cause of an outbreak of M. bovis in predominantly indoor, pure-breed cats (O'Halloran et al. 2019).
The histopathological features of feline M. bovis and M. microti lesions has not been directly compared previously, and in this study, M. bovis lesions were more frequently classified as "organised" granulomas, with structural similarities to typical tuberculous granulomas observed in other species, i.e. a central necrotic core with a macrophage-dominant cellular layer and an outer rim of lymphocytes, with or without an external fibrous capsule (Pagán & Ramakrishnan 2018). In contrast with these descriptions, feline M. bovis-associated lesions often had abundant neutrophils present within zones of granulomatous inflammation, lacked MNGCs and did not show evidence of mineralisation. Conversely, lesions from cats infected with M. microti were more often composed of tightly packed, smaller macrophage clusters with or without neutrophils. These lacked a necrotic core in most cases, and were delimited by intermittent, thin capsules. To differentiate these from the previous presentation they were termed "atypical" granulomas.
These histopathological differences in the arrangement, organisation and structure of granulomas may be due to the host immune response to infection. Studies into the serum cytokine and chemokine profile of cats infected with mycobacteria show increased concentrations of tumour necrosis factor alpha (TNFα) in cats infected with M. bovis compared to those infected with M. microti (O'Halloran et al. 2018). TNFα is a potent pro-inflammatory cytokine, secreted primarily by monocytes, macrophages and dendritic cells (Van Crevel et al. 2002), and it plays an important role in the host immune response to mycobacterial infections through activating macrophages to promote phagocytosis and subsequent killing of mycobacteria, inducing expression of chemokines by macrophages resulting in the recruitment of CD4 + T-lymphocytes, driving necrosis of infected macrophages, and the formation and maintenance of granulomas (Laster et al. 1988, Flynn et al. 1995, Roach et al. 2002, O'Garra et al. 2013, Dorhoi & Kaufmann 2014. It also has been shown to stimulate fibroblasts (Battegay et al. 1995), which may contribute to the deposition of collagen fibres to encapsulate granulomas. However, excessive TNFα may result in disruption of normal granuloma formation, with overactivation of macrophages resulting in increased necrosis with release of intracellular mycobacteria into the necrotic milieu (Dorhoi & Kaufmann 2014). Conversely, cats with M. microti showed increased levels of platelet-derived growth factor-BB (PDGF-BB), which is a potent mitogen of  (Agren et al. 1999, Van Der Kroef et al. 2020). It has also been shown that PDGF-BB enhances TNFα-induced chemokine secretion by fibroblasts and production of collagen (Van Der Kroef et al. 2020). The increased levels of TNFα in cats with M. bovis-infection may contribute to more necrosis within lesions, while increased PDGF-BB in M. microtiinfected cats may result in differences in the pattern of collagen deposition, resulting in smaller granulomas divided by fibrous septa. Differences in the chronicity of the lesion may also play a role, with increased deposition of collagen as lesions progress and become more chronic as reported for cattle (Wangoo et al. 2005). However, due to the clinical nature of the submitted samples it is unknown how long the lesions had been present or when infection occurred. It is therefore not possible to classify samples according to age or know how long the lesions had to develop the structural organisation of the (pyo)granulomatous infiltrate observed. It is possible that the variability in the presentation  British Small Animal Veterinary Association stems from the bacteria themselves, and one can hypothesise that these histological differences could be mediated by key virulence factors. It has been shown that ESAT-6 is a leucocidin resulting in neutrophil necrosis (Francis et al. 2014); this virulence factor may contribute to the increased amount of necrosis in M. bovisassociated lesions. Furthermore, granuloma type and macrophage aggregation is disrupted following infection with RD1 mutant mycobacteria in an experimental setting (Volkman et al. 2004, Davis & Ramakrishnan 2009); the lesions described in this study from cats with M. microti infection, which can be thought of as a natural RD1 deletion mutant (Frota et al. 2004), would appear to support this finding that RD1 proteins play a role in the structure and organisation of mycobacterial granulomas.
AFB were present in 91% of the lesions examined in the current study, in contrast with previous studies which reported that ZN-positive organisms morphologically consistent with mycobacteria were identified in only approximately 30% (37/134) of feline mycobacterial histopathological samples (Gunn-Moore et al. 2013). As has been previously reported, M. microti AFB often presented with an S-shaped morphology (Van Soolingen et al. 1998). It is of note that identification of AFB can be analyst dependent (Gunn-Moore et al. 2011b), and mycobacteria can alter their cell wall composition, resulting in a loss of "acidfastness" on ZN staining (Seiler et al. 2003). Application of the BI scoring system when reporting on ZN-staining of diagnostic samples is recommended for future studies, as it enables more consistent reporting compared to subjective terms to describe the number of ZN-positive organisms present. For example, this is likely to provide a robust reference in the comparative assessment of diagnostic tests, e.g. the likelihood of obtaining a positive PCR result on FFPE tissues with differing BI scores. Of interest, 45% (10/22) of samples had a high BI in this study, and one M. microti sample even scored as grade 5 (100.01 to 1000 AFB/ hpf ). This finding is directly relevant for diagnostics, as feline TB is often thought of as a paucibacillary infection, while multibacillary lesions typically indicate infection with non-tuberculous, and therefore non-zoonotic, mycobacterial species (Davies et al. 2006). The data presented here show that feline MTBC infections can present with abundant AFB, so discriminating between TB and NTM infections based on the number of AFB and BI scoring is not possible.
This study did not identify differences between the area of calprotectin-positivity in M. bovis and M. microti lesions. Calprotectin is expressed by circulating monocytes and those freshly recruited to developing areas of (pyo)granulomatous inflammation, plus granulocytes and some subsets of activated macrophages (Zwadlo et al. 1986). As monocytes differentiate into macrophages there is a loss of calprotectin expression, whereas expression of CD163 is upregulated (Zwadlo et al. 1985). The high proportion of lesions expressing calprotectin suggests active development of the inflammatory infiltrate, with ongoing recruitment of monocytes and neutrophils (Rugtveit et al. 1996). Many macrophages, notably those with an epithelioid phenotype, did not express this molecule, suggesting differentiation and subsequent loss of calprotectin expression; this is consistent with what has been previously reported in cases of feline mycobacteriosis (Kipar et al. 2003). The proportion of calprotectin-positive cells has been shown to decrease in more chronic tuberculous granulomas in other species (García-Jiménez et al. 2012, García-Jiménez et al. 2013, therefore the tissues in this study with lower levels of calprotectin expression could reflect the presence of more chronic lesions. While identification of calprotectin-positive cells can give an insight to the active recruitment of monocytes and granulocytes to the granuloma, it does not fully characterise the macrophage cell populations present, which is a limitation of this study. Alternative pan-macrophage markers, such as ionised calcium-binding adapter molecule 1 (Iba1), may be more specific for detection of macrophage populations within tuberculous lesions (Pierezan et al. 2014), while staining for CD163 could identify anti-inflammatory M2 macrophage populations (McBride et al. 2017). Additionally, markers such as Ki-67 could be used to identify local macrophage proliferation (McBride et al. 2017), providing further description of these lesions above what can be achieved by staining for calprotectin alone.
No differences were observed between M. bovis and M. microti lesions regarding positivity for T-or B-lymphocyte markers. Both T-and B-lymphocytes were located peripherally in tuberculous lesions in the current study and were a minor component of the inflammatory infiltrate. For each cat, the area of positive staining for T-lymphocytes exceeded that for B-lymphocytes, suggesting T-lymphocytes may play a greater role to the composition of the granuloma and to the pathogenesis of feline MTBC infections. In other species, T-lymphocytes expressing CD3 appear to be present within tuberculous lesions at consistent levels across different stages of granuloma development (García-Jiménez et al. 2013, Salguero et al. 2017. The role of T-lymphocytes in the pathogenesis of tuberculous infections, characteristic of a type IV hypersensitivity response, has been well studied and reviewed (Schluger & Rom 1998, O'Garra et al. 2013; CD4 + lymphocytes, and in particular T-helper 1 cells, play an important role in developing protective immunity against mycobacteria as well as stimulating macrophage activity via secretion of IFNγ. CD4 + -deficient individuals have been shown to be more susceptible to mycobacterial infections (Geldmacher et al. 2008), and idiopathic CD4 + lymphocytopenia has been documented in a cat with disseminated M. xenopi infection (Meeks et al. 2008). The role of other T-cell subsets, including CD8+ and γδ TCR+ T-cells, are less well elucidated (Schluger & Rom 1998), although it has been suggested that γδ T-cells may play an important role in the pathogenesis of bovine mycobacterial infections (Guzman et al. 2012). The proportion of B-lymphocytes expressing CD20 or CD79a, appears to increase in some species (García-Jiménez et al. 2012, Vallejo et al. 2018, possibly indicating a shift towards humoral immunity as the disease progresses. The role of B-lymphocytes in the pathogenesis of TB has been often overlooked, although there is gathering evidence that B-lymphocytes may interact with and regulate the response of T-lymphocytes in chronic infections, in addition to driving differentiation of alternatively-activated M2 macrophages, which play a modulatory role in mycobacterial infections (Martinez & Gordon 2014). Plasma cells, which were rarely identified on H&E-stained sections, do not express Pax5 (Barberis et al. 1990 identified by IHC in this study; their contribution to tuberculous lesions in cats could be explored further through use of markers such as immunoglobulin λ light-chain (Mellor et al. 2008). However, the results of the current study appear consistent with findings observed in tuberculous lesions in other species (Kipar et al. 2003, García-Jiménez et al. 2013. Tuberculosis affects a wide range of organs and body systems, but histological and immunological studies have normally been performed on lung or lymph node granulomas (Turner et al. 2003, Wangoo et al. 2005, García-Jiménez et al. 2012, García-Jiménez et al. 2013; since feline tuberculous lesions are predominantly cutaneous, this may explain why previously established granuloma scoring systems did not correspond well to feline lesions. A review of cutaneous mycobacterial lesions in humans identified seven different histopathological patterns, including "organised" tuberculoid granulomas, but it was also noted that multiple patterns could be present within the same biopsy sample examined and that the histopathological presentation of disease could not classify infection with different mycobacterial species (Santa Cruz & Strayer 1982). Neutrophils also appeared to be more frequently identified in cutaneous mycobacterial lesions (Santa Cruz & Strayer 1982). Differences in host-pathogen interactions between host species have also been identified; human M. tuberculosis infection typically results in the formation of caseous granulomas, whereas non-necrotic lesions are more common in some strains of mice (Kramnik et al. 2000). Similarly, M. microti infection in its maintenance host, the field vole, often results in caseous necrosis with associated regions of dystrophic mineralisation, and MNGCs (Kipar et al. 2014), features which were mostly absent from feline lesions. Studies into the histological and immunological appearance of mycobacterial lesions affecting other organs, such as the eye, could provide further insights as to whether lesion development varies due to the tissue or organ infected, the infecting organisms present, or whether the feline immune response differs compared to that in other species. As this study used clinical cases of naturally occurring TB, managed by referring veterinarians, data such as the chronicity of lesions and the duration of formalin-fixation of the sample was unavailable, so it is not possible to exclude that some of the histological and immunological differences could be explained by this.
The results of the histological analyses warrant further investigation on a larger set of confirmed tuberculous samples as the small sample size may be insufficient to support a multivariate model. Furthermore, categorising the continuous variables that did not meet the assumptions for binomial logistic regression results in a loss of statistical power and the interpretation of any statistically significant findings is more complicated. Additionally, multicollinearity of some of the variables in this study was identified; where appropriate, collinear variables were excluded from models which were deemed to not result in a loss of biological explanation of the results. Despite the above limitations, in this study the granuloma type was the most statistically significant finding in suggesting whether a cat was infected with M. bovis or M. microti. This is relevant, as the findings of this study are useful in the diagnostic histopathology setting to infer the likelihood of infection with either M. bovis or M. microti before further testing, i.e. culture, PCR or IGRA.
In conclusion, this study characterised the histological and immunohistochemical appearance of tuberculous lesions in cats naturally infected with M. bovis or M. microti. With ZN staining, M. microti AFB were frequently S-shaped, and numerous AFB were observed in individual cases for both M. bovis and M. microti infections. Therefore, abundant ZN-positive organisms are not a specific diagnostic feature of feline infection with nonzoonotic NTM species. The study identified differences in the histopathological features between these two mycobacterial species that may raise the index of suspicion for infection with either mycobacterial species at an earlier stage of the clinical investigation of these cases.