All canine serum samples were submitted by veterinary practitioners for B. canis serological testing to the Animal Health Diagnostic Center (AHDC) at Cornell University (Figure 1). The samples were tested by both the 2ME-RSAT and the AGID II assay. The 2ME-RSAT is performed with heat-killed whole-cell B. canis (M-) antigen, and the AGID II is performed with B. canis (M-) cytoplasmic antigen, as described below.

A total of 1,582 diagnostic serum sample submissions submitted between August 2020 and August 2022 were tested in parallel with the reference assay and were separated into two populations for the statistical analysis, as described below. To further evaluate sensitivity, 42 archived serum samples from cases with associated B. canis positive blood culture results were evaluated.

Data from serum samples submitted for B. canis serologic testing from January 2023 through December 2024 were retrospectively evaluated to understand the in-use assay performance.

Antigens for both the 2ME-RSAT and AGID II assays were produced by growing large quantities of the B. canis (M-) strain on Brucella agar under BSL3 conditions. The plated organism was harvested, washed with PBS and heat killed at 80 °C for at least 20 min. Once the product was confirmed to be non-viable, it was relocated to the laboratory for further processing. The antigen used for slide agglutination was washed with PBS, stained overnight with a Rose Bengal solution, resuspended in tris-maleate buffer, and diluted to a 6% concentration. The product was then tested in comparison to the current antigen lot and aliquoted for storage. The bacterial suspension designated for AGID II antigen testing was processed through a French Press. The lysate was centrifuged at 8,000 × g, and the supernatant was further clarified by ultracentrifugation at 63,000 × g. The resulting supernatant was tested and diluted as necessary to match the results of the current in-use cytoplasmic antigen.

All diagnostic samples were tested by both the 2ME-RSAT and AGID II tests and test interpretations were based on the evaluation of both results, along with pertinent testing/health history of the animal, if provided, following established standard operating procedures. Briefly, the slide agglutination test was performed by mixing equal amounts of patient serum and 0.2 M 2-ME on a clear slide and allowing that to stand for 30 s. The B. canis slide antigen was then mixed with the treated sample and the slide was rocked for 3 min before reading the agglutination reaction using an inverted microscope with 4× objective. The AGID II test was performed using in-house made AGID agar and a template to punch a seven-well pattern, including a center well for the cytoplasmic antigen and six surrounding wells for alternating positive control and diagnostic samples. The wells were labeled as indicated on the worksheet template and serum was pipetted into the corresponding wells. Positive control serum was added to the indicated outer wells and the cytoplasmic antigen into the center well. The plate was incubated overnight at 25–29 °C, and the reactions were read the following day using a fiber optic light to illuminate the precipitin lines.

Brucella canis antigen preparation, as described above for the 2ME-RSAT, was diluted 1:2 in Laemmli sample buffer (BioRad, Hercules, CA) with 1% 2-ME (Sigma-Aldrich Inc., St. Louis, MO), boiled for 3 min, resolved on a 12% SDS-PAGE gel, and transferred to a PVDF membrane for subsequent immunoblotting. Serum from dogs that were previously positive or negative by the B. canis reference assay, 2ME-RSAT and AGID II (AHDC, Cornell University, Ithaca, NY) were used to probe the immunoblot. Sera were diluted at 1:400, and serum antibody binding was detected with an HRP-conjugated rabbit anti-dog IgG(H + L) (Jackson Immunoresearch Laboratories, West Grove, PA; RRID: AB_2339344), diluted 1:20,000, resolved with ECL western blotting substrate (BioRad, Hercules, CA). A prominent band of apparent molecular weight of ~28 kDa was identified on the immunoblot (data not shown). This band was extracted from a Coomassie stained gel run in parallel and processed for mass spectrometry by the Cornell University Biotechnology Resource Center. Comparison to the protein database showed the predominant protein to be a homolog of the B. abortus BP26 protein.

Full-length BP26 was cloned from B. canis genomic DNA isolated from a B. canis (M-) strain maintained at the AHDC. The primers used for producing the BP26 antigen described here are listed in Table 1. Full-length BP26 (753 bp, GenBank, AQNA01000002) was cloned into a pQE-60 vector (Qiagen Inc., Germantown, MD) digested with NcoI and BamHI, in-frame with a C-terminal 6x Histidine tag. A more soluble version of the BP26 antigen, here on described as BP26, was produced by subcloning a truncated construct into the NcoI/BamHI digested pQE-60 vector to remove the 21 amino acid N-terminal transmembrane domain, thereby improving protein expression.

The BP26 protein was expressed in the SG13009 E. coli expression host (Qiagen Inc., Germantown, MD), grown in LB broth containing kanamycin and ampicillin. Protein expression was induced by the addition of 1 mM isopropyl β-D-1 thiogalactopyranoside (IPTG). The bacteria were suspended in urea lysis buffer (100 mM sodium phosphate, 10 mM Tris, 8 M urea, pH 8.0) and subjected to sonication. The lysate was then cleared by centrifugation, mixed 1:5 with 40 mM imidazole buffer (40 mM imidazole, 20 mM sodium phosphate, 0.5 M NaCl), and filter sterilized through a 0.22 μm Steritop™ filter (MilliporeSigma, Burlington, MA). The His-tagged protein was purified on a FPLC instrument using a HisTrapFF Ni-NTA column (both, GE Healthcare, Piscataway, NJ) and eluted with 500 mM imidazole buffer (500 mM imidazole, 20 mM sodium phosphate, 0.5 M NaCl). Eluted protein fractions were dialyzed against phosphate buffered saline, pH 7.4. Protein concentrations were determined by BCA assay (Pierce, Rockford, IL).

Additional antigen candidates were identified by searching the protein databank for immunoreactive B. canis proteins. One of the candidates investigated was Omp31 (ERU01676.1) (Cassataro et al., 2004). The full-length protein was cloned as described for BP26 (Table 1); however, the recombinant protein was toxic to E. coli in the production platform described above for BP26, and purity of the recombinant protein was difficult to establish (data not shown). This led us to investigate potential immunoreactive peptides from Omp31, starting with evaluation of potential outer-membrane loops.

A transmembrane topology prediction of Omp31 was performed. First, the signal sequence was removed, as defined by SignalP-4.1¹ prediction. Next, the sequence was evaluated with TMpred.² Based on analysis of the mean burial propensity, a region at the C-terminal domain of the protein, representing amino acids 49–77 of the Omp31 protein, was hypothesized to be an externally exposed loop that would have increased exposure for immune recognition. A peptide derived from this region (PO1, [NH2]GKFKHPFSSFDKEDNEQ[COOH]) was synthesized (Lifetein, Hillsborough, NJ), and conjugated to maleimide-activated bovine serum albumin (BSA), through a cysteine amino acid added to the C-terminus of the peptide, following manufacturer instructions (Pierce Biotechnology, Rockford, IL).

In preparation for the multiplex assay, BP26 was coupled to fluorescent bead 33, BSA-conjugated PO1 was coupled to fluorescent bead 34, and for the control for the three-antigen assay, cytoplasmic antigen (CytAg) produced for the AGID II assay was coupled to fluorescent bead 35 (Luminex Corp.).³ The coupling was performed in accordance with manufacturer recommendations using establish laboratory protocols, as previously described (Wagner et al., 2011a). Briefly, all steps were conducted at room temperature, beads were suspended by vortex and sonication, incubations were performed in the dark, and beads were pelleted by centrifugation at 4,700 × g for 4 min. Bead stock (1.0 × 10⁷ beads) were washed with sterile ultrapure water. The beads were activated by first suspending in 100 mM sodium phosphate pH 6.2. Next, 20 μL of 10% (w/v) Sulfo-N-hydroxysulfosuccinimide (Sulfo-NHS, Pierce Biotechnology Inc., Rockford, IL), followed by 20 μL of 10% (w/v) 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide hydrochloride (EDC, Pierce Biotechnology Inc., Rockford, IL) were added and incubated for 20 min. The beads were then washed twice with 50 mM 2-(N-morpholino) ethanesulfonic acid buffer pH 5.0 (MES, Sigma-Aldrich Inc., St. Louis, MO). The activated beads were mixed with the MES buffer and 200 μg of the BP26, BSA-conjugated-PO1 protein, or CytAg in a 1 mL volume and rotated for 3 h to complete the coupling. Next, the beads were incubated in PBN blocking buffer (PBN, phosphate buffered saline with 0.1% bovine serum albumin and 0.5% sodium azide) for 30 min, and finally the beads were washed three times in PBN blocking buffer with 0.02% (v/v) Tween 20. Beads were then counted and stored in the dark at 2–8°C. Each coupling batch was tested in single-plex and compared to the previous coupling batch using previously tested negative, low positive, and high positive sera to ensure comparable results across batches.

Beads 33 and 34 coupled with BP26 and PO1, respectively, for the two-antigen assay, or with the control CytAg coupled bead 35 for the three-antigen assay, were sonicated, mixed, and diluted in the PBN blocking buffer, to a final concentration of 10⁵ beads/mL. Canine serum samples were diluted 1:600 in PBN. Previously tested negative, low positive, and high positive canine sera were set on each assay plate as negative and positive controls. Millipore Multiscreen HTS plates (Millipore, Danvers, MA) were wetted for 10 min with phosphate buffered saline containing 0.05% Tween 20 (PBST). An ELx50 plate washer (Biotek Instruments Inc., Winooski, VT) was used for adding and aspirating PBST for this incubation and subsequent wash steps. After aspirating PBST, 50 μL of each diluted serum or control sample was added to the appropriate wells of the plate. Next, 50 μL of the bead solution was added to each well, and the plate was incubated for 30 min, with shaking, at room temperature. After washing the serum-incubated beads three times, 50 μL of biotinylated rabbit anti-dog IgG(H + L) (Jackson Immunoresearch Laboratories, West Grove, PA; RRID: AB_2339344), diluted 1:3,500 in PBN, was added to each well and incubated for 30 min as above. Following a wash step, 50 μL of streptavidin-phycoerythrin (Invitrogen, Carlsbad, CA), diluted 1:100 in PBN, was added to each well. Plates were incubated for 30 min as above and then washed. Beads were resuspended in 100 μL of PBN and incubated for 15 min, as above. The resuspended beads were then analyzed in a Luminex 200 instrument (Luminex Corp.) (see text footnote 3) using BioPlex software (Bio-Rad Laboratories Inc., Herculese, CA). The data were reported as median fluorescent intensities (MFI).

The performance of the two-antigen, BP26 and PO1, Canine Brucella Multiplex (CBM) assay was evaluated using a Bayesian latent class model (BLCM). This model estimates the sensitivity and specificity of both the CBM assay and the reference assay by utilizing two populations expected to have different prevalence (Cheung et al., 2021). For the model to be identifiable, the number of degrees of freedom must be greater than or equal to the number of unknown parameters in the model. In this case, including two populations with different prevalence, with the assumption that the sensitivity and specificity of the assays were constant across populations, allowed for the model to be identifiable. The two populations (Table 2) were: (1) routine North American diagnostic submissions that are typically received from veterinarians because of clinical signs consistent with B. canis infection or for screening prior to breeding (n = 1,192), and (2) North American submissions requesting documentation of freedom from infection for export of the dog or the dog’s semen from the US to a foreign country (n = 390). Dogs in the latter population are typically free of clinical signs related to B. canis. Prior information about the diagnostic specificity of the reference assay and the population prevalence were modeled using unimodal beta distribution based on published data and laboratory submission history. The specificity of the 2ME-RSAT and AGID II combination test was estimated to be >99%, with a minimum plausible value of 94% (Keid et al., 2009). Based on historic diagnostic submissions to the AHDC for B. canis serology, 2015–2019, the prevalence of positive results in non-export North American diagnostic samples was estimated to be 10%, with a maximum plausible value of 20%, and in export sample submissions, prevalence was estimated to be 0.01%, with a maximum plausible value of 1%. These details were used to derive parameters for prior distributions using the epi.betabuster function in the epiR package (Nunes et al., 2022) in R, version 4.2.1 (R Project, 2022). A uniform prior distribution, beta(1,1), was used for the sensitivity and specificity of the assay under investigation and for the sensitivity of the reference assay. The Bayesian model was run in R using JAGS (2022) through the R2jags package (Su and Yajima, 2021), and diagnostics were visualized using the mcmcplots and coda packages (Curtis et al., 2018; Plummer et al., 2020). To investigate the sensitivity of the model to the defined priors, the model was also evaluated with minimally informative priors, beta(1,1).

ROC curve analysis was performed by evaluating the BLCM model output at 15 different sets of cut-off values for PO1 and BP26 (Supplementary Table S1). Area under the ROC curve was calculated using the trapezoid rule.

For all statistical analyses, samples that were considered “inconclusive” (n = 43, all from the North American diagnostic submissions) on the reference assay were handled as “positive” or “negative,” depending on the specific 2ME-RSAT/AGID II reference assay results: inconclusive samples were classified as “positive” if the 2ME-RSAT results were “positive” and the AGID II results were “suspicious” (n = 16), else (2ME-RSAT “negative”/AGID II “suspicious,” or 2ME-RSAT “positive”/AGID II “negative”) they were considered “negative” (n = 27).

As existing laboratory methods for diagnosing B. canis infection pose significant practical limitations, we sought to develop and explore a novel assay using two immunogenic recombinant antigens, BP26 and PO1, derived from this zoonotic pathogen. Each antigen was evaluated in a singleplex assay and in the combined multiplex assay format with negative, low positive, and high positive sera, and no significant difference in values was observed when the beads were multiplexed. Cut-off values for each antigen were selected to optimize sensitivity and specificity (Supplementary Figure S2); samples with BP26 < 2400MFI and PO1 < 1000MFI were considered negative, and samples with values above these cut-off values were considered non-negative.

As the “gold standard” for diagnosing B. canis infection is bacterial culture, we evaluated archived serum samples from 42 dogs with associated B. canis blood culture confirmation. We found that 39 of these samples were positive on the 2ME-RSAT/AGID II reference assay, while 9 were “inconclusive,” with positive reaction on the 2ME-RSAT only, and suspect (6/9) or negative (3/9) reaction on the AGID II. Thirty-nine of the 42 serum samples were non-negative on the two-antigen CBM assay (Figure 2; Supplementary Figure S3), with 2 dogs producing only BP26 antibody values, 9 dogs producing only PO1 antibody values, and 28 dogs producing antibody values above the cut-off values for both antigens. These results pointed to individual differences of the B. canis antibody response in these confirmed infected dogs. Most dogs (n = 22) had higher PO1 than BP26 antibody values, while 17 dogs showed the opposite trend, and three dogs did not have detectable antibodies against either antigen. Together, these results demonstrated the value of including both antigens for the CBM assay to identify B. canis specific antibodies in dogs with confirmed infection.

Given that the serological reference assay is not a true “gold standard” for diagnosis of B. canis infection, we employed a BLCM analysis to estimate the accuracy of the two-antigen CBM assay. This method allows for the estimation of the sensitivity and specificity of both the test under evaluation as well as the reference assay and estimates the seroprevalence of each population included in the model.

We used prior information about the specificity of the reference assay and the expected populations’ prevalence to help inform the model (as described in the Materials and Methods). Distributions for the prior information are presented in Supplementary Figure S1.

A ROC curve was produced by evaluating the BLCM model at 15 different sets of cut-off values (Supplementary Table S1). The area under the ROC curve (AUC) was 0.913 (95% CI, 0.883–0.917), indicating excellent to outstanding agreement (Mandrekar, 2010) between the reference assay and the two-antigen CBM assay.

The sensitivity and specificity estimations for both the two-antigen CBM assay and the reference assay at the cut-off values of BP26 < 2,400, PO1 < 1,000 is presented in Table 3. The sensitivity of the two-antigen CBM assay was estimated to be 91.6% (95% CI, 85.2–98.0%), which was substantially greater than the estimated sensitivity of the reference assay, estimated to be 66.1% (95% CI, 55.2–81.7%) (Table 3). The specificity of the two-antigen CBM assay was estimated to be 94.9% (95% CI, 92.3–96.9%), slightly lower than the estimated specificity of the reference assay 99.8% (95% CI, 99.0–100.0%). Estimated seroprevalence was determined to be 0.1% (95% CI, 0.0–0.6%) for the export sample submissions, and 16.1% (12.6–19.3%) for the routine diagnostic submissions.

To test the robustness of the BLCM analysis, the model with minimally informative priors (beta(1,1)) for all variables was evaluated. The minimally informed model produced comparable results, with only minor increases in the estimated population prevalence, and associated small decreases in estimated sensitivity, with increased estimates for specificity, indicating that the model results were not strongly influenced by the prior information.

These results, taken together, indicate that the novel two-antigen CBM assay composed of recombinant B. canis BP26 and PO1 antigens, is a robust and reliable tool for the serologic diagnosis of B. canis infection in dogs.

The relative importance of test sensitivity and specificity is defined by a variety of factors, and in the case of diagnosis of a zoonotic pathogen in a companion animal, maximizing sensitivity to reduce false negative test results may be desired; to increase assay sensitivity, we evaluated the addition of a control bead bound to a crude antigen extract, CytAg, which is also used in the AGID II assay. Prior to inclusion in the assay, a set of 150 archived diagnostic serum samples with measurable antibody to PO1 and/or BP26 were evaluated in the CBM assay, and minimal change (%CV < 7) in the quantitative values for PO1 and BP26 was observed. A cut-off value was chosen where samples with CytAg < 1100MFI were considered negative, and samples with values above this cut-off value were considered non-negative; a sample with non-negative results on any of the three beads in the assay was considered non-negative. The inclusion of this control bead resulted in detection of two additional true positive samples, for an overall diagnostic sensitivity of the CBM assay of 97.6% (41/42) (Figure 2; Supplementary Figure S3).

While the addition of the CytAg control bead increases the diagnostic sensitivity of the assay, it also impacts the assay specificity. From January 2023 through December 2024, approximately 10,000 diagnostic samples were tested on the CBM assay, which included the two recombinant antigen beads and the control bead. In total, 1,379 samples produced non-negative results. All samples that tested non-negative were subsequently tested for confirmation on the reference assay. Of the 1,379 samples that tested non-negative, 484 were non-negative only on the CytAg. Of those, only 4.3% (21/484) were confirmed to be positive on the reference assay, and 4.3% (21/484) produced an inconclusive result on the reference assay. Overall, the addition of the CytAg control bead increases assay sensitivity but also decreases assay specificity.