Vaccine-mediated protection against Campylobacter-associated enteric disease

We describe a novel vaccine that provides protection against Campylobacter-associated diarrhea in a natural exposure model.


INTRODUCTION
Campylobacter species cause millions of cases of bacterial gastroenteritis per year and represent one of the most important classes of human pathogens contributing to diarrheal disease throughout the world (1). Human disease occurs via a fecal-oral route of transmission and affects both high-and low-income countries. The medical importance of this pathogen was highlighted in a prospective study involving >22,000 children from seven developing countries in which Campylobacter rated among the top three causes of moderate to severe diarrhea in 24-to 59-month-old children (1). The economic impact of Campylobacter-associated disease can be substantial, with annual costs in the United States alone estimated at up to $5.6 billion (2). Several Campylobacter species have been identified that cause enteric disease, although Campylobacter jejuni and Campylobacter coli are considered the most important pathogens within this genus and account for the majority of all Campylobacter-associated human diseases. Although C. jejuni is the most commonly identified cause of Campylobacter-associated disease, outbreaks in Peru and Egypt have indicated that C. coli is responsible for up to 30 to 37% of Campylobacterassociated diarrhea (3,4). Additional studies have demonstrated a strong correlation between Campylobacter burden and childhood growth faltering (3), indicating that Campylobacter may be a key factor driving poor childhood growth and development outcomes in low resource settings. Together, these studies underscore why this pathogen is recognized as one of the most important global threats in need of targeted vaccine development.
Despite a clear medical need, there is currently no Campylobacter vaccine available for use in humans. One of the primary roadblocks has been the lack of robust and reproducible experimental models (5). While human challenge models represent the most direct approach (6), this is not readily amenable to the experimental manipulation available with animal models. Here, we explored an environmental exposure model of Campylobacter infection in rhesus macaques (RMs) to test the efficacy of potential vaccine candidates. Outdoorhoused RM at the Oregon National Primate Research Center (ONPRC) experience a spectrum of acute and recurrent C. jejuni-and C. coliassociated diarrheal disease that mimics several aspects of Campylobacterassociated disease found among human populations that suffer from conditions of poor sanitation and hygiene, including (i) dysbiotic microbiomes, (ii) symptomatic and asymptomatic carriage, and (iii) higher disease incidence rates among infants with a concomitant increase in disease severity compared with adults (7,8).
We have previously reported on the use of a hydrogen peroxide (H 2 O 2 )-based vaccine platform that is protective against a number of acute and chronic viral infections (9). Here, we have adapted this approach to develop novel inactivated C. coli and C. jejuni vaccines to protect against an enteric bacterial pathogen. Certain strains of Campylobacter express lipooligosaccharide (LOS) that appear to be ganglioside mimics (10) and to mitigate potential safety issues associated with molecular mimicry, we used vaccine strains of Campylobacter [C. coli (NTICC13) and C. jejuni (CG8421) (6)] that lack neuA, neuB, neuC, and cst (sialyltransferase) and are genetically incapable of producing ganglioside mimics (6). H 2 O 2 -inactivated Campylobacter vaccination induced an immunodominant antibody response to bacterial flagellin and provided protective immunity against clinical diarrheal disease in a robust nonhuman primate (NHP) model of naturally occurring C. coli infection despite demonstrating little to no homology within the LOS or capsular polysaccharide (CPS) loci compared to circulating C. coli strains. In contrast to LOS and CPS, the flagellin genes were highly conserved between the C. coli vaccine strain and the circulating strains of C. coli, suggesting that this may be a potential target antigen capable of providing antibacterial immunity across disparate Campylobacter serotypes. These studies not only demonstrate the feasibility of using this natural challenge model but also provide an important proof-of-concept to support the continued development of novel antibacterial vaccines to prevent Campylobacterassociated enteric disease in humans.

Diarrheal burden among outdoor-housed RMs
Most of the ~5000 NHP at the ONPRC are RM, with approximately 75% of the animals housed outdoors in either 1-acre corral breeding groups or smaller shelter breeding groups (7). Diarrheal illness is a substantial concern for captive RM (7,8), and prior studies have shown that animal groups housed in smaller group shelters are particularly prone to high rates of enteric disease (7). Approximately 80% of RM infants are colonized with Campylobacter spp. by 1 month of age, and 69 to 97% of juveniles and adults in the outdoor small breeding groups remain clinically asymptomatic carriers of C. coli and C. jejuni with preliminary unpublished histological evidence indicative of environmental enteropathy. While most animals appear healthy, approximately one quarter of infants will develop acute diarrhea, and half of these animals will progress to chronic/relapsing diarrhea and potentially lethal enteric disease requiring humane euthanasia (7). Similar to humans, enteric disease associated with C. coli or C. jejuni is comparable in these animals, and RM infants and juveniles have higher rates of diarrhea compared to adults (7). RM infants also have more severe disease and weight loss compared to the older age groups with chronic diarrhea cases, resulting in a mortality rate that is nearly double that observed among adults (7). To expand on these studies, we examined the annual rates of diarrheal incidence in shelter-housed RM from 2010 to 2016 (Fig. 1A). Cumulative annual diarrhea incidence that required hospitalization averaged 16.2 ± 3.4% (mean ± SD). Hospitalization refers to the animals that were removed from the group and housed indoors in a veterinary clinic for 1 or more days to receive appropriate diagnostics and treatment for diarrheal illness and dehydration, often including antibiotics and intravenous fluid therapy. When animals were hospitalized due to diarrhea, fecal samples were collected and tested for the presence of pathogenic bacteria, with a focus on Campylobacter and Shigella spp. (Fig. 1B). C. coli was the most common pathogen associated with diarrhea with an incidence of 59 ± 11% of diarrheal cases followed by Shigella (12 ± 4.0%) and C. jejuni (5.9 ± 2.0%). Similar to humans, chronic diarrheal disease associated with Campylobacter in RM resulted in characteristic histopathologic findings in the large intestine including mucosal hyperplasia, separation of glands by large numbers of lymphocytes and plasma cells, neutrophilic infiltration, decrease in goblet cell numbers, and superficial enterocyte erosion and atrophy ( fig. S1). In total, our analysis showed a consistently high burden of C. coli and C. jejuni among outdoor-housed RM, providing the opportunity to perform Campylobacter vaccine field studies under natural fecal-oral exposure conditions.

Campylobacter vaccine development
Many strains of Campylobacter will coexist under hyperendemic conditions since natural infection often does not induce sterilizing immunity in humans or macaques. Successive rounds of reinfection of NHP by multiple strains of C. coli and C. jejuni have also been observed (11), with one study of pig-tailed macaques (Macaca nemestrina) finding 10 serotypes of Campylobacter among 69 isolates with a mean of 8.3 ± 2 different Campylobacter strains identified per infant (11). Isolation of multiple Campylobacter species and serotypes has also been described in humans (12). On the basis of these studies, it was likely that there were multiple strains of C. coli and C. jejuni cocirculating among the outdoor-housed primates at ONPRC. We performed whole-genome sequencing (WGS) of banked C. coli isolates from 2015, 2016, and 2018 and compared them to the vaccine strain isolated in 2013 ( fig. S2). We identified three distinct C. coli strains on the basis of their LOS loci (Fig. 2) and seven distinct strains on the basis of their CPS loci (Fig. 3). Given the burden of diarrhea observed in shelter-housed animals (Fig. 1A) and the substantial negative impact of diarrheal disease on animal health (7,8), we initiated a vaccine development program with an initial focus on C. coli (i.e., H 2 O 2 -Campy C ) since it represented the most common enteric pathogen encountered among hospitalized RM cases (Fig. 1B).
For these studies, we selected a fully sequenced C. coli clone that lacked the genes, neuA, neuB, neuC, and cst (sialyltransferase), but still showed typical smooth, convex white colonies on blood-agar plates and a Gram-negative stain spiral morphology by microscopy ( Fig. 4A, left). Initial inactivation studies using 3% H 2 O 2 resulted in substantial structural damage and cellular lysis of purified C. coli (Fig. 4A, middle). Bearing in mind that microaerophilic bacteria might be more susceptible to oxidative damage, we developed an advanced inactivation approach involving a Fenton-type reaction, using an electronic health record system. To determine incidence rates, only the first instance of clinical diarrhea for any given animal was counted in each calendar year. These unique diarrheal episodes were divided by the average of 1183 ± 44 animals (±SD) in outdoor sheltered group housing each year. (B) For each diarrheal episode (primary or repeat cases), bacterial cultures were tested for the indicated enteric pathogens, C. coli, C. jejuni, and Shigella spp.
in which copper [ cupric ion (Cu 2+ )] is used as a catalyst in an oxidation reaction performed in the presence of lower, less damaging concentrations of H 2 O 2 . Following preliminary small-scale development studies, an optimized approach that maintained the structural integrity of the bacteria was established (Fig. 4A, right). We performed inactivation kinetics by testing samples at regular intervals after incubation with H 2 O 2 + Cu 2+ and found rapid loss of bacterial viability with a half-life (T 1/2 ) = 14.5 min (Fig. 4B). Alum was selected as the vaccine adjuvant on the basis of its established clinical safety record and prior efficacy in conjunction with the H 2 O 2based inactivation platform (9). To examine general reactogenicity before performing NHP experiments, we conducted small-scale studies in BALB/c mice using a full NHP dose of vaccine that was equivalent to ~300-fold higher dose than that planned for the subsequent RM studies on a milligram per kilogram basis (Fig. 4C). Mice that received the H 2 O 2 -Campy C vaccine were comparable to mockvaccinated animals (alum alone) and demonstrated no measurable signs of acute weight loss following vaccination. In contrast, mice that received 10 g of E. coli-derived lipopolysaccharide (LPS) as a positive control demonstrated rapid, albeit transient, weight loss indicative of endotoxin-induced reactogenicity. We then conducted a safety study in RM, vaccinating two adult females with the H 2 O 2 -Campy C vaccine and two with alum-only mock vaccines. Health metrics, including body weight, attitude, appetite, urine output, stool quality/quantity, temperature, and vaccine site, were monitored daily for 14 days and no differences were noted between the vaccinated and unvaccinated pairs. Together, these studies indicated that the H 2 O 2 -Campy C vaccine had an acceptable safety profile for further preclinical assessment and we proceeded to immunize RM to determine vaccine efficacy (VE) in a large-scale field study.

LOS locus
Working cell bank Original isolate Fig. 2. LOS loci demonstrate extensive diversity among RM C. coli isolates. Genome sequencing was performed for 15 primary C. coli isolates, as well as the C. coli vaccine working cell bank. Sequences were queried for the presence of the LOS loci as defined by the waaC and waaF genes. Genes were compiled and categorized into groups based on sequence similarity (≥80% identity), resulting in 39 groups for the LOS. Isolates with partial sequence results are indicated by //. The LOS genomic regions for the C. jejuni CG8421 vaccine strain are also shown. The numbers above each arrow refer to the locus tags in the published sequence (GenBank: CP005388.1), while the numbers within the arrows indicate genes with high similarity (BLAST score > 200) to genes found among the C. coli isolates, with gene segments not meeting this threshold indicated with an "x." Sample isolation dates are provided under each animal ID number. Horizontal dashed lines indicate that there are three distinct strains of C. coli based on genetically similar LOS loci. Note that the strains of C. coli that circulated in 2015 and 2016 during the vaccine field trials were a mismatch to the C. coli and C. jejuni vaccine strains used in these studies. Genome sequence accession numbers are provided in the Data and materials availability section.

Immune responses in vaccinated RMs
We initiated NHP vaccination studies by immunizing animals from outdoor breeding groups with the H 2 O 2 -Campy C vaccine candidate, with 60 animals that were monitored for diarrheal disease for up to 1 year after primary vaccination. There were no significant differences in the demographics (e.g., size of shelter housing unit, sex, or age distribution) of the vaccinated shelters versus the unvaccinated control shelters included in these studies (table S1). No adverse events were associated with vaccination in this large NHP cohort.
To investigate vaccine-mediated immune responses in immunized animals, serum samples were collected just before primary vaccination and at 6 and 12 months after primary vaccination. Serum antibodies were tested against a number of C. coli antigens including a total whole-cell lysate and purified flagellin with group geometric mean titers (GMTs) and 95% confidence intervals (CIs) determined at each time point (Fig. 5 Fig. 3. CPS loci demonstrate extensive diversity among RM C. coli isolates. Genome sequencing was performed for 15 primary C. coli isolates, as well as the C. coli vaccine working cell bank. Sequences were annotated (http://rast.nmpdr.org/) and queried for the presence of the CPS loci as defined by the kpsF and kpsC genes. Genes from all isolates were compiled and categorized into groups on the basis of high sequence similarity (≥80% identity), resulting in a total of 99 groups for the CPS. The genomic organization for each isolate is based on this clustering approach. Isolates with partial sequence results are indicated by //. On the basis of loci organization similarities, isolates were further sorted as shown. For comparison, the CPS genomic regions for the C. jejuni CG8421 vaccine strain are also shown. The numbers above each arrow refer to the locus tags in the published sequence (GenBank: CP005388.1), while the numbers within the arrows indicate genes with high similarity (BLAST score > 200) to genes found among the C. coli isolates. Gene segments from the C. jejuni CG8421 vaccine strain that did not meet this threshold are indicated with an x. The date of sample isolation is provided under each animal ID number. Horizontal dashed lines indicate that there are seven distinct strains of C. coli based on genetically similar CPS loci. Note that the strains of C. coli that circulated in 2015 and 2016 during the vaccine field trials were a mismatch to the C. coli and C. jejuni vaccine strains used in these studies. Genome sequence accession numbers are provided in the Data and materials availability section. 6797) before vaccination to 23,493 (95% CI, 19,008 to 29,035) at 6 months after primary vaccination and reached a titer of 34,888 (95% CI, 28,434 to 42,807) at 12 months (i.e., 6 months after booster vaccination). Serum antibody titers against flagellin showed an even greater increase, starting at a GMT of 18,419 (95% CI, 14,082 to 24,090) at the time of primary vaccination, then increasing to 43,747 (95% CI, 32,466 to 58,949) at 6 months and to 92,042 (95% CI, 76,548 to 110,673) at the 12-month time point. While there was a weak correlation between C. coli whole-cell and flagellin-specific antibody titers before vaccination (R 2 = 0.22), these values became increasingly more correlated following primary (R 2 = 0.32) and booster vaccination (R 2 = 0.90) (Fig. 5C). Since purified flagellin is the only antigen on the y axis and is a component of the whole-cell Campylobacter antigen on the x axis, the increased correlations observed in Fig. 5C indicate that the immune response to flagellin becomes more immunodominant in comparison to the other bacterial antigens after each round of vaccination.
To further evaluate the breadth of the antibacterial antibody response, C. coli lysate was probed by Western blot with prevaccination and postvaccination immune sera from mice and RM (Fig. 5D). As expected, naïve mouse serum was nonreactive against C. coli but showed responses to multiple Campylobacter antigens following H 2 O 2 -Campy C immunization. Sera from unvaccinated RM showed weak reactivity against Campylobacter antigens by Western blot. However, following H 2 O 2 -Campy C vaccination, additional bands of antibody-reactive antigen became more apparent and appeared to share similarity to the breadth of antibody responses observed in the vaccinated mice (Fig. 5D). While the identity of most of the bacterial antigens has yet to be determined, the protein band migrating at approximately 60 kDa was confirmed to be flagellin based on reactivity with an established flagellin-specific CF5 monoclonal antibody (13) and represents an immunodominant antigen following H 2 O 2 -Campy C immunization. In total, these data indicate that Campylobacter vaccine-mediated immunity can be induced to a number of bacterial antigens and results in higher and more consistent levels of antibody than that observed after natural infection alone.

Protection against Campylobacter-associated diarrheal disease
Given the high rates of C. coli-associated enteric disease among the RM (Fig. 1), we first investigated whether intramuscular (IM) administration of a 40-g dose of a C. coli-based vaccine could provide protective mucosal immunity against clinically defined cases of C. coli-associated diarrhea (note that two infants received a 20-g dose of vaccine instead of the full dose at the discretion of the attending veterinarians). Since most human vaccines require at least two immunizations for optimal protective efficacy and the durability of immunological memory is often improved following booster vaccination, we administered a booster dose 6 months later and then continued to monitor the incidence of diarrheal disease among the vaccinated and unvaccinated NHP. In standard direct-challenge animal models, the infection/exposure to the pathogen of interest is not performed until a sufficient period of time has elapsed following vaccination to allow for an immune response to be mounted. Because exposure to enteric pathogens can occur at any time throughout the year among outdoor-housed RM (Fig. 1), this study was prospectively designed to exclude animals that experienced a diarrheal episode during the first 2 weeks following vaccination, since it is unlikely that this would be enough time for vaccine-mediated immunity to reach a protective threshold (see Materials and Methods). In 2015, the H 2 O 2 -Campy C -vaccinated cohort consisted of 60 animals (61 animals minus 1 animal excluded from analysis due to hospitalization 1 day after primary vaccination) including 42 infants/juveniles and 18 adults with a demographic profile that matched the unvaccinated control population (table S1). Disease incidence was monitored for up to 1 year (1 April 2015 to 31 March 2016). Vaccinated animals were followed for an average of 336 days with 49 animals monitored for the entire 365-day observation period, culminating in 20,153 total exposure days in outdoor group housing. During this period of time, a total of two cases of diarrhea were identified among the vaccinated animals and were diagnosed as one case of C. coliassociated diarrhea, zero cases of C. jejuni-associated diarrhea, and one case of non-Campylobacter-associated diarrhea. Unvaccinated contemporaneous control animals (n = 1645) were monitored for 269 days on average with 761 animals followed for the entire 365-day observation period, culminating in 442,393 days of exposure. During this period of observation, 141 total cases of diarrhea were recorded among at-risk control animals. The separate C. coli-associated diarrheal analysis recorded 125 unique cases of disease, while the C. jejuniassociated diarrheal analysis recorded eight cases. Note that there were eight unvaccinated animals housed among the vaccinated NHP cohort in 2015 that either were not available on the day of vaccination or were born after the vaccination date. Of these unvaccinated sentinel animals, two of the eight animals (25%) were hospitalized with C. coli-associated diarrhea, indicating that vaccinated animals had continued exposure to C. coli during the course of the vaccine trial.
Overall, the incidence rate of C. coli-associated diarrhea among the 1645 contemporaneous unvaccinated animals was 9.8% versus 1.96% among the H 2 O 2 -Campy C -immunized animals. VE, as determined through the time-to-event Cox proportional hazards model, was estimated at 83% (95% CI, 1 to 97%, P = 0.048; Fig. 6A). C. jejuniassociated diarrhea cases were rare (0.69% incidence among unvaccinated animals Fig. 6B) and although we found an apparent VE of 100%, this was based on only eight cases identified among the unvaccinated control animals versus zero cases among the 60 H 2 O 2 -Campy Cvaccinated animals and was not statistically significant (P = 0.54). The incidence of all-cause diarrhea (Fig. 6C) was 11.0% among unvaccinated animals compared to 3.69% among H 2 O 2 -Campy C -vaccinated animals, with time-to-event Cox proportional hazard analysis estimating VE of 69% (95% CI, −14 to 92%,) and showed a statistically nonsignificant trend toward reduced enteric disease (P = 0.08), likely due to the underlying reduction in C. coli-associated enteric disease since it is a major contributor to all-cause diarrhea among these animals (Fig. 1B). To our knowledge, this represents the first demonstration of vaccine-mediated protection against Campylobacter using a natural primate model of enteric disease under conditions of repeated exposure through the fecal-oral route of transmission.
Initial results of the H 2 O 2 -Campy C vaccine indicated a trend toward protective immunity during the first 6 months before booster vaccination (P = 0.08). On the basis of the positive results observed with the H 2 O 2 -Campy C vaccine, we initiated a pilot study during the following calendar year (2016, table S1) to (i) determine whether a single Campylobacter vaccination could elicit protective immunity against diarrheal disease and (ii) test whether a related C. jejuni-based vaccine (H 2 O 2 -Campy J ) would provide cross-protective immunity against C. coli. Although the rates of C. jejuni in the sheltered-house animals were likely to be too low to assess homologous protection (Figs. 1B and 6, B and E), this approach was designed to determine whether H 2 O 2 -Campy J could provide heterologous vaccine-mediated cross-protection against C. coli (Fig. 6D). In 2016, the H 2 O 2 -Campy Jvaccinated cohort consisted of 67 animals including 41 infants/ juveniles and 26 adults with demographics that matched the unvaccinated control population (see Materials and Methods and table S1). Similar to the 2015 study, disease incidence was monitored for up to 1 year. Vaccinated animals were followed for an average of 343 days with 54 animals monitored for the entire 365-day observation period, culminating in 22,969 total days of exposure. During this period of time, there were five cases of diarrhea that were all diagnosed as C. coli-associated with zero cases of C. jejuni-associated diarrhea. Unvaccinated contemporaneous control animals (n = 1538) were monitored for an average of 239 days with 759 animals followed for the entire 365-day observation period, culminating in 367,792 days of exposure. During this period of observation, there were 138 total cases of diarrhea recorded including 115 cases of C. coli-associated diarrhea and 9 cases of C. jejuni-associated diarrhea. There were 19 unvaccinated sentinel animals that were housed among the 67 H 2 O 2 -Campy J -immunized cohort, and 1 of the 19 (5.3%) animals was diagnosed with C. coli-associated diarrhea. During the 1-year period following vaccination, the C. coli-associated diarrhea incidence reached 10.8% in the unvaccinated contemporaneous controls versus a rate of 7.8% in the H 2 O 2 -Campy J vaccine recipients (Cox proportional hazard VE, 28%; 95% CI, −70 to 70%, P = 0.46). Similar to the C. coli-based vaccine studies performed in 2015, in these 2016 vaccine studies, there were nine cases of C. jejuni-associated diarrhea among the unvaccinated controls (0.87% incidence rate) with no C. jejuni cases observed among the 67 H 2 O 2 -Campy J -vaccinated animals that were monitored for diarrheal illness for up to 1 year after vaccination (0% incidence of C. jejuni-associated diarrhea, P = 0.46). When comparing all-cause diarrhea, the incidence density among the vaccinated animals reached 7.7%, compared to a rate of 12.6% among controls (Cox proportional hazard VE, 40%; 95% CI, −43 to 75%). Although VE against all-cause diarrhea did not reach statistical The log-rank test was used to determine P values comparing the cumulative risk of diarrhea between vaccinated and unvaccinated groups. Hazard ratios (HRs) and 95% CIs were calculated by inverting the partial-likelihood score test under the Cox proportional hazards model (40), with VE defined as 1 − HR × 100%. U.I. indicates undefined CIs. For both vaccine cohort studies, unvaccinated control animals were contemporaneously followed. Arrows represent the dates that vaccinations were performed. significance (P = 0.26), it showed a trend that is consistent with the results observed after H 2 O 2 -Campy C vaccination.
Analysis of the published genome for the C. jejuni strain used in the H 2 O 2 -Campy J vaccine [strain CG8421 (6)] indicated that similar to the H 2 O 2 -Campy C vaccine, it did not match the CPS and LOS loci relative to the C. coli strains that circulated in the NHP colony in 2015 or 2016 ( Fig. 2 and Fig. 3). However, the major flagellin gene products (flagellins A and B) demonstrated 89 to 95% sequence identity to the primary C. coli isolates circulating between 2015 and 2016 (table S2) and suggest that flagellin and/or other conserved surface proteins may provide a mechanism to explain potential cross-protective immunity. In total, these studies show statistically significant vaccinemediated protection against C. coli-associated disease with the potential for at least partial cross-species protection against heterologous species of Campylobacter that should be further explored.

DISCUSSION
Campylobacteriosis is one of the most common bacterial infections worldwide, with an annual estimated burden of >25,000 deaths in children under 5 years of age (14). In addition to acute gastrointestinal symptoms, recent studies have revealed that infection with Campylobacter is associated with poor growth outcomes for children in low-income settings (3). In the studies presented here, we used a natural acquisition model of C. coli and C. jejuni infection among RM to test novel H 2 O 2inactivated Campylobacter vaccines. Our experimental Campylobacter vaccine provided up to 83% VE against C. coli-associated diarrheal disease. These results indicate that an H 2 O 2 -inactivated Campylobacter vaccine may be a viable option toward the development of a human vaccine against this important enteric pathogen.
The NHP model of Campylobacter-associated disease described here is unique in the sense that it relies on natural exposure and fecaloral routes of transmission. This model may also be useful for determining potential immune correlates that are associated with protection against Campylobacter-associated diarrhea. In our experiments, we had only a single breakthrough case C. coli-associated diarrhea after C. coli vaccination, and so the sample size was too small to determine correlates of immunity. Nevertheless, future studies, preferably using graded doses of vaccine to elicit varying levels of vaccinemediated immunity, could be useful for identifying an immune correlate against this important diarrheal disease. Poultry represents another commonly used model of Campylobacter infection, but C. jejuni is often considered a commensal organism that generally does not cause overt disease (15). Direct gastric challenge models using ferrets have been reported, although the need for compounds that inhibit peristalsis, the use of high challenge doses [10 10 to 10 11 colony-forming unit (CFU)] and the relatively mild Campylobacterassociated disease, limits the relevance of this approach (16). Oral challenge studies in rats have demonstrated colonization but reported a lack of clinical symptoms (17). A more recent report used high-dose oral challenge in rats to induce an irritable bowel syndromelike disease (18), but infections were cleared rapidly (≤14 days) and no animals had watery stool, which is inconsistent with human disease. Gnotobiotic piglets are susceptible to Campylobacter infection, whereas standard-housed piglets are generally resistant to overt disease (16). Similarly, both oral and intraperitoneal Campylobacter challenges in laboratory mice typically result in subclinical infection (16) with the exception of the recently developed antibiotic pretreatment model in zinc-deficient animals that show several characteris-tics of human campylobacteriosis after oral challenge (19). Infection of MyD88 knockout mice, in conjunction with vancomycin antibiotic treatment, has shown the ability to induce colonic pathology but failed to provoke overt diarrhea (20). Rabbit models require surgery and direct application of bacteria to the intestinal tract, but the relevance to human disease is uncertain due to the artificial nature of the infection and the trauma associated with the surgery (16). Direct intragastric challenge of RMs with C. jejuni may induce clinical signs of diarrhea (5), but results have been variable and require doses as high as 10 11 CFU to elicit disease. Even under these conditions, about one in five animals may fail to show any signs of diarrhea (5). The New World monkey, Aotus nancymaae, has also been proposed as an alternative NHP challenge model but requires doses as high as 5 × 10 12 CFU of C. jejuni to induce relatively mild illness in ~80% of animals (21). The limitations observed with direct challenge studies in NHP have also been encountered in human challenge studies. A meta-analysis of direct human and NHP challenge studies demonstrated that both species require similarly high doses of C. jejuni to cause overt disease, with doses that are 100-fold to 10 million-fold higher than those found during natural infection (22). While the reason for these discrepancies is unclear, these results suggest that direct gastric challenge of primates with laboratory-cultured Campylobacter may not be optimal compared to a natural exposure model for pathogenesis and vaccine development studies.
Most small laboratory animals are maintained under specific pathogen-free conditions, whereas the RM model provides the opportunity to study more complex enteric disease under conditions that involve coinfections more similar to those observed in humans living in endemic low-and middle-income countries. Likewise, our NHP colony is endemic for several human enteric pathogens including Campylobacter, Shigella flexneri, Shigella dysenteriae, Cryptosporidium, and Giardia among other bacterial and parasitic pathogens. A recent study also found the RM gut microbiome was more closely related to established microbiome datasets for Malawian and Amerindian populations than to an American microbiome (23). We believe that development of a Campylobacter vaccine that is effective under these conditions of comorbidity is more likely to be successful in field trials than vaccines against Campylobacter that are studied only in isolation. Although the correlates of immunity for Campylobacter remain largely undefined, T cells are unlikely to play a direct role in protection since these are extracellular bacterium. Studies in several animal models (24,25) and studies of naturally infected humans (26,27) have demonstrated that flagellin is a dominant target of the humoral immune response. Although results from an initial phase 1 human clinical trial involving intranasal vaccination with a subunit flagellinbased vaccine were suboptimal (28), in poultry vaccination trials, flagellin-specific antibody responses have been associated with a 100-to 1000-fold reduction in C. jejuni cecal colonization (29), underscoring the potential importance of mounting strong antibody responses to this and possibly other bacterial surface antigens. A recombinant flagellin subunit vaccine was also able to demonstrate heterologous protection in a mouse intranasal challenge model of C. jejuni (30). In our studies, vaccination of RM with a whole-cell vaccine resulted in Campylobacter-specific antibody levels that were higher than that observed following natural exposure and greatly enhanced the levels of anti-flagellin antibodies (Fig. 5). It is possible that the high immunogenicity of flagellin in our vaccine formulation is due, at least in part, to maintaining conformational and structural integrity of the antigen on the intact bacterial cell surface in a manner that is more similar to the antigen presentation that would occur during active infection. The vaccine-mediated antibody responses to flagellin observed in these NHP appears to be analogous to earlier studies of naturally infected humans which indicated that "This protein has a particularly high ratio of antigenicity to molar representation on the outer membrane." (27). The protection data against circulating bacterial strains that differed at the LOS and CPS loci (Figs. 2 and 3) also suggest that these elevated antibody responses to the more highly conserved flagellin protein, and potentially other conserved surface proteins (28), may be important in vaccine-mediated control of Campylobacter-associated enteric disease.
Although Campylobacter is a mucosal pathogen, mucosal vaccination was not required to induce protective immunity. Similarly, outer membrane vesicle (OMV) vaccines against Neisseria meningitidis, a Gram-negative bacterium, are also administered intramuscularly and have demonstrated up to 80 to 85% efficacy against invasive forms of disease (31). Besides providing broad immunity at the species level, the N. meningitidis vaccine approach also appears to hold promise even at the genus level. In a recent retrospective case-control study in New Zealand, an OMV-based N. meningitidis vaccine showed 31% efficacy (95% CI, 21 to 39%) against Neisseria gonorrhoeae (32). While this level of protection was modest, this result suggests that vaccines that contain multiple bacterial surface antigens provide broad immune responses and enhanced protective immunity, even against more distantly related mucosal bacteria. For the Campylobacter vaccine approach presented here, it is notable that the C. coli and C. jejuni vaccine strains did not match the LOS or CPS of the bacterial strains isolated during 2015 and 2016, the years in which vaccine field trials were conducted (Figs. 2 and 3). The central domains of the Campylobacter flagellin proteins can vary across strains and species, but the N and C termini maintain a high level of amino acid homology (33) and this sequence homology (or perhaps similarity among other surface proteins) may have contributed to the potential heterologous cross-protection against C. coli observed using the H 2 O 2 -Campy J vaccine, similar to the results observed for N. meningitidis vaccine-mediated protection against N. gonorrhoeae (32).
Although the natural exposure model of Campylobacter-associated diarrhea among outdoor-housed RM has advantages over other animal models, it also has several limitations. Unlike experimental directchallenge models used for other pathogens, this natural exposure model does not have 100% disease penetrance. With annual C. coliassociated diarrhea incidence of 9.8 to 10.8% during the 2015 and 2016 trials, respectively, a relatively large sample size is needed to obtain statistically significant data. A single breakthrough case of C. coli-associated diarrhea among the 60 H 2 O 2 -Campy C -immunized animals reduced VE from 100 to 83% VE when compared to the rates of disease observed among the unvaccinated control animals. Likewise, despite achieving what appeared to be 100% VE against C. jejuni-associated diarrheal disease, the incidence rate among the controls (0.69 and 0.83% disease incidence in 2015 and 2016, respectively) precluded definitive statistical analysis. Another challenge to this natural exposure model is that animals are at risk of enteric infection during and shortly after vaccination before a protective immune response can be mounted. We anticipated this issue and designed the experimental approach to exclude diarrheal animals that were diagnosed during the first 2 weeks after primary vaccination of our two-dose vaccination series. This resulted in the exclusion of one animal from the H 2 O 2 -Campy C -immunized cohort due to diarrhea-associated hospitalization within 1 day after vaccination.
This animal was likely beginning to be symptomatic before vaccination but reached the threshold for requiring hospitalization based on degree of dehydration, lethargy, and body score, as well as loose stool (8). If it was included in the experimental analysis, then the apparent VE would be 65% (P = 0.12). However, this is not an appropriate comparison because this current study focused on the development of a preventative vaccine against Campylobacter-associated diarrhea and is not designed to function as a therapeutic vaccine and is therefore unlikely to provide protection against disease once clinical manifestations of the enteric infection have already begun. Further studies to refine preclinical NHP models should be explored and eventual development of an optimized direct-challenge approach will greatly facilitate future testing of safe and effective vaccines against Campylobacter-associated disease.
To date, only a limited number of Campylobacter vaccine strategies have been examined in humans. Preliminary studies with a formalininactivated vaccine were conducted using an oral vaccination route and showed no protection and signs of diarrheal disease after ingestion of high vaccine doses (28). An intranasal vaccine approach was attempted for a recombinant flagellin candidate, but the vaccine has not advanced due to poor immunogenicity (28). ACE393 is a more recent Campylobacter vaccine candidate focused on a single protein identified through proteomic screens (34). This vaccine was administered as an IM immunization but provided no clinical protection following controlled live challenge in humans (28). Currently, a CPS vaccine candidate represents the most advanced C. jejuni vaccine approach being pursued in the clinical setting (10). In studies performed in A. nancymaae monkeys, a three-dose subcutaneous regimen was able to protect against homologous gastric challenge (35). As noted in the study, it is unlikely that a single defined CPS will be protective against a large range of C. jejuni strains (10), and this approach may require multiple CPS antigens for broad vaccine-mediated protection. Multiple serotypes for C. coli and C. jejuni have been identified on the basis of two serotyping protocols, Penner and Lior. Both serotyping approaches were developed as cataloging methods to study the epidemiology of Campylobacter outbreaks before the implementation of modern genetic tools, and their importance in relation to cross-protection and VE remain unclear. In the absence of specific Campylobacter strain pre-adsorption, reference sera demonstrate substantial cross-reactivity, suggesting that the wholecell immunization approach used to develop these reference sera actually elicits a relatively broad response to a number of different serotypes. For instance, in a ferret model of C. jejuni disease, heterologous cross-protection was observed between different Lior strains, ranging from 67 to 84% VE against overt signs of diarrhea (36). Other studies in small-animal models have likewise demonstrated heterologous protection to different Campylobacter serotypes using bacterial whole-cell approaches (37). Successful whole-cell inactivated Campylobacter vaccines against C. jejuni and Campylobacter fetus fetus have been used commercially in sheep vaccines for many years (e.g., Campylobacter Fetus-Jejuni Bacterin and CampyVax). Campylovexin was developed to prevent Campylobacter-associated abortions in sheep, with ~80% efficacy, and studies indicate that it is protective against 26 different circulating strains of C. fetus fetus (Campylovexin Product Profile, Virbac Animal Health, New Zealand, 2015). Together, these reports support our Campylobacter vaccine approach, which uses an advanced whole-cell formulation developed to preserve antigenic structures and achieve antibacterial antibody titers that exceed those observed following natural exposure.
In this study, we demonstrate significant VE against naturally occurring Campylobacter infection using a novel peroxide-inactivated whole-cell vaccine. Vaccine-mediated protection reached 83% in our field trials despite the vaccine strain expressing CPS and LOS that were genetically mismatched from the circulating strains of Campylobacter. Regardless of the Campylobacter vaccine strain(s) used for future clinical development, our results demonstrate that significant vaccinemediated protection against Campylobacter-associated diarrhea is feasible and supports the continued development of new and advanced vaccines against these globally relevant enteric pathogens.

Vaccine production
Production lots of C. coli (NTICC13) or C. jejuni [CG8421 strain provided by P. Guerry at the Naval Medical Research Center (6)] were grown to late-log phase in shaker flasks of supplemented growth medium and concentrated/purified by tangential flow filtration. Inactivation was initiated with the addition of 0.10% H 2 O 2 and 2 M CuCl 2 , with a total inactivation time of 20 to 22 hours at room temperature (RT). Inactivation was stopped by the addition of excess EDTA and catalase before being adsorbed to 0.10% alum. Five percent of the bulk material for each lot was tested for residual live bacteria under both microaerophilic and standard atmosphere conditions, for a total testing volume of 10%. Each lot was tested for LOS content using an end point chromogenic limulus amebocyte lysate assay (Lonza), with an average value of 9000 endotoxin units (EU) ± 3598 (±SEM) per 40-g dose determined among four production lots. While there is no precise relationship between EU and endotoxin mass, a range of 5 to 10 EU/ng has been reported for established endotoxin standards (38). Using an average value of 7.5 EU/ng, we estimate ~1.2 g of bacterial LOS per dose, which is similar to the vaccine against N. meningitidis (BEXSERO) with a range of 1 to 3 g of LOS per dose (39).

Campylobacter WGS and phylogenetic analysis
Campylobacter fecal isolates were collected over a 3-year period from a sampling of healthy asymptomatic RM carriers and animals hospitalized with Campylobacter-associated diarrhea. All isolates were further streaked for purity before a final expansion and DNA extractions. Genomic DNA libraries were generated using iGenomX Riptide (Carlsbad, CA) high throughput rapid library prep following the manufacturer's suggested protocol using 50 ng of input DNA per sample. Each library was prepared using a unique molecular identifier. The resulting multiplexed library was sequenced on an Illumina MiSeq (San Diego, CA) using MiSeq v2 reagents paired end 2 × 150base pair reads. Reads were demultiplexed using fulcrum genomics DemuxFastqs tool (--read-structures = 8B12M130T 8M142T --maxmismatches = 0). Trimmed, demultiplexed reads were assembled into contigs using SPAdes genome assembler (cab.spbu.ru/software/ spades/; default settings) and submitted to the Pathosystems Resource Integration Center (PATRIC; www.patricbrc.org) for annotation via the RAST toolkit (rast.theseed.org). Reference and assembled genomes were placed in a phylogenetic tree using PATRICs' Phylogenetic Tree Building service with a final alignment via FastTree (microbesonline.org/fasttree/) using no automated progressive refinement ( fig. S2). To allow for expanded flagellin gene sequence analysis, a subset of samples was also sequenced using the long-read PacBio Sequel platform (GENEWIZ, South Plainfield, NJ).

Campylobacter antigen ELISAs
For C. coli specific enzyme-linked immunosorbent assays (ELISAs), whole-cell lysate, purified flagellin, and purified LOS were produced. For whole-cell lysate, bacteria were grown in suspension and inactivated using the optimized H 2 O 2 -based approach as described above. Samples were then sonicated, aliquoted, and held at ≤−65°C. To produce purified flagellin, C. coli was grown and concentrated by tangential flow filtration, resuspended in phosphate-buffered saline (PBS), and homogenized by sonication. Homogenates were centrifuged at 2000g for 30 min and resuspended in PBS, followed by ultracentrifugation for 3 hours at 100,000g. Purified flagellin pellets were resuspended in PBS, aliquoted, and stored at ≤−65°C. Silverstained SDS-polyacrylamide gel electrophoresis gels run under reduced conditions showed a major band migrating at ~60 kDa that reacted with the flagellin-specific CF5 monoclonal antibody (13). For ELISA tests, serum samples were first preadsorbed to reduce nonspecific binding by diluting serum 1:100 into suspensions of heat-killed (56°C for 1 hour) S. flexneri (containing approximately 10 9 CFU/ml) for 60 min at RT, followed by brief centrifugation to collect the preadsorbed serum samples.
Serum ELISA was performed as described in (9). Briefly, ELISA plates were coated overnight at 2° to 8°C using optimal concentrations of each antigen as determined through small-scale pilot studies. Unbound antigen was removed, and plates were treated with blocking buffer [5% nonfat dry milk in PBS-T (PBS supplemented with 0.05% Tween 20)] for 1 hour at RT. Plates were rinsed one time with PBS-T and incubated for 1 hour with serial dilutions of the adsorbed serum samples. Plates were washed five times with PBS-T and incubated with an optimal dilution of a goat anti-monkey immunoglobulin G-horseradish peroxidase antibody (Rockland, Limerick, PA) for 1 hour at RT. After a final wash, plates were developed with o-phenylenediamine dihydrochloride substrate for 20 min, with development stopped by the addition of an equal volume of 1 M HCl. Optical densities (ODs) were measured at 490 nm, and a log-log transformation of the OD versus reciprocal serum dilution was performed. End point titers were determined as the reciprocal of the serum dilution needed to reach an OD of 0.10. Each plate contained a serum standard to allow normalization between experiments, and each sample was tested in duplicate, with the average value between duplicates taken as the final titer.

Safety assessment in mice and RMs
For preliminary safety studies, 8-week-old female BALB/c mice were obtained from the Jackson laboratory (Bar Harbor, ME) and immunized by the intraperitoneal route with a 40-g dose of H 2 O 2 -Campy C . In a separate control study, mice were treated with 10 g of E. coli LPS (O111:B4, List Biological Laboratories, Campbell, CA). Animal weights were followed for up to 1 week after immunization. The second phase of our safety assessment was conducted in RM. Two adult females received an IM vaccination with the H 2 O 2 -Campy C vaccine, and two adult females received a mock, alum-only IM vaccination. Health metrics, including weight, attitude, appetite, urine, stool, temperature, and vaccine site, were monitored daily for 14 days.

Immunization of mice and RMs
To produce Campylobacter hyper-immune sera for Western blot staining, mice were immunized with 40 g of H 2 O 2 -Campy C at 0, 2, and 3 months. Serum samples were collected at 1 month following the final immunization. The ONPRC maintains an active RM breeding colony, and for the H 2 O 2 -Campy C vaccine study, animals were vaccinated intramuscularly with alum-adjuvanted, inactivated C. coli vaccine in March 2015 (table S1). At the discretion of the attending veterinarian, two infants received a 0.5-ml (20 g) primary dose, with all other animals receiving a 1.0-ml (40 g) primary dose. Animals that were hospitalized with diarrhea during the first 14 days after the first vaccination were a priori excluded from analysis, since these animals are housed in a high exposure setting and 14 days is unlikely to be enough time for the vaccine to elicit a protective immune response. Among the first RM cohort, one adult female experienced a diarrheal episode 1 day after H 2 O 2 -Campy C vaccination, leaving a total of 60 immunized animals that were monitored for diarrheal disease for up to 1 year beginning on 1 April, 2015. Of these animals, 53 were available to receive a 40-g 6-month booster vaccination, and 49 had matched serum samples available at all time points for serological analysis of antibacterial immunity. In July 2016, a second cohort of RM was vaccinated with a single IM 40-g dose of the alum-adjuvanted C. jejuni vaccine, H 2 O 2 -Campy J (table S1). Three animals experienced diarrhea within 2 weeks of primary vaccination, resulting in a final cohort of 67 animals that were monitored for diarrheal disease for up to 1 year beginning on 3 August 2016.

Diarrheal disease incidence
Diarrhea incidence rates among vaccinated animals were tracked for a 1-year period after primary vaccination using a centralized electronic health record system (7) and compared contemporaneously to unvaccinated control animals during that same year of disease activity. Any animals exhibiting signs of severe diarrheal illness were transported to the veterinary hospital for further evaluation. If a case of diarrhea was confirmed, then rectal fecal cultures were submitted to the ONPRC Clinical Pathology Laboratory for C. coli, C. jejuni, and Shigella spp. diagnosis and the results were documented in a searchable central database (PRIMe). RMs were observed daily for the entire year by trained husbandry staff, including personnel who were blinded to the animal's vaccination history. For all-cause diarrhea analysis, animals were considered to have met the end point with their first recorded case of any diarrheal episode. For C. coliassociated diarrhea, animals met the end point with their first recorded case of C. coli-associated diarrhea, regardless of whether they were previously diagnosed with C. jejuni-associated diarrhea or with unrelated/all-cause diarrhea. For C. jejuni-associated diarrhea, animals met the end point with their first recorded case of C. jejuni-associated diarrhea, regardless of whether they were previously diagnosed with C. coli-associated diarrhea or with unrelated/all-cause diarrhea. C. coli-and C. jejuni-associated diarrhea diagnoses were performed by blinded personnel in the ONPRC Clinical Pathology Laboratory who were not aware of the animal's vaccination history.

Statistics
Figures show group averages ± SEM. Correlation between Campylobacter whole-cell lysate and flagellin ELISA titers were determined by linear regression following logarithmic transformation. Statistical comparisons of antibacterial antibody levels were made on logarithm transformed titers using repeated measures ANOVA with Tukey's multiple test correction. In the experimental design stage, power analysis was performed to estimate necessary cohort sizes using Fisher's twotailed exact test (G*Power, version 3.1.9.2, Heinrich-Heine-Universität Düsseldorf). An annual C. coli-associated diarrhea rate of 13% was assumed for unvaccinated animals, with an 18:1 allocation of unvaccinated to vaccinated animals. Using these parameters, we estimated that ≥50 vaccinated animals, compared to at least 900 unvaccinated animals, would provide ≥80% power to detect an effect size of 0.10 at the 0.05 significance level. Similar to human clinical trials/field studies, any animal removed from the study for unrelated reasons was considered lost to follow-up, but their data were included in the final dataset up to the date of removal. Given that each shelter unit typically houses 30 to 40 animals, we chose to vaccinate two shelters and compare diarrheal incidence rates relative to a small number of unvaccinated animals within the two shelters (n = 8 unvaccinated animals in 2015 and n = 19 unvaccinated animals in 2016 were housed with the vaccinated cohorts), together with the >1500 unvaccinated control animals housed in adjacent shelters (table S1). The unvaccinated sentinel animals that were housed among the vaccinated cohorts were not by design but occurred because the animals were either temporarily located elsewhere on the day of vaccination or were born after the vaccination date. For both vaccine cohorts, all animals were considered at risk while located in shelter housing during each 1-year period following vaccination. Ideally, statistical modeling for VE should account for animal clustering. To examine the effect of reduction in effective sample size, we performed a mixed effects Cox regression model to account for the correlation between outcome measurements collected for monkeys from the same shelter. This approach yielded a marginally nonsignificant result: H 2 O 2 -Campy C vaccination provided 87% VE (95% CI, −18 to 99%, P = 0.07) against C. coli-associated diarrhea. However, this approach is problematic due to husbandry and logistical needs among the breeding groups that resulted in animals being moved from one shelter to another or being moved from one breeding group to another, throughout the course of the year. As a result, shelter membership may not be a proper representation of clustering structure. Therefore, for disease acquisition studies in Fig. 6, the log-rank test was used for comparing the cumulative risk of diarrhea between vaccinated and unvaccinated groups. In addition, VE was based on time to disease and defined as [1 -hazard ratio (HR)] × 100%. The HR and associated 95% CIs were obtained by inverting the partial-likelihood score test under the Cox proportional hazards model as described in (40). Statistical analyses were performed using R3.6.0 (R Foundation for Statistical Computing, Vienna, Austria, 2019) and SAS9.4 (SAS Institute, Cary, NC).

Study approval
All animal studies were overseen and approved by the Oregon Health and Science University Institutional Animal Care and Use Committee in accordance with the National Institutes of Health guide for the care and use of laboratory animals. Animals were housed in accordance with standards established by the U.S. Federal Animal Welfare Act and the Guide for the Care and Use of Laboratory Animals.

SUPPLEMENTARY MATERIALS
Supplementary material for this article is available at http://advances.sciencemag.org/cgi/ content/full/6/26/eaba4511/DC1 View/request a protocol for this paper from Bio-protocol.