Introduction

Mucosa-associated E. coli are found in Crohn’s disease adherent to the intestinal epithelial surface [1], within granulomas [2], mesenteric lymph nodes [3], and in peripheral blood [4]. In the laboratory, they translocate across the microfold (M) cells that overlie Peyer’s patches and colonic lymphoid follicles [5, 6]. They replicate within macrophage vesicles [7, 8] and induce granulomas [9]. They lack any consistent genotype [10], and it is unclear whether they are part of the pathogenic process or merely secondary bystanders.

Proof of a pathogenic role depends on showing that therapy targeted to eradicate E. coli is beneficial. This entails killing the bacteria replicating within macrophage vesicles [11]. This is not straightforward for some antibiotics such as penicillin and gentamicin are unable to penetrate macrophages. In Crohn’s disease, meta-analysis has not shown convincing benefit with antibiotics [12]. However, there are positive signals with rifaximin [13], ciprofloxacin (as single agent in one study [14] but usually in combination with metronidazole) [15], and azithromycin with metronidazole [16]. None of these effects have been sufficiently clear for antibiotic therapy to become “mainstream”.

Whipple’s disease, caused by Tropheryma whipplei, and Q-fever, caused by Coxiella burnetii, are both unequivocally due to replication of bacteria within intestinal macrophages and are difficult to cure even with long-term antibiotics. One effective strategy is to combine long-term doxycycline with the antimalarial hydroxychloroquine [17, 18]. The rationale for hydroxychloroquine is that many bacteria, including Crohn’s disease mucosal E. coli, require an acid environment to replicate [8]. Hydroxychloroquine is a mild base that enters macrophage vesicles and reverses their normal acidification [19]. In the laboratory, this substantially enhances killing of E. coli by macrophages [20].

We have therefore assessed the efficacy and safety of a combination of short-term ciprofloxacin and longer-term doxycycline with hydroxychloroquine in the treatment of active Crohn’s disease. Budesonide was prescribed as comparator as this was standard care for patients with moderately active Crohn’s disease at the inception of this trial (2013) [21].

Materials and Methods

Study Design

This was an open-label investigator-led randomized controlled trial in adults with moderately severe Crohn’s disease. The aim was to provide an unbiased estimate of the remission rate for the antibiotic/hydroxychloroquine combination to inform subsequent phase 3 trial design. Patients were recruited from 8 hospitals in England between 2013 and 2018.

The study protocol conforms to the ethical guidelines of the 1975 Declaration of Helsinki as reflected in a priori approval by the North West-Liverpool East Research Ethics Committee of the NHS Health Research Authority on February 6, 2012 (08/H1002/71). Ethics approval required use of standard of care at the time of application as a comparator rather than placebo. The study was co-sponsored by University of Liverpool and the Royal Liverpool and Broadgreen Hospitals University Trust and adopted onto the National Institute for Health Research portfolio of supported clinical research studies. An independent Data Monitoring Committee monitored the trial at approximately 6-month intervals. All patients gave informed written consent.

Participants

Eligible patients were aged ≥ 18 with moderately active Crohn’s disease (CDAI > 220–450) involving small intestine, colon, or both, with serum C reactive protein (CRP) ≥ 5 mg/l and/or fecal calprotectin > 250 µg/gram. Patients receiving thiopurines had to have stable dose for ≥ 3 months. Patients receiving anti-TNF, other biologics, or methotrexate within previous 3 months were excluded. Concurrent use of prednisolone > 5mg/day or budesonide > 3 mg/day was also excluded as were patients receiving antibiotics within the previous 4 weeks. Patients receiving mesalazine must have a stable dose for at least 1 month.

Randomization and Stratification

Patients were randomized centrally 1:1 to either antibiotic/hydroxychloroquine combination or budesonide (Entocort CR©). Patients were stratified prior to randomization into four groups: isolated colonic Crohn’s disease versus small intestine ± colon, and immunosuppression (thiopurines) versus no immunosuppression. Randomization was by random number generation in Microsoft Excel, in blocks of 10 patients (by sequence of joining trial) within these four groups. Opaque allocation envelopes labeled by patient number were filled by the trial statistician at a remote site and posted to the trial coordinator.

Interventions and Crossover

Patients randomized to antibiotics/hydroxychloroquine received oral ciprofloxacin 500 mg bd, doxycycline 100 mg bd, hydroxychloroquine 200 mg tds before meals for 4 weeks followed by 20 weeks continued therapy with doxycycline 100 mg bd and hydroxychloroquine 200 mg tds. Patients randomized to budesonide received oral budesonide (Entocort CR©) 9 mg/day taken in the morning for 8 weeks followed by 6 mg/day for 2 weeks as per Rutgeerts et al. [22] followed by 3 mg/day for a further 2 weeks as per current standard steroid tailing practice. Patients failing to respond (CDAI fall by > 70) by 10 weeks were offered crossover onto the alternative treatment (Fig. 1).

Fig. 1
figure 1

Study design

Full size image

Outcomes

The primary endpoint was remission (CDAI ≤ 150) without the addition of any other treatment for Crohn’s disease and assessed at three time points: 10 weeks; remission at 10 weeks maintained through to 24 weeks; and remission at 10 weeks maintained through to 52 weeks. The 10-week time point was chosen to conform to the comparator budesonide trial data [22], 24 weeks to correspond with the end of the antibiotic/hydroxychloroquine therapy period, and 52 weeks to determine whether any therapeutic effect might persist after cessation. Secondary endpoints included remission (CDAI ≤ 150) at 4 weeks, remission or response (CDAI fall by > 70 points) at 10 weeks, patient global assessment of symptom severity by IBD10 (10 cm visual analogue score) [23] at 4, 10, 24, and 52 weeks, and fecal calprotectin (assayed using whichever assay was in routine use at the recruiting site) at 10, 24, and 52 weeks. Adverse events were assessed at each visit.

Power Calculation and Early Closure

Budesonide was predicted to have a 53% remission rate (95% CI 43%, 64%) at 10 weeks [22]. A sample size of 50 in each group was estimated to have 80% power at 5% significance level to detect a difference between the predicted 53% remission rate for budesonide and a remission rate of 80% in the antibiotic/hydroxychloroquine group. Recruitment commenced late 2013 and initially progressed reasonably but gradually fell below target probably due to increasing availability of biologics. With 59 patients recruited and 78 treatment courses with crossover, over the 5-year recruitment period, the Data Monitoring Committee considered that we had data sufficient to address the primary aims of the trial and that trial closure was appropriate.

Serology

Serum samples were obtained on recruitment, blind-labeled, and tested for relevant antibodies: OmpC (outer membrane porin C of E. coli and other Gram-negative bacteria), FlaX, Fla2, CBir1 (all flagellar antigens present in Clostridiales), and anti-Saccharomyces cerevisiae antibodies (ASCA) by quantitative ELISA performed by Prometheus® Laboratories Inc., San Diego, as part of the Prometheus IBD sgi Diagnostic® Cat.#1800 and with reference to their pre-established normal ranges.

Statistical Analysis

The primary analysis of categorical outcomes of remission at 10 weeks and maintenance of remission at 24 weeks and 52 weeks was by Fisher’s exact test. This was also performed with crossover data included. A preplanned subgroup analysis assessed the possibility that serology, particularly anti-Saccharomyces cerevisiae antibody (ASCA) status, might predict response to antibiotics/hydroxychloroquine. To account for multiple testing in these secondary analyses, significance was set at P values of 0.02 (Fisher’s test including crossover) and 0.01 for the other tests, implying an overall type I error risk of approximately 0.05 (Bonferroni calculations). Secondary outcomes were analyzed using Fisher’s exact test for categorical outcomes and Wilcoxon Rank Sign test for numerical values (patient global index [IBD10] and fecal calprotectin).

Results

Eighty-three patients were screened and 61 randomized of whom 59 took the study medication and were eligible for participation (Supplementary Figure 1). There were no significant differences at baseline between the treatment groups in demographics or disease activity (Table 1).

Table 1 Demographic data
Full size table

Response to Therapy

Initial Therapy

Twenty-seven patients received antibiotics and hydroxychloroquine as the initial therapy, and 32 received budesonide. Response was not significantly different between antibiotics/hydroxychloroquine and budesonide at any of the three primary outcome time points: 10 weeks, 24 weeks, and 52 weeks (Table 2). Only patients who were in remission at 10 weeks were eligible for the primary analyses of maintenance in remission at 24 and 52 weeks. However, a post hoc analysis that includes 3 patients who responded (CDAI fall by > 70 but still > 150) by 10 weeks and had gone into remission by 24 weeks on continued therapy showed a trend toward benefit with antibiotics/hydroxychloroquine, particularly when assessed per protocol (5/17 in remission at 24 weeks versus 1/26 for budesonide; P = 0.028 [not reaching predetermined significance threshold of 0.01]. There was a nonsignificant tendency toward a higher remission rate with budesonide at 10 weeks (8/32 versus 2/27 on intention to treat (ITT) analysis (P = 0.092), but by 24 weeks only one budesonide-treated patient remained in remission. Only one patient in each treatment group remained in remission through to 52 weeks—plus one additional patient receiving antibiotics/hydroxychloroquine who had been a responder at 10 weeks.

Table 2 Response to initial therapy course (primary endpoint data in bold)
Full size table

Initial Therapy and Crossover Combined

Including crossover, there were 39 treatment courses in each group. Intention to treat analysis showed no significant differences in primary endpoint data at any of the three time points, 10, 24, and 52 weeks (Table 3, Fig. 2).

Table 3 Response to therapy including crossover (primary endpoint data in bold)
Full size table
Fig. 2
figure 2

Remission rates at 10, 24, and 52 weeks, including crossover data: a intention to treat data; b per protocol data

Full size image

When analyzed per protocol, and including (post hoc) the three 10-week responders who were in remission by 24 weeks, 9 of 24 patients receiving antibiotics/hydroxychloroquine were in remission at 24 weeks compared with 1 of 32 receiving budesonide (P = 0.001). By 52 weeks, only one of 31 who had received budesonide per protocol remained in remission compared with three of 23 patients who received antibiotics/hydroxychloroquine plus one additional patient who had responded (CDAI fall > 70 but still > 150) by 10 weeks.

Impact of Phenotype and Serology

There was no signal that patients with isolated colonic Crohn’s disease (L2) responded differently to the antibiotic/hydroxychloroquine combination (Supplementary Table 1; P = 1.0).

Out of 59 patients recruited, 50 had sera available for testing, 21 were positive for ASCA, 6 were positive for OmpC, and 26 were positive for either ASCA or OmpC. ASCA, OmpC, separately or in combination, and FlaX and CBir serology were not associated with response to the antibiotic/hydroxychloroquine combination (Supplementary Table 2).

The phenotypic profile of the 9 patients who were in remission at 24 weeks on antibiotics/hydroxychloroquine showed no obvious features that might be used to predict response (Table 4).

Table 4 Phenotypic and serological features of the 9 patients who achieved remission at 24 weeks on antibiotics/hydroxychloroquine
Full size table

Secondary Endpoints

Rates of remission at 4 weeks and the combined endpoint of remission or response at 10 weeks were similar in both treatment groups, on initial therapy (Table 2) and including crossover (Table 3).

Patient global index of severity (IBD10) correlated well with CDAI (r = − 0.745; P < 0.0001 based on 219 data sets from the 59 patients) with IBD10 > 8 equivalent to CDAI ≤ 150 (remission) as previously reported23 (Supplementary Figure 2). Patients showed similar improvement in both treatment groups. IBD10 improved from week 0 median (IQR) 3.3 (2.1–4.9) to week 10 5.9 (4.2–8.1) with antibiotics/hydroxychloroquine (P = 0.001) and from week 0 median (IQR) 4.1 (2.2–5.3) to week 10 6.35 (2.7–9.1) with budesonide (P = 0.006) (Supplementary Figure 3). Again there was evidence of better response at 24 weeks with antibiotics/hydroxychloroquine median (IQR) 8.75 (6.75–9.3) compared with budesonide median 6 (3.3–6.5) but numbers at this endpoint were small (n = 12 antibiotics/hydroxychloroquine; n = 6 budesonide).

Fecal calprotectin correlated less closely with CDAI (r = 0.331; P < 0.0001 based on 136 datasets from all 59 patients) (Supplementary Fig. 4). Improvements in both treatment groups were similar but modest—falling from median (IQR) 497 (265–726) μg/g at week 0 to 331.5 (89–600) μg/g at week 10 with antibiotics/hydroxychloroquine (P = 0.03) and from 573.5 (307–924) μg/g at week 0 to 454 (167–634) μg/g at week 10 with budesonide (P = 0.15) (Supplementary Fig. 5). Based on per protocol analysis, 10/17 patients receiving antibiotics/hydroxychloroquine achieved fecal calprotectin < 250 μg/g at week 10 compared with 8/27 receiving budesonide (P = 0.07).

Adverse Events

There were 100 adverse events affecting 34 of 39 (87%) patients receiving antibiotics/hydroxychloroquine and 55 adverse events in 31 of 39 (79%) patients receiving budesonide (Table 5). Withdrawals by 10 weeks for adverse events were seen in 15 receiving antibiotics/hydroxychloroquine and 6 budesonide. Nausea affected 24 of 39 (62%) receiving antibiotics/hydroxychloroquine but was usually transient and only led to discontinuation of therapy in two; photosensitivity affected 11 of 39 (28%) even though all patients were warned to use high sun protective factor creams and minimize exposure and led to discontinuation of therapy in three; tendinopathy or arthralgia affected 7 of 39 (18%) and tendinopathy led to discontinuation in two but resolved without tendon rupture; and 6 of 39 (15%) developed candidiasis of whom two received topical antifungals. Five patients receiving antibiotics/hydroxychloroquine suffered severe adverse events including two drug-related (vomiting and photosensitivity) and three unrelated—a fall in an elderly patient complicated by rhabdomyolysis, perianal abscess, and non-fatal pulmonary embolism. There was one severe adverse event (incisional hernia) in a patient receiving budesonide.

Table 5 Adverse events (AEs)
Full size table

Discussion

In this open-label study, we found no significant difference in remission or response rates at 10, 24, or 52 weeks between the antibiotic/hydroxychloroquine combination and a standard 12-week course of budesonide. The study was not powered for non-inferiority. On initial therapy, there was a nonsignificant trend toward better results at 10 weeks with budesonide; however, with crossover data included, similar remission and response rates were seen with both treatments. The 24-week data also showed no significant difference between treatment groups when assessed on intention to treat using predefined thresholds for statistical significance. However, when assessed per protocol, the results were encouraging for the antibiotic/hydroxychloroquine combination with 9 of 24 in remission at 24 weeks. Sustained remission to 24 weeks was only seen in one of 32 (per protocol) budesonide-treated patients—who had of course tailed off their therapy by 12 weeks. This might sound “unfair” given that the doxycycline and hydroxychloroquine were continued through to 24 weeks but is justified given that the efficacy of budesonide is only sustained for around 3 months, with longer maintenance ineffective [24]. The 52-week data showed a modest signal to support prolonged efficacy with the antibiotic/hydroxychloroquine combination, with 4/23 in remission at that time point on per protocol analysis. Patients who crossed over to the antibiotic/hydroxychloroquine combination fared better (remission at 10 weeks in 5/12, 42%; remission maintained to 24 weeks in 4/12, 33%; and remission maintained to 52 weeks in 2/12, 17%) raising the possibility that prior steroids followed by antibiotics/hydroxychloroquine might have been better than either treatment alone.

Remission rates with budesonide were lower than predicted, 8/32 (25%) at 10 weeks compared with 53% in the original trial by Rutgeerts et al. [22]. The reason for this is unclear, but might in part be due to the inclusion of patients with isolated colonic Crohn’s disease in the current study (Supplementary Table 1). When these are excluded, the remission rate, including crossover, with budesonide was 9/24 (38%) when analyzed per protocol at 10 weeks. Average CDAI at baseline was 277 which is almost identical to that in the Rutgeerts et al. trial (275). CRP was lower (median 8.5 mg/l compared with mean 25 mg/l in Rutgeerts et al.). Moreover, most (7/8; 87.5%) of the patients who were in remission at 10 weeks with budesonide had relapsed by 24 weeks off therapy.

The patient global assessment (IBD10) correlated very well with CDAI scores, confirming that an IBD10 score of 8 is equivalent to a CDAI of 150 [23], and showed treatment responses comparable to the primary outcome data, with improvements by 10 weeks similar for both treatment groups. There was a much weaker correlation between fecal calprotectin and CDAI, and improvements seen in both treatment arms were modest. Fecal calprotectin is known to correlate better with mucosal healing, and 4 of 9 patients in remission at 24 weeks had fecal calprotectin < 70 μg/g, one of the lower proposed cutoff values for indicating likely endoscopic remission.

The prevalence of mucosal E. coli in Crohn’s disease ranges from 21 to 63% of cases [1]. The antibiotic/hydroxychloroquine combination would also be effective against other organisms, particularly Gram-negative bacteria, that could invade the gut and replicate within macrophages. We had hoped that serological testing might identify patients more likely to be affected by E. coli and thus predict response to antibiotics/hydroxychloroquine. OmpC is an outer membrane porin protein found in Gram-negative organisms including E. coli but only 6 of 50 patients tested positive for OmpC antibody. Anti-Saccharomyces cerevisiae (baker’s yeast) antibody (ASCA) is the best established antibody associated with Crohn’s disease. Its origin is unclear as baker’s yeast is not the only source of its antigen, an oligomannan with a specific mannose alpha 1,3 mannose alpha 1,2 mannose terminal sequence [25, 26]. Other possible microbial sources include Candida albicans and Mycobacterium avium subspecies paratuberculosis [27, 28]. There is an intriguing possibility that a key ASCA antigen could be the mannan receptor for Gram-negative bacterial fimH on glycoprotein 2 (GP2) on microfold (M) cells [29]. M cells are the initial sites for bacterial invasion and are also thought to be the sites of the initial lesions in Crohn’s disease. There is considerable overlap between ASCA and anti-GP2 serology in Crohn’s disease sera [30]. Moreover, although the mannan on GP2 has yet to be characterized, the bladder equivalent of GP2 is uroplakin and its mannan receptor for fimH expressed by uropathogenic E. coli is oligomannose-3 which contains the same terminal mannose alpha 1,3 mannose sequence as the ASCA epitope [31]. It therefore seemed plausible that ASCA positivity might identify Crohn’s disease patients with E. coli invasion. However, in this study there was no correlation between ASCA, either separately, or combined with OmpC positivity, and response to the antibiotic/hydroxychloroquine combination.

There is a good case for considering isolated colonic Crohn’s disease as a separate condition from Crohn’s disease affecting the small intestine [32, 33]. There has also been a suggestion that it might respond better to antibiotics [32]. In this study, patients with isolated colonic disease showed no significant difference in response to antibiotics/hydroxychloroquine. Careful analysis of the nine patients who were in remission at 24 weeks on the antibiotic/hydroxychloroquine combination again failed to show association with phenotype or serotype. It is possible that a more specific bacterial DNA-based technique is needed, possibly based on the detection of circulating Gram-negative bacterial DNA which has been shown to predict relapse [4].

Side effects were an issue with the ciprofloxacin, doxycycline, and hydroxychloroquine combination, particularly in the first month when patients were taking all three drugs. Nausea was common but was usually very transient but photosensitivity (all three drugs are known to cause this) and Achilles tendon pain were also issues. Hydroxychloroquine is associated with the risk of retinopathy, but this is related to cumulative dosage, usually over years of therapy. All patients were screened prior to trial entry for any history of visual impairment not corrected by wearing glasses and by visual acuity testing repeated at 10 weeks or early withdrawal. No patient failed screening on these grounds. One patient was withdrawn because of transient blurred vision, but this was thought unlikely to be related to the hydroxychloroquine. Six patients developed candidiasis, but none stopped trial medication because of this.

We believe that the signals of efficacy seen with the antibiotic/hydroxychloroquine combination, particularly the longer-term remissions, justify further study. It is unfortunate that neither phenotypic information nor serology seems to identify patients likely to respond. It is possible that a DNA-based approach probing for E. coli or other Gram-negative organisms on fecal or peripheral blood samples might better identify patients likely to respond, but this was beyond the funding available for this study. Alternatives deserving further study include serological testing for serum anti-E. coli/salivary CEACAM6 index [34] and quantification of fecal IgA-coated E. coli [35].

The choice of antibiotics is also an issue. Ciprofloxacin was selected mainly because it showed the greatest in vitro potency against E. coli replicating within macrophages [11]. It does, however, seem to have added considerably to side effects. A future trial might be confined to long-term therapy with the doxycycline–hydroxychloroquine combination. This has proved successful in Whipple’s disease [19] and Q-fever [18]. Alternative strategies targeting the altered Crohn’s disease microbiota could include bacteriophages or bacterial function-editing substrates [36].

Conclusion

Further study of long-term doxycycline plus hydroxychloroquine as an adjunct to biologic therapy in Crohn’s disease should be considered.