Personalized AZithromycin/metronidAZole + dietary therapy in pediatric Crohn disease: Results of a pilot study

Current treatment strategies of Crohn's disease (CD) mainly focus on suppressing inflammation by modulating the immune system.1, 2 In keeping with recent European Crohn's and Colitis Organisation - European Society for Paediatric Gastroenterology Hepatology and Nutrition guidelines, most pediatric CD patients require starting of biologics as, or soon after, induction of remission to mitigate the impact of disease while trying to optimize vaccination status before starting immune suppression.1 Changes in gut microbiome composition have been linked to CD development, disease course, and ability to achieve and maintain remission.3, 4 We have recently characterized the changes of the CD microbiome associated with remission induced by nutritional therapy with the Crohn's disease exclusion diet (CDED) with partial enteral nutrition (PEN).5, 6 Nutritional therapy is also associated with changes in the microbiome towards more healthy controls.5, 7 In turn, treatment with antibiotics (e.g., azithromycin + metronidazole) has previously been shown to be useful in altering the microbiome to induce remission in CD patients.8 By using Bayesian methods, Dunn and colleagues previously identified community-level microbiome signatures predictive of CD remission status at 6 months postnutritional-therapy using only pretreatment stool samples (File S1). We have applied this approach with the aim of predicting treatment success over a range of treatment modalities within the previously published RISK cohort.9-11 Here, we report the results of a pilot feasibility trial prospectively integrating this prediction model in the early treatment of children with mild-to-moderate pediatric CD with nutritional therapy. This study was approved by METC-AMC (2020-803) and CCMO (NL71847.018.19). We designed a multicenter, randomized, controlled, open-label, add-on pilot trial in 20 pediatric patients (3–17 years) with mild-to-moderate CD (10 < PCDAI ≤ 37.5) from August 2021 to August 2022. We used our pre-trained prediction model with the aim of predicting disease outcome at 12 months within the first 4 weeks of treatment (PAZAZ study; clinicaltrials.gov: NCT04186247).12 We hypothesized that patients with predicted relapse-associated signatures would benefit from add-on targeted antibiotic treatment following induction treatment. After 4 weeks of nutritional therapy, patients were allocated to further treatment groups based on their baseline microbiome signature (Figure 1). We chose the timepoint at Week 4, as samples were sent from the Netherlands to Canada and required processing and bioinformatic analysis (see below and File S1). Relapse-associated microbiome signatures were evaluated using 16S rRNA gene sequencing and our Bayesian predictive model (BioMiCo—File S1) based on baseline stool of included children. All children received CDED + PEN from Week 0. At Week 4, patients in remission with baseline relapse-associated signatures (group A) were scheduled to be randomized to CDED+antibiotics (A2) or CDED + PEN alone (A1). Patients in remission without this signature continued CDED + PEN alone (B). Patients with active disease after 4 weeks received CDED+antibiotics regardless of their baseline microbiome signature (C). Add-on therapy for patients in groups A2 and C consisted of a combination of azithromycin 7.5 mg/kg (Weeks 4–8: 5 days/week; Weeks 9–12: 3 days/week) with metronidazole 20 mg/kg/day (Weeks 4–12).13 The primary outcome of this pilot study was to evaluate the feasibility of treatment allocation based on baseline microbiome sequencing results determined at a central lab (Dalhousie University, Canada). In addition, we aimed to describe the efficacy of this personalized treatment strategy (in addition to standard-of-care treatment) in sustaining remission up to 52 weeks (PCDAI ≤ 10, no need for re-induction) in patients with a relapse-associated microbiome signature. We generated 16S rDNA from the V4V5 region according to the SOP in Comeau et al 2017,14 with updates from Comeau and Kwawukume.15 Sequence data were processed using QIIME2 following the exact same procedure as outlined above for the existing data, including VSEARCH to join paired end reads, and deblur with trim length set to 252 so that the same sequence lengths was used to denoise the reads into ASVs and assign taxonomy. Using the final trained model (generated previously from the truly treatment naïve RISK dataset of responders and nonresponders, see Figures S1 and S2) and the ASVs identified in each pilot pediatric RCT sample as the test data, BioMiCo was used to assign responder/non-responder status to each RCT sample. This testing phase of BioMiCo was run for 1000 iterations after a 100 iteration burn-in period.16 We integrated the trained prediction model into a pilot clinical trial to study feasibility of microbiome-stratification and to assess if we could optimize nutritional induction treatment (using CDED + PEN) by additional microbiome-modulation (with azithromycin + metronidazole) in patients predicted to be at risk for relapse within 1 year after start of treatment. We included 13 patients at the Amsterdam UMC within our pilot before the study was stopped prematurely by the Data Safety Monitoring Board because of reaching our primary feasibility endpoint, as is further described below. Our primary feasibility endpoint to obtain timely microbiome baseline results within 4 weeks after start of treatment, was reached in 12/13 patients (92%). Of note, the missing result was due to the low sequencing quality of the sample (available at Week 4) which had to be re-sequenced, with results becoming available at Week 5. Of the 13 included children, n = 6 reached remission at Week 4 with CDED + PEN. Of those, none of these patients carried the relapse-associated microbiome according to our risk model. There were n = 6 primary nonresponders who failed to reach remission by week 4 and entered group C regardless of their microbiome results. Of those, 2/6 had a relapse-associated microbiome according to our model. Both started with additional antibiotics, of which one patient achieved clinical remission and another was escalated to anti-TNF at Week 8 because disease activity persisted. Of the remaining primary non-responders, 4/6 did not carry the high-risk signature. One patient dropped out because the antibiotics were not available in her dosage at that time, one patient continued CDED + PEN only at the discretion of the treating physician due to significant clinical response to treatment (Δ PCDAI 22.5; total PCDAI 12.5 at 4 weeks) and two decided to leave the study because they didn't want to continue CDED and switched/escalated to prednisone/anti-TNF. One participant dropped out of the study within 1 week after inclusion, and therefore, the clinical condition at Week 4 was unknown, but they had a high-risk microbiome signature. None of the subjects with relapse-associated microbiome entered group A (randomization arm) of the study. An overview of included patients, timely microbiome results, predicted outcomes, and subsequent treatment groups is found in Table 1. The study was stopped prematurely in consultation with the Data Safety Monitoring Board due to reaching the feasibility endpoint. Furthermore, patient distribution at that timepoint across treatment groups showed that a full-scale trial evaluating the efficacy would not be feasible with this study design as carriers of the relapse-associated microbiome signature did not respond to nutritional induction therapy within the first 4 weeks of treatment. At Week 52, only three children remained in the study. All other participants had either declined to continue with the CDED + PEN regimen or required escalation of medical therapy. One patient, classified as a predicted and clinical non-responder at Week 4 (assigned to Group C and treated with azithromycin/metronidazole), achieved clinical remission by Week 52, with a fecal calprotectin level of 36 μg/g. However, this patient had required an additional 3-week course of antibiotics after Week 24 (while following CDED Phase 3). Mesalazine and a tapering course of budesonide were initiated after Week 36, alongside improved adherence to CDED Phase 2. The second patient, a predicted responder at Week 4, was in clinical remission at week 52 despite an elevated calprotectin level of 534 μg/g. This patient had initiated mesalazine and azathioprine after Week 4, due to persistently elevated calprotectin levels and clinical symptoms during CDED Phase 2. The third patient, also a predicted responder at Week 4, achieved clinical remission during CDED + PEN Phase 1 and remained in remission at Week 52 with a calprotectin level of 100 μg/g. The patient had continued with CDED phase 3 until Week 24 in combination with methotrexate, which had been started at the end of CDED phase 2 due to increasing symptoms. Adverse events were mild and related to disease activity (anemia n = 2, which resolved after iron infusion) or treatment (headache after iron infusion which resolved spontaneously, nausea after MTX for which a split MTX dose over 2 days was prescribed). Azithromycin/metronidazole were possibly related to a change in taste (n = 1) and dizziness (n = 1), both resolved. We report the results of our pilot feasibility study in pediatric CD using microbiome-based prediction prospectively for the first time. By centrally assessing the microbiome (Dalhousie, Canada) based on truly treatment-naïve samples, we aimed to standardize extraction, data preparation and analysis of a complex microbiome community in time for a treatment decision at Week 4. We met our microbiome-assessment feasibility target in 12/13 children at Week 4. Our study design of positioning antibiotics as add-on therapy to help sustain remission showed to be not feasible: children with mild-to-moderate CD with a high-risk-of-relapse microbiome did not achieve remission by week 4 with nutritional therapy. This may suggest that the success of (in this case) nutritional therapy may be in part the result of the microbiome community present, and associated with metabolomic profiles described after starting our pilot study.17, 18 However, this pilot study primarily demonstrated feasibility of sequencing and classification turnaround, and was not powered to study efficacy of microbiome-guided therapy. In taking this first step toward personalized microbiome-informed therapy, we have moved beyond post-hoc prediction modeling in retrospective analysis and have shown the feasibility of microbiome-based treatment allocation for further clinical studies. The authors would like to thank all participating children and their families, as well as the many colleagues in pediatric gastroenterology and pediatric nutrition who made the PAZAZ pilot study possible (Dr. Koot, Dr. Tabbers, Prof. Benninga, Dr. Kindermann, and Dr. Van Wijk) at Amsterdam UMC and Prof. Rashid at the IWK Health Centre, Halifax, NS, Canada). We want to thank Prof. Mel Heyman (University of California San Francisco) and Dr. Whitney Sunseri (Pittsburgh, PA, USA), Prof. Jeffrey Hyams (Hartford, CT, USA), Prof. Subra Kugathasan (Atlanta, GA, USA), Prof. Dror Shouval (Petah Tikva, Israel), Prof. Arie Levine (Holon, Israel) and Mrs. Rotem Sigall Boneh (Holon, Israel) for their support at the beginning of the study. In addition, the authors want to express their gratitude to Prof. David Wilson (Edinburgh, UK), Dr. Amit Assa (Jerusalem, Israel), Prof. Michael Rosen (Stanford, CA, USA) and Prof. Lee Denson (Cincinnati, OH, USA) for taking place in the data safety monitoring board (DSMB) of our pilot study. We also want to thank Ms Monica Coudurier (UNC) for her invaluable support in reporting the trial results to clinicaltrials.gov. Johan Van Limbergen was supported by CIHR-SPOR-Chronic Diseases grant (Inflammation, Microbiome, and Alimentation: Gastro-Intestinal and Neuropsychiatric Effects: the IMAGINE-SPOR chronic disease network), the Canadian Institutes of Health Research (CIHR)-Canadian Association of Gastroenterology-Crohn's Colitis Canada New Investigator Award (2015–2019), Crohn and Colitis Foundation of America, Pro-Kiids award number 585718, by the Wetenschappelijke Adviesraad Emma Kinderziekenhuis and by a Health Holland TKI grant in partnership with Nestlé Health Sciences (TKI-LSH-ADT-2021-AMC-26344). Charlotte M. Verburgt MD PhD, Nikki van der Kruk MD, Katherine A. Dunn PhD, Joseph P. Bielawski PhD, Anthony R. Otley MD, Tim de Meij MD PhD, Francisco Sylvester PhD, Andre M. Comeau PhD, Morgan G. I. Langille PhD, Wouter J de Jonge PhD, and Johan E. Van Limbergen MD PhD. The authors declare no conflicts of interest. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.