Statement regarding federally-sponsored research

[001] This work was supported in part by NIH-supported Cincinnati Children's Hospital Research Foundation Digestive Health Center (1P30DK078392-01). The U.S. Government may have rights in this invention.

Backgrou nd

[002] Crohn's disease (CD) is a chronic inflammatory condition of the gastrointestinal tract characterized by a relapsing and remitting course. Current evidence suggests that host genetics and microbial dysbiosis play a fundamental role in CD pathogenesis. Pediatric onset CD is the fastest growing incident age group and exhibits a more aggressive course than adult onset disease. The majority of children present with an inflammatory (non-penetrating, non- stricturing) phenotype. A sub-group rapidly progress to complicated disease behaviors with stricturing and possible bowel obstruction and/or internal penetrating fistulas often resulting in intra-abdominal sepsis. Previous reports on the natural history of CD have shown rates of complicated disease ranging from 48% to 52% at 5 years following diagnosis. Factors associated with complicated disease behaviors include age at diagnosis, ileal disease location, serologic responses to a variety of microbial antigens, and possibly cumulative genetic risk.

[003] Despite substantial progress in understanding the immune pathogenesis of CD, little is known about the precise mechanisms responsible for disease complications. Wound healing triggered by inflammation may lead to tissue repair or fibrosis depending on the balance between production and degradation of extracellular matrix (ECM) proteins. I n CD, stricturing occurs when regeneration and repair fail to restore normal tissue architecture, and bowel wall thickening leads to luminal narrowing. Internal penetration develops as a result of active transmural bowel wall inflammation with or without distal luminal narrowing. The marked heterogeneity in patient course suggests a strong host biological component with conditioning by environmental and intestinal microbial factors. [004] The impact of current therapies, specifically anti-tumor necrosis factor alpha (anti- TNFa) agents, on the natural history of disease remains unclear. Discovery of factors contributing to disease complications and predicting risk of such complications in children presenting with non-stricturing and non-penetrating CD is crucial in guiding therapeutic decisions. In 2008 the RISK (Risk Stratification and Identification of Immunogenetic and

Microbial Markers of Rapid Disease Progression in Children with Crohn's Disease) Study was initiated to prospectively characterize the natural history of newly diagnosed CD in children presenting in an uncomplicated disease state. A risk stratification model was derived based on clinical, host biology, and microbial factors defined at diagnosis as well as treatments including anti-TN Fa therapy.

[005] Evidence-based models for rapidly progressive Crohn's Disease (CD) to guide the use of expensive medical therapies such as anti-TNFa agents are urgently needed. In preparation for this study we reviewed the scientific literature regarding both population-based evidence and prospective and retrospective cohort studies which defined the natural history of disease complications in pediatric CD, and associated clinical, serologic, and genetic predictors. These prior studies were limited by relatively small sample size, precluding validation of predictors, and most did not follow patients from the time of diagnosis prospectively. While several studies developed models which included clinical, demographic, genetic, and serologic factors, none included data regarding the global pattern of gene expression in the affected gut at diagnosis, or the associated microbiota. The existing predictive models were therefore limited both in terms of validation, and insight into the biology of refractory disease.

Summary

[006] To address this need, we enrolled 913 children with CD at diagnosis at 28 sites in North America between 2008 and 2012 and completed 36 months of prospective follow-up. We derived and validated a risk stratification model based on clinical and serologic factors defined at diagnosis which has very high negative predictive value (NPV). Our comparative effectiveness analysis demonstrated that early anti-TN Fa therapy (i.e., within 90 days of diagnosis) was associated with reduced rates of internal penetrating, but not stricturing, disease complications. We detected a novel ileal ECM (Extra-Cellular Matrix) gene signature present at diagnosis associated with future stricturing complications. This improved the discriminant power of the model with respect to specificity and PPV (positive predictive value) and highlighted the importance of fibrogenesis in this sub-group who did not benefit from early anti-TN Fa therapy.

[007] The diverging rates of stricturing versus penetrating behavior in our cohort over time suggested potential differences in disease pathogenesis, and we show that early anti-TNFa will prevent internal penetrating but not stricturing complications. We provide a mechanistic basis for this observation by defining a novel ileal pro-fibrotic signature detectable at diagnosis. Collectively, these data will facilitate both appropriate treatment choices (early anti-TNFa to prevent penetrating complications) and rationale testing of novel anti-fibrotic approaches in patients at high risk for stricturing disease in future clinic trials.

Brief description of the drawings

[008] Figure 1. Development of Stricturing or Penetrating Complications during Follow-u p .

Panel A shows the observed complication-free survival probabilities during 36-month follow-up from diagnosis for stricturing (B2) or penetrating (B3) behavior in the entire cohort of 948 pediatric Crohn's Disease (CD) patients based on a competing risk model analysis. This includes 35 CD patients who either developed complications during the first 90 days after diagnosis, or lacked complete information for disease location at diagnosis, and were therefore excluded from the primary analysis. A total of 97 patients developed complications, 63 with stricturing (B2) and 34 with penetrating (B3) behavior.

[009] Panel B shows the competing risk model-based complication-free survival probabilities observed in the propensity-score matched cohort (n=382) stratified by early anti-TNFa therapy. Early ant-TNFa therapy was defined as exposure within 90 days of diagnosis, and the survival probabilities were computed among those staying complication free by 90 days. Among those treated with the early anti-TN Fa therapy, a total of 18 patients developed complications, 14 with stricturing (B2) and 4 with penetrating (B3) behavior. 23 patients developed complications, 12 with stricturing (B2) and 11 with penetrating (B3) behavior, among the matched group not treated with early anti-TNFa therapy.

[010] Figure 2. Ileal Gene Signatures Associated with Development of Disease

Complications. Panel A shows the proportion of ileal genes upregulated in patients who developed stricturing (B2, blue circles) versus penetrating (B3, red circles) complications for 19 Gene Ontology (GO) pathways significantly enriched for genes upregulated in pairwise comparisons between the subgroups (see Supplementary methods section 'Pathway analysis of Disease Complications' for more details). Names of pathways along with the total number of genes included in each pathway (in brackets) are shown in the left side of the panel. Sample sizes were n=18 (B2), n=ll (B3), and n=214 (Bl). The pathways were selected as highly significant in the enrichment analyses of B2vsBl or B3vsBl (see Tables 4-5).

[Oil] Panel B shows a combined scatterplot and density plot of log2 fold changes in ileal gene expression for patients who exhibited stricturing (B2) vs. complication-free (Bl, X-axis) and stricturing (B2) vs. penetrating (B3, Y-axis) behavior, with higher average expression in B2 either to the right or to the top (and correspondingly, higher expression in Bl to the left, or B3 to the bottom). Genes involved in ECM remodeling (n=70, GO pathway:0005201) are depicted with blue filled dots and blue lines. Background values for all genes (n=17,081) are shown with grey dots and black lines.

[012] Panel C shows a volcano plot of significance (negative logarithm of the p-value on the Y- axis) against difference in average ileal gene expression (log2 scale, X-axis) between CD patients low risk for complications who nevertheless developed stricturing behavior (B2 Low Probability) and CD patients high risk for complications who nevertheless remained complication-free (Bl Protected). Genes involved in the mitochondrial respiratory chain (n=179, GO

pathway:0022900 and GO pathway:0045333; in dark red) are almost all upregulated in "Bl Protected", whereas genes involved in remodeling of the ECM (n=68, GO pathway:0005201; in light blue) are upregulated in "B2 Low Probability.""B2 Low Probability" refers to nine patients who developed B2 despite a 3-year probability of complication from the competing risk model below the median of all B2 (average: 0-048, range: 0-029-0-072); "Bl Protected" refers to 22 patients who did not develop any complications despite being among the top 10% of 3-year probability of complication according to the competing risk model (average: 0-152, range:

0-121-0-241). The small sample size results in modest significance values, but the coherence of the two pathways is strongly indicated by the polarity of the genes in the two signatures on either side of zero.

[013] Figure 3- Consort Diagram Depicting the Study Design and Patient Outcomes. Patient outcomes were defined based on Montreal Classification: Bl - inflammatory behavior with no stricturing or luminal penetrating complications; B2 - stricturing behavior with no luminal penetrating complications; B3 - luminal penetrating behavior with or without concurrent stricturing complications.

[014] Figure 4 - Differentially Abundant Taxa associated with Crohn's Disease and the Development of disease complications. (A) Differentially abundant microbial organisms associated with diagnosis were identified through multivariate statistical analysis (MaAsLin) at the genus level. Fold changes for each genus were calculated by dividing the mean abundance in CD patients by that of non-IBD controls. Samples were analyzed concurrently, however, fold changes are displayed by sample-type to identify whether the signal is sample-type specific. All displayed genera were statistically significant with all FDR values below 0.02. Taxa in brackets represent recommended groupings by the Greengenes database based on whole genome phylogeny. The genera at the top were increased in CD patients, whereas the genera at the bottom were decreased in CD patients.

[015] (B) Two genera associated with the development of B2 were identified by contrasting CD patients that stayed Bl with those that developed B2 (MaAsLin, FDR a < 0.2). Fold changes for each genus were calculated by dividing the mean abundance in Bl patients by that of B2 patients. The top genus was increased in B2 patients compared to Bl, whereas the bottom genus was decreased.

[016] (C) Analogously to (B), patients that developed B3 were compared to patients that remained Bl. Two genera were significantly implicated in the development of B3 (MaAsLin, FDR a< 0.2).

[017] Figure 5. Risk model for disease complications and extracellular matrix remodeling (A) or respiratory chain (B) signatures provide two uncorrected classifiers of patient complications. Each circle represents the two scores, with blue filled circles representing B2 individuals. Note that just one individual develops B2 complication among the one third of patients who have an ECM PCI value < 0, and 3-year probability of B2 < 0Ό8 (yellow shaded quadrant). Similarly, children with a Respiratory Chain PCI > 0, and low probability of B2 are also protected. Since these two gene expression scores are strongly correlated, we do not include them jointly in the final model in Table 3 as the number of cases is too small to reliably distinguish their multivariate contributions.

DETAILED DESCRIPTION

[018] Abbreviations: AUC, Area under the curve; ASCA, cerevisiae antibodies; CBirl, CBirl flagellin; CI, Confidence Interval; GM-CSF, Granulocyte Macrophage Colony Stimulating Factor; HR, Hazard Ratio; IQR, interquartile range; NPV, Negative predictive value; OmpC, Escherichia coli outer membrane porin C; pANCA, Perinuclear anti-neutrophil cytoplasmic antibodies; PC, principal component; PCDAI, Pediatric Crohn's Disease Activity Index; PPV, Positive predictive value; SD, standard deviation; SNP, Single-nucleotide

polymorphism. Definitions: "early" administration of anti-TN Fa means within 90 days of diagnosis.

EXAMPLES

Methods

Study Population and Outcome Classification

[019] 1813 subjects <18 years old with suspected inflammatory bowel disease (IBD) were enrolled at 28 North American sites from November 2008 to June 2012 (ClinicalTrials.gov

Identifier: NCT00790543, Fig. 3). 402 lacked gut inflammation on endoscopy and served as non- IBD controls for the microbial and gene expression studies. Samples for microbial and gene expression studies were obtained prior to therapy. IBD diagnoses included CD in 1096, ulcerative colitis (UC) in 204, and IBD-unspecified (IBD-U) in 111. Eighteen CD subjects had incomplete information on disease location, 59 experienced complications either at the time of diagnosis (42) or within 90 days (17), and 106 lacked at least one follow-up visit and were excluded. 913 CD patients who were complication-free through 90 days after diagnosis were included in the final analysis.

[020] Patients were managed according to the dictates of their physicians, not by

standardized protocols. Early anti-TNFa exposure was defined by initiation of therapy within 90 days of diagnosis while complication-free and successful completion of both induction (3 doses infliximab, two doses adalimumab) and at least one maintenance dose. Disease behavior and location were defined based on the Montreal classification system. Stricturing disease (B2) was defined as persistent luminal narrowing with pre-stenotic dilation observed by contrast small bowel imaging. Internal penetrating disease (B3) was defined as intra-abdominal fistulizing disease resulting in intra-abdominal or pelvic abscess or fistula to an adjacent organ excluding the vagina or perianal region. Bl refers to an uncomplicated state.

Laboratory Methods and Analyses

[021] Serology: Serologic determination of pANCA, ASCA IgG, ASCA IgA, anti-CBirl, and anti- OmpC was performed at Cedars-Sinai Hospital, Los Angeles, California. GM-CSF auto-antibodies were measured at Cincinnati Children's Hospital Medical Center.

[022] Genotyping: NOD2 genotypes (rs2066844, rs2066845 and rs2066847) were extracted from previously published data or determined using TaqMan ^® SNP genotyping assays. We utilized the "score" routine available in PLI NK to generate a weighted genetic risk score using Immunochip data for 137 SNPs associated with CD and IBD.

[023] Microbial Community Profiling: DNA was isolated from ileal and rectal biopsies and stool samples and subjected to 16S rRNA amplicon sequencing. The data were subsequently reanalyzed using advanced software tools and analysis strategies.

[024] RNA Sequencing (RNASeq): Biopsies were taken from the ileum, rectum and other colonic locations. However, the results presented here only used ileal biopsies. RNA was isolated from ileal biopsies and global patterns of gene expression were determined using RNASeq as previously reported. Statistical Analysis and Reporting

[025] Sample Size: We planned to enroll 1100 subjects anticipating 9% drop-outs and 15% complication rate over three-year follow-up projected by historical cohort data (n=583). This was expected to sufficiently power identification of a total of 10 risk factors for B2 and B3 outcomes. The planned cohort was also anticipated to predict complication risk with sufficient precision estimated through split sample validation (projected SD of 3% for PPV and 1% for NPV based on 1,000 simulated data).

[026] Missing Data: In the cohort of n=913 disease outcomes were completely observed with potential risk factor data missing either by study design or in small fractions of the cohort. By study design genotyping, microbial community profiling, and RNA sequencing were conducted in different subsets of the cohort. The respective subsets with complete data were analyzed. Serology samples were not analyzable in 28 (3-1%) patients and missing cases were imputed as sero-negative to be conservative.

[027] Competing Risk Model: We considered time to complication as the response and analyzed the data using a competing risk model. Whereas standa rd Cox- proportional hazard (PH) regression is concerned with only one type of event (event vs no event), in this case the response time was defined by the two complication outcomes and three mutually exclusively defined disease behavior states (stricturing (B2)/penetrating (B3)/no event (Bl)) were considered. Therefore, a competing risk model was utilized to extend the PH model and estimate complication-specific hazards. Clinical variables tested included age at diagnosis, gender, race, disease location, disease severity measured by the Pediatric Crohn's Disease Activity Index, perianal disease, height z-score, weight z-score, and BMI z-score. Laboratory variables tested included NOD2 genotype and the CD genetic risk score, albumin, hemoglobin, erythrocyte sedimentation rate (ESR), C-reactive protein (CRP), GM-CSF auto-antibody, antimicrobial serologies, pANCA, and ileal gene expression. Ileal gene expression data for each subject was reduced to principal component 1 (PCI) for the ECM production biologic pathway prior to testing in the model. Variables selected to remain in the model included those with p- value <0.1 for either the stricturing or penetrating outcome, or in the case of isolated ileal location, prior evidence linking this factor to disease complications. [028] Propensity Score Analysis: 197 subjects received anti-TNFa within 90 days of diagnosis. Of these, 191 met the criteria for medication exposure and were included in the analysis.

Propensity of early anti-TNFa exposure was modeled by regressing the effects of baseline factors affecting anti-TNFa use (age, race, gender, disease location, perianal disease, height z- score, weight z-score, PCDAI score at diagnosis, any deep ulcer on colonoscopy, and interaction between age and PCDAI) on early anti-TNFa use. Scores representing the propensity of early anti-TNFa exposure were computed and used to match each of the patients who received early anti-TNFa with a patient who did not. We used the greedy-matching algorithm with caliper of 0.1SD and obtained n=191 matched pairs.

[029] The study was reported as per the STROBE statement (http://www.st.robe- statement.org/index.php?id=strobe-home) for observational cohort studies. R package cmprsk (Version 2.2-7) ¹⁷ was used in R 3.2.2 ¹⁸ for competing risk modeling. Descriptive statistics were computed in SAS 9.3 (SAS Institute, Cary NC) ¹⁹ .

Supplementary Methods

[030] Risk Prediction Validation: We validated the risk prediction models for the accuracy of the risk predication of complications within three years. The power for the validation depends on both the sample size and the number of complicating events, but more critically on the latter. Whereas baseline clinical/demographic data were available for the entire cohort, other data were available in different subsets. The predication accuracy measures were computed by split sample method when the sample size supported splitting data into half and using one for independent data validation. Otherwise we used leave-one-out method.

[031] Microbial Community Profiling and Statistical Analysis: DNA was isolated from ileal and rectal biopsies and stool samples and subjected to 16S rRNA amplicon sequencing. We analyzed 929 microbial samples from 625 patients (422 CD, 203 non-IBD). Samples with at least 3,000 reads were re-analyzed with a focus on noise-reduction, including quality trimming and adapter removal (trimmomatic2 - sliding window: 4bp, minimum quality: 15, minimum length: 50bp), read overlapping (PandaSeq3 - minimum overlap: 20bp) and OTU calling followed by chimeraremoval, to achieve a detailed characterization of differentially abundant genera in CD patients based on multivariate statistical analyses. Briefly, reads were dereplicated, sorted by abundance and singletons were discarded (vsearch,www.github.com/torognes/vsearch). OUT clustering at 97% identity was followed by chimera removal through comparison against the GOLD database (usearch). Subsequently, the sequencing reads from each dataset were mapped back against the OTUs and taxonomy was assigned based on the Greengenes database. The resulting OTU table was further analyzed at genus level, where each OTU was required to occur at a relative abundance of at least 0-001% across all samples and be present in at least 3 samples. Relative abundances were renormalized after filtering. Differential abundance testing (CD versus non-IBD) was done with MaAsLin including the following covariates: subject, sample location, age at diagnosis, ethnicity, gender, diagnosis and antibiotics use, where subject was treated as a random covariate. For the analysis of differentially abundant taxa associated with the development of disease complications we further included disease location and B2/B3 as covariates.

[032] RNA Sequencing (RNASeq): RNA was isolated from ileal biopsies and global patterns of gene expression were determined using RNASeq as previously reported. Biopsies were taken from the location of disease, so some heterogeneity of expression may be attributable to this source, or to differences in tissue content captured in the sample. The reported comparisons include 214 Bl stable, and 18 B2 or 11 B3 (6 of whom also had signs of stricturing) progressing patients. RNA was not available for the remaining individuals included in Table 1 (namely 621 Bl, 36 B2, and 13 B3 patients). Reads were mapped to the human genome (hgl9) with TopHat 2.0.13,6 and the number of aligned reads was quantified with HTSeq-0-6-1. Only genes with counts per million (CPM) >1 were considered and following TMM normalization differential expression was assessed in edgeR7 using the Benjamini-Hochberg procedure to calculate the false discovery rate for each comparison. The comparison of B2 relative to Bl yielded 116 and 1 genes up and down regulated, respectively (FOl.2 at FDR<0.01). Since B3 is heterogeneous and has a smaller sample size, in order to include more genes for the gene set enrichment analysis we simply adopted FC>2, yielding 137 and 152 genes were up and downregulated, respectively in B3 relative to Bl. However, this analysis is conservative and since there is considerable covariance of genes with related biological functions as seen in Figure 2B,C, there is a high false negative rate which obscures the tendency for similar pathways to be upregulated in both stricturing and penetrating disease (see Pathway analysis of Disease Complications below). Figure 2A reports the biases in a number of pathways found to be differentially expressed using the ToppFun tool in ToppGene Suite8 and DAVID Bioinformatics Database9 were used for functional annotation enrichment analyses. Statistical analysis and plotting were performed using the R software ν·3·2.

[033] Pathway analysis of Disease Complications: A variety of pairwise group comparisons involving B2 and B3 groups as defined in the propensity score analysis, as well as subsets of these individuals with penetrating only ("B3-only", 5 individuals), penetrating and structuring ("B3 + B2", 6 individuals), or stricturing only ("B2", 18 individuals), were performed. This identified 19 pathways with significant gene set enrichment at Bonferroni-corrected p<0.01, from which a representative sample was selected as follows: 9 from Table S3 based only on genes up-regulated in B2; 6 from Table S4 based only on genes up-regulated in all B3 (11 individuals); 3 from the comparison of B2 and "B3-only" in Table S5, and the anti-TN Fa pathway (given its centrality to the analyses). Figure 2A shows the bias for each pathway simply as the proportion of genes independently annotated to the pathway in the GO database, which are higher in the B2 or "all B3" groups. This analysis is conservative since the comparison of "B3- only" to "B2" would result in even more extreme biases, given that the "B2+B3" profiles are more B2-like. For consistency with the B2 and B3 definition in the text following the Montreal recommendations, where all individuals who develop penetrating disease are classified as B3, we adopted a liberal 2-fold differential expression inclusion criterion for the B3 vs Bl comparison (compared with FDR < 0.01 for much more highly powered B2 vs Bl comparison) in Table S4. Each of these pathways is significant in the "B3-only" vs Bl comparison.

Results

Progression to complications

[034] Characteristics of the subjects included in the final analysis are shown in Table 1.

Median (Interquartile range, IQR) age at diagnosis was 12.4 (10Ό-14-7) years, 565 (62%) were males, 681(75%) were white, and 121 (13%) African-American or of mixed races. Stricturing disease was associated with ASCA, CBirl, and GM-CSF seropositivity and penetrating disease was associated with older age ("older" being > 12.9 years of age), African American race, as well as ASCA and CBirl seropositivity. Patients with perianal involvement were more likely to be male, to receive anti-TNFa therapy within 90 days of diagnosis, and to be positive for both antimicrobial and GM-CSF auto-antibody serologies. Survival curves illustrating the time to complicating disease behaviors are shown in Figure 1A. A total of 54 subjects experienced stricturing complications and 24 subjects penetrating complications during follow-up

(p=0-0007).

Table 1. Clinical and Demographic Characteristics of the Study Cohort Stratified by Disease

Behavior during Follow-up.

Disease Behavior at Last Follow-up

(Montreal Classification System)*

Inflammatory Stricturing (B2) Penetrating (B3)

Number of Subjects 835 54 24

Demographics and Follow-up

Median age at diagnosis-years (IQR) 12-3 (9-9-14-5) 12-9 (10-6-15-1)15-6 (12-9-

Female gender (%) 316 (37-8) 19 (35-2) 13 (54-2)

African American or mixed race (%) 103 (12-3) 9 (16-7) 9 (37-5) ^††

Median duration of follow-up-months (IQR) 47 (36-55) 41 (36-50) ^† 40 (34-49) ^†

Median time to behavior change-days (IQR) - 520 (301-722) 497 (260-680) Disease Activity and Treatment Exposures

Moderate-to-severe disease activity (%) 388 (46-5) 23 (42-6) 12 (50-0)

Median height z-score (IQR) -0-25 (-0-94-0-45) -0-6 (-1-34- -0-35 (-1-23-

Median BMI z-score (IQR) -0-66 (-1-63-0-15) -0-85 (-1-88- -0.76 (-1-84-

Anti-TN Fa within 90 days of diagnosis (%) 173 (20-7) 14 (25-9) 4 (16-7)

Immune-modulator within 90 days (%) 378 (45-3) 21 (38-9) 14 (58-3)

Small bowel imaging within 6 months (%) 608 (72-8) 47 (87-0) ^† 19 (79-2)

Disease Location at Diagnosis

Isolated Terminal Ileum +/- cecum (%) 166 (19-9) 16 (29-6) 7 (29-2)

Isolated Colonic (%) 210 (25-1) 10 (18-5) 3 (12-5) lleo-colonic (%) 459 (55) 28 (51-9) 14 (58-3)

Perianal disease (%) 115 (13-8) 7 (13.0) 4 (16-7)

Serological Reactivity Status at Diagnosis ASCA IgA (%) 182 (21-8) 22 (40-7) ^†† 14 (58-3) ASCA IgG (%) 182 (21-8) in 35 2) ^' 9 (37-5)

CBirl (%) 293 (35-1) 32 (59-3) ^††† 16 (66-7) ^††

GM-CSF auto-antibody ( ) 381 (45-6) 35 (64-8) ^'■ 14 (58-3)

OmpC (%) 54 (6-47) 4 (7-41) 2 (8-33)

DANCA (%) 129 (15-4) 5 (9-26) 3 (12-5)

*Montreal Classification ⁸ of Disease Behavior: Bl- inflammatory behavior with no stricturing or luminal penetrating complications; B2 - stricturing behavior with no luminal penetrating complications; B3 - luminal penetrating behavior with or without concurrent stricturing complications.† p-value <0.05 for comparisons of B2 vs. Bl and B3 vs. Bl.†† p-value < 0-01 for comparisons of B2 vs. Bl and B3 vs. Bl.††† p-value < 0-001 for comparisons of B2 vs. Bl and B3 vs. Bl.

[035] Within 90 days of diagnosis, 191/913 subjects (20-9% of total) received anti-TNFa therapy (180 (94%) infliximab, 11 (6%) adalimumab) and 413 were treated with an

immunomodulator (IM)) (thiopurine (TP) or methotrexate (MTX)), out of which 32 were also treated with anti-TNFa. Propensity-score matching was used to address differences in baseline patient characteristics, and a sample of n=191 pairs of one patient who received early anti TNFa and one who did not was obtained (Table 2).

Table 2. Clinical and Demographic Characteristics of the Study Cohort Stratified by Perianal Disease Status at Baseline in the Overall Cohort.

Disease Location at Diagnosis

Isolated Terminal Ileum +/- cecum (%) 163 (20.7) 26 (20.6)

Isolated Colonic (%) 191 (24.3) 32 (25.4)

lleo-colonic (%) 433 (55.0) 68 (54.0)

Serological Reactivity Status at Diagnosis

ASCA IgA (%) 169 (21.5) 49 (38.9)†††

ASCA IgG (%) 163 (20.7) 47 (37.3)†††

CBirl (%) 282 (35.8) 59 (46.8)†

GM-CSF auto-antibody (%) 353 (44.9) 77 (61.1)†††

OmpC (%) 45 ( 5.7) 15 (11.9)††

pANCA (%) 118 (15.0) 19 (15.1)

† p-value < 0.05,†† p-value < 0 1,††† p-value < 0Ό01 for perianal vs no perianal involvement

[036] Survival curve analysis of this matched cohort demonstrated similar progression to stricturing behavior in patients irrespective of early anti-TNFa exposure (Figure IB). By comparison, progression to penetrating behavior was reduced three-fold in those who received early anti-TNFa, although this did not reach significance in the unadjusted analysis (ρ=0 675). We were limited by the small sample size of patients who received combination therapy (both anti-TNF and IMM, n=32) in performing further analysis of this important subgroup. Response to early anti-TN Fa defined by achieving steroid-free remission six months after diagnosis was achieved in 124(71%) of subjects with available data for this outcome. We did not observe a difference in rates of B2 or B3 complications in month six anti-TNFa responders versus non-responders, although the small sample size of these sub-groups precluded drawing firm conclusions.

Validation of a Competing Risk Model for Disease Complications

[037] The competing risk model utilizing clinical and serologic variables is shown in Table 2. Neither NOD2 genotype nor a polygenic CD risk score reached significance (ρ=0·14 and ρ=0·46 for stricturing disease; ρ=0·39 and ρ=0·27 for penetrating disease). Ileal disease location and ASCA IgA and CBirl seropositivity were associated with stricturing behavior. Older age (i.e., > 12.9 years of age), African American race, and ASCA IgA and CBirl seropositivity were associated with penetrating behavior. We tested the discriminant power of the model by randomly splitting the cohort into equal test and validation groups and averaging the performance over 1000 iterations. These test characteristics included a sensitivity of 66% (95% CI, 51-82), specificity of 63% (95% CI, 55-71), PPV of 14% (95% CI, 12-17), and NPV of 95% (95% CT, ^" 4-9T) T e

differences in factors in the competing risk model. Early anti-TNFa was not associated with a reduction in stricturing behavior (HR (95%CI): 1-13 (0.51-2.51), ρ=0·76) but was associated with a reduction in penetrating behavior (HR (95%CI): 0-30 (0-10, 0-89), p=0.0296) (Table 3).

Table 3. Competing Risk Model for Disease Complications and Early Anti-TNFa Comparative Effectiveness Analysis.

Competing Risk Model Stricturing Behavior (B2) Penetrating Behavior (B3)

HR (95%CI> D-value HR ( 95%C0 D-value

Overall Cohort (N=913)*

Age at diagnosis 1-07 (0-97-1-17) 0-16 1-45 (1-17-1-8) 0.0008

African American race 1-08 (0-52-2-22) 0-84 3-19 (1-39-7-31) 0-0061

Isolated Ileal Location (LI) 1-60 (0-88-2-91) 0 12 1-23 (0-51-2-95) 0-64

ASCA IgA + 1- 69 (0-94-3-07) 0-0816 2- 68 (1-19-6-04) 0-0171 CBirl + 2- 30 (1.26-4.20) 0-0070 3- 01 (1-31-6-93) 0-0097

* The test characteristics (95%CI) for the competing risk model in the overall cohort n=913 were: Area Under the Receiver Operator Characteristic Curve (AUC), 0-7 (0-64-0-76); Sensitivity, 66% (51-82); Specificity, 63% (55-71); PPV, 14% (12-17); NPV, 95% (94-97); prevalence of complications, 8-5%. These characteristics were calculated from 1000 random split samples, each using one half to validate predictions based on the other half. The cutoff point set to balance sensitivity with specificity was 0-077. NPV denotes negative predictive value and PPV positive predictive value. In contrast the test characteristics observed for the model based on clinical factors alone model were: AUC, 0-6 (0-56-0-69); Sensitivity, 56% (38-67); Specificity 63% (56-71); PPV, 12% (10-15); NPV 94% (92-95).

Distinct Microbiota are Associated with Disease Complications

[038] We identified 14 genera associated with pediatric CD (Figure 4A). In addition to organisms previously implicated including Clostridiales, Pasteurellaceae, Veillonellaceae, Erysipelotrichaceae and Bacteroidales, we identified new organisms, such as Campylobacter, Akkermansia, Collinsella and Desulfovibrio. The largest increase was detected for Aggregatibacter, while the greatest decrease was observed for Roseburia. Several organisms were associated with disease complications (Figure 4B and 4C). Rothia and Ruminococcus were implicated in the development of strictures. Collinsella was elevated in patients who developed penetrating disease (B3) whereas Veillonella was increased specifically in the ileum. However, the sub-cohort with observed microbial data differed from the whole cohort in key baseline clinical features, and so the individual taxa were not incorporated in the risk prediction model for further evaluation.

Ileal Gene Expression Improves the Discriminant Power of the Model

[039] While there was considerable heterogeneity in the ileal global pattern of gene expression, comparisons between groups revealed significant differences in gene expression. Genes regulating ECM accumulation were induced at diagnosis in patients who developed strictures (Table 4) yet genes regulating the acute inflammatory response to microbes were induced in those developing penetrating disease (Table 5 and 6). The balance between antimicrobial acute inflammatory and ECM accumulation pathways within patients who developed penetrating (B3) versus stricturing (B2) complications, respectively, is shown in Figure 2A. As illustrated in Figure 2B, the ECM structural constituent molecular function was induced to a greater degree in patients who developed stricturing complications relative to both those who remained complication-free (Bl), and those progressing to penetrating disease.

Table 4. Clinical and Demographic Characteristics of the Propensity-Score Matched Cohort.

Isolated Colonic (%) 45 (23-6%) 49 (25-7%) lleo-colonic (%) 106 (55-5%) 103 (53-9%)

Perianal disease (%) 49 (25-7%) 49 (25-7%)

Moderate-to-severe disease activity (%) 111 (58-1%) 116 (60-7%)

Median height z-score (IQR) -1-1 (-0-34-0-24) -1-14 (-0-29-0-45)

Median BMI z-score (IQR) -1-84 (-0-95-0-05) -1-82 (-0-96-0-06)

Median (IQR) albumin, g/dL 3-3 (2-8-3-8) 3-3 (2-9-3-8)

Median (IQR) CRP, mg/L 1-6 (4-0-14-3) 1-9 (4-7-12)

Serological Reactivity Status at Diagnosis

ASCA IgA (%) 54 (28-3%) 57 (29-8%)

ASCA IgG (%) 49 (25-7%) 55 (28-8%)

CBirl (%) 70 (36-6%) 79 (41-4%)

GM-CSF auto-antibody (%) 77 (40-3%) 96 (50-3%)

OmpC (%) 14 (7-33%) 19 (9-95%) pANCA (%) 33 (17-3%) 29 (15-2%)

Small bowel imaging within 6 mos (%) 132 (69-1%) 148 (77-5%)

Early IM Therapy 94 (49-2%) 32 (16-8%)

Outcome by 36 mos after diagnosis (%)

Remained complication-free 168 (88-0%) 173 (90-6%)

Developed stricturing 12 (6-3%) 14 (7-3%) complication

Developed penetrating 11 (5-8%) 4 (2-1%) complication

[040] Table 5. Pathway Enrichment Analysis of Ileal Genes Upregulated in Patients Who Developed Stricturing Complications. a) Molecular functions:

b) Biological processes:

Gene Ontology (GO) terms detected with Toppfun (ToppGene Suite8 ) as significantly enriched among the 116 loci upregulated in B2 vs. Bl (fold change >1·2 at FDR<0-01). All terms detected as enriched at Bonferroni corrected P<0.05 (or top 10 only if more than ten are detected) are shown for both GO Molecular function and Biological process categories. For each GO term, details about the number of upregulated loci and total number of loci annotated are provided (only terms with =15 loci are considered). Pathway names in bold are included in Figure 2.

[041] Table 6. Pathway Enrichment Analysis of Ileal Genes Upregulated in Patients Who Developed Penetrating Complications. a) Molecular functions:

b) Biological processes:

Gene Ontology (GO) terms detected with Toppfun (ToppGene Suite8 ) as significantly enriched among the 137 loci detected as upregulated in B3 vs. Bl (fold change >2). All terms detected as enriched at Bonferroni corrected Ρ<0·05 (or top 10 only if more than ten are detected) are shown for both GO Molecular function and Biological process categories. For each GO term, details about the number of upregulated loci and total number of loci annotated are provided (only terms with =15 loci are considered). Pathway names in bold are included in Figure 2.

[042] We then tested for differences in ileal gene expression between patients who were predicted to be at low risk for a stricturing complication, but nevertheless did progress, and those who were predicted to be at high risk for stricturing disease but remained complication- free at 36 months. This analysis identified enrichment for a mitochondrial function gene signature in patients in the at-risk group remaining complication free (Figure 2C, Table 7).

Table 7. Pathway Enrichment Analysis of Ileal Genes Upregulated in Patients Who Developed Penetrating Complications compared to Patients Who Developed Stricturing Complications.

Biological processes:

GO:0050727 regulation of inflammatory 8 (345) 2E-03 IL23A,GGT1,S100A12 response

GO:0030593 neutrophil chemotaxis 5 (92) 4E-03 S100A8,SAA1,IL23A

GO:0050832 defense response to 4 (42) 5E-03 ADAM,S100A9,S100A12 fungus

GO:0006954 inflammatory response 10 (714) 7E-03 VNN1,TFF2,HP

GO:1990266 neutrophil migration 5 (103) 7E-03 S100A9,S100A12,IL23A

Gene Ontology (GO) terms detected with Toppfun (ToppGene Suite8 ) as significantly enriched among the 52 loci detected as upregulated in B3-only vs. B2 (fold change >1.2 at P<0.01). All terms detected as enriched at Bonferroni corrected Ρ<0·05 (or top 10 only if more than ten are detected) are shown for GO Biological process category. For each GO term, details about the number of upregulated loci and total number of loci annotated are provided (only terms with =15 loci are considered). Pathway names in bold are included in Figure 2.

[043] Conversely, we also observed enhancement of the ECM gene signature in predicted low risk patients who nevertheless progressed to a stricture (Figure 2C, Table 8).

Table 8. Pathway Enrichment Analysis of Ileal Genes Downregulated in B2 "Low Probability" Compared to Bl "Protected"

Molecular functions

b) Biological processes: GO accession Pathway Name Genes Bonferroni Example genes

upregulated

[Genes in

pathway)

GO:0022904 respiratory 44 (131) 9E-43 COX6A, NDUFA13, COQ9 electron transport

chain

GO:0022900 electron transport 44 (134) 3E-42 CYC1,ETFA1, UQCR10 chain

GO:0045333 cellular respiration 47 (203) 2E-37 SLC37A2,C0X8A,PPARGC1A

GO:0055114 oxidation- 83 (1107) 6E-30 DBI,ECl,DDO

reduction process

GO:0006119 oxidative 28 (91) 6E-25 ATP5D,COX5A,NDUFS4 phosphorylation

GO:0015980 energy derivation 48 (393) 9E-25 ACADVL,SUCLG1,ETFB by oxidation of

organic compounds

GO:0042775 mitochondrial ATP 25 (67) 2E-24 UQCC3,UQCRQ,COQ9 synthesis coupled

electron transport

GO:0042773 ATP synthesis 25 (69) 5E-24 NDUFA3, NDUFCl, COX6C coupled electron

transport

GO:0006091 generation of 50 (488) 2E-22 HCHD10,CYC1,ATP5D precursor

metabolites and

energy

GO:0046034 ATP metabolic 36 (225) 6E-22 ATP5G1,ATP5I,TSP0

process

Gene Ontology (GO) terms detected with Toppfun (ToppGene Suite8 ) as significantly enriched among the 345 loci detected as downregulated in B2 "Low Probability" compared to Bl "Protected" (fold change <0·83 at FDR<0-05). All terms detected as enriched at Bonferroni corrected Ρ<0·05 (or top 10 only if more than ten are detected) are shown for both GO

Molecular function and Biological process categories. For each GO term, details about the number of downregulated loci and total number of loci annotated are provided (only terms with =15 loci are considered).

[044] The first principal component of the ECM structural constituent molecular function gene signature was associated with stricturing behavior in the risk model (HR (95% CI): 1-70

(1-12,2-57), ρ=0 120). Leave-one-out cross validation showed that the model performance increased as this gene signature was added, with an AUC of 0-72, sensitivity of 69% and specificity of 71% (Table 9). These improved statistics were evident despite having one quarter of the sample size of the full cohort available for gene expression, in large part because the transcriptome and serology risk factors were almost orthogonal (Figure 5). In fact, eight additional patients who experienced a B2 complication were classified as high risk based on the ECM gene signature (lower right sector of Figure 5A) who otherwise would have been classified as low risk in the model based only on clinical factors including ileal location and serology.

Table 9. Competing Risk Model Including the Extra-Cellular Matrix Gene Signature

Sub-cohort (n=243)* Stricturing Behavior (B2) Penetrating Behavior (B3)

HR (95%CI) p-value HR (95%CI) p-value

Age at diagnosis 1-07 (0-91-1-27) 0-42 1-45 (0-98-2-14) 0-0606

African American race 0-3 (0-04-2-47) 0-27 2-31 (0-4-13-27) 0-35

Isolated Ileal Location 1-09 (0-39-2-99) 0-87 1-36 (0-37-4-93) 0-64

ASCA IgA + 1-48 (0-58-3-75) 0-41 2-92 (0-81-10-48) 0-10

CBirl + 2-14 (0-84-5-44) 0-11 7-99 (1-89-33-77) 0-0047

ECM gene signature 1-7 (1-12-2-57) 0-0120 1-21 (0-53-2-73) 0-65

The test characteristics for the competing risk model excluding the ECM gene signature within the sub-cohort of 243 patients with ileal gene expression data were: Area Under the Receiver Operator Characteristic Curve (AUC), 0-66; Sensitivity, 69%; Specificity, 66%; PPV, 22%; NPV, 94%; prevalence of complications was 11-9% (7.4% for B2; 4.5% for B3). The addition of the ECM gene 541 signature yielded the following test characteristics: AUC, 0-72; Sensitivity, 69%; Specificity, 71%; PPV, 24%; NPV, 94%. The discriminant power was computed via leave-one-out cross-validation. ECM: extra-cellular matrix.

Discussion

[045] Patients with CD typically exhibit an inflammatory disease behavior at diagnosis, with some then progressing to stricturing or internal penetrating complications. Despite increased use of anti-TNFa therapy over the past decade, a population-level decline in rates of obstructive complications and surgeries has not been observed. While a Danish cohort study did identify a reduction in CD surgeries from 1979-2011, an association with specific medications was not established. Disease complications, in particular penetrating ones, account for significantly higher health care costs. Additionally, anti-TNFa therapy itself is costly. Therefore, models to predict which CD patients are at highest risk of complications and estimates of the relative benefit of early anti-TNFa exposure, are urgently needed. [046] We found that older age, African-American race, ileal disease location, and ASCA and CBirl seropositivity were associated with disease complications. The strongest effect was observed for CBirl seropositivity to bacterial flagellin, which is induced at an early age, and may be involved in CD pathogenesis. Flagellin is a dominant antigen driving mucosal T cell responses in CD, and both flagellin-specific T cells and anti-f lagellin antibodies exacerbate inflammation in murine colitis. We therefore tested for an association between ileal microbiota and

complications. We identified 14 genera associated with pediatric CD. The largest increase was detected for Aggregatibacter, a bacterial taxon associated with a decrease in mucosal pattern recognition receptors.

[047] The greatest decrease was observed for Roseburia, including butyrate-producing bacteria which promote the epithelial barrier. Distinct taxa were in turn associated with progression to complications, suggesting a role in modulating host biology. These observations will require validation to determine their utility in a predictive model.

[048] Largely because of the low prevalence of complications in the cohort, the PPV of the clinical and serologic model was low. We hypothesized that the ileal global pattern of gene expression, as an integrated read-out for local host biology, might improve the specificity and PPV of the model. The mucosa overlying established CD strictures exhibits a pro-fibrotic pattern of gene expression. It was unknown whether ileal gene signatures would reveal these processes at an uncomplicated stage of disease. We detected a pronounced ECM gene signature in patients who ultimately progressed to strictures. This gene signature was associated with stricturing complications and improved the discriminant power of the model with respect to specificity and PPV. The ECM signature detected in the stricturing sub-group may inform more efficient enrollment into trials of anti-fibrotic therapies.

[049] This study was ideally suited to test the comparative effectiveness of early anti-TN Fa in reducing complication rates. Early anti-TNFa exposure was defined in a per-protocol, rather than intent-to-treat, manner, to assess the biologic effect in the matched groups. However, only six subjects who received anti-TNFa within 90 days did not meet the per-protocol exposure definition. We first conducted propensity-score matching to account for clinical factors associated with early anti-TNFa use, and then adjusted for risk factors for complications. We observed a substantial reduction in penetrating, but not fibrostenotic complications, with early anti-TNFa in the propensity-matched cohort. This is consistent with prior reports, although only for penetrating complications. Therefore, earlier introduction of anti-TNFa may reduce progression to internal abscesses and fistulas, with attendant high morbidity and costs.

[050] There is significant morbidity associated with CD complications and hence tools that help communicate the risk of the disease itself allow for more informed risk benefit discussions about therapies. We have conducted a large multi-center inception cohort study and derived and validated a prognostic model for disease complications which is suitable for use in clinical practice. Older age at diagnosis, African-American race, ileal disease location, and ASCA and CBirl seropositivity were associated with increased risk for disease complications. These patients may be prioritized for early anti-TNFa therapy, with a benefit of reducing internal penetrating complications which account for high morbidity and health care costs in CD.

Conversely, the high NPV of the model may be used to classify patients at low risk for complications with a high degree of confidence. Based on the validated competing risk model including clinical and serological factors, n=108 (56.5%) patients who received early anti-TNFa were projected at low risk and would have been de-prioritized for this approach if the treatment decision was based on the risk model. For these patients, treatment decisions may be guided by other considerations including steroid dependent or refractory inflammatory disease behavior, or growth failure. In this regard, our prior report in the RISK cohort demonstrated that early anti-TNFa therapy was associated with higher rates of steroid-free remission and improved growth one year after diagnosis. We now extend those findings by demonstrating a reduction in rates of penetrating complications with early anti-TNFa. The clinical utility and PPV of our risk model was improved by the addition of ileal gene signatures. Collectively these results advance our understanding of the pathogenesis of disease complications and inform more personalized approaches for children newly diagnosed with Crohn's disease. anti-TNFa Comparative Effectiveness Analysis of Propensity Score Matched Cohort

Age at diagnosis 143 (0-97-1-31) 041 1-37 (1-03-1-81) 0-0278

African American race 1-25 I ^• 43-3- 63) 0-68 3-02 (0-97-9-39) 0-0555

Isolated Ileal Location (LI) 1-66 (0-65-4-26) 0-29 1-26 (0-36-4-43) 0-72

7 -R7 M -71 -fi u-R o?__\ J D-D1 fi^ 7 -DQ ίΠ·71 7 \ ΠΊ R

CBirl + 1-52 (0-63-3-7) 0-35 4-82 (1-53-15-2) 0-0072

Earlv anti-TNFa 1-13 (0-51-2-51) 0-76 0-30 (0-10-0-89) 0-0296

** Propensity of early anti-TNFa therapy use within three months of diagnosis in individual patients was analyzed and estimated propensity scores were used to obtain n=191 matched pairs, each consisting of one patient who received early anti-TNFa therapy and one patient who did not with similar baseline characteristics. Prevalence of complications was 10-7% (6-8% for B2 and 3-9% for B3).