Serum Analyte Profiles Associated With Crohn’s Disease and Disease Location

Abstract

Background

Crohn’s disease (CD) can affect any segment of the digestive tract but is most often localized in the ileal, ileocolonic, and colorectal regions of the intestines. It is believed that the chronic inflammation in CD is a result of an imbalance between the epithelial barrier, the immune system, and the intestinal microbiota. The aim of the study was to identify circulating markers associated with CD and/or disease location in CD patients.

Methods

We tested 49 cytokines, chemokines, and growth factors in serum samples from 300 patients with CD and 300 controls. After quality control, analyte levels were tested for association with CD and disease location.

Results

We identified 13 analytes that were higher in CD patients relative to healthy controls and that remained significant after conservative Bonferroni correction (P < 0.0015). In particular, CXCL9, CXCL1, and interleukin IL-6 had the greatest effect and were highly significant (P < 5 × 10^–7). We also identified 9 analytes that were associated with disease location, with VEGF, IL-12p70, and IL-6 being elevated in patients with colorectal disease (P < 3 × 10^–4).

Conclusions

Multiple serum analytes are elevated in CD. These implicate the involvement of multiple cell types from the immune, epithelial, and endothelial systems, suggesting that circulating analytes reflect the inflammatory processes that are ongoing within the gut. Moreover, the identification of distinct profiles according to disease location supports the existence of a biological difference between ileal and colonic CD, consistent with previous genetic and clinical observations.

Introduction

The inflammatory bowel diseases, the major forms of which are Crohn’s disease (CD) and ulcerative colitis (UC), are characterized by chronic inflammation of the gastrointestinal tract. These are complex diseases where multiple genetic and nongenetic risk factors contribute to disease susceptibility, with components of the gut microbiota and history of smoking being major environmental contributors. These observations have led to a model whereby microbial components in a genetically susceptible host lead to perturbed intestinal homeostasis which manifests as gut dysbiosis, local dysregulation of the gut-associated immune system, and impaired barrier and defense mechanisms of the intestinal epithelium.¹ As is to be expected for any complex trait, CD and UC exhibit heterogeneity in terms of disease onset, presentation, location, progression, and response to therapy.²

Genome wide association studies (GWASs) have identified more than 200 genomic loci associated with IBD.^{3, 4} These studies have led to the identification of causal genes that implicate a wide variety of biological functions such as microbial pattern recognition, autophagy, and intestinal barrier function.^5–7 A predominant role for immune signaling in IBD pathophysiology is supported by the fact that many IBD-associated genetic variants are linked with innate and adaptative immune response, such as genes that are involved in primary immunodeficiencies, T-cell function, cytokine production, and immune signaling.^{3, 5} In addition, a large body of evidence from mouse models of IBD demonstrates that cytokine/chemokine production and signaling play an important role in disease pathogenesis, regardless of whether the models are based on infection, chemical injury, or genetic manipulation.⁸ Furthermore, multiple studies of blood and mucosal biopsy samples in cohorts ranging in size from roughly 20 to 100 patients have identified differences in RNA and/or protein expression levels of specific cytokines and chemokines between patients and controls and between IBD patients with active vs quiescent disease.⁹ We therefore hypothesized that in cohorts that are sufficiently well-powered, it should be possible to identify a robust signature of circulating cytokines and chemokines that differs between patients with CD and control subjects.

Recently, genotype-phenotype analyses of nearly 35,000 IBD patients found that genetic risk scores associate mainly with disease location, which is essentially fixed over time and with age at onset.¹⁰ In contrast, little or no genetic association with disease behavior was detected after correcting for disease location and age at onset. Thus, disease location is an intrinsic aspect of a patient’s disease and the major driver behind changes in disease behavior over time.¹⁰ In fact, this large international genetic study found that the CD group was much better explained by 2 groups, ileal CD and colonic CD, with ileocolonic CD having an intermediate genetic profile. The identification of genetic variants associated with disease location in CD has raised the possibility that genetic and nongenetic biomarkers could be used to better define and understand biological heterogeneity in the IBD population, potentially opening up the exciting possibility of a more personalized approach to treatment and disease management. We therefore hypothesized that at least part of the biological heterogeneity in the IBD population could be detectable in the patterns of cytokines and chemokines in patient sera.

To test these 2 hypotheses, we performed a multiplex analysis of 49 analytes, representing a broad range of cytokines, chemokines, and growth factors in 300 CD patients and 300 controls matched for age and sex, all collected by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) IBD Genetics Consortium (IBDGC) as part of our genetic studies. Our analyses identified at least 13 cytokines, chemokines, and growth factors that were significantly and robustly associated with CD and/or disease location.

Materials & Methods

Study Population

Patients and controls were selected from the National Institute of Diabetes and Digestive and Kidney Diseases IBD Genetics Consortium Repository. Recruitment of patients and healthy unrelated controls (spouses and “best friends” of index cases) for the IBDGC Repository was performed by the 6 genetic research centers (GRCs) of the IBDGC (http://ibdgc.uchicago.edu/about/grcs/). Phenotyping at each GRC was performed by clinicians with experience in managing patients with IBD and was conducted according to an established phenotype operating manual. Inflammatory bowel disease phenotype classification using this NIDDK IBDGC protocol has been validated.¹¹ Only patients with an established diagnosis of CD were included. The samples were selected from the IBDGC repository and processed in 2 separate phases, referred to as IBDGC-1 and IBDGC-2 (detailed in Table 1). For the first phase (IBDGC-1), we selected 100 CD patients with a more complex behavior (stricturing or penetrating behavior; B2 or B3, respectively) and 100 controls. For the second phase, we selected 200 CD patients with a more representative proportion of disease behavior and location (IBDGC-2) and 200 controls. In addition, CD patients were selected based on availability of serum samples and availability of relevantly matched control subjects with available serum samples. Controls were matched for sex, age, and ethnicity. For both phases, we prioritized samples collected more recently and patients with at least 3 years of follow-up since diagnosis. For CD patients, information on smoking habits, disease behavior, disease location, surgery, and age at diagnosis were collected (Table 1).

TABLE 1.

Characteristics of Crohn’s Disease (CD) Patients and Controls

		CD patients			Controls
Phenotypes	Categories	IBDGC-1	IBDGC-2	Total	IBDGC-1	IBDGC-2	Total
		n	n	n	n	n	n
Sex	Male	46	88	134	46	88	134
	Female	52	109	161	53	111	164
Ethnicity	White Non-Jewish	85	154	239	86	156	242
	Jewish	9	43	52	9	43	52
	Black/AA	4	0	4	4	0	4
Smokinga	Yes	21	45	66	9	21	30
	Ex-smoker	8	20	28	13	29	42
	No	68	129	197	76	146	222
	Missing	1	3	4	1	3	4
Age at Diagnosis	Range	1–52	7–67	1–67	—	—	—
	Median	19	23	22	—	—	—
	Quartiles: Q1-Q3	14–26	17–32	16–31	—	—	—
Age at recruitment	Range	19–70	16–87	16–87	19–70	16–76	16–76
	Median	41	37	38	40	35	37
	Quartiles: Q1-Q3	27–51	28–46	28–47	28–51	26–50	26–50
Disease Location	Ileal (L1)	31	52	83	—	—	—
	Ileocolonic (L3)	58	99	157	—	—	—
	Colorectal (L2)	8	46	54	—	—	—
Upper GI	Yes	9	15	24	—	—	—
	No	80	150	230	—	—	—
	Missing	9	32	41	—	—	—
Perianal	Yes	38	52	90	—	—	—
	No	59	143	202	—	—	—
	Missing	1	2	3	—	—	—
Disease Behavior	Inflammatory (B1)	0	105	105	—	—	—
	Stricturing (B2)	51	51	102	—	—	—
	Penetrating (B3)	47	41	88	—	—	—
Surgery	Yes	86	102	188	—	—	—
	No	11	95	106	—	—	—
	Missing	1	0	1	—	—	—

^aSmoking was ascertained differently for patients (at diagnosis) and controls (at recruitment). For age at recruitment, sex, and ethnicity, there were no significant differences between cases and controls (lowest P value > 0.6).

Ethics Statement

The collection of serum sample and clinical data was approved by the ethics committees at each of the recruitment sites of the IBD Genetics Consortium. The current study has been approved by the ethics committee of the Montreal Heart Institute, known as the Comité d’éthique de la recherche et du développement des nouvelles technologies (CERDNT).

Measurement of Serum Analytes

Serum samples from the 300 adult patients with CD and 300 healthy unrelated donors were collected by the IBDGC between 2004 and 2011. Specifically, blood was drawn in 10 mL vacutainer tubes. Tubes were left in vertical position for 30 minutes at room temperature to allow blood to clot. Tubes were then spun at 1100 to 1300 × g for 15 minutes. at approximately 20°C in a swing-type centrifuge. The serum layer was removed manually with a pipette, aliquoted into 8 separate 0.5-mL bar-coded cryovials, frozen at −80^oC, shipped to the IBDGC central facility (Fisher Bioservices) on dry ice, and then stored at −80°C until used. Samples were thawed just before analysis with no prior freeze-thaw cycles.

We selected 3 panels that capture a maximal set of analytes, cytokines, chemokines, or growth factors. Specifically, these analytes were analyzed using Bio-Plex Pro human cytokine, chemokine, and growth factor 27-Plex, 21-Plex, and sCD40L Single-plex assays (Bio-Rad Laboratories, Irvine, California, USA). These assays were conducted according to the manufacturer’s specifications and were read with a Bio-Plex MAGPIX (Bio-Rad Laboratories) at 525 nm and 625 nm. Analytes were identified and quantified using the CCD imager. Together, these panels assess the expression level of 49 serum analytes. The complete list of analytes tested can be found in Table S1. Ten months separated the processing of IBDGC-1 and IBDGC-2 samples.

Plate Design and Data Processing

Samples were processed on different plates and dates using a balanced design. The IBDGC-1 samples were processed on 4 plates and included 80 replicated samples. The IBDGC-2 samples were processed on 6 plates and included 60 replicates. Data from the 2 phases were analyzed independently. Principal component analysis (PCA) was used for inspection of plate effects and outliers. Visual inspection identified outlier samples with values far above the range of other samples, which were removed before data processing. Fluorescence intensity (FI) distributions were then normalized between plates using ComBat (Library SVA, R 3.2.0), a Bayesian tool for batch adjustment effect initially designed for microarray expression data. For each analyte, the method applies a correction for mean and variance on log-transformed data, taking into account the information from all the analytes of the multiplex. Standard curves were fitted using 5-parameter logistic curves on standard fluorescence intensities pooled from all the plates. Detection threshold was defined as the lowest value of the standard curve. Analytes with more than 50% of samples having fluorescence intensities below the detection threshold were removed from the analysis. To be noted, for the analytes that were undetected in >50% of samples; we investigated for possible association between “detection status” and phenotype, but no evidence of association was detected. Most of the remaining analytes had more than 75% of samples within the range of reliable quantification, defined by a recovery of standard concentration between 70% and 130%. Correlation (Spearman and Pearson) between replicate samples was used as a measure of reproducibility. Analytes were classified regarding the quality of their quantification. Analytes with correlation above 60% between replicates and at least 50% samples within reliable quantification range, mostly within linear part of the standard curve, were classified as “reliable quantification.” The remaining analytes were classified as “noisy quantification” and excluded from final analyses. The IBDGC-1 phase was used to filter analytes based on quantification, given the IBDGC-2 data set had fewer replicates distributed across more plates. It is worth noting that 2 analytes (IL-4 and GM-CSF) were excluded from IBDGC-2 because they failed to reach detection threshold in that data set. The final data set included 35 analytes, all listed in Table S1.

Before proceeding to statistical analyses, normalized log-transformed fluorescence intensity distributions were truncated at 2.5 interquartile range (IQR) from the lower and upper quartiles; samples were not removed, but extreme values were reduced to minimize impact of outliers at the next steps. A correction for collection center and collection date was then applied, with ComBat and linear regression, respectively. Concentrations were computed from the standard curve. Samples below detection threshold were imputed at the minimal observed value. Distribution of concentrations was truncated at 1.5 IQR from the lower and upper quartiles to minimize impact of possible outliers on analyses.

Statistical Analysis

Analysis for association with phenotype was performed using linear regression including sex and age at sample collection. The 2 data sets (IBDGC-1 and IBDGC-2) were analyzed separately, and the results combined with inverse variance weight. The IBDGC-1 was not used for disease location and behavior, as it only included B2 and B3 patients. For analyses of disease location, we used complementary approaches testing for subtypes vs controls, and for colorectal vs ileal, encoding ileocolonic patients as intermediate in effect sizes to those with ileal and colorectal phenotypes. When testing for association between serum analyte levels and disease status/disease phenotypes, we used a threshold of P < 0.05. We also used a conservative Bonferroni correction for the number of detectable analytes tested (n = 35) and used a threshold of P <0.0015 to correct for multiple testing.

Results

Identification of Robust Serum Markers Significantly Elevated in Crohn’s Disease

As a first step to identify analytes that differ between CD patients and healthy controls, we selected 100 CD patients with a complicated disease behavior (stricturing/B2 or internal penetrating/B3 disease) and 100 healthy controls matched for sex and age (IBDGC-1). The demographic and clinical phenotypes of these subjects are presented in Table 1. Serum from patients and controls were arrayed in plates with a balanced design and tested for 49 different analytes that represent products of genes within IBD loci, ligands for proteins encoded by genes within IBD loci, and a broad selection of cytokines, chemokines, and growth factors known to play a role in a variety of immune functions (Table S1). After quality control analyses, 3 samples were excluded for poor quality, and 35 of the analytes were deemed detectable with reliable quantification in both controls and patients (Table S1). Subsequent statistical analyses determined that 11 analytes (P < 0.05) had a higher concentration in serum from CD patients compared with controls (Table 2). These included pro-inflammatory cytokines such as interleukin (IL)-6 and IL-16, in addition to prototypical T helper (Th)-1, interferon (IFN)-γ, and Th2 (IL-5) cytokines. Also included in this list were 5 chemokines that primarily recruit or target monocytes and neutrophils (CXCL1/GRO-α; CXCL8/IL-8), lymphocytes (CXCL9/MIG), or a combination of lymphoid and myeloid cells (CCL11/Eotaxin; CXCL12/SDF-1α). The soluble low affinity receptor for interleukin-2 (sIL2-Rα) was among this list of analytes with the strongest effect sizes and is a cytokine that has been reported to enhance Th17 responses through its ability to sequester local IL-2.¹²

TABLE 2.

Serum Analytes That Differ Significantly Between CD Patients and Controls

	IBDGC-1 (B2B3) vs Controls			IBDGC-2 (B2B3) vs Controls			Combined (B2B3) vs Controls			Combined (ALL CD) vs Controls
ANALYTES	BETA	SE	P	BETA	SE	P	BETA	SE	P	BETA	SE	P
CXCL9/MIG	0.364	0.137	8.40E-03	0.422	0.124	7.82E-04	0.395	0.092	1.68E-05	*0.441	*0.084	*1.40E-07
CXCL1/GRO-α	0.360	0.162	2.79E-02	0.473	0.093	7.19E-07	0.445	0.081	3.73E-08	*0.437	*0.066	*4.78E-11
IL-6	0.495	0.180	6.47E-03	0.386	0.115	9.34E-04	0.417	0.097	1.69E-05	*0.419	*0.082	*2.87E-07
VEGF	0.411	0.248	1.00E-01	0.263	0.160	1.01E-01	0.307	0.135	2.28E-02	*0.359	*0.112	*1.39E-03
IL-2Rα	0.498	0.131	1.89E-04	0.215	0.100	3.16E-02	0.319	0.079	5.74E-05	*0.333	*0.069	*1.38E-06
IL-10	0.245	0.250	3.29E-01	0.340	0.209	1.04E-01	0.301	0.160	6.02E-02	0.318	0.131	1.55E-02
CXCL8/IL-8	0.275	0.077	4.32E-04	0.317	0.072	1.51E-05	0.297	0.052	1.50E-08	*0.293	*0.046	*1.62E-10
IL-7	0.470	0.165	4.87E-03	0.231	0.067	6.70E-04	0.265	0.062	2.06E-05	*0.280	*0.051	*3.11E-08
HGF	0.229	0.120	5.68E-02	0.297	0.083	4.09E-04	0.275	0.068	5.56E-05	*0.265	*0.058	*5.14E-06
IL-18	0.191	0.120	1.15E-01	0.271	0.100	7.22E-03	0.238	0.077	1.99E-03	*0.261	*0.068	*1.11E-04
CCL11/Eotaxin	0.261	0.097	7.71E-03	0.148	0.092	1.09E-01	0.201	0.067	2.53E-03	*0.245	*0.058	*2.61E-05
MIF	0.046	0.151	7.59E-01	0.282	0.121	2.11E-02	0.189	0.095	4.56E-02	0.209	0.083	1.20E-02
IL-1RA	0.097	0.133	4.66E-01	0.272	0.097	5.57E-03	0.211	0.079	7.26E-03	0.206	0.066	1.94E-03
CXCL12/SDF-1α	0.153	0.066	2.24E-02	0.240	0.058	4.27E-05	0.202	0.044	3.37E-06	*0.192	*0.037	*2.80E-07
IL-5	0.167	0.062	7.92E-03	0.140	0.071	5.14E-02	0.155	0.047	9.39E-04	*0.168	*0.041	*4.80E-05
CD40L	0.096	0.103	3.54E-01	0.112	0.092	2.25E-01	0.104	0.068	1.27E-01	0.134	0.059	2.24E-02
G-CSF	0.064	0.071	3.68E-01	0.140	0.061	2.17E-02	0.108	0.046	1.94E-02	0.125	0.040	1.70E-03
IL-16	0.434	0.171	1.18E-02	0.092	0.091	3.12E-01	0.167	0.080	3.68E-02	0.123	0.065	5.79E-02
IFN-γ	0.102	0.051	4.82E-02	0.119	0.050	1.80E-02	0.111	0.036	1.99E-03	*0.121	*0.031	*1.22E-04
PDGF-ββ	0.143	0.086	9.72E-02	0.052	0.097	5.92E-01	0.103	0.064	1.09E-01	0.120	0.057	3.55E-02
TNF-α	0.071	0.067	2.89E-01	0.107	0.085	2.10E-01	0.085	0.053	1.07E-01	0.114	0.047	1.63E-02
IL-1β	0.092	0.058	1.14E-01	0.078	0.048	1.02E-01	0.084	0.037	2.29E-02	0.091	0.031	3.91E-03

Note: 49 analytes were tested in patients and controls; 14 failed our criteria to be included in analyses and 13 were analyzed but not associated with CD. For the “all CD” combined analysis, *P-values are significant after Bonferroni correction (P < 0.0015).

Given these promising results, we set out to test the same analytes but in a larger, independent set of 200 CD patients and 200 matched controls (IBDGC-2). As can be seen in Table 1, the IBDGC-2 samples had roughly equivalent numbers of cases in the B1, B2, and B3 behavior categories. To be comparable to the analyses performed in IBDGC-1, our initial analysis of IBDGC-2 focused only on CD patients with either B2 or B3 phenotypes. Using the same quality control measures (4 samples excluded) and significance threshold (P < 0.05), we identified 13 analytes that were associated with a higher concentration in serum from CD patients compared with controls (Table 2, Table S1). Importantly, this list contained 8 of 11 analytes found to be associated in IBDGC-1; only IL-5 (P = 0.05), IL-16 (P = 0.31), and CCL11/Eotaxin (P = 0.11) did not meet the significance threshold. The analytes that were novel to this analysis were (1) the pro-inflammatory Th1 cytokine IL-18 that stimulates IFN-γ production in T-helper type 1 cells; (2) the interleukin-1 receptor antagonist IL-1Rα/IL-1RN; (3) the macrophage migration inhibitory factor (MIF), which is a pro-inflammatory lymphokine released by white blood cells in response to bacterial antigens; (4) the hepatocyte growth factor (HGF) that acts as a multi-functional cytokine on cells of mainly epithelial origin—although when it forms a heterodimer with interleukin-7 (IL-7), it functions as a pre-pro-B cell growth-stimulating factor; and (5) the granulocyte colony-stimulating factor (G-CSF) that functions in maintaining neutrophil levels and activity in vivo.

Given the consistency in the results from the analyses from these 2 phases, we then performed a combined analysis of the results from IBDGC-1 and IBDGC-2, representing a total of 190 CD patients with either B2 or B3 phenotypes and 298 controls to have maximal statistical power to detect analytes associated with the B2/B3 CD phenotypes. All 16 analytes that were identified in the separate analyses of the IBDGC-1 and/or IBDGC-2 samples reached the threshold of P < 0.05 (Table 2, Table S1). Two additional analytes were identified in this meta-analysis: the pro-inflammatory cytokine IL-1β and the vascular endothelial growth factor (VEGF) for a total of 18.

Next, we combined data from all patients (295) and controls in a combined analysis and identified 21 analytes that were significantly higher (P < 0.05) in CD patients relative to healthy controls, with MIG, GRO-α, and IL-6 displaying the largest effect sizes (Table 2; Table S1; Fig. 1). This list of analytes elevated in CD patients compared with controls was similar regardless of whether the combined analysis included all CD patients or only those with B2 or B3 phenotypes, with TNF-α, CD40L, and PDGF-ββ reaching the 0.05 threshold. If we implemented the conservative Bonferroni correction for the number of detectable analytes tested (n = 35) and set the significance threshold at P < 0.0015, we found 13 analytes with highly significant association with CD: CXCL1/GRO-α, CXCL8/IL-8, IL-7, CXCL9/MIG, IL-2Rα, HGF, CCL11/Eotaxin, IL-5, IL-18, VEGF, IL-6, CXCL12/SDF-1α, and IFN-γ. The analytes with the strongest effects (β = 0.44) were CXCL9/MIG and CXCL1/GRO-α, the latter being an antimicrobial chemokine that acts as a chemoattractant for neutrophils.¹³ Nine of these analytes are implicated by the known genetic risk factors for IBD. Specifically, genes within IBD susceptibility loci encoded 5 of these analytes (CCL11/Eotaxin, IL-2Rα, IL-10, IFN-γ, and IL-1β) or their corresponding ligand or receptor (IL-2, IL-7R, IL-18R1, CD40, IFNGR2, and IL-1R2).³

FIGURE 1.

Effect sizes for association to CD with confidence intervals. Confidence intervals at 95% (black) and 99.85% (gray, Bonferroni correction) for the combined analysis including all CD patients are shown for each analyte that passed our quality threshold in both data sets. The effect is shown on the log₂ scale, so the absence of effect is at 0, and the fold change on concentrations is given by 2^effect (eg, an effect of 0.2 is a 1.15-fold change).

Identification of Robust Serum Markers Significantly Associated With Disease Location

Next we examined if the analytes segregated with disease location. To address this question, we examined the distribution of CD cases with isolated ileal (L1), ileocolonic (L3), or isolated colonic diseases (L2) in our IBDGC-1 and IBDGC-2 samples. Generally, ileocolonic disease represents roughly 50% of CD patients, whereas isolated ileal and isolated colonic disease each comprise approximately 25% of CD patients.² As these proportions were reflected in IBDGC-2 but not IBDGC-1, we limited our disease location analyses to IBDGC-2. When using a P < 0.05 threshold for suggestive association, a distinct analyte signature emerged based on disease location when comparing patients with specific disease locations (L1, L2, L3) with controls—and when comparing colorectal with ileal (Table 3, Table S1).

TABLE 3.

Serum Analytes With Significant Association to CD(all) and/or Disease Location

	Disease Location Association
	Ileal vs Control			Colorectal vs Control			Colorectal vs Ileal
ANALYTE	BETA	SE	P	BETA	SE	P	BETA	SE	P
IL-13	-0.137	0.093	1.41E-01	0.216	0.097	2.72E-02	0.354	0.112	1.88E-03
CCL3/MIP-1α	-0.106	0.106	3.20E-01	0.286	0.111	1.07E-02	0.387	0.129	3.08E-03
TNF-α*	-0.043	0.104	6.76E-01	0.334	0.109	2.23E-03	0.369	0.129	4.62E-03
IL-6	0.136	0.141	3.37E-01	0.634	0.150	3.02E-05	0.488	0.174	5.48E-03
IL-12p70	0.018	0.139	8.99E-01	0.531	0.145	2.90E-04	0.487	0.174	5.66E-03
G-CSF*	0.052	0.075	4.85E-01	0.308	0.078	1.02E-04	0.244	0.097	1.27E-02
IL-7**	0.154	0.082	6.32E-02	0.387	0.086	9.87E-06	0.221	0.102	3.19E-02
IL-1β*	-0.007	0.058	9.02E-01	0.142	0.061	2.01E-02	0.147	0.071	4.11E-02
VEGF	0.260	0.195	1.84E-01	0.763	0.203	1.92E-04	0.488	0.252	5.42E-02
IL-10*	0.162	0.245	5.10E-01	0.628	0.249	1.21E-02	0.443	0.283	1.19E-01
IL-5**	0.087	0.086	3.13E-01	0.257	0.090	4.41E-03	0.161	0.108	1.39E-01
CXCL8/IL-8**	0.165	0.089	6.30E-02	0.325	0.093	5.13E-04	0.157	0.113	1.66E-01
IL-1RA*	0.179	0.119	1.34E-01	0.369	0.125	3.31E-03	0.184	0.152	2.27E-01
IFN-γ**	0.056	0.062	3.65E-01	0.155	0.065	1.70E-02	0.094	0.080	2.43E-01
CXCL12/SDF-1α**	0.242	0.070	6.56E-04	0.148	0.074	4.60E-02	-0.094	0.085	2.68E-01
CXCL1/GRO-α**	0.369	0.113	1.22E-03	0.520	0.120	1.82E-05	0.143	0.139	3.07E-01
CCL11/Eotaxin**	0.170	0.114	1.34E-01	0.284	0.119	1.75E-02	0.106	0.146	4.68E-01
MIF*	0.265	0.155	8.81E-02	0.139	0.162	3.94E-01	-0.110	0.202	5.87E-01
CD40L*	0.188	0.113	9.54E-02	0.149	0.118	2.09E-01	-0.054	0.144	7.06E-01
PDGF-ββ*	0.059	0.119	6.23E-01	0.111	0.124	3.72E-01	0.050	0.153	7.42E-01
IL-2Rα**	0.196	0.125	1.17E-01	0.244	0.132	6.47E-02	0.054	0.167	7.48E-01
IL-18**	0.308	0.127	1.61E-02	0.249	0.133	6.24E-02	-0.048	0.164	7.69E-01
HGF**	0.294	0.104	4.95E-03	0.272	0.109	1.28E-02	-0.021	0.132	8.73E-01
CXCL9/MIG**	0.303	0.164	6.54E-02	0.284	0.172	9.89E-02	-0.006	0.235	9.79E-01

See Table 2.

* Analytes associated at P < 0.05 in CD vs controls.

**Analytes associated at P < 0.0015 in CD vs controls.

To better interpret the relationship between all of the analytes tested and disease location, we plotted the effect size for each analyte with respect to disease status and disease location on separate axes, together with 95% confidence intervals (CIs) of effects by disease location (Fig. 2). With this representation, the analytes fell into 4 clusters: (1) analytes that have similar effect for ileal and colorectal locations (HGF, CXCL9/MIG, IL-2Rα, MIF, PDGF-ββ, IL-18, CD40L, CCL11/Eotaxin); (2) an analyte with mild and suggestively stronger effect in ileal location (CXCL12/SDF-1α); (3) multiple analytes with a much stronger or specific effect on colorectal location (VEGF, IL-6, IL-1β, IL-12p70, TNF-α, IL-13, CCL3, IL-7, G-CSF); and (4) analytes with modestly/suggestively stronger effect on colorectal location (IL-5, IFN-γ, CXCL8/IL-8, CXCL1/GRO-α, IL-1Rα, and IL-10). Interestingly, 3 analytes (IL-13, CCL3, and IL-12p70) were associated with a strong effect size on colorectal location but showed no association to ileal vs controls and thus were not significantly associated with CD status. For most cytokines, effect sizes for ileocolonic disease location are intermediate to those of ileal and colorectal only, sometimes closer to one or the other (Fig. 2C and Table S1). Notably, for CXCL9/MIG, the effect for ileocolonic is suggestively larger than for colorectal and ileal only. In addition, it should be noted that there was no association between any of the analytes tested and disease behavior as classified as inflammatory (B1), stricturing (B2) or penetrating (B3) disease (Table S1). This was true whether we controlled for disease location or not. We also investigated the robustness of our results when controlling for smoking status, even though this information was missing for some participants and not ascertained in the same way for patients (at diagnosis) and controls (at recruitment). We observed no notable differences in the results (Table S2).

FIGURE 2.

Representation of effect sizes for disease location. Each ellipse represents a cytokine associated to disease status and/or disease location at P < 0.05. Analytes associated at P < 0.0015 in the combined CD vs controls analyses are indicated in bold and larger font size. Ellipses color is solely for the purpose of identifying the different cytokines on the plot. (A) Effect sizes are shown for CD vs control (y axis) and colorectal vs ileal (x axis), with the width and height of the ellipse illustrating standard error of the effect for disease location and disease status, respectively. Analytes higher on the y axis are increased in CD compared with controls. Analytes more to the right side of graph are increased in colorectal patients compared with ileal patients. (B) Effect sizes are shown for colorectal vs controls (y axis) and ileal vs controls (x axis). Width and height of the ellipse illustrate standard error of the effect for colorectal and ileal vs controls, respectively. Analytes higher on the y axis are increased in ileal patients compared with controls. Analytes more to the right are increased in colorectal patients compared with controls. Analytes on the diagonal show similar effect sizes for ileal and colorectal. To be noted, 95% confidence intervals correspond to twice the dimension of the ellipses. (C) Effect sizes are shown with 95% confidence intervals. The left panel shows effect for patient vs control by disease location; colorectal (blue), ileocolonic (purple), and ileal (red). The right panel shows the effect of colorectal vs ileal (black). Names of analytes associated to CD vs controls in the main analysis (Table 2) are shown in bold if significant (P < 0.0015) after Bonferroni correction, or underlined if only nominally significant (P < 0.05).

Discussion

Alteration of the gut immune-epithelium-microbiota balance is a hallmark of IBD. This notion is supported by many studies in humans and model systems that demonstrate alterations in the innate and acquired immune systems, impaired intestinal barrier function, and gut dysbiosis in IBD.¹ The specific alterations in homeostatic mechanisms, however, are likely to vary from one patient to another given the heterogeneity within the IBD population in terms of disease onset, disease presentation and progression, and response to therapy. Research to better define and understand this heterogeneity and identify its impact on clinical outcomes is at the basis of precision medicine. In terms of genetic contribution to IBD, it has been well established that over 200 genomic regions influence an individual’s predisposition to developing IBD, and extensive analysis of approximately half of these loci in roughly 35,000 IBD patients revealed that the strongest association between genetic variation and clinical features was disease location.¹⁰ In fact, these genetic analyses found that disease location in CD was an intrinsic aspect of a patient’s disease and that biologically ileal Crohn’s disease and colonic Crohn’s disease are distinct. Given the importance of the immune system in IBD, it is conceivable that serum analytes may be informative in terms of disease heterogeneity, even though the chronic inflammatory processes are occurring in the periphery, with a predominant role for the gut-associated lymphoid tissue.

As a first step to address this question, we wanted to identify serum analytes that were robustly associated with disease. To achieve this, it was important to optimize statistical power to minimize type-1 error, and thus we focused on a single disease, CD, and tested a large cohort of approximately 300 cases and an equivalent number of controls. These analyses enabled the identification of 13 serum analytes that were strongly associated with CD in this cohort (P < 0.0015) and another 8 that were suggestively associated with CD (P < 0.05). Results were very consistent between the 2 phases of this project, with samples that were tested 10 months apart, suggesting that these findings are reproducible; although these analytes should be tested in equally large cohorts by other research groups to formally test reproducibility. Having said this, we performed a search for publications reporting association for 1 or more of these analytes in IBD and have found that all but 1 of these analytes had some degree of association to CD, although in more modestly sized cohorts (60 different CD studies with an average of 45 CD patients/study; Table S3). Only CXCL12/SDF-1α has not previously been reported to be elevated in the serum of CD patients, although its RNA expression was found to be elevated in intestinal biopsies.¹⁴ Taken together, this provides strong evidence that the serum levels for these 13 analytes are truly associated with CD.

The analyte with the most statistically significant association with CD was CXCL1/GRO-α, a chemokine produced by myeloid cells (P < 5 × 10^–11). Evidence of association of CXCL1/GRO-α serum levels with CD and UC has been reported in 2 studies of adult and pediatric cohorts^{15, 16} and with UC alone in another,¹⁷ with cohorts ranging in size from 24 to 42 CD patients and 18 to 60 UC patients. CXCL1/GRO-α is 1 of 7 chemokines (CXCL1–3, CXCL5–CXCL8) that bind to 2 receptors, CXCR1 and CXCR2 that control neutrophil function.¹⁸CXCL1/GRO-α is known to be a potent activator (agonist) for the CXCR2 receptor. The second most CD-associated analyte, CXCL8/IL-8, is a chemokine produced by myeloid and epithelial cells and is a potent activator of the CXCR1 receptor. There have been 7 previous reports examining serum levels of CXCL8/IL-8, and all found some level of association with CD, thus supporting this finding (Table S3). Together, this establishes that serum levels of these 2 neutrophil chemoattractants are convincingly associated with CD, which is entirely consistent with the important role for neutrophils in the establishment and maintenance of intestinal inflammatory processes in CD.¹⁹ Interestingly, there are therapeutic antagonists for both of these receptors, such as reparixin, navarixin, danirixin, and elubirixin, that have been developed for the treatment of breast cancer or chronic inflammatory diseases such as chronic obstructive pulmonary disease (COPD) and psoriasis.^20–23 This suggests that these or other CXCR1/CXCR2 antagonists might play a beneficial role in the treatment of patients with CD potentially to control disease flares that are associated with neutrophil infiltration and intestinal crypt abscesses.²⁴ Such a therapeutic strategy would be complementary to other approaches currently under consideration that are aimed at controlling neutrophil-mediated inflammation in IBD, such as blocking the glycoprotein CD11b/CD18 that is known to play an important role in regulating polymorphonuclear leukocyte transepithelial migration.²⁴

It should be noted, however, that IL-6 and CXCL9 had equivalent effects as CXCL1, although they were not as significantly associated (P < 3 × 10^–7 vs P < 5 × 10^–11). Interleukin-6 is a well-characterized cytokine that is produced during viral infection by a diverse set of immune and nonimmune cells and has pleiotropic effects.²⁵ Its production is typically transient and leads to significant increases in acute phase proteins.²⁵ Elevated serum levels of IL-6 have previously been associated with numerous chronic inflammatory conditions such as rheumatoid arthritis (RA), systemic lupus erythematosus, systemic sclerosis, and many others, with IL-6 blockade used as therapy in RA and juvenile idiopathic arthritis.²⁵ Twenty-three previous studies have noted elevated serum levels in CD patients (Table S3); although given that this observation is not disease-specific, it would seem that it is more likely a marker of disease activity, consistent with previous reports of a correlation between serum IL-6 levels and endoscopic activity and response to anti-TNF therapy.²⁶ In the case of CXCL9, IFN-γ induces its expression in epithelial cells, and its expression is also elevated in colonic biopsies taken from CD and UC patients.^{27, 28} Moreover, levels of CXCL9 have been recently found to be elevated in the serum of CD and UC patients, supporting the observation that was made in the current study.²⁹

The other 9 analytes that are significantly elevated in the serum of CD patients are produced by multiple cell types that include lymphoid cells (IL-2Rα, IL-5, IFN-γ), myeloid cells (IL-7, IL-5, IL-18, VEGF, IFN-γ), epithelial cells (CCL11/Eotaxin, IL-7), and a variety of other cell types (HGF, VEGF, CXCL12/SDF-1α). Moreover, the cells targeted by these analytes include lymphoid cells (IL-7, CCL11/Eotaxin, IL-5, IL-18, CXCL12/SDF-1α), myeloid cells (IFN-γ, CXCL12/SDF-1α), epithelial cells (HGF), and endothelial cells (VEGF). The broad set of cell types implicated by this group of serum analytes (Fig. 3) seems to reflect the complex nature of the inflammatory processes involving an important interplay between immune and nonimmune cells that is characteristic of the chronic intestinal inflammation of CD.³⁰ Although these results lack the finer details of the cellular context and functions that can be provided by studying intestinal tissues, such as using a variety of single-cell technologies, these results do suggest that circulating analytes provide relevant biological information on inflammatory processes occurring in the periphery (Fig. 3). For example, the elevated sCD25 in circulation likely reflects high levels in the periphery, suggesting that in CD patients there are enhanced local pathogenic Th17 responses due to the ability of sCD25 to act as a decoy receptor for IL-2, thus sequestering local IL-2.¹² As such, this information can be used to guide future functional studies and/or explore novel avenues for therapeutic development. For example, there are already anti-interleukin-18 antibodies being developed for Type 2 diabetes and recombinant human IL-18 binding protein currently being investigated for the treatment of a variety of autoimmune diseases that could potentially be repositioned for the treatment of CD.^{31, 32} This possibility is further supported by the fact that IL-18R1 and IL-18RAP are associated with CD.³ In addition, although IL-18 itself has not been associated to CD at genome-wide significance, a recent report supported a role for IBD-associated genetic variants controlling serum IL-18 levels and being causally related to IBD.³³ Another example is that of the IL-7/IL-7R pathway, as it has recently been reported that an elevation in local IL-7 receptor (IL-7R) pathway is strongly associated with nonresponsiveness to anti-TNF therapy, supporting the notion that a blockade of IL-7/IL-7R could be an effective approach in this clinically important patient population.³⁴

FIGURE 3.

A biological pathway perspective for the 13 serum analytes that were strongly associated with CD in this study (P < 0.0015). These 13 analytes were found to be significantly elevated in the sera of CD patients and reflect the complex nature of the inflammatory processes that are ongoing in these patients. Many of these are chemoattractants, produced and secreted by a variety of cells such as epithelial cells (CCL11), neutrophils (CXCL8, CXCL9), monocytes/macrophages (CXCL1), and play an important role in the recruitment (dashed arrows) of neutrophils, eosinophils, and T cells. Many of the other analytes play important roles in regulation and crosstalk of immune cells/response (IL-7, soluble IL-2Rα, IFN-γ, IL-18) tissue regeneration (IL-6, HGF), angiogenesis (VEGF, HGF), and cellular differentiation (IL-5). Though most of these analytes are common to all CD subgroups (based on disease behavior and disease location), some seem to be more strongly associated with ileal (red) or colorectal location (green). Analytes are positioned next to the cells that are primarily responsible for their production and secretion. Dashed arrows represent chemoattractant properties of denoted cytokine/chemokine.

Given that our data support that serum cytokines, chemokines and other analytes provide relevant biological information on intestinal inflammation, we examined whether the differences between ileal and colonic CD, based on observations made with genetic data,¹⁰ would also be reflected in patterns of circulating analytes. Interestingly, we observed a continuum of effects in relation to serum analytes and their association with disease location (Fig. 2). The strongest of these was the association between VEGF and colonic disease. There have been previous reports of an elevated serum VEGF association with CD, but only 1 study examined association with disease location; this was negative, although this was only examined in 24 patients and thus likely underpowered (Table S3). In terms of biological function, VEGF is a well-known angiogenic factor. The role of angiogenesis in IBD is complex: it contributes to the pathogenic processes by enabling increased recruitment of inflammatory cells to sites of inflammation, increased supply of nutrients and oxygen, and wound healing.³⁵ Although targeting the angiogenesis pathway per se might be too risky given the complex nature of the biological effects, it is interesting to note that the α ₄ β ₇ integrins that are currently targeted by IBD therapies are upregulated during angiogenesis.³⁵ The inflammatory angiogenesis that occurs in IBD, however, is not believed to be limited to the colon.

Only 1 analyte had suggestively stronger effect on ileal location (CXCL12/SDF-1α), and its serum levels have not been previously associated with CD. CXCL12/SDF-1α, a homeostatic chemokine produced by multiple cell types, is the ligand for both the CXCR4 and ACKR3/CXCR7 receptors, and has been implicated in multiple autoimmune diseases.³⁶ Its role in IBD has been shown to be in part due to an effect on the migration of memory T cells to the intestinal mucosa.³⁷ Although the exact mechanism to explain the observed difference in serum expression of CXCL12/SDF-1α in ileal CD patients remains to be elucidated, a difference in immune cell infiltrates in these locations in IBD has been previously observed.³⁸

In these analyses, we also identified 4 analytes associated with colorectal disease that lacked significant association (P > 0.1) to CD per se (IL-13, CCL3/MIP-1α, IL-12p70) or that only had modest association to CD (TNF-α, P = 0.02). This is particularly relevant given the fact that 2 of these are targets of current biologic therapies in CD: (1) TNF-α is the target of the most commonly used biologic therapy in IBD, and (2) anti-IL-12p40 reagent targets both the bioactive form of IL-12 (IL-12p70) and IL-23. In addition, CCL3 is a ligand for 3 different receptors, CCR1, CCR4, and CCR5, for which there are bioavailable antagonists under clinical development for rheumatoid arthritis, chronic obstructive pulmonary disease, and other inflammatory diseases that could potentially be interesting to pursue within the context of IBD.^{39, 40} Moreover, our findings with regard to TNF-α association with colorectal disease but not ileal disease could shed some light as to why clinical outcomes of TNF blockade therapy are poorer in patients with ileal disease.⁴¹

The primary limitation of the current study is its retrospective design, as the cohort used was established to advance gene discovery and identification of disease mechanisms and should be considered a cross-sectional cohort representative of a general IBD population. Nonetheless, our study examined the serum profile of 35 analytes in 300 CD patients and 300 healthy donors and identified 13 analytes that were significantly and robustly associated with CD. Future studies of these 13 analytes in large prospective cohorts would enable the examination of whether their serum levels are associated with other clinical characteristics, outcomes, or inflammatory markers (eg, C-reactive protein), and identify any potential confounders such as medical therapy, disease activity, or other parameters that may influence inflammatory status such as obesity. In addition, our findings suggest that colonic inflammation is both genetically and biologically distinct from ileal inflammation. This raises the exciting possibility of a more personalized approach to patients’ treatment and disease management. Indeed, as we found evidence that the serum analytes are reflective of the complex inflammatory processes occurring at local sites of inflammation in the digestive tract, we examined our findings with respect to implications on potential therapies, particularly regarding repurposing and/or focusing the development of existing molecules for the treatment of CD. Finally, measuring serum analytes in the context of prospective clinical studies of existing therapies and those under development will further aid our understanding of disease heterogeneity in terms of important clinical outcomes and hopefully lead to effective, minimally invasive biomarkers that will aid in disease management.

Funding

This work was supported by the National Institutes of Diabetes, Digestive and Kidney Diseases (grant numbers DK062431 to SRB, DK062422 and DK062429 to JHC, DK062420 to RHD, DK062423 to MS, DK062413 to DPBM, and DK062432 to JDR). This work was also supported by a grant to the iGenoMed Consortium (grant numbers 4521; GPH-129341 to AB and JDR). The work is cofunded by Génome Québec, Genome Canada, the Government of Canada, the Ministère de l’Enseignement Supérieur de la Recherche, de la Science et de la Technologie du Québec, the Canadian Institutes of Health Research (with contributions from the Institute of Infection and Immunity, the Institute of Genetics, and the Institute of Nutrition, Metabolism and Diabetes), Genome British Columbia, Crohn’s Colitis Canada, and Agilent Technologies. SL holds a Research Scholars Emeritus Award from the Fonds de Recherche Santé–Québec. JDR holds a Canada Research Chair (grant 230625). This project also benefited from infrastructure supported by the Canada Foundation for Innovation (grant numbers 202695, 218944, and 20415 to JDR).

Author Contribution

GB, SL, and JDR contributed to the study design and concept. AB, NIGC, ML, SB, RD, DM, MS, and JC contributed to the sample acquisition. AP and GC acquired the data. GB, LC, PS, CDR, ML, SL, and JDR analyzed and interpreted the data. GB, LC, SL, and JDR wrote the manuscript. AB, JC, and JDR obtained funding. SL and JDR supervised the study.

Data Availability Statement

The data underlying this article are available in the article and in its online supplementary material.

Group Contributors

Members of the iGenoMed consortium: Alain Bitton, Gabrielle Boucher, Guy Charron, Christine Des Rosiers, Anik Forest, Philippe Goyette, Sabine Ivinson, Lawrence Joseph, Rita Kohen, Jean Lachaine, Sylvie Lesage, Megan Levings, John D. Rioux, Julie Thompson-Legault, Luc Vachon, Sophie Veilleux, and Brian White-Guay.

Members of NIDDK IBD Genetics Consortium: Manisha Bajpai, Sondra Birch, Alain Bitton, Krzysztof Borowski, Gregory Botwin, Gabrielle Boucher, Steven R. Brant, Wei Chen, Judy H. Cho, Roberto Cordero, Justin Côté-Daigneault, Mark J. Daly, Lisa Datta, Richard H. Duerr, Melissa Filice, Philip Fleshner, Kyle Gettler, Mamta Giri, Philippe Goyette, Ke Hao, Talin Haritunians, Yuval Itan, Elyse Johnston, Liza Konnikova, Carol Landers, Mark Lazarev, Dalin Li, Dermot P. B. McGovern, Emebet Mengesha, Miriam Merad, Vessela Miladinova, Shadi Nayeri, Siobhan Proksell, Milgrom Raquel, John D. Rioux, Klaudia Rymaszewski, Ksenija Sabic, Bruce Sands, L. Philip Schumm, Marc B. Schwartz, Mark S. Silverberg, Claire L. Simpson, Joanne M. Stempak, Christine Stevens, Stephan R. Targan, and Ramnik Xavier.