Serum proteomic and metabolomic analyses from patients with IBD identify biological pathways associated with treatment success with anti-integrin therapy

Abstract

Crohn's disease (CD) and ulcerative colitis (UC) are chronic inflammatory diseases of the gastrointestinal tract believed to arise from an imbalance between its epithelial, immune and microbial components. It has been shown that biological differences (e.g. genetic, epigenetic, microbial, environmental) exist between patients with IBD. It is also known that there is important heterogeneity in the response to therapies that target very specific biological pathways (e.g. TNF-alpha signaling, IL-23R signaling, immune cell trafficking). The aim of this study was to identify potential biological differences associated with differential treatment response to the anti α4β7 integrin therapy known as vedolizumab. We performed targeted analyses of > 150 proteins and metabolites, and nontargeted analyses of > 1100 lipid entities in serum samples from 92 IBD patients (42 CD, 50 UC) immediately prior to initiation of therapy with vedolizumab (baseline samples) and at their first clinical assessment (week 14 samples). We detected that the baseline levels of multiple serum cytokines, amino acids, acylcarnitines and triglycerides were different between responders and nonresponders to treatment with vedolizumab. We also noted changes in serum analytes between baseline and week 14 samples that were different between these two groups of patients. Many of these serum analytes are markers of biological pathways that are involved in the activation, proliferation and metabolism of pro-inflammatory cells. This study provides support for the hypothesis that biological differences between individuals not only impact the risk to develop IBD and IBD-related clinical phenotypes but also an IBD patient's likelihood of responding to a biological therapy.

Keywords: Crohn's disease; biomarkers; treatment response; ulcerative colitis; vedolizumab.

Figure 1

Targeted serum protein analyte levels associated with response to vedolizumab treatment. Responders to vedolizumab treatment had higher baseline serum levels of PDGF‐ββ than nonresponders (a). In contrast, responders had lower serum levels of SDF‐1a at baseline as compared with nonresponders (b). In responders, CXCL9/MIG levels increased 61% (FC = 1.61) between baseline and week 14, whereas there was a 66% decrease (FC = 0.44) in nonresponders (c). In contrast, serum levels of IL‐10 decreased 15% (FC = 0.85) in responders, with a 27% increase (FC = 1.27) in nonresponders (d). Filled symbols are for individuals where samples were only available at baseline. All plots shown are for “UC.” See also Supplementary table 2.

Figure 2

Targeted serum metabolite levels associated with response to vedolizumab treatment. A selected set of analytes that were found to be associated with response at 14 weeks are presented. Elevated baseline levels of the amino acid lysine (a) and the organic acid lactate (b) were associated with response to treatment in patients with UC (P = 2.1 × 10⁻³ and P = 1.31 × 10⁻³, respectively). We also observed that a ~2‐fold increase in serum levels of the organic acid citrate was associated with response in patients with CD (c). Baseline serum levels of multiple acyl carnitines, including AC 3:0, were also associated (P = 2.9 × 10⁻³) with response (d). Filled symbols are for individuals where samples were only available at baseline. All plots shown are for “UC” except for citrate (c), where the plot is shown for “CD.” See also Supplementary table 3.

Figure 3

Untargeted serum LC–MS‐based lipidomic analyses. We identified 107 lipid features that were significantly (P < 0.05) associated with response (a). Colors in the volcano plot represent the 18 clusters of metabolites identified by WGCNA. The first component (PC1) from the PCA of the 1111 lipid features was significantly associated (P = 0.017) with response (b). This PC1 was not only associated with response to therapy but was also associated with BMI (P = 5.1 × 10⁻⁵) (c); dotted lines represent SE of loess fit. Importantly, we did not detect any significant association between BMI and response. The PCA loadings value for each lipid feature was plotted, illustrating that PC1 is primarily driven by the “blue” WGCNA cluster, which is also associated with response (P = 5.8 × 10⁻³) (d). We also found an opposite impact for features from another module (named turquoise), which was not significantly associated with response to therapy (P = 0.44). See also Supplementary tables 4 and 5.