Peripheral Blood DNA Methylation Signatures and Response to Tofacitinib in Moderate-to-severe Ulcerative Colitis

Abstract

Introduction: Predictive biomarkers for treatment efficacy of ulcerative colitis [UC] treatments are lacking. Here, we performed a longitudinal study investigating the association and potential predictive power of genome-wide peripheral blood [PB] DNA methylation signatures and response to tofacitinib treatment in UC.

Methods: We recruited moderate-to-severe UC patients starting tofacitinib treatment, and measured PB DNA methylation profiles at baseline [T1], after 8 weeks [T2], and in a subset [n = 8] after a median of 20 weeks [T3] using the Illumina Infinium HumanMethylation EPIC BeadChip. After 8 weeks, we distinguished responders [R] from non-responders [NR] based on a centrally read endoscopic response [decrease in endoscopic Mayo score ≥1 or Ulcerative Colitis Endoscopic Index of Severity ≥2] combined with corticosteroid-free clinical and/or biochemical response. T1 PB samples were used for biomarker identification, and T2 and publicly available intraclass correlation [ICC] data were used for stability analyses. RNA-sequencing was performed to understand the downstream effects of the predictor CpG loci.

Results: In total, 16 R and 15 NR patients, with a median disease duration of 7 [4-12] years and overall comparable patient characteristics at baseline, were analysed. We identified a panel of 53 differentially methylated positions [DMPs] associated with response to tofacitinib [AUROC 0.74]. Most DMPs [77%] demonstrated both short- and long-term hyperstability [ICC ≥0.90], irrespective of inflammatory status. Gene expression analysis showed lower FGFR2 [pBH = 0.011] and LRPAP1 [pBH = 0.020], and higher OR2L13 [pBH = 0.016] expression at T1 in R compared with NR.

Conclusion: Our observations demonstrate the utility of genome-wide PB DNA methylation signatures to predict response to tofacitinib.

Keywords: Epigenetics; biomarkers; personalised medicine.

Graphical Abstract

Figure 1.

Predictive model using stability selected gradient boosting for response to tofacitinib. A] Machine learning setup and model generation to predict response to tofacitinib. B] Receiver operating characteristics plot showing mean area under the curve [AUC] performance upon internal validation on 20% withheld test set. C] Radar plot presenting the difference in methylation between response [red] and non-response [blue] for the top 25 predictor CpG loci. D] Variable feature importance of the top 25 predictor CpG loci. E] Functional overrepresentation using GO-term analysis of genes annotated to the 53 predictor CpG loci.

Figure 2.

Expression of FGFR2, OR2L13, and LRPAP1. Visualisation of the level of expression (log₂[counts+1]) for each individual sample before [T1] and at 8 weeks into tofacitinib treatment [T2] time coloured by response. Red represents responders and blue represents non-responders. The top plot represents a scatterplot of the patient over time, with dashed lines connecting samples obtained from the same patient. The bottom plots represent boxplots for T1 and T2 separately, comparing responders and non-responders annotated with the differential expression p-value.

Figure 3.

Stability analyses and estimated cell proportion. A] Spearman correlation plot showing the correlation of differential DNA methylation between responders [R] and non-responders [NR] over time. Dots in grey represent CpG loci located on the Illumina EPIC array, dots in red represent the identified predictor CpG loci. The four quadrants represent the associated differential methylation across time. CpGs in the blue quadrant are hypermethylated in both T1 and T2, whereas CpGs in the red quadrant are both hypomethylated in T1 and T2, indicating time stability. CpGs in the green quadrant are, whereas CpGs in the yellow quadrant are, indicating time instability. B] Intraclass correlation coefficients of the 53 predictor CpGs obtained from previous long-term stability analyses. Grey dashes represent the classification boundaries introduced by Koo and Li, with blocks representing poor [ICC <0.5], moderate [0.5 ≤ ICC < 0.75], good [0.75 ≤ ICC < 0.9], and excellent [0.9 ≥ICC]. C] Estimated blood cell distribution of the monocytes, NK cells, CD8+ T cells, B cells, CD4+ T cells, and neutrophils using Houseman. The x-axis of each box indicates the difference between responders [R] [red] vs non-responders [NR] [blue]. The y-axis of each box shows the proportion of that particular cell type. A significantly lower proportion of monocytes at T2 were observed for patients responding to tofacitinib.