Tissue-based Gene Expression as Potential Biomarkers for IBD Course

Abstract

Inflammatory bowel diseases (IBDs) are highly heterogeneous in disease phenotype, behavior, and response to therapy. Diagnostic and therapeutic decisions in IBD are based primarily on clinical and endoscopic severity and histopathologic analysis of intestinal biopsies. With this approach, however, only a minority of patients experience durable remission. This may be due to substantial heterogeneity in disease pathogenicity that is not accounted for by current classification systems. Patients can present with similar clinical and endoscopic severity and receive similar therapy but show divergent response ranging from mucosal/transmural healing to nonresponse. Using mucosal biopsy samples that are already obtained as part of the clinical practice to support the diagnosis and state-of-the-art high throughput sequencing approaches can detect the widest range in host gene expression in the actual lining of the affected gut. These analyses can better dissect disease heterogeneity and guide potential treatment response. Here we review studies that use gut tissue–based gene expression profiles to predict disease outcome in IBD.

Inflammatory bowel diseases are highly heterogeneous in response to therapy. Gut mucosal biopsy can detect host gene expression in the affected gut. This article reviews the use of tissue-based gene expression profiles to direct treatment response and IBD outcome.

BACKGROUND AND UNMET NEED

Inflammatory bowel diseases (IBD), Crohn’s cisease (CD), and ulcerative colitis (UC) cause gut tissue chronic inflammation, have no cure, and require life-long therapy. They are highly heterogeneous in disease phenotype, behavior, and—most importantly—in response to therapy.^1–3 Even with best available therapy, the overall rate of lasting remission is only <50%, and a substantial fraction undergo surgical resection. Patients with similar clinical and endoscopic severity can have different outcomes, ranging from mucosal healing to persistent refractory disease, illustrating the lack of sufficient power of clinical factors alone to predict disease outcome. Because the disease primarily involves gut tissue, tissue-based factors may better capture patient-specific drivers of disease and can be used as potential biomarkers. To be able to account for disease heterogeneity, large patient cohorts with mucosal biopsies are needed to explore which biomarkers can predict a predefined specific disease outcome. Important studies focused on IBD genetic risk variants and how those are associated with disease pathogenesis,⁴ type, and natural history,⁵ but those will not be covered in this review. This review will try to cover several attempts of using gut tissue–based gene expression profiles alone or in conjunction with clinical factors to predict several disease outcomes in both CD and UC.

TECHNOLOGY

High throughput transcriptomics or gene expression technologies (RNAseq and microarrays) are used to capture the total genes and transcripts expressed in a specific condition and tissue. Microarrays measure the abundances of a defined set of transcripts via their hybridization to an array of complementary probes, whereas RNAseq is accomplished by reverse transcribing RNA in vitro and sequencing the resulting cDNA. Bulk freshly isolated whole mucosal biopsies of RNA-seq and microarray technologies have been used to study gene expression patterns. Bulk RNASeq cannot identify with certainty cell-type specific pathways, which would be identified by single-cell RNA sequencing (scRNA-seq) technologies and complementary immune cell phenotyping including CyTOF⁶ (a system that combines elemental metal isotopes conjugated to antibodies to evaluate more parameters simultaneously on individual cells than used in traditional flow-cytometry). However, bulk biopsy use is more feasible technically and economically than using single-cell transcriptomics because those are used in the clinical setting for diagnosis and follow-up and can be used “as is” without further immediate processing (after placement in specific RNA preservative solution) as a potential future diagnostic/prognostic tool. Use of fresh biopsies can potentially be substituted by formalin-fixed, paraffin-embedded (FFPE) tissue samples that are routinely stored in clinical pathology archives, and its use for transcriptomics was previously tested with good results.^{7, 8}In addition, more targeted gene panels⁹ can be applied as biomarkers instead of high throughput RNAseq transcriptomics, depending on cost effectiveness.

TISSUE-BASED GENE EXPRESSION PANELS IN CD

RISK¹⁰ is a pediatric CD inception cohort study including 28 sites in the United States and Canada. Nine hundred thirteen CD patients with inflammatory disease behavior (B1) who were younger than 18 years old at diagnosis were enrolled between 2008 and 2012 (ClinicalTrials.gov identifier NCT00790543). We used RISK to define baseline factors associated with pathogenesis,^{11, 12} type (cCD [L2 colon only CD] and iCD [L1 ileal/L3 ileal colonic disease]), severity (presence of deep ulcers), steroid-free clinical remission after 6 month from diagnosis,¹³ and progression to disease complications during follow-up¹⁰ ( Fig. 1).

FIGURE 1.

Cartoon highlighting the RISK Chon’s disease treatment–naïve cohort setting and aims (A), and the PROTECT ulcerative colitis treatment–naïve cohort setting and aims (B).

Using RISK, we were able to show that while endoscopic appearance lacks classic inflammatory process in the ilea of cCD patients, changes in gene expression and microbial community seen in iCD were remarkably preserved within the cCD subgroup. We defined an APOA1 gene coexpression ileal signature that is CD specific and can potentially be used to differentiate cCD from UC. Using the Support Vector Machine, we developed a classification model that was built upon 45 UC and 37 cCD patients, and then tested its accuracy in an independent validation cohort (26 cCD and 28 UC patients), which resulted in 76% overall accuracy (accurate classification of 41 of 54 patients) in the validation, demonstrating potential feasibility and a reasonable level of accuracy to classify the 2 colon-only forms of IBD¹³ using the ileal gene expression data.

In agreement with the discordance between clinical (Pediatric Crohn’s disease activity index) and mucosal, we were unable to identify shared microbial shift and mucosal-based gene signature for both tissue injury and clinical severity. We were able to characterize robust gene expression signature associated with mucosal severity (deep ulcers), linked with Th1 polarization and oxidative stress pathways. In contrast, we detect asociation between depletion of specific Bacteroidetes and Firmicutes taxa and expansion of Proteobacteria with clinical severity (Pediatric Crohn’s disease activity index). This would suggest that ileal gene expression is more linked with mucosal healing, whereas microbial community can be served as a biomarker for clinical remission.¹³ Indeed, a prediction model for steroid- and surgery-free remission (SSFR) that included clinical, microbial factors, and gene expression was superior (area under the curve [AUC] of 0.760; P = 0.0043 [likelihood ratio test]) to a model that included only clinical factors (AUC of 0.705).¹³ These results need to be validated independently. Subsequently, single-cell RNAseq (scRNASeq) defined a mononuclear phagocyte, activated T- cell, and fibroblast module in adult CD ileal resection specimens,⁶ that when applied to the prior RISK bulk RNASeq data set, was induced in patients who did not achieve durable steroid-free remission with antitumor necrosis factor (anti-TNF).

Ileal stricturing behavior (B2) is the major indication for resective surgery in Crohn’s disease (CD). We detected a pronounced extracellular matrix (ECM) gene signature in patients who ultimately progressed to strictures.¹⁰ We reported that anti-Saccharomyces cerevisiae antibodies and CBir1 serology and ileal extracellular matrix gene expression were associated with future stricturing complications within 3 years from diagnosis.⁴ The ECM signature improved the discriminant power of the serology clinical model with respect to specificity and positive predictive value. This model with the ECM receiver operating characteristic (ROC) was 0.72, sensitivity of 69%, specificity of 71%, positive predictive value of 24%, and negative predictive value of 94%. In a work in progress, we included 237 CD subjects from RISK, of which 48% (113) were not included in the previous analyses; we were able show ROC of 0.82, sensitivity of 88%, specificity of 70%, positive predictive value of 18%, and negative predictive value of 99% for future stricturing complications within 5 years from diagnosis using a subset of the ECM signature.

TISSUE-BASED GENE EXPRESSION PANELS IN UC

The Predicting Response to Standardized Pediatric Colitis Therapy (PROTECT, Clinical Trials: NCT 01536535) study is the largest prospective inception cohort study to examine factors associated with early responses to standardized, first-line therapy in pediatric UC.¹ The PROTECT study included 428 newly diagnosed pediatric UC patients from 29 pediatric gastroenterology centers in North America. In this study, we systematically examined response to consensus-defined disease severity–based treatment regimens guided by the Pediatric Ulcerative Colitis Activity Index (PUCAI).¹ Pretreatment rectal gene expression was defined by mRNA-Seq for a representative discovery group of 206 UC PROTECT patients, 20 age- and sex-matched non-IBD controls, and a validation group of 50 UC PROTECT patients.¹⁴ At diagnosis, demographic and clinical data were recorded, disease was clinically and endoscopically graded, and rectal biopsy histology was centrally read.¹⁵ We captured robust gene expression and pathways that are linked to UC pathogenesis, severity, early response to corticosteroid (CS) therapy after 4 weeks,¹⁴ week-52 CS-free remission with mesalazine alone, and escalation to anti-TNF-α by week 52 for moderate/severe patients¹⁶ (Fig. 1 and Table 1). Interestingly, we captured a profound suppression of mitochondrial genes and functions in colonic mucosa at diagnosis across cohorts in UC; however in RISK, CD ileum mitochondrial gene signature was protective from stricturing complications (patients at risk of stricturing with higher mitochondrial genes remained complication free). In PROTECT¹ and in previous studies in children and adults,^{2, 17, 18} higher baseline disease severity identified patients less likely to achieve remission with corticosteroids. We were able to show that adding baseline gene expression data¹⁴ can supplement and improve those models. We defined baseline UC severity genes and pathways and prioritized those regulating innate CXCR-mediated leukocyte recruitment and epithelial transformation. We identified a corticosteroid response gene signature and validated it in an independent subset of UC patients. The corticosteroid response gene signature is enriched for chemokines CXCL/6/8/10/11/17 and cytokines CXCR1/2, which promote recruitment of neutrophils, activation of the innate immune system, and response to bacteria and external stimuli. Notably, the corticosteroid response gene signature showed a substantial overlap with genes previously associated with anti-TNF response and exhibited a similar consistent difference between responders and nonresponders to anti-integrin α ₄β ₇or anti-TNF therapies. A multivariable analysis combining the corticosteroid response gene signature PC1 and ALOX15 gene expression with clinical variables better predicted corticosteroid responsiveness than clinical factors alone, with an AUC of 0.777 (95% CI, 0.692–0.848), sensitivity of 62.7%, (95% CI, 52.8–72.5%), specificity of 76.6% (95% CI, 0.68.8%–84.4%), positive predictive value of 72.3%, and negative predictive value of 67.8% (AUC cutoff at ≥0.5). Interestingly and in alignment with previous reports^{19, 20} linking Oncostatin M (OSM) and Triggering Receptor Expressed On Monocytes 1 to anti-TNF response, colon UC scRNAseq analyses demonstrated that inflammatory monocytes and fibroblast may mediate resistance to anti-TNF via expression of OSM and OSMR, respectively.²¹

TABLE 1.

Examples of Gut-based Gene Signatures Linked with UC Outcome in PROTECT

Early response (W4) to corticosteroid (CS) in PROTECT UC	ABCG2, ADAMTS4, ADGRF1, AGT, ANGPTL4, APOBEC3A, AQP8, AQP9, BEAN1, C2CD4A, C4BPA, CA1, CA2, CD177, CD274, CD300E, CHP2, CKB, CLDN1, CLDN14, CLEC4D, CLEC5A, CMTM2, CSF2, CSF3, CSF3R, CXCL10, CXCL11, CXCL17, CXCL6, CXCL8, CXCR1, CXCR2, DEFB4A, ENKUR, FABP1, FAM83A, FCAR, FCER1A, FCGR3B, FCN3, FOSL1, FPR1, FPR2, FRMD1, GAL, GLDN, GLRA2, GLT1D1, GPR84, GUCA2A, GUCA2B, HCAR2, HCAR3, HMGCS2, HP, IFNG, IGSF9, IL11, IL1A, IL1B, IL6, INHBA, ITGA2, KCNJ15, KRT6A, KRT6B, LILRA6, LYPD1, MCEMP1, MMP1, MMP10, MMP3, NFE2, OSM, OTOP2, PCK1, PLA2G12B, PLAU, PLLP, PPBP, PROK2, RBP2, REG1A, RND1, S100A12, S100A8, S100A9, SAA1, SAA2, SAA2-SAA4, SAA4,, SELE, SFRP2, SLC11A1, SLC26A2, SLC26A3,, SLC30A10, SLC6A14, SPRR1B, SPRR2A, SPRR3, STC1, SULT1A2, TCN1, TMEM72, TMIGD1, TNIP3, TREM1, TRPM6, UGT1A8, USP2, VSIG1, VSTM2A, WISP1
33 genes that include epithelial transporters and channels, and anti-microbial peptides associated with escalation to anti-TNF-α by week 52 in PROTECT UC	ABCG2, AQP8, BEAN1, C4BPA, CA1, CHP2, CKB, CTSG, DEFA5, DEFA6, DEFB4A, GLRA2, GUCA2A, GUCA2B, HAVCR1, KRT6A, KRT6B, LYPD1, PCK1, PLA2G12B, PRSS2, REG1A, SAA2, SAA4, SLC26A2, SLC26A3, SLC30A10, SLC6A19, SPRR1B, SPRR2A, SULT1A2, TCN1, TMIGD1
Antimicrobial peptide signature in W52 SFR in PROTECT UC	CD4, DEFA5, DEFA6, PRSS2, PRSS3

Week-52 steroid-free remission (SFR) is defined as clinical remission (PUCAI <10), no corticosteroids for ≥4 weeks immediately before week 52, and no medical therapy beyond mesalazine or colectomy.¹⁶ An antimicrobial peptide signature including alpha-defensins was significantly lower in patients achieving week-52 SFR in comparison with those who did not (P = 0.0002). When including the antimicrobial peptide signature principal component analysis to a clinical model that included week-4 remission and also Ruminococcaceae (560535) operational taxonomic unit and Sutterella (589923) operational taxonomic unit abundance in 177 PROTECT subjects, the model AUC was 0.75 (95% CI, 0.68–0.83), sensitivity was 49% (95% CI, 38%–61%), specificity was 78% (95% CI, 70%–85%), PPV was 58% (95% CI, 46%–71%), and NPV was 71% (95% CI, 62%–79%). Additionally, 40% of moderate/severe patients escalated to anti-TNF-α in PROTECT by week 52. A panel of 33 genes that included epithelial transporters and channels and antimicrobial peptides was significantly associated with escalation to anti-TNF-α (P < 0.0001) by week 52. When added to a model that included categorical 25-hydroxy vitamin D, hemoglobin ≥10 g/dL, week-4 remission, the model significantly improved in characteristics, with AUC of 0.88 (95% CI, 0.81–0.94), sensitivity of 66% (95% CI, 52%–80%), specificity of 83% (95% CI, 74%–91%), PPV of 71% (95% CI, 57%–85%), and NPV of 79% (95% CI, 70%–85%).

Another attempt for tissue-based biomarker discovery was implemented in the ACT1 infliximab study, optimized in PURSUIT, and then tested in PROgECT (NCT01988961).²² Colon biopsies collected in the ACT1 infliximab study were used to generate the predictive gene expression signature, and baseline expression levels of 109 probe (81 genes) distinguished responders with complete mucosal healing and histologic normalization from nonresponders at week 8 with >90% sensitivity and specificity.²³ The predictive gene panel was then refined using the the PURSUIT golimumab colon biopsies,²⁴ resulting in a signature of 13 genes that achieved the maximum area under the receiver operating characteristic curve (AUCROC) value for predicting mucosal healing response in week 6 in PURSUIT (AUCROC of 0.768). Using the signature of 13 genes was referred to as the molecular prediction signature (MPS) and was then prospectively evaluated for prediction of mucosal healing at week 6 and week 30 in 103 patients treated with golimumab in the open-label study PROgECT. Their results indicated that MPS was significantly better than chance at predicting mucosal healing at both weeks 6 (AUC 0.688) and 30 (AUC 0.671), with high sensitivity (87%) but low specificity (34%) in PROgECT in comparison with PURSUIT, reflecting a high false-positive rate or overprediction of mucosal healing responders. Based on this, they concluded that the test is not yet clinically applicable as a tool to predict response and that further efforts should be implemented to to identify robust response predictors for IBD.

FUTURE DIRECTIONS FOR IMPROVED DEVELOPMENT OF TISSUE GENE EXPRESSION PANELS

Tissue-based gene expression panels have shown some feasibility to predict and guide disease course. However, as was shown in the different studies, several matters need to be acknowledged during their implementation in practice, including definition of biomarkers-specific outcome, use of technology, and ensuring validation of the results in independent cohorts aiming for both high sensitivity and specificity of the tested biomarker.

Outcome is a key part that needs to be defined for biomarkers. Therapeutic goals can include clinical improvement/remission, endoscopic healing (EH), transmural healing (TH) defined by magnetic resonance enterography (MRE), and complete/deep healing determined by both ileocolonoscopy and MRE. Other potential outcomes may be linked with disease complications, surgeries, and patient-reported outcomes. Although all these seem important, perhaps the most important outcome in the personalization of care, in the eyes of the reviewer, is the ability to predict if a specific given treatment will improve a patient’s condition (early response) and lead to complication-free, steroid-free, long-lasting remission.

Choosing technology is another key factor. There are clear advantages in using clinical archived pathology FFPE section from mucosal biopsies obtained at the time of endoscopy (retrospectively and prospectively), as those are already collected routinely and are used for diagnosis and follow-up. However, their robust utility for tissue-based biomarkers needs to be further defined. Alternatively, fresh samples that are collected in conditions that preserve RNA in ambient condition can be utilized and then sent to the analytic laboratory. Those can be analyzed using high throughput methodologies or using a predefined gene panel sequencing or quantitative polymerase chain reaction (qPCR) technologies. To enable emergence of tissue-based biomarkers in clinical practice, these need to be tested in independent cohorts. Replication and validation in independent data sets is crucially needed and requires financial investment. Ways to facilitate such efforts can emerge from federal and industrial efforts and can be further facilitated by national and international collaboration of human-based cohorts and bio-specimens. The Crohn’s Colitis Foundation (CCF) IBD plexus initiative is one such example, aiming to centralize and link data from diverse research studies to facilitate sharing across the research community. The RISK study described here is already embedded in IBD plexus, in addition to other ongoing studies such as the Study of a Prospective Adult Research Cohort with IBD (SPARC IBD), the Quality of Care Program (IBD Qorus), and IBD Partners: Online Patient Survey Study.

Glossary

Abbreviations

B1

inflammatory disease behavior

B2

stricturing behavior

B3

penetrating behavior

CCF

the Crohn’s Colitis Foundation

CD

Crohn’s disease

cCD

L2 colon only CD

CS

corticosteroid

EH

endoscopic healing

FFPE

formalin-fixed, paraffin-embedded

IBD

inflammatory bowel diseases

iCD

L1 ileal/L3 ileal colonic disease

MRE

magnetic resonance enterography

scRNA-seq

single-cell RNA sequencing

SFR

steroid free remission

TH

transmural healing

UC

ulcerative colitis

Supported by: The RISK cohort was supported by the Crohn’s and Colitis Foundation, and the PROTECT was supported by NIDDK 5U01DK095745 and by the Gene Analysis and Integrative Morphology cores of the National Institutes of Health (NIH)-supported Cincinnati Children’s Hospital Research Foundation Digestive Health Center (1P30DK078392-01). The author is also supported by the Israel Science Foundation (908/15), the I-CORE program (41/11), Bill and Melinda Gates Foundation (OPP1144149), the Helmsley Charitable Trust, and the ERC (758313).

Conflicts of Interest: The author has no conflict of interests to disclose.