AbstractThe expanding portfolio of targeted therapies for ulcerative colitis (UC) suggests that a more precise approach to defining disease activity will aid clinical decision-making. This prospective study used genome-wide microarrays to characterize gene expression in biopsies from the most inflamed colon segments from patients with UC and analyzed associations between molecular changes and short-term outcomes while on standard-of-care treatment. We analyzed 141 biopsies—128 biopsies from 112 UC patients and 13 biopsies from eight inflammatory bowel disease unclassified (IBDU) patients. Endoscopic disease was associated with expression of innate immunity transcripts, e.g. complement factor B (CFB); inflammasome genes (ZBP1 and PIM2); calprotectin (S100A8 and S100A9); and inflammation-, injury-, and innate immunity-associated pathway analysis terms. A cross-validated molecular machine learning classifier trained on the endoscopic Mayo subscore predicted the endoscopic Mayo subscore with area-under-the-curve of 0.85. A molecular calprotectin transcript score showed strong associations with fecal calprotectin and the endoscopic Mayo subscore. Logistic regression models showed that molecular features (e.g. molecular classifier and molecular calprotectin scores) improved the prediction of disease progression over conventional, clinical features alone (e.g. total Mayo score, fecal calprotectin, physician global assessment). The molecular features of UC showed strong correlations with disease activity and permitted development of machine-learning predictive disease classifiers that can be applied to expanded testing in diverse cohorts.
Background & aimsThe goal of treatment in ulcerative colitis (UC) is to suppress disease activity and improve patient outcomes. This is balanced against the risk of medication side effects and over-treatment that can lead to adverse outcomes and unnecessary expense. Currently, disease assessment is hindered by limitations of current clinical classification systems1, and the suboptimal understanding of the relationship between disease phenotype and patient outcomes2. Conventional assessment of UC uses a combination of clinical parameters (e.g. partial Mayo score), biochemical markers (e.g. fecal calprotectin), and endoscopic evaluation (e.g. endoscopic Mayo subscore)2. Fecal calprotectin is non-invasive but has limited sensitivity (78%) and specificity (73%) for prediction of patient outcomes3. The endoscopic Mayo subscore is widely accepted4,5 but has substantial intra- and interobserver variability6 and the categorical scores (0–3) have limited ability to reflect disease heterogeneity. Histology scoring (i.e. Geboes Score) is labor-intensive, requires specialized pathologists, and is also subject to substantial interobserver variability7,8.There is an unmet need that can be addressed using molecular tools capable of assessing UC and extending the current classifications via mucosal biopsies obtained during endoscopy. Many studies have focused on assessing differentially-expressed genes between UC biopsies and normal colon tissue9,10,11, or severe UC versus mild/moderate UC using spatially-resolved single-cell sequencing12, but faced difficulties in predicting disease severity and/or were under powered. Recent studies have also speculated at the interplay between ‘epithelially-activated’ and ‘immune-activated’ disease to explain the varying response to biologics such as golimumab, infliximab, and vedolizumab13. We previously used microarrays to make genome-wide measurements of gene expression in UC biopsies14, showing a large-scale disturbance involving inflammation and parenchymal injury and dedifferentiation, with similarities to transcript changes seen in other chronic inflammatory conditions15,16,17,18,19. Expression of transcript sets previously associated with T cell activity and parenchymal injury correlated with the endoscopic Mayo subscore and the presence of lamina propria lymphoplasmacytic infiltrate in colon biopsies14.The present study aimed to develop a molecular scoring system that would add insight and value to existing systems. We selected microarrays as the molecular platform rather than sequencing because microarrays are highly established and can be standardized for operation in testing centers and interpreted by machine learning algorithms20,21. We documented the molecular changes that correlate with UC disease activity (i.e. endoscopic Mayo subscores > 1 vs. ≤1) to understand the biological mechanisms operating at the mucosal level, developed cross-validated molecular classifiers to reduce UC disease activity to a molecular scoring system that would provide continuous numbers rather than categorical classes, and evaluated the relationship of these scores to future disease activity and clinical outcomes. An overview of the study workflow is shown in Fig. 1.Fig. 1Study flowchart.Full size imageResultsPatient population and demographics141 colon biopsies from 120 patients with UC based on current guidelines22,23 were prospectively collected from two quaternary academic referral centers in Canada and the United States. A small number of biopsies (N = 13 biopsies from 8 patients) initially labeled as UC had some indeterminate features (such as non-confluent disease on endoscopy and minimal bleeding) on review. Our review relabeled these biopsies as inflammatory bowel disease unclassified (IBDU). These patients were originally characterized as UC previously and were clinically managed as such. Table 1 summarizes patient and biopsy demographics. Of note, biopsies used in this study were collected from the most inflamed area, as described in the methods.Table 1 Demographics of the patients (N = 120) and biopsies (N = 141).Full size tableTranscripts associated with UC disease activityTotal RNA was extracted from biopsies and processed for Affymetrix PrimeView GeneChip microarrays (see methods). We analyzed the transcripts associated with UC disease activity, high (> 1) vs. low (≤ 1) endoscopic Mayo subscore, by ranking each transcript’s association strength (unadjusted P value) within the UC biopsies (N = 128). Molecules of special interest in UC were flagged (Fig. 2). At P < 0.05, we found 21,768 probe sets (11,658 unique transcripts) with increased or decreased expression in association with a high endoscopic Mayo subscore.Fig. 2Molecular landscape of ulcerative colitis in 128 biopsies as shown by a volcano plot. Probe sets towards the upper right have high association and fold change, indicating a strong relationship with UC disease activity as represented by increased endoscopic Mayo subscore (> 1 vs. ≤ 1). Probe sets towards the middle and further left have moderate to lower associations and fold change, indicating a lack of relationship with UC activity. The FDR (false discovery rate) of < 0.05 is indicated by the grey, dashed line. Transcripts of interest were annotated. Abbreviations: UC, ulcerative colitis.Full size imageThe top 30 transcripts (by P value) increased in UC with a high endoscopic Mayo subscore are shown as blue dots in Fig. 2. The top transcript was cell division cycle 25B (CDC25B, unadjusted P = 4.6 × 10−16), a phosphatase and regulator of G2/M phases of the cell cycle with broad expression24. Other top transcripts included components of innate immunity expressed in macrophages e.g. complement factor B (CFB, P = 1.8 × 10−13) and macrophage gene CHI3L3; regulators of the NLRP3 inflammasome, e.g. serine/threonine kinase PIM2 (P = 3.9 × 10−13); and markers of endothelial injury and matrix remodeling, e.g. lysyl oxidase-like 2 (LOXL2, P = 4.3 × 10−13). A number of these transcripts including VWF25, CFB26, and LOXL227 have previously been shown to be related to UC activity. Calprotectin transcripts S100A8 and S100A9 were also associated with disease activity (P = 1.5 × 10−9 and 6.2 × 10−9), but not within the top 30.Targets of advanced therapy used in UC were associated with disease activity (Fig. 2, green dots): tumor necrosis factor alpha (TNF, P = 5.0 × 10−5); IL12B (p40 subunit, P = 6.9 × 10−4), which forms heterodimers with IL12A (P = 1.8 × 10−3) or IL23A (p19 subunit, P = 5.2 × 10−6); integrin alpha 4 (ITGA4, P = 4.9 × 10−6) and ITGB7 (P = 1.0 × 10−3); and Janus kinase JAK2 (P = 3.7 × 10−10), JAK3 (P = 6.6 × 10−9), and JAK1 (P = 2.0 × 10−8).Inflammasome transcripts associated with UC disease activity included caspase recruitment domain family, member 8 (CARD8, P = 7.3 × 10−9), nucleotide-binding oligomerization domain containing 2 (NOD2, P = 2.4 × 10−7), NLRP3 (P = 2.1 × 10−3) and NLRP9 (P = 3.0 × 10−2) transcripts (Fig. 2, yellow dots), significant in UC for their relationship to pathogenesis but also their potential role in colonocyte wound repair28,29. Three inflammasome transcripts implicated in epithelial stem cell (EpSC) reprogramming during injury30,31 were moderately associated with the endoscopic Mayo subscore: interferon-inducible protein (AIM2, P = 5.7 × 10−8), interleukin 1 beta (IL1B, P = 2.1 × 10−7), and caspase 1 (CASP1, P = 3.3 × 10−6).Solute carrier transcripts highly expressed in normal colon14 were decreased in active UC e.g. SLA26A2 (involved in matrix organization) and bicarbonate transporter SLC4A4 (Fig. 2, tan dots).Details of the top 30 transcripts increased in biopsies with high endoscopic Mayo subscores are shown in Table 2, and the top 30 transcripts with decreased expression in biopsies with high endoscopic Mayo subscore are outlined in Table 3.Table 2 Top 30 unique transcripts by P value with increased expression in biopsies with endoscopic Mayo subcore >1 vs. ≤1 (N = 128 biopsies).Full size tableUC activity transcripts are primarily expressed in inflammatory cells and colonocytesWe analyzed the expression of the top transcripts (increased or decreased in biopsies with high endoscopic Mayo subscore) in colonic tissue using the Human Protein Atlas32 (Tables 2 and 3). Many of the top increased transcripts showed elevated expression in granulocytes and colonocytes (e.g. CDC25B, VWF, ZBP1), suggesting nonspecific inflammation and tissue injury. Some top increased transcripts showed significantly increased expression in B or T cells (e.g. PIM2, LOXL2, IFITM2), suggesting adaptive immune activity contributing to the disease.Table 3 Top 30 unique transcripts by P value with decreased expression in biopsies with endoscopic Mayo subscore >1 vs. ≤1 (N = 128 biopsies).Full size tableTop decreased transcripts showed predominant expression in colonocytes, intestinal endocrine cells, undifferentiated cells, and Paneth cells – indicating a loss of parenchymal function and structural integrity during UC flares (endoscopic Mayo subscore > 1).Pathway analysis of UC disease activity emphasize tissue injury and prominence of innate immunityTop Gene Ontology (GO) terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways (P < 0.05) are summarized in Table 4.Table 4 Top 10 pathways from overrepresentation analysis of top 150 genes increased in the comparison of UC biopsies with endoscopic Mayo subscore >1 vs. ≤1 (N = 128 UC biopsies).Full size tableThe top GO terms classified as Biological Process (BP) and Cellular Compartment (CC) were mainly associated with matrix and remodeling, e.g. extracellular matrix organization (P = 3.5 × 10−11) and basement membrane (P = 1.5 × 10−8). Top Molecular Function (MF) terms were associated with matrix and cellular structure, e.g. extracellular matrix structural constituent (P = 5.6 × 10−11). Top KEGG pathways included protein digestion and absorption (P = 4 × 10−4), and peroxisome proliferator-activated receptors (PPAR) signaling pathway (P = 4.8 × 10−4), which is involved in cellular differentiation, development, metabolism, and tumorigenesis. The TNF signaling pathway was ranked 17th in the KEGG analysis (P = 0.003).The GO terms are visually represented in Fig. 3. BP terms were separated into three interconnected groups: the purple group, terms mostly associated with inflammation and immunity e.g. ‘innate immune response’; the blue group, terms associated with cell regulation, cytokine production, and protein generation/metabolism e.g. ‘protein metabolism’ and ‘protein maturation’; and the pink group, terms associated with the response to wounding. CC terms (yellow group) were dominated by matrix (e.g. ‘extracellular matrix’), membrane (e.g. ‘basement membrane’), and collagen-associated terms (e.g. ‘fibrillar collagen’). MF terms (green group) focused on matrix structure (e.g. ‘extracellular matrix structure’) and protein binding (e.g. ‘receptor binding’) but also highlighted the significance of heat shock proteins (e.g. ‘HSP protein binding’), implicated in protecting against progression of UC disease33.Fig. 3Overrepresentation analysis of top 150 transcripts differentially expressed between biopsies with endoscopic Mayo subscore > 1 vs. ≤ 1. Nodes and edges were generated using Cytoscape and BiNGO. Nodes (circles) are colored by significance (darker colors = higher P value), and are grouped according to pathway term type (i.e. biological process – blue, purple, and pink; cellular compartment - yellow, molecular function - green). Redundant relationships were removed to simplify the visual output.Full size imageMolecular classifiers predict UC disease activityWe developed two molecular classifiers using ten-fold cross-validation and an ensemble of 12 machine learning algorithms for predicting disease activity, trained on the endoscopic Mayo subscore (> 1 vs. ≤1). The first classifier, MayoProb1, used only UC biopsies (N = 128), and the second, MayoProb2, used a slightly larger population of both UC and IBDU biopsies (N = 141). Both classifiers predicted endoscopic Mayo subscore > 1 with an area-under-the-curve (AUC) of 0.85 (Fig. 4A).Fig. 4Molecular classifier performance. (a) estimated performance of both classifiers shown by area-under-the-curve (AUC). Biopsies were also plotted in beeswarm-boxplots showing their (b) MayoProb1 classifier score vs. endoscopic Mayo subscore, and (c) their MayoProb2 classifier score vs. endoscopic Mayo subscore, showing a mean increase in classifier score as the endoscopic Mayo subscore increases. (d) Random forest showing relative variable importance in prediction of 3–6 month post-biopsy status. Molecular (molecular calprotectin, both classifier scores) and standard-of-care variables (fecal calprotectin, partial Mayo score, total Mayo score, endoscopic Mayo subscore) were compared for their importance in predicting 3–6 month post-biopsy patient status.Full size imageClassifier scores for both MayoProb1 (Fig. 4B) and MayoProb2 (Fig. 4C) increased with a rise in the endoscopic Mayo subscore in the biopsies.Molecular scores correlated with clinical variablesWe assessed the correlation between the two molecular classifiers (MayoProb1 and MayoProb2), calprotectin transcript set score (MCalpro, the geometric mean of the standardized expression scores of the S100A8 and S100A9 transcripts, see Methods), and several clinical variables: total Mayo score, endoscopic Mayo subscore, and fecal calprotectin (Table 5). Clinical variables were ordinal and categorical while molecular scores were continuous numbers.Table 5 Spearman correlation coefficients – molecular classifier scores compared with clinical features in UC and IBDU biopsies (N = 141 biopsies).Full size tableThe MayoProb1 classifier score correlated with endoscopic Mayo subscore (correlation coefficient 0.67), the physician’s global assessment at biopsy (0.69), and the total Mayo Score (0.61). The MayoProb2 classifier correlations were similar to those of MayoProb1.MCalpro correlated with fecal calprotectin (correlation coefficient 0.63), endoscopic Mayo subscore (0.59), total Mayo score (0.62), and physician’s global assessment (0.67).Molecular features correlated only moderately overall with the partial Mayo subscoreMolecular features correlated with future patient status. Of the 141 biopsies (colonoscopies), 80 patients had follow-up at 3–6 months (Supplementary Table S1). To assess the relationship between molecular UC disease features at time of biopsy to short-term clinical outcomes, we developed a status code classification that summarized the change between the patient’s disease on the day of biopsy and 3–6 months post-biopsy: patients with poor outcomes (status code > 1) vs. those with favorable outcomes (status code ≤ 1) using t-tests (Table 6). This revealed significant differences in MCalpro, MayoProb1, and MayoProb2 scores (P = 0.01, 0.005, and 0.002, respectively). The endoscopic Mayo subscore also differed significantly, though with a weaker P value (P = 0.02) than the molecular scores. Other clinical variables did not demonstrate differences between those with poor or favorable outcomes.We compared the importance of standard-of-care clinical variables (endoscopic Mayo subscore, partial Mayo score, total Mayo score, and fecal calprotectin) to the MCalpro and classifier scores in random forests for the prediction of the 3–6 month status code (Fig. 4D). The most important variables for this prediction were molecular features (MCalpro and the MayoProb2 classifier score).Table 6 Relationship between various scores at time zero and future status code (assessed at 3–6 months post-biopsy, status code >1 vs. ≤1) for all UC and IBDU biopsies (N = 141 biopsies).Full size tableMolecular scores improve the prediction of future patient status in logistic regression modelsWe used logistic regression to compare two models predicting 3–6 month post-biopsy clinical status (Table 7). Model 1 included standard-of-care variables (fecal calprotectin, partial Mayo score, and endoscopic Mayo subscore). Model 2 included the same standard-of-care variables plus molecular variables: MCalpro, MayoProb1, and MayoProb2. The full model with both molecular and clinical features was better than the model using clinical features alone (P = 0.035).Table 7 Predicting 3–6 month post-biopsy status considering only standard-of-care clinical features and molecular features developed in these analyses (UC and IBDU, N = 141 biopsies).Full size tableDiscussionThis study was designed to define the molecular processes occurring in the mucosa of active UC and assessed whether molecular classifiers could add to the clinician’s ability to predict disease trajectory. We found that the molecular landscape of UC is dominated by innate immune processes, particularly complement factors, and transcripts associated with parenchymal tissue injury. Transcripts were highly expressed in colonocytes and granulocytes, but less so in T and B cells. Molecular classifiers were developed using transcripts most differentially expressed between biopsies with high and low disease activity (endoscopic Mayo subscore > 1 vs. ≤1). A molecular calprotectin score (using S100A8 and S100A9 transcripts) showed strong relationships to fecal calprotectin proteins at time of biopsy. Notably, the molecular classifiers (MayoProb1 and MayoProb2) and the molecular calprotectin score was important for the prediction of poor patient status at 3–6 months post-biopsy, and both added significantly to a model predicting 3–6-month disease status vs. conventional UC disease variables (as per logistic regression). This indicates molecular classifiers have the potential to add to clinical assessment at time of biopsy by anticipating future disease status during standard-of-care management. This may help identify patients who will require intensive therapy early on (i.e. escalated dosing, adjunct steroids, or change of therapy).Because patient response to UC therapies such as anti-TNF, anti-IL12/23, anti-integrin alpha 4 beta 7, and JAK inhibition remains highly variable, prediction at time of biopsy of patients who may have a poor outcome is important – not only to indicate the need for intensified therapeutic approach, but also for escalating monitoring approaches. Endoscopic and histologic healing have presented alternative endpoints for predicting ‘good’ vs. ‘poor’ outcomes in patients, but both represent outcomes that eventually become predictive. Conversely, molecular scoring is predictive at the time of initial endoscopy: endoscopic and histologic healing are the targets. As initial assessment tools, both histology and endoscopy remain challenging and have not been standardized: endoscopic appearance has a less than optimal relationship with outcomes34, while histologic scoring requires complete histology performed at every endoscopy (not currently the standard-of-care at every center), is subject to interobserver variation35, and is heterogenous across centers36.Machine-learning-derived classifiers present the advantage of being rapidly accessible and highly reproducible when the molecular platform is standardized: analyzing the same mRNA sample twice on microarrays gives the same results with > 99% reproducibility21. Classifier output (scores) are continuous numerical values, capable of describing greater heterogeneity within the diverse UC patient population compared to any ordinal endoscopic or histologic assessment. A molecular report with all scores for a new biopsy can be generated within 48 h of biopsy37, faster than average histologic or fecal calprotectin testing. Overall, classifiers and molecular calprotectin scores add a new, highly reproducible dimension to conventional variables, potentially replacing fecal calprotectin when endoscopy is performed, leaving fecal calprotectin for non-endoscopic assessment.The molecular landscape of UC indicates multiple innate immune processes but is not definitive for a primary role for adaptive immunity e.g. cognate effector T cell activity, despite considerable interest in T cell processes in UC38,39,40. While our previous work found that lymphoplasmacytic infiltrate in UC biopsies correlated with the molecular disturbance14 and is a predictor of relapse in patients41, this may be related to the response-to-wounding and chronicity, rather than active disease. Mucosal injury leads to a breakdown in the epithelial barrier, and microbiota and bacterial products likely perpetuate an ongoing neutrophilic inflammatory process, even if the initial injury process has diminished. One possibility is that initiating and sustaining mechanisms are separate: injury to the epithelium – e.g. a cognate T cell event – can be remembered through reprogramming of epithelial stem cells (EpSCs, through chromatin changes in genomic regions associated with inflammation), rendering the mucosa susceptible to otherwise innocuous inflammatory stimuli30,31. Of interest, inflammasome transcripts correlating with UC activity are among those encoded by genes in open chromatin domains induced by damage in EpSCs (e.g. AIM2, IL-1B, and CASP1)30,31. This finding is compatible with the association between UC activity and the inflammasome, including strong associations of NOD2, IL18, and caspases with UC activity42. Thus cognate T cell-mediated autoimmunity could set the stage for continuing UC activity39,43,44,45 via alterations in EpSCs, reprogramming the epithelium to be vulnerable to local influences such as microbes and their products that are usually innocuous30,31,46. This model may explain the inconsistent efficacy of empirically derived therapies for UC, and why such therapies differ strikingly in their efficacy from a pure cognate T cell driven disease process, like organ transplant rejection.Limitations to this study include the use of standard-of-care management rather than a strict management protocol, but this also provides a real-world environment for our conclusions in preparation for larger studies exploring specific therapeutic protocols or various disease phenotypes, creating a future in which the choice of therapy is guided by the molecular features, as is often the case in oncology. Machine learning creates precise algorithms even when trained on gold-standard clinical assessments that are inherent with errors or variability20. While we selected microarrays for the many benefits of this technology, we acknowledge the inherent limitations compared to other possible methods (e.g. RNA sequencing) for analysis: microarrays limited our analyses to the 19,462 unique transcripts present on the chip, in some cases may limit sensitivity for low-expression genes, and does not provide the same level of data as RNA sequencing for machine learning applications. Therapeutics were not used as a variable in these analyses due to heterogeneity preventing meaningful analysis but may be considered in the future with a larger dataset and more defined groups of therapeutic strategies. Future analyses looking at later time points will better define the predictive ability of molecular scores, and we also plan to assess cases with discrepancy between molecular scores, endoscopy activity, and fecal calprotectin, with a further goal of developing a prospectively followed cohort, with set time points of clinical, biochemical, and endoscopic assessment. Multivariable analysis was unavailable due to the limited dataset size.We acknowledge a desire for less invasive tests (e.g. of peripheral blood), but this must be balanced with deficiencies in this approach. Such tests may offer an improved safety profile with more frequent monitoring, but lose specificity as they correlate with different disease processes19,47,48,49. Recent development of a blood-based gene-expression test for IBD showed some improvement over C-reactive protein, but did not add additional information to fecal calprotectin or endoscopy50. Our goal was to build on and enhance the accepted gold standard-of-care – endoscopic assessment and histology.Management of UC is based on assessments that have suboptimal granularity and reproducibility – standardized molecular assessments can potentially address this need. Most standard-of-care methods are either subjective with high interobserver variation (Mayo scores, physician global assessment), or serve as a binary tool for predicting disease severity (fecal calprotectin) rather than as a means for predicting patient outcomes. The molecular phenotype of UC assessed using standardized microarrays and machine learning classifiers offers a granular approach to assessing disease severity and activity. Furthermore, the molecular toolset developed here provides an additional source of rapidly available, reproducible data when making management decisions, added to standard-of-care assessment, and has the potential for improving patient outcomes.MethodsPatients, biopsy collection, and diagnosesPatients (N = 120) age ≥ 18 years were prospectively enrolled based on an established clinical diagnosis of UC during standard-of-care colonoscopies. 1–4 bites per patient were collected from the most endoscopically inflamed area (as determined by the endoscopist), combined as a single ‘biopsy’ for a total of 141 biopsies, as per protocols at the Center of Excellence for Gastrointestinal Inflammation and Immunity Research (CEGIIR, University of Alberta Hospital, Edmonton, Canada) and at Cedars-Sinai Medical Center (Los Angeles, USA). During patient enrollment, 13 (of the 120) patients underwent more than one colonoscopy at separate time points, hence why a total of 141 biopsies were included. Normal biopsies were also collected from 17 non-UC patients undergoing colonoscopies with no known history of IBD or other gastrointestinal disease. These control biopsies were only used in select analyses, when indicated. All biopsies were placed in RNAlater™, and stored at −20 °C for isolation of RNA.Demographics (age, sex, date of diagnosis), disease status (partial Mayo score)51,52,53, medications at the time of biopsy, and endoscopic data (extent of disease, endoscopic Mayo subscore for colon segments) were all collected at the time of endoscopy and verified with the electronic medical records for all patients. All IBD medications were permitted within this population, including 5-aminosalicylic acid formulations, corticosteroids, immunomodulators, antibiotics, and advanced therapies (biologics and small molecules).All diagnoses were established by experienced IBD gastroenterologists using standard-of-care methods (Supplementary Table S2)14.Study approvalThe study was reviewed and approved by the University of Alberta Health Research Ethics Board (HREB) Biomedical Panel (Edmonton, Canada), and Cedars-Sinai Institutional Review Board (Los Angeles, USA). Participants provided written informed consent to research study staff and received a copy of their signed informed consent. This study was performed in accordance with relevant guidelines, regulations, and with the Declaration of Helsinki.Biopsy processingBiopsies in RNAlater™ were transferred to the Alberta Transplant Applied Genomics Center (ATAGC, Edmonton, Canada) for processing. Total RNA was extracted, cleaned, and labeled using established methods17. Extracted RNA was high quality and yield (mean RNA integrity number (RIN) 7.7, average yield 1.78 µg). Purified total RNA was labeled with the IVT Express labeling kit or IVT Plus labeling kit (Affymetrix, Santa Clara, USA) and hybridized to human PrimeView arrays (Affymetrix) according to manufacturer protocols21.The Affymetrix PrimeView GeneChip microarrays include 19,462 unique (nonredundant) transcripts from 49,495 probe sets available on the array. Microarrays were scanned, ‘CEL’ files were obtained using GeneChip Operating Software (Affymetrix), and robust multiarray averaging was used to normalize the CEL files17. Once all CEL files were available, a correction factor was calculated to normalize CEL file expression data between biopsies processed using different labeling kits.Overrepresentation analysisOverrepresentation of UC activity-associated transcripts in GO and in KEGG pathways was analyzed using the “enrichGO” function from the “clusterProfiler”54 package in R55.Pathway terms were visualized using Cytoscape56 and the associated BiNGO57 and stringAPP58 applications. The top 150 transcripts increased in high endoscopic Mayo subscore biopsies were used (all had an adjusted P < 0.05). This allowed many significant transcripts to be represented, but simplified the visual result by limiting the number of transcripts used.Human protein atlasCellular expression in colon was assessed using the Human Protein Atlas32. Each transcript was searched, and the single-cell expression examined in colonic tissue. The top five cell types were listed when applicable, otherwise only reported cellular expression was listed.Classifier developmentA classifier was developed for predicting disease activity (endoscopic Mayo subscore > 1 vs. ≤1). Upon chart review, it was revealed that a minority of patients were originally labeled with indeterminate features (clinically and endoscopically) indicating a diagnosis of inflammatory bowel disease unclassified (IBDU), so the first classifier was trained on a population that excluded these samples (MayoProb1). The second classifier (MayoProb2) used the larger dataset of all biopsies (UC and IBDU), as patients had complete endoscopic Mayo subscores and had been clinically managed as UC. Both classifiers were used throughout these analyses to best represent real-life clinical practice in UC and to observe the benefit of increased statistical power in classifier development.Using 10-fold cross validation we randomly split the biopsy set used for each classifier (MayoProb1, MayoProb2) into a training set (90% of all samples) and a test set (10% of all samples). This was repeated 10 times for all 10 folds per previously established protocols, resulting in 10 different t-tests and 10 different lists of top 20 probe sets59. This process is summarized in Supplementary Fig. S1. The classifier was trained on binary classes of endoscopic Mayo subscores (disease class = scores > 1, and without disease = scores ≤ 1) as reported by the endoscopist during endoscopy and generated ‘probability of active disease’ estimates ranging from 0.0 to 1.0.Each algorithm used the top 20 probe sets selected by t-test between the classes. For stability, each classifier was trained in 12 machine learning algorithms, and the median of all 12 algorithms was used as an ensemble classifier score, as previously published20.The 12 different machine learning algorithms were: linear discriminant analysis (lda), regularized discriminant analysis (rda), mixture discriminant analysis (mda), flexible discriminant analysis (fda), gradient boosting machine (gbm), radial support vector machine (SVMR), linear support vector machine (SVML), random forest (rf), C5.0, neural networks (nnet), Bayes glm (bayesglm), and generalized linear model elastic-net (glmnet). The output value of the classifier was the median score from 12 separate machine learning algorithms, as this provides increased stability of the estimate.Development of a transcript set score for calprotectin expression in the biopsyTranscript set scores are a set of transcripts represented on the microarray and are chosen by the end user to represent a particular biology or disease mechanism. A transcript set score is calculated as the geometric mean of the standardized expression scores (standardized against a control population). We developed a transcript set to represent calprotectin-associated expression, using the probe sets for S100A8 and S100A9 (as the gene products of S100A8 and S100A9 represent the calprotectin heterodimer). This became the molecular calprotectin score (MCalpro) – the geometric mean expression of S100A8 and S100A9 in a defined population vs. the 17 normal (control) colon biopsies.Development of a status code to describe disease outcomes 3–6 months post-biopsyTo assess the relationship between molecular UC disease features at time of biopsy to short-term clinical outcomes, we developed a status code classification that summarized the change between the patient’s disease on the day of biopsy and 3–6 months post-biopsy (Supplementary Table S3). The status code was assigned by an IBD clinician (BPH), who was blinded to the molecular data. ‘Good’ outcomes were assigned if the patient was in remission 3–6 months post-biopsy (status 0) or had significant improvement with some ongoing disease activity (status 1). ‘Poor’ outcomes were assigned if the patient was having ongoing disease activity with no response to clinical management (status 2) or worsening disease activity (status 3).Derivation of the status code (outcome) was based on disease activity and severity (including biochemical parameters (C-reactive protein, fecal calprotectin), endoscopic Mayo subscores, partial Mayo scores), any change in therapy shortly after the biopsy, therapy response from the time of the biopsy to the 3–6-month reassessment, and assessment of current disease as documented by the endoscopist or treating physician. In cases where follow-up endoscopic evaluation did not occur within 3–6 months, the partial Mayo score was used. The electronic medical record was reviewed to ensure the patient did initiate the prescribed new or escalated treatment(s) between the time of biopsy and the 3–6 month reassess. Total and partial Mayo criteria are described in Supplementary Table S4.This scoring system was created to represent common clinical assessments for a treating physician during standard-of-care reassessment of patients. A 3–6 month time frame was used to provide a reasonable window for disease improvement and response to any management changes shortly after biopsy.AuthorsAll authors had access to the study data and had reviewed and approved the final manuscript. StatisticsStatistical analysis and graphics were done in the “R” software package, version 4.055 with various libraries from Bioconductor 3.260, and in Microsoft Excel version 16 (Redmond, WA). Differentially expressed genes were generated using the ‘limma’ Bioconductor package61. Significance of probe set expression is given as unadjusted P values (Bayesian t-test), except in cases where the false discovery rate (FDR) is specified. Volcano plots were generated using Excel and colored by previously assigned transcript sets of interest.
Data availability
Microarray (CEL) files relevant to this study will be made available on the Gene Expression Omnibus (GEO) website when the paper is published. Deidentified data and methods can be shared on reasonable request to the corresponding author.
ReferencesSilverberg, M. S. et al. Toward an integrated clinical, molecular and serological classification of inflammatory bowel disease: Report of a working party of the 2005 Montreal World Congress of Gastroenterology. Can. J. Gastroenterol. 19(Suppl A), 5A–36A. https://doi.org/10.1155/2005/269076 (2005).Turner, D. et al. STRIDE-II: An update on the selecting therapeutic targets in inflammatory bowel Disease (STRIDE) Initiative of the International Organization for the study of IBD (IOIBD): Determining therapeutic goals for treat-to-target strategies in IBD. Gastroenterology. https://doi.org/10.1053/j.gastro.2020.12.031 (2021).Article
PubMed
Google Scholar
Mao, R. et al. Fecal calprotectin in predicting relapse of inflammatory bowel diseases: A meta-analysis of prospective studies. Inflamm. Bowel Dis. 18, 1894–1899. https://doi.org/10.1002/ibd.22861 (2012).Article
PubMed
MATH
Google Scholar
Sharara, A. I., Malaeb, M., Lenfant, M. & Ferrante, M. Assessment of endoscopic disease activity in ulcerative colitis: Is simplicity the ultimate sophistication? Inflamm. Intest Dis. 7, 7–12. https://doi.org/10.1159/000518131 (2022).Article
PubMed
Google Scholar
Mohammed Vashist, N. et al. Endoscopic scoring indices for evaluation of disease activity in ulcerative colitis. Cochrane Database Syst. Rev. 1, CD011450. https://doi.org/10.1002/14651858.CD011450.pub2 (2018).Article
PubMed
Google Scholar
Venner, J. M. & Bernstein, C. N. Current endoscopic scoring systems in inflammatory bowel disease: Strengths and limitations. Gastrointest. Endosc Clin. N Am. https://doi.org/10.1016/j.giec.2024.04.014 (2024).Article
PubMed
MATH
Google Scholar
Geboes, K. et al. A reproducible grading scale for histological assessment of inflammation in ulcerative colitis. Gut 47, 404–409. https://doi.org/10.1136/gut.47.3.404 (2000).Article
PubMed
PubMed Central
Google Scholar
Jauregui-Amezaga, A. et al. A simplified Geboes score for ulcerative colitis. J. Crohns Colitis 11, 305–313. https://doi.org/10.1093/ecco-jcc/jjw154 (2017).Article
PubMed
Google Scholar
Shi, W. et al. Analysis of genes involved in ulcerative colitis activity and tumorigenesis through systematic mining of gene co-expression networks. Front. Physiol. 10, 662. https://doi.org/10.3389/fphys.2019.00662 (2019).Article
PubMed
PubMed Central
MATH
Google Scholar
Pang, X. et al. Transcriptomic analyses of treatment-naive pediatric ulcerative colitis patients and exploration of underlying disease pathogenesis. J. Transl. Med. 21, 30. https://doi.org/10.1186/s12967-023-03881-6 (2023).Article
PubMed
PubMed Central
MATH
Google Scholar
Huang, J., Zhang, J., Wang, F., Zhang, B. & Tang, X. Revealing immune infiltrate characteristics and potential diagnostic value of immune-related genes in ulcerative colitis: An integrative genomic analysis. Front. Public. Health 10, 1003002. https://doi.org/10.3389/fpubh.2022.1003002 (2022).Article
PubMed
PubMed Central
Google Scholar
Mayer, A. T. et al. A tissue atlas of ulcerative colitis revealing evidence of sex-dependent differences in disease-driving inflammatory cell types and resistance to TNF inhibitor therapy. Sci. Adv. 9, eadd1166. https://doi.org/10.1126/sciadv.add1166 (2023).Article
PubMed
PubMed Central
Google Scholar
Chang, M. J. et al. A compendium of mucosal molecular characteristics provides novel perspectives on the treatment of ulcerative colitis. J. Crohns Colitis 17, 909–918. https://doi.org/10.1093/ecco-jcc/jjad011 (2023).Article
PubMed
PubMed Central
MATH
Google Scholar
Halloran, B. et al. Molecular patterns in human ulcerative colitis and correlation with response to infliximab. Inflamm. Bowel Dis. 20, 2353–2363. https://doi.org/10.1097/MIB.0000000000000239 (2014).Article
PubMed
MATH
Google Scholar
Venner, J., Famulski, K. & Halloran, P. The molecular landscape of T cell-mediated rejection: The T cell synapse emerges. Am. J. Transplant. 13, 376–376. https://doi.org/10.1111/ajt.12266 (2013).Article
Google Scholar
Madill-Thomsen, K. et al. The molecular diagnosis of rejection in liver transplant biopsies: First results of the INTERLIVER study. Am. J. Transpl. 20, 2156–2172. https://doi.org/10.1111/ajt.15828 (2020).Article
MATH
Google Scholar
Halloran, P. F. et al. Real time central assessment of kidney transplant indication biopsies by microarrays: The INTERCOMEX study. Am. J. Transpl. 17, 2851–2862. https://doi.org/10.1111/ajt.14329 (2017).Article
MATH
Google Scholar
Halloran, P. F. et al. Building a tissue-based molecular diagnostic system in heart transplant rejection: The heart molecular microscope diagnostic (MMDx) system. J. Heart Lung Transpl. 36, 1192–1200. https://doi.org/10.1016/j.healun.2017.05.029 (2017).Article
MATH
Google Scholar
Halloran, P. F. et al. Review: The transcripts associated with organ allograft rejection. Am. J. Transpl. 18, 785–795. https://doi.org/10.1111/ajt.14600 (2018).Article
MathSciNet
MATH
Google Scholar
Reeve, J. et al. Generating automated kidney transplant biopsy reports combining molecular measurements with ensembles of machine learning classifiers. Am. J. Transpl. 19, 2719–2731. https://doi.org/10.1111/ajt.15351 (2019).Article
MATH
Google Scholar
Affymetrix. In http://www.affymetrix.com/support/technical/datasheets/primeview_array_cartridge_datasheet.pdf 1–4.Sturm, A. et al. ECCO-ESGAR guideline for diagnostic assessment in IBD part 2: IBD scores and general principles and technical aspects. J. Crohn’s Colitis 13, 273–284. https://doi.org/10.1093/ecco-jcc/jjy114 (2019).Article
MATH
Google Scholar
Maaser, C. et al. ECCO-ESGAR guideline for diagnostic assessment in IBD part 1: Initial diagnosis, monitoring of known IBD, detection of complications. J. Crohn’s Colitis 13, 144–164. https://doi.org/10.1093/ecco-jcc/jjy113 (2019).Article
MATH
Google Scholar
Pira, G. et al. Landscape of transcriptome variations uncovering known and novel driver events in colorectal carcinoma. Sci. Rep. 10, 432. https://doi.org/10.1038/s41598-019-57311-z (2020).Article
ADS
PubMed
PubMed Central
Google Scholar
Zezos, P. et al. Elevated plasma von Willebrand factor levels in patients with active ulcerative colitis reflect endothelial perturbation due to systemic inflammation. World J. Gastroenterol. 11, 7639–7645. https://doi.org/10.3748/wjg.v11.i48.7639 (2005).Article
PubMed
PubMed Central
MATH
Google Scholar
Ostvik, A. E. et al. Mucosal toll-like receptor 3-dependent synthesis of complement factor B and systemic complement activation in inflammatory bowel disease. Inflamm. Bowel Dis. 20, 995–1003. https://doi.org/10.1097/MIB.0000000000000035 (2014).Article
PubMed
Google Scholar
Du, J., Yin, J., Du, H. & Zhang, J. Revisiting an expression dataset of discordant inflammatory bowel disease twin pairs using a mutation Burden Test reveals CYP2C18 as a novel marker. Front. Genet. 12, 680125. https://doi.org/10.3389/fgene.2021.680125 (2021).Article
PubMed
PubMed Central
Google Scholar
Kanneganti, T. D. Inflammatory bowel Disease and the NLRP3 inflammasome. N Engl. J. Med. 377, 694–696. https://doi.org/10.1056/NEJMcibr1706536 (2017).Article
PubMed
MATH
Google Scholar
Mullins, B. & Chen, J. NLRP9 in innate immunity and inflammation. Immunology 162, 262–267. https://doi.org/10.1111/imm.13290 (2021).Article
PubMed
MATH
Google Scholar
Naik, S. et al. Inflammatory memory sensitizes skin epithelial stem cells to tissue damage. Nature 550, 475–480. https://doi.org/10.1038/nature24271 (2017).Article
ADS
PubMed
PubMed Central
Google Scholar
Dai, X. & Medzhitov, R. Inflammation: Memory beyond immunity. Nature 550, 460–461. https://doi.org/10.1038/nature24154 (2017).Article
ADS
PubMed
MATH
Google Scholar
Uhlen, M. et al. Proteomics. Tissue-based map of the human proteome. Science 347, 1260419. https://doi.org/10.1126/science.1260419 (2015).Article
PubMed
MATH
Google Scholar
Gong, M., Zhang, F., Miao, Y. & Niu, J. Advances of heat shock family in ulcerative colitis. Front. Pharmacol. 13, 869930. https://doi.org/10.3389/fphar.2022.869930 (2022).Article
PubMed
PubMed Central
MATH
Google Scholar
Shah, S. C., Colombel, J. F., Sands, B. E. & Narula, N. Mucosal healing is associated with improved long-term outcomes of patients with ulcerative colitis: A systematic review and meta-analysis. Clin. Gastroenterol. Hepatol. 14, 1245–1255. https://doi.org/10.1016/j.cgh.2016.01.015 (2016).Mosli, M. H. et al. Histologic scoring indices for evaluation of disease activity in ulcerative colitis. Cochrane Database Syst. Rev. 5, CD011256. https://doi.org/10.1002/14651858.CD011256.pub2 (2017).Article
PubMed
Google Scholar
Long, M. D. & Rubin, D. T. Histologic remission in ulcerative colitis: Are we there yet? Am. J. Gastroenterol. 114, 713–715. https://doi.org/10.14309/ajg.0000000000000235 (2019).Article
PubMed
Google Scholar
Madill-Thomsen, K. S. & Halloran, P. F. Precision diagnostics in transplanted organs using microarray-assessed gene expression: Concepts and technical methods of the molecular microscope (R) diagnostic system (MMDx). Clin. Sci. (Lond.) 138, 663–685. https://doi.org/10.1042/CS20220530 (2024).Article
PubMed
Google Scholar
Larmonier, C. B., Shehab, K. W., Ghishan, F. K. & Kiela, P. R. T lymphocyte dynamics in inflammatory bowel diseases: Role of the microbiome. Biomed. Res. Int. 2015, 504638. https://doi.org/10.1155/2015/504638 (2015).Kappeler, A. & Mueller, C. The role of activated cytotoxic T cells in inflammatory bowel disease. Histol. Histopathol. 15, 167–172. https://doi.org/10.14670/HH-15.167 (2000).Article
PubMed
MATH
Google Scholar
Casalegno Garduno, R. & Dabritz, J. New insights on CD8(+) T cells in inflammatory bowel disease and therapeutic approaches. Front. Immunol. 12, 738762. https://doi.org/10.3389/fimmu.2021.738762 (2021).Article
PubMed
PubMed Central
Google Scholar
Feagins, L. A., Melton, S. D., Iqbal, R., Dunbar, K. B. & Spechler, S. J. Clinical implications of histologic abnormalities in colonic biopsy specimens from patients with ulcerative colitis in clinical remission. Inflamm. Bowel Dis. 19, 1477–1482. https://doi.org/10.1097/MIB.0b013e318281f4ae (2013).Article
PubMed
Google Scholar
Hornung, V. et al. AIM2 recognizes cytosolic dsDNA and forms a caspase-1-activating inflammasome with ASC. Nature 458, 514–518. https://doi.org/10.1038/nature07725 (2009).Article
ADS
PubMed
PubMed Central
MATH
Google Scholar
Harbour, S. N., Maynard, C. L., Zindl, C. L., Schoeb, T. R. & Weaver, C. T. Th17 cells give rise to Th1 cells that are required for the pathogenesis of colitis. Proc. Natl. Acad. Sci. U. S. A. 112, 7061–7066. https://doi.org/10.1073/pnas.1415675112 (2015).Article
ADS
PubMed
PubMed Central
Google Scholar
Lamb, C. A. et al. alphaEbeta7 integrin identifies subsets of pro-inflammatory colonic CD4 + T lymphocytes in ulcerative colitis. J. Crohns Colitis (2016).Croitoru, K. & Zhou, P. T-cell-induced mucosal damage in the intestine. Curr. Opin. Gastroenterol. 20, 581–586. https://doi.org/10.1097/00001574-200411000-00013 (2004).Article
PubMed
MATH
Google Scholar
Ungaro, R., Mehandru, S., Allen, P. B., Peyrin-Biroulet, L. & Colombel, J. F. Ulcerative colitis. Lancet 389, 1756–1770. https://doi.org/10.1016/S0140-6736(16)32126-2 (2017).Article
PubMed
Google Scholar
Chechlinska, M., Kowalewska, M. & Nowak, R. Systemic inflammation as a confounding factor in cancer biomarker discovery and validation. Nat. Rev. Cancer 10, 2–3. https://doi.org/10.1038/nrc2782 (2010).Article
PubMed
Google Scholar
Pan, F. et al. The potential impact of clinical factors on blood-based biomarkers for Alzheimer’s disease. Transl. Neurodegener. 12, 39. https://doi.org/10.1186/s40035-023-00371-z (2023).Article
PubMed
PubMed Central
MATH
Google Scholar
Gauthier, P. T. et al. Distinct molecular processes mediate donor-derived cell-free DNA release from kidney transplants in different disease states. Transplantation 108, 898–910. https://doi.org/10.1097/TP.0000000000004877 (2024).Article
PubMed
MATH
Google Scholar
Argmann, C. et al. Biopsy and blood-based molecular biomarker of inflammation in IBD. Gut 72, 1271–1287. https://doi.org/10.1136/gutjnl-2021-326451 (2023).Article
PubMed
MATH
Google Scholar
Colombel, J. F. et al. Early mucosal healing with infliximab is associated with improved long-term clinical outcomes in ulcerative colitis. Gastroenterology 141, 1194–1201. https://doi.org/10.1053/j.gastro.2011.06.054 (2011).Article
PubMed
MATH
Google Scholar
Paine, E. R. Colonoscopic evaluation in ulcerative colitis. Gastroenterol. Rep. 2, 161–168. https://doi.org/10.1093/gastro/gou028 (2014).Article
MATH
Google Scholar
Schroeder, K. W., Tremaine, W. J. & Ilstrup, D. M. Coated oral 5-aminosalicylic acid therapy for mildly to moderately active ulcerative colitis. A randomized study. N Engl. J. Med. 317, 1625–1629. https://doi.org/10.1056/NEJM198712243172603 (1987).Article
PubMed
Google Scholar
Yu, G., Wang, L. G., Han, Y. & He, Q. Y. clusterProfiler: An R package for comparing biological themes among gene clusters. OMICS 16, 284–287. https://doi.org/10.1089/omi.2011.0118 (2012).Article
PubMed
PubMed Central
MATH
Google Scholar
R. C. T. R: A Language and Environment for Statistical Computing (2018). http://www.r-project.org/ (2018).Shannon, P. et al. Cytoscape: A software environment for integrated models of biomolecular interaction networks. Genome Res. 13, 2498–2504. https://doi.org/10.1101/gr.1239303 (2003).Article
PubMed
PubMed Central
MATH
Google Scholar
Maere, S., Heymans, K. & Kuiper, M. BiNGO: A Cytoscape plugin to assess overrepresentation of gene ontology categories in biological networks. Bioinformatics 21, 3448–3449. https://doi.org/10.1093/bioinformatics/bti551 (2005).Article
PubMed
MATH
Google Scholar
Doncheva, N. T., Morris, J. H., Gorodkin, J. & Jensen, L. J. Cytoscape StringApp: Network analysis and visualization of proteomics data. J. Proteome Res. 18, 623–632. https://doi.org/10.1021/acs.jproteome.8b00702 (2019).Article
PubMed
Google Scholar
Reeve, J. et al. Assessing rejection-related disease in kidney transplant biopsies based on archetypal analysis of molecular phenotypes. JCI Insight 2, e94197. https://doi.org/10.1172/jci.insight.94197 (2017).Article
PubMed
PubMed Central
MATH
Google Scholar
Huber, W. et al. Orchestrating high-throughput genomic analysis with bioconductor. Nat. Methods. 12, 115–121. https://doi.org/10.1038/nmeth.3252 (2015).Article
ADS
PubMed
PubMed Central
MATH
Google Scholar
Ritchie, M. E. et al. Limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res. 43, e47. https://doi.org/10.1093/nar/gkv007 (2015).Article
PubMed
PubMed Central
MATH
Google Scholar
Download referencesAcknowledgementsThis research has been supported by funding and/or resources from Genome Canada, Canada Foundation for Innovation, University of Alberta Hospital Foundation, University of Alberta Division of Gastroenterology, Alberta Ministry of Advanced Education and Technology, Mendez National Institute of Transplantation Foundation, Industrial Research Assistance Program, and the Roche Organ Transplant Research Foundation. Partial support was also provided by funding from a licensing agreement with the One Lambda division of Thermo Fisher. Dr. P.F. Halloran held a Canada Research Chair in Transplant Immunology until 2008 and currently holds the Muttart Chair in Clinical Immunology. We are grateful to Simone Withecomb and the IBD unit at the University of Alberta for their continued advice and support of the project.Author informationAuthor notesKatelynn S. Madill-Thomsen and Jeffery M. Venner contributed equally to this work.Authors and AffiliationsDepartment of Medicine, University of Alberta, Edmonton, AB, CanadaKatelynn S. Madill-Thomsen, Konrad S. Famulski & Philip F. HalloranAlberta Transplant Applied Genomics Center, University of Alberta, Edmonton, AB, CanadaKatelynn S. Madill-Thomsen, Jeffery M. Venner, Konrad S. Famulski & Philip F. HalloranCedars-Sinai Medical Center, Los Angeles, CA, USAJeffery M. VennerDepartment of Medicine, Division of Gastroenterology, University of Alberta, 8540 – 112 Street NW, Edmonton, AB, T6G 2X8, CanadaDenise E. Parsons, Sami Hoque, Karen I. Kroeker, Karen Wong, Farhad Peerani, Levinus A. Dieleman, Frank Hoentjen, Daniel C. Baumgart & Brendan P. HalloranDepartment of Laboratory Medicine and Pathology, University of Alberta, Edmonton, AB, CanadaAducio L. ThiesenAuthorsKatelynn S. Madill-ThomsenView author publicationsYou can also search for this author in
PubMed Google ScholarJeffery M. VennerView author publicationsYou can also search for this author in
PubMed Google ScholarDenise E. ParsonsView author publicationsYou can also search for this author in
PubMed Google ScholarKonrad S. FamulskiView author publicationsYou can also search for this author in
PubMed Google ScholarAducio L. ThiesenView author publicationsYou can also search for this author in
PubMed Google ScholarSami HoqueView author publicationsYou can also search for this author in
PubMed Google ScholarKaren I. KroekerView author publicationsYou can also search for this author in
PubMed Google ScholarKaren WongView author publicationsYou can also search for this author in
PubMed Google ScholarFarhad PeeraniView author publicationsYou can also search for this author in
PubMed Google ScholarLevinus A. DielemanView author publicationsYou can also search for this author in
PubMed Google ScholarFrank HoentjenView author publicationsYou can also search for this author in
PubMed Google ScholarDaniel C. BaumgartView author publicationsYou can also search for this author in
PubMed Google ScholarPhilip F. HalloranView author publicationsYou can also search for this author in
PubMed Google ScholarBrendan P. HalloranView author publicationsYou can also search for this author in
PubMed Google ScholarContributionsKSMT – study design, data curation, analysis, writing original draft; JMV – study design, analysis, figure generation, writing original draft, reviewing and editing; DEP – supporting resources, project administration, reviewing and editing; KSF – study design, reviewing and editing; ALT – data curation, supporting resources, reviewing and editing; SH – data curation, project administration, reviewing and editing; KK – funding acquisition, supporting resources, reviewing and editingKW – funding acquisition, supporting resources, reviewing and editing; FP – funding acquisition, supporting resources, reviewing and editing; LD – funding acquisition, supporting resources, reviewing and editing; FH – funding acquisition, supporting resources, reviewing and editing; DB – funding acquisition, supporting resources, reviewing and editing; PFH – study design, data curation, funding acquisition, supporting supervision, writing original draft; BPH – study design, data curation, funding acquisition, supporting supervision, writing original draft, reviewing and editing.Corresponding authorCorrespondence to
Brendan P. Halloran.Ethics declarations
Competing interests
P.F. Halloran holds shares in Transcriptome Sciences Inc (TSI), a University of Alberta research company dedicated to developing molecular diagnostics, supported in part by a licensing agreement between TSI and Thermo Fisher, and by a research grant from Natera. P.F. Halloran is a consultant to Natera. Thermo Fisher and Natera had no role in the study design, analysis, or reporting of results. The other authors have nothing to declare.
Additional informationPublisher’s noteSpringer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.Electronic supplementary materialBelow is the link to the electronic supplementary material.Supplementary Material 1Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and permissionsAbout this articleCite this articleMadill-Thomsen, K.S., Venner, J.M., Parsons, D.E. et al. Relating the molecular phenotype of ulcerative colitis to the clinical course.
Sci Rep 15, 8342 (2025). https://doi.org/10.1038/s41598-025-90618-8Download citationReceived: 29 September 2024Accepted: 14 February 2025Published: 11 March 2025DOI: https://doi.org/10.1038/s41598-025-90618-8Share this articleAnyone you share the following link with will be able to read this content:Get shareable linkSorry, a shareable link is not currently available for this article.Copy to clipboard
Provided by the Springer Nature SharedIt content-sharing initiative
KeywordsInflammatory bowel diseaseMolecular classifierMachine learningMolecular landscapeDisease activity