Identification and Validation of MicroRNA Profiles in Fecal Samples for Detection of Colorectal Cancer

Journal Pre-proof Identification and validation of microRNA profiles in fecal samples for detection of colorectal cancer Saray Duran-Sanchon, Lorena Moreno, Josep M. Augé, Miquel Serra-Burriel, Míriam Cuatrecasas, Leticia Moreira, Agatha Martín, Anna Serradesanferm, Àngels Pozo, Rosa Costa, Antonio Lacy, Maria Pellisé, Juan José Lozano, Meritxell Gironella, Antoni Castells PII: DOI: Reference:

S0016-5085(19)41439-X https://doi.org/10.1053/j.gastro.2019.10.005 YGAST 62942

To appear in: Gastroenterology Accepted Date: 5 October 2019 Please cite this article as: Duran-Sanchon S, Moreno L, Augé JM, Serra-Burriel M, Cuatrecasas M, Moreira L, Martín A, Serradesanferm A, Pozo À, Costa R, Lacy A, Pellisé M, Lozano JJ, Gironella M, Castells A, Identification and validation of microRNA profiles in fecal samples for detection of colorectal cancer, Gastroenterology (2019), doi: https://doi.org/10.1053/j.gastro.2019.10.005. This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2019 by the AGA Institute

Discovery phase (tissue samples)

Validation phase (fecal samples) CRC AA N

Patients undergoing surgical or endoscopic resection

Technical validation

Fecal miRNA-based predictive model Gradient boosting machine algorithm

Nine out of 21 miRNA up-regulated in feces Next generation sequencing miRNA analysis Clinical validation FIT-positive participants in CRC screening program

N

Selection of 21 miRNA commonly up-regulated in both CRC and AA

NAA vs.

CRC+AA

N+NAA

AA

CRC

↑ miR-421 ↑ miR-27a-3p ↑ miR-130b-3p

Abbreviations: CRC, colorectal cancer; AA, advanced adenoma; NAA, non-advanced adenoma; N, normal colonoscopy; FIT, fecal immunochemical test.

Variables

miR-421 miR-27a-3p Fecal hemoglobin concentration Age and gender

Title: Identification and validation of microRNA profiles in fecal samples for detection of colorectal cancer

Short Title: Fecal miRNAs as non-invasive biomarkers for CRC screening.

Saray Duran-Sanchon1, Lorena Moreno1, Josep M. Augé2, Miquel SerraBurriel3, Míriam Cuatrecasas4, Leticia Moreira1, Agatha Martín1, Anna Serradesanferm5, Àngels Pozo5, Rosa Costa1, Antonio Lacy6, Maria Pellisé1, Juan José Lozano7, Meritxell Gironella1,*, Antoni Castells1,*.

1

Gastroenterology

Department,

Hospital

Clinic

of

Barcelona,

Institut

d’Investigacions Biomèdiques August Pi i Sunyer (IDIBAPS), University of Barcelona, Centro de Investigación Biomédica en Red de Enfermedades Hepáticas y Digestivas (CIBEREHD), Barcelona, Catalonia, Spain. 2

Biochemistry and Molecular Genetics Department, Hospital Clinic of

Barcelona, IDIBAPS, Barcelona, Catalonia, Spain. 3

Center for Research in Health and Economics, Universitat Pompeu Fabra,

Barcelona, Catalonia, Spain. 4

Pathology Department and Tumour Bank-Biobank, Hospital Clínic of

Barcelona, IDIBAPS, CIBEREHD, Barcelona, Catalonia, Spain. 5

Preventive Medicine and Hospital Epidemiology Department, Hospital Clínic,

Barcelona, Catalonia, Spain. 6

Gastrointestinal Surgery Department, Hospital Clínic of Barcelona, IDIBAPS,

University of Barcelona, CIBEREHD, Barcelona, Catalonia, Spain. 7

Bioinformatics Platform, CIBEREHD, Barcelona, Catalonia, Spain. 1

*Both authors share senior authorship.

Funding: The present work was supported by Ministerio de Economía y Competitividad (SAF2014-54453-R), Fundación Científica de la Asociación Española contra el Cáncer (GCB13131592CAST), Agència de Gestió d’Ajuts Universitaris i de Recerca (2017SGR653), and Instituto de Salud Carlos III (PI17/01003). CIBEREHD (Centro de Investigación Biomédica en Red de Enfermedades Hepáticas y Digestivas) is funded by the Instituto de Salud Carlos III.

Abbreviations: CRC, Colorectal cancer; AA, Advanced adenomas; NAA, Nonadvanced adenomas;

FIT, fecal immunochemical testing;

NGS,

Next

Generation Sequencing.

Corresponding author: Dr. Antoni Castells, Gastroenterology Department, Hospital Clinic of Barcelona; Villarroel 170, 08036 Barcelona, Catalonia, Spain. Phone number: +34 93 227 5703; e-mail: [email protected].

Conflicts of interest: There are no conflicts of interest to disclose.

Author contributions: Conception and design (AC, MG); collection and assembly of data (SDS, LM, JA, AM); data analysis and interpretation (SDS, MSB, JJL, MG, AC); provision of study materials or patients (JA, MC, LM, AS, AP, RC, AL,

2

MP, AC); manuscript writing (SDS, MG, AC); manuscript editing (AC); all authors gave final approval of the version to be published.

Acknowledgements We are indebted to the IDIBAPS Biobank, integrated in the Spanish National Biobank Network, for samples procurement. We also appreciate the collaboration of nurses at the Endoscopy Unit for helping us to collect the samples. This work was developed at the Centro Esther Koplowitz, Barcelona, Catalonia, Spain.

3

Abstract Background & Aims: Screening for colorectal cancer (CRC) is effective in average-risk population. The most extended strategy in organized programs involves the fecal immunochemical test, which is limited by low sensitivity for detection of advanced adenomas (AAs). We aimed to identify microRNA (miRNA) signatures in fecal samples that identify patients with AAs or CRC and might be used in non-invasive screening. Methods: Our study comprised 4 stages. In the discovery phase, we performed genome-wide miRNA expression profiling of 124 fresh, paired colorectal tumor and non-tumor samples (30 colorectal tumors; 32 AAs) from patients in Spain. In the technical validation stage, miRNAs with altered expression levels in tumor vs non-tumor tissues were quantified by reverse transcription PCR in fecal samples from a subset of patients included in the discovery phase (n=39) and individuals without colorectal neoplasms (controls, n=39). In the clinical validation stage, the miRNAs found to be most significantly up-regulated by quantitative reverse transcription PCR analysis were measured in an independent set of fecal samples (n=767) from patients with positive results from fecal immunochemical tests in a CRC screening program. Finally, we developed model to identify patients with advanced neoplasms (CRCs or AAs) based on their miRNA profiles, using findings from colonoscopy as the reference standard. Results: Among 200 and 324 miRNAs significantly deregulated in CRC and AA tissues, respectively, 7 and 5 of these miRNAs were also found to be deregulated in feces (technical validation). Of them, MIR421, MIR130b-3p, and MIR27a-3p were confirmed to be upregulated in fecal samples from patients

4

with advanced neoplasms. In our model, the combination of fecal level of MIR421, MIR27a-3p, and hemoglobin identified patients with CRC with an area under the curve (AUC) of 0.93, compared to an AUC of 0.67 for fecal hemoglobin concentration alone. The combination of markers identified patients with AA with an AUC value of 0.64, compared to an AUC of 0.59 for fecal hemoglobin concentration alone. Conclusions: We found that increased levels of 2 miRNAs and hemoglobin in feces can identify patients with AAs or CRC more accurately than fecal hemoglobin concentration alone. Assays for these miRNAs might be added to fecal tests for detection of CRC or AAs.

KEY WORDS: biomarker, FIT, miR-421, miR-130b

5

Introduction Colorectal cancer (CRC) is the third most common incident cancer and the second leading cause of cancer-related death in the world, accounting for almost 1.8 million of new cases and 800.000 deaths in 20181. Although 90% of patients diagnosed at early stages have an overall survival of more than 5 years, this figure decreases to 10% in patients diagnosed at advanced stages with distant metastasis2. In that sense, evidences from several studies have shown that CRC screening is effective and cost-effective in average-risk population3. Recommended CRC screening strategies fall in two broad categories: stool tests (occult blood -guaiac test and fecal immunochemical test (FIT) - or exfoliated

DNA

tests)

and

structural

exams

(flexible

sigmoidoscopy,

colonoscopy, and CT colonography)4. While structural exams detect both cancer and premalignant lesions, stool tests primarily identify cancer because of their limited sensitivity to detect advanced adenomas (AA)5. Moreover, in a twostep screening scenario, which is the most extended worldwide2, the relatively low specificity of the initial examination results in a high false-positive rate, leading to a significant number of unnecessary colonoscopies6. The use of biomarkers for screening purposes appears as an appealing approach to overcome the above mentioned limitations7. MicroRNAs (miRNA) are small endogenous non-coding RNAs of 18 to 25 nucleotides that negatively regulate gene expression at post-transcriptional level by either repressing transcript translation or inducing the degradation of target mRNAs8. Importantly, they are involved in several biological processes, i.e. carcinogenesis9,10. Indeed, deregulated miRNAs reflect physiopathological states and allow

6

distinguishing different stages and subtypes of cancer, including CRC11, in addition to discriminating neoplastic patients from healthy individuals12. Interestingly, it seems that precursor lesions also show an altered miRNA pattern that could be, in part, secreted to the extracellular milieu13,14. As a consequence, miRNAs detected in different body fluids, i.e. feces15, have been suggested as potential biomarkers for CRC screening. The aim of this study was to identify a fecal miRNA signature able to act as a new non-invasive, effective and efficient strategy for CRC screening in a population-based setting. For this purpose, we followed a discovery approach by genome-wide miRNA expression profiling in tissue samples, followed by technical and clinical validation of miRNA candidates in feces from participants in an organized CRC screening program.

7

Patients and methods The study comprised four main stages: 1) miRNA discovery phase by next generation sequencing (NGS) in a tissue set of CRC and AA samples, along with their paired normal mucosa; 2) technical validation of miRNA candidates by quantitative reverse-transcription PCR (qRT-PCR) in fecal samples from a subset of patients included in the discovery phase as well as control individuals; 3) clinical validation of the most significantly up-regulated miRNAs by qRT-PCR in an independent set of fecal samples obtained from participants in the Barcelona's CRC screening programme16; and 4) development of a miRNA-based predictive model to discriminate patients with advanced neoplasms (i.e. CRC or AA) of those with non-relevant findings at colonoscopy. The outline of study is shown in Figure S1. For the miRNA discovery phase, 124 fresh colorectal tissue samples were prospectively collected at the Hospital Clinic of Barcelona. Paired neoplastic and normal mucosa samples were obtained from 30 patients with CRC submitted to surgery; none of these patients had received either neoadjuvant chemotherapy or radiation therapy. In addition, paired neoplastic and normal mucosa samples were obtained from 32 patients with AA after endoscopic resection. All tissue samples were preserved in RNAlater® (Invitrogen, Carlsbad, CA) and frozen at -80ºC until RNA extraction. For the technical validation of miRNA candidates in fecal samples, a subset of 39 patients included in the previous stage (11 with CRC and 28 with AA) and 39 control individuals with normal colonoscopy were analyzed. Finally, clinical validation of selected miRNA was performed in 767 fecal samples recruited among FIT-positive participants in the Barcelona’s CRC

8

screening program between March 2011 and May 2017. Each participant provided one fecal sample using a specimen collection device (OC-Sensor, Eiken Chemical Co., Japan), which collects 10 mg feces with a serrated probe attached to the cap into 2 mL buffer. All samples were kept at -80ºC until RNA extraction. Patients in whom only serrated lesions were detected were excluded from this analysis in order to diminish study heterogeneity. Characteristics of the population-based, organized screening program are described elsewhere16. Clinico-pathological features of all individuals included in the study are shown in Table 1. The study was approved by the Institutional Ethics Committee of Hospital Clinic of Barcelona, and written informed consent was obtained from all participants in accordance with the Declaration of Helsinki.

MicroRNA extraction Total RNA, including miRNAs, was isolated from tissues or feces (500 µL buffer) using the miRNeasy® mini kit (Qiagen, Valencia, CA) according to the manufacturer’s protocol. For tissue samples, after the RNA extraction, an RNeasy MinElute Cleanup® kit (Qiagen, Valencia, CA) was used to warrant removal of contaminants and to concentrate samples in a final elution volume of 12 µL. The purity of RNA tissue samples was analyzed using Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA) and concentration was determined

by

NanoDrop®

1000

spectrophotometer

(Wilmington,

DE).

Concentration of total miRNA present in fecal samples was measured by Quant-iT® microRNA assay kit (Invitrogen, Carlsbad, CA).

9

Genome-wide miRNA profiling by NGS Generation of small RNA libraries was performed from 1 µg of total RNA from tissue using TruSeq® small RNA sample prep kit (Illumina, San Diego, CA) according to the manufacturer’s protocol. First, 3’ and 5’ RNA adapters were ligated at the small RNA ends. Thereafter, cDNA constructs were synthetized by reverse transcription using SuperScript® II reverse transcriptase (Invitrogen, Carlsbad, CA) with specific primers complementary to the 3’RNA adapter. cDNA was further amplified by PCR using indexed adapters supplied in the kit. This step selectively enriches RNA fragments with adapter molecules in both ends. Finally, the amplified cDNA constructs from libraries were isolated on a 6% Novex® TBE gels (Life Technologies, Carlsbad, CA). The area representing the band size of 18 to 36 base pairs (bp) was cut from the gel, and DNA was precipitated and eluted in 10 µL of elution buffer. The cDNA libraries generated were analyzed using Agilent High Sensitivity DNA® kit (Agilent Technologies, CA) to ensure acceptable quantity and check size distribution. The highthroughput sequencing of the cDNA libraries was run in HiSeq® 2000 (Illumina, CA) with 1x50 bp single-end reads to obtain >15M reads per sample. Quality control was made by image analysis and assignation of bases by Real Time Analysis software. We discarded reads with low confidence values in the first 25 bases. Quality control reads were aligned with a reference genome with GEM program. The error percentage was below 2.5% calculated from a Genome of PhiX added to the samples before sequencing.

10

Analysis of fecal miRNA expression by qRT-PCR qRT-PCR was done using singleplex TaqMan® microRNA assays (Applied Biosystems Inc., Foster City, CA). Briefly, 5 ng of total miRNA were used to perform retro-transcription. Thereafter, a pre-amplification step was performed in some miRNAs before quantitative PCR due to the low miRNA levels in feces. Finally, quantitative PCR was performed in a Viia7® Real Time PCR system (Applied Biosystems Inc., Foster City, CA) using 2 µL of cDNA in a final volume of 10 µL. Each point was assessed in triplicate. Due to the lack of a reliable endogenous control to normalize fecal miRNAs, absolute quantification was carried out. Thus, the average Ct for each sample was converted to ng of specific miRNA / g of total miRNA using standard curves made by serial dilutions of known quantities of each specific synthetic miRNA (Integrated DNA Technologies, IA).

Bioinformatics and statistical analysis Sequencing analysis was done by using the sRNAbench package17. Briefly, after adapter trimming and unique read grouping, reads were aligned to the human genome (UCSC hg19) using Bowtie 1.1.218 allowing for one mismatch. To provide annotations for RNA elements that mapped to the human genome, miRBase (version 21) for mature and pre-miRNA sequences was used. Count data were voom-transformed to log2-counts per million (logCPM) and normalized by cyclic-loess method19. To identify differentially expressed miRNAs, moderated-t statistics were applied. Differential expression, fold change and expression mean as log2 of the difference were analyzed between colorectal neoplastic tissue and paired normal mucosa. P-values were adjusted

11

for multiple testing by Benjamini and Hochberg method. MiRNAs with false discovery rate <0.05 were considered significant. Principal component analysis plots were made to visualize high dimensional data in a 2D graph in which the areas delimited by the ellipses represents 95% of the binormal distribution of the sample scores on the first and second axes. Venn diagrams considering significant miRNAs were also performed. Selection of miRNA candidates from NGS results was made based on the following criteria: false discovery rate <0.05, fold change ≥1.5, mean expression >3.5, and up-regulation in both CRC and AA. For quantitative variables, Student's t test was used. Discriminative capacity of individual miRNAs was evaluated by multivariate logistic regression, adjusted by age and gender. P-value ≤0.05 was regarded as significant. Area under the receiver operating characteristics (AUC) curve and the derived cutpoints were computed using pROC R-package considering each miRNA as a continuous variable. Sensitivity and specificity were calculated from the optimal cut-point associated with the minimum error rate.

Fecal miRNA-based predictive modelling The predictive model to discriminate between different groups of individuals was performed considering miRNAs results obtained in the clinical validation phase, along with age and gender. Primary endpoint of this analysis was to discriminate patients with advanced neoplasm (i.e. CRC or AA) of those with non-relevant findings at colonoscopy (i.e. NAA or normal examination). Secondary endpoints were to distinguish patients with advanced neoplasm, CRC and AA from individuals with normal colonoscopy, respectively.

12

For predictive model generation, samples were randomly split into training and test sets as 75%-25% proportion. Then, numeric data were preprocessed by centering and scaling. To address the imbalance of sample groups, synthetic minority over-sampling technique20 was used. In the training set, a 10-fold cross-validation was performed. Algorithms tested for the generation of the model were: C-tree, random forest, linear discriminant analysis, gradient boosting machine21, support vector machine, and K-nearest neighbor. P-value <0.05 was regarded as significant. Figure S2 shows the analytic strategy. Discrimination measures were AUC, sensitivity, specificity, and positive and negative predictive values. A post-processing calibration was made with 15 predicted versus observed probability bins22 in order to analyze the error distribution of the predictive model. All analyses were performed with R under CARET package23. Finally, since clinical validation of fecal miRNA signature was performed in a cohort of FIT-positive individuals, results obtained with this predictive model were compared to the one that would be achieved by using fecal hemoglobin concentration, a reliable biomarker of patients with advanced neoplasm among FIT-positive participants, in the same population16. Finally, performance of a predictive model combining both miRNA signature and fecal hemoglobin concentration was also assessed.

13

Results MiRNA discovery phase by NGS In the tissue set of colorectal samples, expression of 1640 miRNAs was detected and 637 miRNAs had more than 100 counts. Of them, 200 and 324 miRNAs were significantly deregulated in CRC and AA, respectively, in comparison to their paired normal mucosa, with a fold change ≥1.5. Expression of the 50 most significant deregulated miRNAs is depicted in Figure 1. Between-group analysis showed that miRNA expression profiling could distinguish CRC or AA tissue samples from their paired normal mucosa (Figure 2A), as well CRC from AA. Moreover, we found that 72 and 56 miRNAs were commonly up- or down-regulated, respectively, in both neoplastic lesions (Figure 2B). According to the selection criteria (i.e. false discovery rate <0.05, fold change ≥1.5, mean expression >3.5, and up-regulation in both CRC and AA), 21 miRNA candidates (Table 2) were selected for technical validation in fecal samples.

MiRNA analysis on fecal samples by qRT-PCR In order to elucidate whether miRNA patterns in colorectal tissues could be reproduced in fecal samples, we first analyzed by qRT-PCR the abovementioned 21 miRNAs in a subset of 39 patients included in the discovery phase (11 patients with CRC and 28 with AA) as well as 39 control individuals with normal colonoscopy. Results of this technical validation phase showed that 7 miRNAs were significantly up-regulated in fecal samples from patients with CRC (miR-130b-

14

3p, miR-21-5p, miR-221-5p, miR-25-3p, miR-27a-3p, miR-34a-5p, and miR421) (Table S1). In addition, four of them (miR-130b-3p, miR-21-5p, miR-27a3p and miR-421) as well as miR-335-3p were significantly up-regulated in fecal samples from patients with AA. These 8 miRNAs, along with miR-29a-3p, which was also up-regulated in fecal samples from patients with CRC and AA (AUC of 0.84 and 0.71, respectively; p-value <0.1), were selected to be clinically validated in an independent cohort of participants in the CRC screening program.

Clinical validation of miRNA candidates The above-mentioned 9 up-regulated miRNAs were validated in fecal samples from a prospectively collected cohort of 767 FIT-positive individuals, which included 67 patients with CRC, 347 patients with AA, 136 patients with NAA, and 217 individuals with normal colonoscopy (Table 1). In this set of samples, up-regulation of miR-25-3p, miR-27a-3p, miR-29a3p, miR-34a-5p, miR-130b-3p, miR-221-3p, and miR-421 was confirmed in CRC patients (AUCs ranging from 0.69 to 0.77), whereas up-regulation of miR130b-3p and miR-421 was confirmed in patients with AA (AUCs were 0.69 and 0.71, respectively), in comparison with the control group. It is important to note that none of these 7 fecal miRNAs showed significant differences between patients with NAA and individuals with normal colonoscopy (Table 3). Finally, with respect to the primary endpoint of study, miR-421, miR-27a3p and miR-130b-3p were significantly selected as the most discriminant fecal miRNAs to distinguish patients with advanced neoplasm from those with nonrelevant findings at colonoscopy (Table 4).

15

Development and validation of a fecal miRNA-based predictive model for colorectal cancer screening Predictive modelling to discriminate between different groups of patients was performed considering fecal miRNAs significantly up-regulated in patients with advanced neoplasm, along with age and gender. According to the principal component analysis (Figure S3), miR-421 and miR-27a-3p were finally selected for this purpose because they were shown not to be redundant among the three most discriminant fecal miRNAs. For model generation, individuals were randomly split into training (n=578) and test sets (n=189), with a 10-fold crossvalidation in the development stage to reduce bias and variability (Table S2). As previously mentioned, different algorithms were tested in order to choose the one that fitted better with respect to the primary endpoint of the study, being the gradient boosting machine algorithm finally selected based on its highest accuracy (data not shown). As it is shown in Table 5, the resulting predictive model combining miR421, miR-27a-3p, age and gender was highly accurate (AUC=0.63) to identify patients with advanced neoplasm among FIT-positive participants in the CRC screening program. Interestingly, this result was due not only to a high accuracy for recognizing patients with CRC (AUC=0.74; sensitivity, 96%), but also to distinguishing patients with AA from control individuals (AUC=0.64; sensitivity, 59%). The results achieved in the fecal miRNA-based predictive model were superior to those obtained using fecal hemoglobin concentration as classifier, with respect to detection of advanced neoplasm (AUC=0.62 when compared 16

with subjects with non-relevant findings at colonoscopy; AUC=0.59 when compared with individuals with normal colonoscopy), CRC (AUC=0.67) and AA (AUC=0.59) (Table 5). Finally, combination of both miRNA signature and fecal hemoglobin concentration allowed the highest accuracy for identifying patients with advanced neoplasm (AUC=0.70 when compared with subjects with nonrelevant findings at colonoscopy; AUC=0.67 when compared with individuals with normal colonoscopy), CRC (AUC=0.93) and AA (AUC=0.64) (Table 5). Indeed, calibration curves of the combined model indicated that predictions were closer to the observed outcomes, and the error distribution was lower, in comparison to the results obtained using fecal hemoglobin concentration alone (Figure S4).

17

Discussion Following a genome-wide miRNA profiling approach in a large collection of colorectal tissue samples, we have been able to demonstrate that miRNA expression can discriminate between CRC, AA and normal colorectal mucosa. More importantly, tissue miRNA pattern was reflected in fecal samples exhibiting significant differential expression between patients with advanced neoplasm and subjects with non-relevant findings at colonoscopy. Finally, these results allowed us to design a fecal miRNA signature highly accurate to distinguish patients with either CRC or AA among FIT-positive participants in a screening program. Population-based, organized screening programs for CRC are effective and cost-effective2–4. The vast majority of them follow a two-step strategy, in which the front-line test selects those individuals who should undergo colonoscopy. In such a context, FIT-based programs, which are the most extended worldwide, are limited by a sensitivity for detecting patients with AA (<25%), thus favoring early detection rather than CRC prevention6. Combination of fecal DNA and occult blood testing has emerged as a potential useful approach to overcome this limitation, since it achieved higher sensitivity for advanced neoplasm than FIT, at expenses of more false-positive results4. This low specificity, along with its high cost and logistic difficulties due to the large amount of fresh feces needed, makes it difficult to be use in a population-based scenario. Hence, there is a need to develop new non-invasive strategies with better performance for detecting both CRC and AA, associated with a high compliance and adherence, affordable and widely distributable24.

18

In the last few years, strong evidence reinforces the use of miRNA detection in body fluids as a novel and promising approach for CRC screening25–27. Interestingly, it has been described that deregulation of miRNAs could be a premature step in the development of different cancers9,28. Therefore, we have hypothesized that miRNAs could be found deregulated in feces from patients with AA, as well as from those with CRC. In order to select potential miRNA candidates, we performed miRNome analysis by NGS in tissue samples from patients with CRC or AA, and their paired normal colorectal mucosa. We confirmed that CRC shows a significant deregulation of miRNA profile, and observed that this deregulation is also present in its precursor lesion, thus indicating that miRNA deregulation is an early step in the colocarcinogenic process. To our knowledge, although miRNA profiling has been evaluated in CRC29,30, this is the first study in which AA lesions have been miR-sequenced. The use of miRNA as a novel screening strategy has been mainly focused on plasma and serum samples26,31. However, there were some studies evaluating individual miRNA candidates selected from the literature15,32–35 that demonstrated their stability and reproducibility when they were analyzed in feces. It is considered that the vast majority of fecal miRNAs come from the continuous exfoliation of tumor cells into the colonic lumen35,36, thus making them excellent biomarkers in this setting. In such a context, miR-92a and miR135b were suggested as the most accurate ones in small cohorts of patients and lacking of an adequate validation33,37. It is important to note, however, that feces are a very heterogeneous biological material difficult to normalize between individuals. Moreover, the lack of a robust endogenous control for

19

miRNA analysis may induce some bias, thus making difficult to compare results from different studies. In order to overcome this limitation, we carried out a quantitative measure of fecal miRNAs by extrapolating qRT-PCR values into standard curves. More importantly, our study was designed as a systematic, discovery-validation approach in which tissue miRNA profiling in CRC and AA were used to select the most commonly up-regulated miRNA candidates to be technically and clinically validated in fecal samples. We have demonstrated that some tissue miRNAs significantly upregulated in CRC can be detected in feces from patients with such a neoplasm. Indeed, we found seven fecal miRNAs significantly up-regulated in patients with CRC, some of them already being involved in CRC pathogenesis: miR-34a was described as a regulator of multiple pathways implicated in migration, invasion and metastasis38; miR-130b and miR-221 promote tumor development though PPARγ and STAT3, respectively39,40; miR-25 was described as a metastasispromoting miRNA in CRC41; and miR-27a promotes proliferation, migration, invasion and metastasis through the Wnt/β-catenin pathway and MAPK/ERK pathway42,43. Although miR-421 is nearly unexplored in CRC, it has been described to be involved in other tumors inhibiting apoptosis and promoting metastasis though CASP3 and CDH1 interaction, as well as regulating ATM and SMAD4 genes44–46. In that sense, the results obtained in our study highlight miR-421 as highly up-regulated in both CRC and AA, thus unveiling a new potentially important miRNA in colorectal carcinogenesis. Only miR-221-3p was previously reported to be significantly up-regulated in fecal samples from patients with CRC and, in accordance with our results, it was not observed in fecal samples from patients with AA47. Finally, in contrast to previous

20

publications, we have not been able to validate up-regulation of fecal miR-21-5p in CRC patients32,34. To our knowledge, it is also the first time that miR-130b-3p and miR-421could be detected in fecal samples from patients with AA. In that sense, it is important to mention that among the 21 miRNA candidates significantly upregulated in AA tissue samples, we were able to validate a significant upregulation in feces of 5 out of them. This fact could be due to a small amount of cell exfoliation or total miRNA secretion delivered into the lumen in comparison to what occurred in patients with CRC. Among the seven fecal miRNAs clinically validated for CRC in our study, miR-421 and miR-27a-3p were selected to generate a miRNA-based predictive model for screening purposes. In our cohort of FIT-positive participants in an organized screening program, this model can accurately identify patients with advanced neoplasm (sensitivity for CRC and AA, 96% and 59%, respectively). Interestingly, when we assessed the diagnostic performance of fecal hemoglobin concentration -the biomarker targeted by FIT- in the same cohort of individuals, its accuracy in distinguishing both subgroups of patients was lower, especially with respect to sensitivity and specificity for AA (Table 5). Finally, in a combined predictive model based on both fecal miRNA signature and hemoglobin concentration, its diagnostic performance was maximized. It is important to highlight that improvement was observed in the capacity of identifying patients with CRC (AUC=0.93, in comparison to an AUC=0.67 obtained using fecal hemoglobin concentration alone) but not in distinguishing patients with AA in which scenario fecal miRNA and combined models

21

performed similarly although superior to hemoglobin concentration alone (AUC=0.64 and AUC=0.59, respectively) (Table 5). Strengths of this study rely on several aspects. First, its design included both an unbiased discovery phase using genome-wide miRNA profiling by NGS in tissue samples from CRC and AA patients, and a subsequent technical and clinical validation in fecal samples from an independent cohort of average-risk individuals participating in an organized screening program. Second, the large number of patients and control subjects prospectively recruited and included in both stages of the study represent the major series evaluating the usefulness of fecal miRNA for screening purposes published so far. Third, miRNA detection was feasible using a small amount of feces (10 mg) collected and preserved into an original FIT collection device, thus making it widely distributable for population-based screening purposes. We are aware, however, of a limitation of this study, which is the use of FIT-positive individuals for evaluating the accuracy of fecal miRNA signature, since it precludes to directly comparing these results with those achieved in naive subjects participating in screening programs. To overcome this limitation, we assessed the accuracy of fecal hemoglobin concentration -a reliable biomarker

of

patients

with

advanced

neoplasm

among

FIT-positive

participants16- in the same cohort of individuals, and demonstrated that the miRNA signature performed better than fecal hemoglobin concentration in this subgroup. Interestingly, fecal hemoglobin concentration improved the accuracy of miRNA signature when both parameters were combined. On the other hand, the discovered signature could be used as a triage test among FIT-positive individuals, in order to increase specificity of FIT and, consequently, diminish

22

the number of unnecessary colonoscopies. However, the present analysis was not designed to evaluate this potential utility and, accordingly, comparison of predictive models was based on the corresponding AUC. In that sense, it is important to note that performance characteristics reported in Table 5 were calculated considering the optimal cut-point associated with the minimum error rate rather than prioritizing specificity. In conclusion, results of this study demonstrated that patients with advanced neoplasm can be identified through a miRNA signature in feces, which accuracy seems to be superior to the one achieved using fecal hemoglobin concentration among FIT-positive individuals. In this latter scenario, the fecal miRNA-based model might contribute to improve effectiveness of current FIT-based screening programs by favoring the use of a lower fecal hemoglobin cut-off level, which would increase sensitivity of FIT for AA, while potentially maintaining an adequate specificity because of its combination with the fecal miRNA signature. Finally, confirmation of the results in a naive average-risk population would position fecal miRNA detection as a new noninvasive, widely distributable and cost-effective front-line test for populationbased, organized CRC screening.

23

References 1.

Bray F, Ferlay J, Soerjomataram I, et al. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2018;68:394–424.

2.

Young GP, Rabeneck L, Winawer SJ. The Global Paradigm Shift in Screening for Colorectal Cancer. Gastroenterology 2019;156:843– 851.e2.

3.

Steele R, Rey J-F, Lambert R. European guidelines for quality assurance in colorectal cancer screening and diagnosis. First Edition – Professional requirements and training. Endoscopy 2012;44:SE106-SE115.

4.

Bibbins-Domingo K, Grossman DC, Curry SJ, et al. Screening for Colorectal Cancer. JAMA 2016;315:2564.

5.

Gupta S, Halm EA, Rockey DC, et al. Comparative Effectiveness of Fecal Immunochemical Test Outreach, Colonoscopy Outreach, and Usual Care for Boosting Colorectal Cancer Screening Among the Underserved. JAMA Intern Med 2013;173:1725–1732.

6.

Robertson DJ, Lee JK, Boland CR, et al. Recommendations on Fecal Immunochemical Testing to Screen for Colorectal Neoplasia: A Consensus Statement by the US Multi-Society Task Force on Colorectal Cancer. Gastroenterology 2017;152:1217–1237.e3.

7.

Quintero E, Castells A, Bujanda L, et al. Colonoscopy versus Fecal Immunochemical Testing in Colorectal-Cancer Screening. N Engl J Med 2012;366:697–706.

8.

Bartel DP. MicroRNAs: Genomics, Biogenesis, Mechanism, and Function. Cell 2004;116:281–297.

24

9.

Esquela-Kerscher A, Slack FJ. Oncomirs — microRNAs with a role in cancer. Nat Rev Cancer 2006;6:259–269.

10.

Bushati N, Cohen SM. microRNA Functions. Annu Rev Cell Dev Biol 2007;23:175–205.

11.

Schetter AJ, Okayama H, Harris CC. The Role of MicroRNAs in Colorectal Cancer. Cancer J 2012;18:244–252.

12.

Leva G Di, Croce CM. miRNA profiling of cancer. Curr Opin Genet Dev 2013;23:3–11.

13.

Moridikia A, Mirzaei H, Sahebkar A, et al. MicroRNAs: Potential candidates for diagnosis and treatment of colorectal cancer. J Cell Physiol 2018;233:901–913.

14.

Larrea E, Sole C, Manterola L, et al. New Concepts in Cancer Biomarkers: Circulating miRNAs in Liquid Biopsies. Int J Mol Sci 2016;17:627.

15.

Link A, Balaguer F, Shen Y, et al. Fecal MicroRNAs as Novel Biomarkers for Colon Cancer Screening. Cancer Epidemiol Biomarkers Prev 2010;19:1766–1774.

16.

Auge JM, Pellise M, Escudero JM, et al. Risk Stratification for Advanced Colorectal Neoplasia According to Fecal Hemoglobin Concentration in a Colorectal Cancer Screening Program. Gastroenterology 2014;147:628– 636.e1.

17.

Barturen G, Rueda A, Hamberg M, et al. sRNAbench: profiling of small RNAs and its sequence variants in single or multi-species high-throughput experiments. Methods Next Gener Seq 2014;1.

18.

Langmead B, Trapnell C, Pop M, et al. Ultrafast and memory-efficient

25

alignment of short DNA sequences to the human genome. Genome Biol 2009;10:R25. 19.

Ritchie ME, Phipson B, Wu D, et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res 2015;43:e47–e47.

20.

Chawla N V., Bowyer KW, Hall LO, et al. SMOTE: Synthetic Minority Over-sampling Technique. J Artif Intell Res 2002;16:321–357.

21.

Friedman JH. Stochastic gradient boosting. Comput Stat Data Anal 2002;38:367–378.

22.

Austin PC, Steyerberg EW. Graphical assessment of internal and external calibration of logistic regression models by using loess smoothers. Stat Med 2014;33:517–535.

23.

Kuhn M. Building Predictive Models in R Using the caret Package. J Stat Softw 2008;28.

24.

Issa IA, Noureddine M. Colorectal cancer screening: An updated review of the available options. World J Gastroenterol 2017;23:5086.

25.

Vishnoi A, Rani S. MiRNA Biogenesis and Regulation of Diseases: An Overview. In: Rani S, ed. MicroRNA Profiling. Methods in Molecular Biology. Volume 1509. New York, NY: Springer Science; 2017:1–10.

26.

Herreros-Villanueva M, Duran-Sanchon S, Martín AC, et al. Plasma MicroRNA Signature Validation for Early Detection of Colorectal Cancer. Clin Transl Gastroenterol 2019;10:e00003.

27.

Slaby O. Non-coding RNAs as Biomarkers for Colorectal Cancer Screening and Early Detection. In: Slaby O, Calin GA, eds. Non-coding RNAs in Colorectal Cancer. Advances in Experimental Medicine and

26

Biology. Vol 937. Cham: Springer; 2016:153–170. 28.

Vila-Navarro E, Vila-Casadesús M, Moreira L, et al. MicroRNAs for Detection of Pancreatic Neoplasia. Ann Surg 2017;265:1226–1234.

29.

Neerincx M, Sie DLS, Wiel MA van de, et al. MiR expression profiles of paired primary colorectal cancer and metastases by next-generation sequencing. Oncogenesis 2015;4:e170–e170.

30.

Hamfjord J, Stangeland AM, Hughes T, et al. Differential Expression of miRNAs in Colorectal Cancer: Comparison of Paired Tumor Tissue and Adjacent Normal Mucosa Using High-Throughput Sequencing. PLoS One 2012;7:e34150.

31.

Vila-Navarro E, Duran-Sanchon S, Vila-Casadesús M, et al. Novel Circulating miRNA Signatures for Early Detection of Pancreatic Neoplasia. Clin Transl Gastroenterol 2019;10:e00029.

32.

Yau TO, Wu CW, Tang C-M, et al. microRNA-20a in human faeces as a non-invasive biomarker for colorectal cancer. Oncotarget 2016;7:1559– 1568.

33.

Wu CW, Ng SSM, Dong YJ, et al. Detection of miR-92a and miR-21 in stool samples as potential screening biomarkers for colorectal cancer and polyps. Gut 2012;61:739–745.

34.

Chang P-Y, Chen C-C, Chang Y-S, et al. MicroRNA-223 and microRNA92a in stool and plasma samples act as complementary biomarkers to increase colorectal cancer detection. Oncotarget 2016;7:10663–10675.

35.

Koga Y, Yasunaga M, Takahashi A, et al. MicroRNA Expression Profiling of Exfoliated Colonocytes Isolated from Feces for Colorectal Cancer Screening. Cancer Prev Res 2010;3:1435–1442.

27

36.

Matsushita H, Matsumura Y, Moriya Y, et al. A New Method for Isolating Colonocytes From Naturally Evacuated Feces and Its Clinical Application to Colorectal Cancer Diagnosis. Gastroenterology 2005;129:1918–1927.

37.

Wu CW, Ng SC, Dong Y, et al. Identification of microRNA-135b in Stool as a Potential Noninvasive Biomarker for Colorectal Cancer and Adenoma. Clin Cancer Res 2014;20:2994–3002.

38.

Gao J, Li N, Dong Y, et al. miR-34a-5p suppresses colorectal cancer metastasis and predicts recurrence in patients with stage II/III colorectal cancer. Oncogene 2015;34:4142–4152.

39.

Liu S, Sun X, Wang M, et al. A microRNA 221– and 222–Mediated Feedback Loop Maintains Constitutive Activation of NFκB and STAT3 in Colorectal Cancer Cells. Gastroenterology 2014;147:847–859.e11.

40.

Colangelo T, Fucci A, Votino C, et al. MicroRNA-130b Promotes Tumor Development and Is Associated with Poor Prognosis in Colorectal Cancer. Neoplasia 2013;15:1086–1099.

41.

Zeng Z, Li Y, Pan Y, et al. Cancer-derived exosomal miR-25-3p promotes

pre-metastatic

niche

formation

by

inducing

vascular

permeability and angiogenesis. Nat Commun 2018;9:5395. 42.

Liang J, Tang J, Shi H, et al. miR-27a-3p targeting RXRα promotes colorectal cancer progression by activating Wnt/β-catenin pathway. Oncotarget 2017;8:82991–83008.

43.

Pan W, Wang H, Jianwei R, et al. MicroRNA-27a Promotes Proliferation, Migration and Invasion by Targeting MAP2K4 in Human Osteosarcoma Cells. Cell Physiol Biochem 2014;33:402–412.

44.

Ge X, Liu X, Lin F, et al. MicroRNA-421 regulated by HIF-1; promotes

28

metastasis, inhibits apoptosis, and induces cisplatin resistance by targeting E-cadherin and caspase-3 in gastric cancer. Oncotarget 2016;7:24466–82. 45.

Hao J, Zhang S, Zhou Y, et al. MicroRNA 421 suppresses DPC4/Smad4 in pancreatic cancer. Biochem Biophys Res Commun 2011;406:552–557.

46.

Hu H, Du L, Nagabayashi G, et al. ATM is down-regulated by N-Mycregulated microRNA-421. Proc Natl Acad Sci 2010;107:1506–1511.

47.

Yau TO, Wu CW, Dong Y, et al. microRNA-221 and microRNA-18a identification in stool as potential biomarkers for the non-invasive diagnosis of colorectal carcinoma. Br J Cancer 2014;111:1765–1771.

Author names in bold designate shared co-first authorship.

29

Figure legends

Figure 1. Heatmap from NGS results in a total of 124 tissue samples. Expression of the 50 most significant deregulated miRNAs (i.e. highest fold change and lowest false discovery rate) for CRC or AA with respect to their paired normal tissue is shown. Red pixels correspond to an increased abundance of the specific miRNA in the indicated sample, whereas green pixels indicate decreased miRNA levels.

Figure 2. Panel A. Between-group analysis plot depicting sample clustering based on miRNA expression profile. CRC, colorectal cancer; AA, advanced adenoma; C, paired normal tissue. Panel B. Venn diagram generated from NGS results of 124 tissue samples. Inside circles, miRNAs with false discovery rate <0.05 and fold change ≥1.5 or ≤-1.5 are shown. Convergence between circles is the common significant deregulated miRNAs in both neoplastic lesions. Red, number of up-regulated miRNAs; green, number of down-regulated miRNAs.

30

31

Table 1. Clinico-pathological characteristics of patients included in the study. Discovery phase (n=124)

Age, mean (SD) Gender, no. (%) Female Male

Technical validation phase (n=78)

Clinical validation phase (n=767)

CRC (n=30)

AA (n=32)

CRC (n=11)

AA (n=28)

Control (n=39)

CRC (n=67)

AA (n=347)

NAA (n=136)

Control (n=217)

72 (12.2)

59.6 (5.5)

68.3 (12.5)

59.7 (5.4)

59.2 (6.1)

63 (7.8)

59.9 (5.9)

59.8 (5.5)

59.4 (5.7)

16 (53.3) 14 (46.7)

13 (40.6) 19 (59.4)

5 (45.5) 6 (54.5)

12 (42.9) 16 (57.1)

29 (74.4) 10 (25.6)

27 (40.3) 40 (59.7)

115 (33.1) 232 (66.9)

60 (44.1) 76 (55.9)

131 (60.4) 86 (39.6)

12 (40.0) 18 (60.0)

-

6 (54.5) 5 (45.5)

-

-

20 (29.9) 47 (70.1)

-

-

-

9 (30.0) 12 (40.0) 8 (26.7) 1 (3.3) -

-

5 (45.5) 4 (36.4) 2 (18.2) -

-

-

23 (34.3) 16 (23.9) 18 (26.8) 5 (7.5) 5 (7.5)

-

-

-

-

32 (100.0) 2 (6.3) 6 (18.8) 1 (3.1)

-

18 (64.3) 28 (100.0) 1 (3.6) 6 (21.4) -

-

-

224 (64.5) 275 (79.3) 75 (21.6) 93 (26.8) 17 (4.9)

-

-

CRC characteristics Tumor location, no. (%) Proximal Distal Tumor stage, no. (%) I II III IV Unknown AA characteristics (%) Three or more polyps, no. (%) Size ≥10 mm, no. (%) HGD, no. (%) Villous component, no. (%) Carcinoma in situ, no. (%)

CRC, colorectal cancer; AA, advanced adenoma; NAA, non-advanced adenoma; SD, standard deviation; HGD, high grade dysplasia.

Table 2. MiRNAs selected in the discovery phase in tissue samples to be further validated in fecal samples. Discovery phase (NGS) CRC vs. normal mucosa MicroRNA miR-106b-5p miR-130b-3p miR-17-5p miR-182-5p miR-183-5p miR-18a-3p miR-203a miR-20a-5p miR-21-5p miR-221-3p miR-24-3p miR-25-3p miR-27a-3p miR-29a-3p miR-335-3p miR-345-5p miR-34a-5p miR-421 miR-424-3p miR-92a-3p miR-95-3p

FC 1.72 1.96 2.07 6.32 6.25 2.00 2.73 2.21 1.54 2.11 1.72 1.46 1.70 1.89 2.64 1.79 2.85 2.08 1.92 1.64 2.40

FDR 3.28E-11 1.56E-10 1.52E-10 7.56E-20 5.42E-19 4.82E-08 4.93E-13 6.07E-10 1.95E-08 4.80E-14 2.79E-09 1.54E-06 5.59E-08 2.63E-09 2.22E-12 6.07E-10 2.73E-14 1.00E-08 1.20E-07 7.89E-07 5.50E-09

AA vs. normal mucosa FC 1.72 1.96 2.07 6.32 6.25 2.00 2.73 2.21 1.54 2.11 1.72 1.46 1.70 1.89 2.64 1.79 2.85 2.08 1.92 1.64 2.40

CRC, colorectal cancer; AA, advanced adenoma, FC, fold change; FDR, false discovery rate.

FDR 3.28E-11 1.56E-10 1.52E-10 7.56E-20 5.42E-19 4.82E-08 4.93E-13 6.07E-10 1.95E-08 4.80E-14 2.79E-09 1.54E-06 5.59E-08 2.63E-09 2.22E-12 6.07E-10 2.73E-14 1.00E-08 1.20E-07 7.89E-07 5.50E-09

Expression mean 5.9 6.1 8.4 10.4 7.9 3.8 7.7 8.7 12.5 9.4 7.5 11.8 7.4 10.1 6.1 8.3 8.3 5.2 4.2 15.8 4.8

Table 3. Clinical validation of miRNA candidates by qRT-PCR in fecal samples. Clinical validation phase (n=767) CRC (n=67) vs. control (n=217)

AA (n=347) vs. control (n=217)

1

NAA (n=136) vs. control (n=217)

MicroRNA

P-value

AUC

P-value

AUC

P-value

AUC

miR-130b-3p

1.06E-02

0.71

6.86E-03

0.69

2.64E-01

0.62

miR-21-5p

1.16E-01

0.69

6.15E-01

0.65

6.68E-01

0.59

miR-221-3p

7.26E-03

0.70

6.96E-01

0.64

7.24E-01

0.60

miR-25-3p

3.47E-02

0.70

1.42E-01

0.65

1.51E-01

0.61

miR-27a-3p

2.78E-02

0.69

4.86E-01

0.65

1.23E-01

0.61

miR-29a-3p

4.44E-02

0.69

2.70E-01

0.64

8.91E-02

0.60

miR-335-3p

5.77E-01

0.67

3.13E-01

0.65

1.64E-01

0.63

miR-34a-5p

7.30E-03

0.71

6.44E-01

0.64

7.94E-01

0.59

miR-421

1.27E-06

0.77

1.18E-04

0.71

1.66E-01

0.61

CRC, colorectal cancer; AA, advanced adenoma; NAA, non-advanced adenoma; AUC: area under the receiver operating characteristics curve. 1 Results were adjusted by age and gender.

Table 4. Performance characteristics of fecal miRNA candidates to identify patients with advanced neoplasm. Clinical validation phase (n=767)

1

Advanced neoplasm (n=414) vs. non-advanced neoplasm (n=3 MicroRNA

2

2

p-value

Sensitivity

Specificity

AUC

0.003

82

39

0.64

miR-21-5p

ns

72

48

0.62

miR-221-3p

ns

70

51

0.62

miR-25-3p

ns

72

47

0.62

miR-27a-3p

0.02

69

52

0.63

miR-29a-3p

ns

68

53

0.62

miR-335-3p

ns

74

45

0.62

miR-34a-5p

ns

71

49

0.62

0.0001

81

43

0.68

miR-130b-3p

miR-421

AUC, area under the receiver operating characteristics curve; ns, not significant. 1 Results were adjusted by age and gender. 2 Sensitivity and specificity was calculated considering the optimal cut-point associated with the minimum error rate.

Table 5. Discriminative capacity of fecal miRNA-based predictive model1.

ndpoint

AUC

Training set (n=578) 2 2 95%CI Sn Sp

2

2

AUC

Test set (n=189) 2 2 95%CI Sn Sp

PPV

2

PPV

NPV

0.40 0.30 0.47 0.47

0.88 0.93 0.94 0.81

0.63 0.63 0.74 0.64

0.55 0.54 0.58 0.54

0.71 0.72 0.91 0.74

0.67 0.42 0.96 0.59

0.60 0.73 0.33 0.69

0.34 0.15 0.41 0.41

0.32 0.19 0.45 0.45

0.81 0.93 0.82 0.72

0.62 0.59 0.67 0.59

0.54 0.50 0.49 0.49

0.70 0.68 0.85 0.69

0.62 0.45 1.00 0.43

0.58 0.74 0.31 0.63

0.27 0.17 0.33 0.33

miRNA-based predictive model 3

CRC) 4 CRC)

4

0.74 0.74 0.86 0.71

0.70 0.69 0.80 0.66

0.78 0.78 0.92 0.76

0.74 0.64 0.96 0.61

0.63 0.77 0.36 0.71

hemoglobin concentration-based predictive model 3

CRC) 4 CRC)

4

0.61 0.67 0.70 0.64

0.57 0.62 0.61 0.59

0.66 0.72 0.78 0.69

0.60 0.53 0.89 0.50

0.59 0.75 0.33 0.68

ined fecal miRNA and hemoglobin concentration-based predictive model 3 CRC) 0.72 0.68 0.76 0.67 0.66 0.54 0.78 0.70 0.63 0.78 0.68 0.64 4 CRC) 0.74 0.69 0.78 0.63 0.79 0.40 0.91 0.67 0.58 0.76 0.48 0.75 4 0.90 0.86 0.94 0.96 0.48 0.70 0.90 0.93 0.87 0.99 0.97 0.43 0.70 0.65 0.75 0.50 0.75 0.70 0.56 0.64 0.54 0.74 0.49 0.71 1 MiRNA-based predictive model included miR-421 and miR-27a-3p, along with age and gender. 2 All parameters were calculated considering the optimal cut-point associated with the minimum error rate. 3 Negative category: patients with non-advanced adenomas and individuals with normal colonoscopy. 4 Negative category: individuals with normal colonoscopy. CRC, colorectal cancer, AA, advanced adenoma; NAA, non-advanced adenoma, AUC, area under the receiver operating characteristics curve; CI, confidence interval; Sn, sensitivity; Sp, specificity; PPV, positive predictive value; NPV, negative predictive value.

0.47 0.22 0.63 0.63

What you need to know: Background and context: MicroRNA (miRNA) signatures in fecal samples might be used to identify individuals with advanced adenomas (AAs) or colorectal cancer (CRC) and increase the accuracy of non-invasive screening methods. New findings: Increased levels of the microRNAs MIR421 and MIR27a-3p, along with increased hemoglobin concentration, in fecal samples identified individuals with CRC or AA among subjects with a positive result from a fecal immunohistochemical test in a screening program. Limitations: This was a study of individuals with a positive result from a fecal immunohistochemical test; further studies are needed of subjects naïve to CRC screening. Impact: Analysis of fecal miRNA profiles might be used as a non-invasive, widely distributable, and cost-effective approach for population-based screening. Lay Summary: This study showed that molecules detected in fecal samples can identify individuals with advanced colorectal neoplasms. This test might be used to increase the accuracy of non-invasive colorectal cancer screening methods.

Table S1. Fecal miRNA qRT-PCR results from the technical validation phase. Technical validation phase (n=78) CRC (n=11) vs. control (n=39) MicroRNA hsa-miR-106b-5p hsa-miR-130b-3p hsa-miR-17-5p hsa-miR-182-5p hsa-miR-183-5p hsa-miR-18a-3p hsa-miR-203a hsa-miR-20a-5p hsa-miR-21-5p hsa-miR-221-3p hsa-miR-24-3p hsa-miR-25-3p hsa-miR-27a-3p hsa-miR-29a-3p hsa-miR-335-3p hsa-miR-345-5p hsa-miR-34a-5p hsa-miR-421 hsa-miR-424-3p hsa-miR-92a-3p hsa-miR-95-3p

1

AA (n=28) vs. control (n=39)

p-value

AUC

p-value

AUC

0.10 0.03 0.08 0.99 0.06 0.22 0.18 0.54 0.03 0.03 0.10 0.04 0.03 0.07 0.11 0.14 0.02 0.02 0.11 0.14 0.09

0.82 0.86 0.83 0.76 0.83 0.77 0.80 0.75 0.87 0.85 0.82 0.85 0.87 0.84 0.81 0.79 0.87 0.90 0.81 0.80 0.83

0.10 0.05 0.95 0.39 0.13 0.63 0.48 0.68 0.01 0.20 0.33 0.07 0.01 0.10 0.02 0.83 0.26 0.002 0.41 0.65 0.98

0.74 0.79 0.65 0.69 0.70 0.66 0.67 0.67 0.76 0.69 0.68 0.72 0.77 0.71 0.75 0.66 0.70 0.83 0.70 0.66 0.68

CRC, colorectal cancer; AA, advanced adenoma; AUC, area under the receiver operating characteristics curve. 1 Results were adjusted by age and gender.

Table S2. Characteristics of patients included in training and test sets used in the development and validation of the predictive model.

2

Age [IQR]

2

Hemoglobin concentration [IQR]

1

Training set (n=578)

Test set (n=189)

P-value

60.0 [55.2;65.0]

60.0 [55.0;64.0]

0.33

290 [159;756]

275 [160;897]

0.95

Gender (%)

0.27

Female

258 (44.6)

75 (39.7)

Male

320 (55.4)

114 (60.3)

miR-21-5p

18.3 [8.88;45.7]

15.7 [7.94;39.7]

0.35

miR-25-3p

2.01 [0.94;4.23]

2.19 [0.98;3.91]

0.42

miR-27a-3p

0.92 [0.38;2.13]

0.92 [0.35;2.10]

0.91

miR-29a-3p

0.83 [0.34;1.98]

0.71 [0.34;1.72]

0.70

MiRNA [IQR]

2

miR-34a-5p

165 [92.6;294]

166 [104;278]

0.91

miR-130b-3p

4.54 [2.43;8.19]

4.73 [2.59;9.16]

0.32

miR-221-3p

14.7 [5.70;35.3]

14.0 [6.30;29.2]

0.98

miR-335-3p

77.6 [44.9;125]

80.1 [45.9;126]

0.57

miR-421

103 [53.2;210]

111 [54.8;211]

0.83

Group (%)

0.93

Colorectal cancer

49 (8.48)

18 (9.52)

Advanced adenomas

260 (45.0)

87 (46.0)

Non-advanced adenomas

105 (18.2)

31 (16.4)

Control (normal colonoscopy)

164 (28.4)

53 (28.0)

1

P-value was obtained by Mann-Whitney's U test (continuous variables) and Fisher's test (categorical variables). Variables are expressed as average with the interquartile range (IQR).

2

Figure S1. Outline of the study. CRC, colorectal cancer; AA, advanced adenomas; NAA, non-advanced adenomas; FDR, false discovery rate; FC, fold change.

Figure S2. Analytic strategy for miRNA predictive model generation with results obtained in the clinical validation phase (n=767). Training and test sets had 75% (n=578) and 25% (n=189) of samples, respectively. CRC, colorectal cancer; AA, advanced adenomas; NAA, non-advanced adenomas; SMOTE, synthetic minority oversampling technique; GBM, gradient boosting machine; RF, random forest; CT, C-tree; SVM, support-vector machine; KNN, K-nearest neighbor; LDA, linear discriminant analysis.

Figure S3. Principal component analysis. Analyzed fecal miRNAs were clustered in two groups, indicating that miRNAs in the same group were redundant. PC1, principal component 1; PC2, principal component 2.

Figure S4. Calibration curves made with 15 predicted versus observed probability bins for the primary endpoint (i.e. identification of patients with advanced neoplasm). A. Results obtained considering only fecal hemoglobin concentration, adjusted by age and gender. B. Results obtained considering combination of fecal miRNA signature and fecal hemoglobin concentration, adjusted by age and gender. Superior panels show results obtained in the training set (n=578), whereas inferior panels show those achieved in the test set (n=189).