A Distinctive microRNA (miRNA) Signature in the Blood of Colorectal Cancer (CRC) Patients at Surgery

Background: Liquid biopsy (LB) provides an examination of the peripheral blood of cancer patients for circulating tumor cells, cell-free nucleic acids and microRNAs (miRNAs) and is an established tool of precision medicine. Unlike most previous LB studies that focused on advanced metastatic colorectal cancer (CRC), we assessed miRNA dysregulation in blood samples obtained on the day of surgery from patients with primary CRC lesions but no clinical evidence of extra-colonic diffusion. In this study, plasma preparation included miRNAs associated to exosomes, but excluded large macrovesicles from the preparation. Methods: The miRNA profile in plasma isolated from a cohort of 35 CRC patients at the day of surgery was analyzed by Next Generation Sequencing (NGS) and further confirmed by droplet digital RT-PCR (dd-RT-PCR). Results: A miR-141-3p/miR-221-3p/miR-222-3p upregulation signature previously described in advanced CRC did not discriminate the analyzed early-CRC cohort from six tumor-free donors (Tf-D). In contrast, NGS-based miRNome analysis of a training cohort of five CRC and three tumor-free donors identified a novel, distinct nine miRNA signature comprising five up-regulated and four down-regulated miRNAs, six of which could be confirmed in the full CRC and tumor-free donor validation dataset by dd-RT-PCR. Additionally, a KRAS (Kirsten Rat Sarcoma Viral Oncogene Homolog) mutant status was correlated with the plasma content of three identified miRNAs. Conclusions: When the data obtained were comparatively evaluated, at least one of the miRNAs belonging to the signature list was found to be dysregulated in 34/35 (97.1%) of our early-CRC plasma samples. The miRNA list provides diagnostic markers as well as possible molecular targets for protocols focusing on “microRNA therapeutics”.


Introduction
Colorectal cancer (CRC) is the third most common malignancy and the fourth leading cause of cancer death worldwide [1][2][3][4]. Its burden is expected to increase by 60%, with more than 2.2 million new cases and 1.1 million cancer deaths by 2030 [5,6]. Prevention and early diagnosis are key to patients at advanced disease stages [35][36][37]. On this basis, we set out to determine, in a single study, whether these three miRNAs (alone and in aggregate) might be similarly elevated at earlier disease stages-i.e., in the presence of a more limited tumor burden. To this end, we enrolled 35 newly diagnosed CRC patients. Blood was obtained right before (hours to minutes) surgical removal of the primary tumor, and plasma miRNAs were quantitatively assessed by dd-RT-PCR assays specifically designed for each miRNA. Blood from six tumor-free donors (Tf-D) was assessed in parallel. The results are shown in graphical form ( Figure 1A), as representative dd-RT-PCR plots ( Figure 1B), and in tabular form ( Figure 1C). The plasma levels of miR-141-3p, miR-221-3p and miR-222-3p exceeded the highest levels seen in the controls (boxed values) in only 11/35 (31.4%), 11/35 (31.4%) and 8/35 (22.8%) of the tested patients, respectively (bolded in Figure 1C). Even when cumulatively considered, the three miRNAs were poorly informative on our samples, since only 20/35 (57.1%) CRC patients displayed at least one up-regulated miRNA, whereas the remaining patients displayed no up-regulation at all ( Figure 1C). We conclude that neither single miRNAs nor their combination efficiently discriminate CRC patients from tumor-free donors (Tf-D) suggesting that miR-141-3p, miR-221-3p and miR-222-3p may find limited application at early CRC stages, unless combined with the analysis of other miRNAs. In view of this, we were prompted to undergo a systematic miRNome screening to look for more robust miRNA dys-regulation events in the pre-operative stage.

miRNome Assessment by NGS: Experimental Training Set
Five and three plasma samples were randomly selected from CRC patients and tumor-free donors (Tf-D), respectively. These eight samples were used for NGS-based miRNome assessment as described in the Methods. Data were analyzed to highlight clusters, if any, with possible biological significance.
The unsupervised heatmap of log transformed normalized miRNA expression levels, the unsupervised Principal Component Analysis (PCA), as well as the heatmaps of CRC vs. tumor-free donors (at Fold Changes >2) are shown in Figure 2-panels A, B and C, respectively. The heatmap of CRC vs. tumor-free samples at Fold Changes >1.5 is shown in Figure S1. All these elaborations

miRNome Assessment by NGS: Experimental Training Set
Five and three plasma samples were randomly selected from CRC patients and tumor-free donors (Tf-D), respectively. These eight samples were used for NGS-based miRNome assessment as described in the Methods. Data were analyzed to highlight clusters, if any, with possible biological significance.
The unsupervised heatmap of log transformed normalized miRNA expression levels, the unsupervised Principal Component Analysis (PCA), as well as the heatmaps of CRC vs. tumor-free donors (at Fold Changes >2) are shown in Figure 2-panels A, B and C, respectively. The heatmap of CRC vs. tumor-free samples at Fold Changes >1.5 is shown in Figure S1. All these elaborations provided evidence for distinct clusters of CRC vs. tumor-free donors (Tf-D), although clusters were not entirely resolved. As to individual miRNAs, setting the FC cutoff at >1.5 resulted in the identification of 77 differentially expressed miRNAs ( Figure S1), a number that was refined to 26 by repeating the analysis at FC value >2.0 ( Figure 2C). Next, we excluded miRNAs that (a) exhibited the greatest variation (ranges) across Tf-D samples (SD >2.5 fold the average values), and/or (b) were expressed at very low levels in dd-RT-PCR (content <0.2 copies/µL) or for which an amplification kit was not available. This resulted in a short list of candidate miRNAs in our training set ( Figure 3A) comprising the final 9-miRNA signature, including five up-regulated and four down-regulated miRNAs ( Figure 3B). MicroRNA-specific dd-RT-PCR assays were developed to independently confirm this signature in a larger cohort of CRC samples. Representative plots of dd-RT-PCR analysis of up-( Figure 3C) and down-( Figure 3D) regulated miRNAs are shown in Figure 3. provided evidence for distinct clusters of CRC vs. tumor-free donors (Tf-D), although clusters were not entirely resolved. As to individual miRNAs, setting the FC cutoff at >1.5 resulted in the identification of 77 differentially expressed miRNAs ( Figure S1), a number that was refined to 26 by repeating the analysis at FC value >2.0 ( Figure 2C). Next, we excluded miRNAs that (a) exhibited the greatest variation (ranges) across Tf-D samples (SD >2.5 fold the average values), and/or (b) were expressed at very low levels in dd-RT-PCR (content <0.2 copies/µL) or for which an amplification kit was not available. This resulted in a short list of candidate miRNAs in our training set ( Figure 3A) comprising the final 9-miRNA signature, including five up-regulated and four down-regulated miRNAs ( Figure 3B). MicroRNA-specific dd-RT-PCR assays were developed to independently confirm this signature in a larger cohort of CRC samples. Representative plots of dd-RT-PCR analysis of up-( Figure 3C) and down-( Figure 3D) regulated miRNAs are shown in Figure 3.   Figure S1 (FC > 1.5) and in Figure 2C (FC > 2).  Figure  S1 (FC >1.5) and in Figure 2C (FC >2).    The dotted green and red boxes in the control samples indicate the plasma lowest (green) and highest (red) miRNA levels used for comparison with the CRC samples. In the CRC samples, the solid green boxes show down-regulated miRNA values (lower values with respect to the lowest values found in controls), while the solid red boxes show up-regulated miRNA values (higher values with respect to the highest values found in controls). The number of dysregulated miRNAs is indicated from each CRC sample in the last right columns (this numbers in several CRC samples increase considering the analysis of miR-141-3p, miR-221-3p and miR-222-3p shown in Figure 1C). A further analysis comparing CRC values with the average values found in controls is reported in Figure S2. A comparison of miR-15b-5p, miR-584-5p and miR-425-3p (displaying an upregulation trend in CRC samples), and miR-141-3p, miR-221-3p and miR-222-3p, is shown in Table S3.

dd-RT-PCR Assessment of the 9-miRNA CRC Signature on the Full Experimental Dataset
Figures 4 and 5A summarize the complete quantitative analysis of the 9-miRNA signature in the plasma of tumor-free donors (Tf-D) and CRC samples (n = 6 and n = 35, respectively). When compared to the training set, the full dataset revealed a wider scatter in the observed values, as expected. Nevertheless, overall trends (increases in miR-584-5p, miR-15b-5p, and miR-425-3p and decreases in miR-144-3p, miR-144-5p, and miR-486-5p in CRC compared to normal) could be confirmed in at least 6/9 miRNAs. The two most significant changes in miRNA levels were seen for miR-584-5p (up-regulated) and miR-144-3p (down-regulated), Mann-Whitney p = 0.016 and p = 0.046, respectively. Data are analytically displayed in Figure 4 as copies/uL of miRNA in plasma, and the dysregulation (up-or down-regulation relative to control) is individually noted for each patient. Although single miRNAs in the 9-miRNA CRC signature may not be informative in individual patients (the most informative was miR-584-5p, that was found upregulated in 25/35 patients), it is readily evident from Figure 4 that, in all but one (CRC30) of the 35 tested CRC samples, there are from 2 to 8 dysregulated miRNAs. Therefore, the 9-miRNA signature is informative as to the CRC disease status at surgery in 97.1% of the CRC patients tested in this study. The data depicted in Figure 4 were supported by a further analysis conducted comparing the values in CRC samples with the average values found in controls and reported in Figure S2. Of interest, some individual CRC plasma samples behaved as outliers either because of the number of dysregulated miRNAs or because of very high down-or up-regulation. In summary, there are some CRC cases with strong signatures. This can be a very useful information in the road of personalized treatments based of the so called "miRNA therapeutics".
Concerning the high levels of CRC heterogeneity with respect to miRNA content ( Figure 4 and Table S2), further analyses are necessary, since the analyses of tumor location, grading and the presence of a metastatic phenotype create a need for larger numbers of recruited patients.

Discussion
In this study, we show that three blood miRNAs (miR-141, miR-221 and miR-222), long known to be upregulated in advanced CRC, are weakly associated with early CRC. These three miRNAs were found to be upregulated in 65.7% of CRC patients at surgery (Figure 1C), whereas the remaining

Discussion
In this study, we show that three blood miRNAs (miR-141, miR-221 and miR-222), long known to be upregulated in advanced CRC, are weakly associated with early CRC. These three miRNAs were found to be upregulated in 65.7% of CRC patients at surgery (Figure 1C), whereas the remaining miRNA levels are very similar (or lower) in samples from tumor-free donors (Tf-D). Based on these results, we went on and identified a novel 9-miRNA signature from the blood miRNomes of CRC patients with localized disease, on the day of surgical removal of their primary tumor, days to weeks after the initial diagnosis. This early-CRC signature, identified by NGS, was subsequently validated by dd-RT-PCR, and displayed 97.1% specificity (Figure 4 and Figure S2. The 9-miRNA is independent of the above mentioned miR-141-3p, miR-221-3p and miR-222-3p but can complement the diagnostic power of these three miRNAs (Table S1). Altogether, these results highlight the importance of implementing stage-specific CRC biomarkers. Improved association with a disease status held, even when the analysis was restricted to the upregulated miRNAs of the newly identified signature.
There are several caveats to these findings, including the limited number of clinical samples: tumor-free donors (Tf-D) and CRC patients were 3 and 5, and 6 and 35, in the training and experimental datasets, respectively. Therefore, it is not expected that this signature might be self-standing, but rather that it represents a useful source of biomarkers to be added to other classifiers and predictors. In addition, further studies will clarify the role (if any) of several miRNAs that have been excluded by our analysis (for instance due to their low expression in the training set or due to the lack of commercial PCR kits). Notwithstanding these limitations, we provide evidence that the 6-miRNA signature is robust, is consistent with a large body of published evidence, and captures early dysregulation events in CRC. As assessed by PCA, dysregulation appears to involve a minimum of three distinct miRNA sets of related biological functions mapped by miR-144-5p, miR-144-3p and miR-486-5p (Set 1), miR-15b-5p, miR-221-3p, miR-425-3p, miR-584-5p and miR-10a-5p (Set 2), miR-483-5p, miR-141-3p, miR-222-3p and miR-1247-5p (Set 3). Of special interest, although only one of these miRNAs was previously associated in blood with CRC, and none with an early-stage disease, many were suspected to be strongly involved in CRC pathobiology [42][43][44][45]. For instance, miR-144-3p (down-regulated in 25/35 of our CRC patients) was also expressed at low levels in the SW837 and SW1463 cell lines, and its overexpression suppresses CRC cell viability, migration and proliferation [46]. It is tempting to speculate that low miR-144-3p blood levels are due to the adaptive selection of aggressive CRC cell populations. Potential targets of miR-144-3p in CRC are anoctamin 1 (ANO1) [47], MAD Homolog 4 (SMAD4) [48] and Rho-associated coiled-coil containing protein kinase 1 (ROCK1) [46,49]. Likewise, miR-486-5p down-regulation in blood mirrors its low expression (relative to normal), previously seen in tumor tissues from early as well as advanced CRC stages [50]. Pleomorphic adenoma gene like-2 (PLAGL2) [51], a zinc finger protein transcription factor displaying oncogenic function in colorectal cancer (CRC) has been proposed as a target of miR-486-5p. As to the down-regulation of miR-1247 in 11/35 of our CRC patients, it is similarly consistent with a predominant tumor suppressor function that is well established in a variety of tumors, including CRC [52][53][54][55][56][57]. The MYCBP2/c-myc axis may underlie the anti-tumor activities of miR-1247 [58]. However, we also observed up-regulation in seven cases, a finding that deserves further functional studies.
The upregulation of miRNAs in our blood collection is also in keeping with previous studies. For instance, miR-483-5p has been shown to be strongly oncogenic by acting through a complex interplay involving Insulin Growth Factor (IGF2) and Deleted in Liver Cancer (DLC-1) [52]. Interestingly, this miRNA was already known to be higher in the blood of CRC patients as compared to normal controls [52]. As to miR-10a-5p and miR-425-3p, little is known about their biological role, but they are highly expressed in tumors of the right colon [59], and BRAF-mutated CRCs [60], respectively. Thus, with the exception of miR-144-5p, miR-584-5p, and miR-15b-5p, that, to our knowledge, were not previously associated with CRC, all the miRNAs identified by the present study are contained in the known CRC tissue miRNome, and behave as established CRC onco-suppressors (when down-regulated) or oncomiRNAs (when up-regulated). Concordance between the presently detected levels in blood and the previously noted levels in CRC tissues suggests, albeit in the absence of direct testing, a simple straightforward mechanisms whereby direct release from CRC tumors results in roughly proportional miRNA abundance in CRC tissues and blood. This is remarkable, since several confounding factors appear to operate on the fine tuning of miRNA and ctDNA release, as assessed by mouse CRC xenotransplant models [61]. Possible molecular targets of the miRNAs belonging to the CRC-9-miRNA-signature are shown in Table S3.
In analogy with many published studies [62][63][64], some of our up-regulated miRNAs may be useful therapeutic targets. In this respect, it is important to note that KRAS mutations are predictive of resistance to epidermal growth factor receptor (EGFR)-blockade with therapeutic antibodies in metastatic CRC. Furthermore, KRAS mutations are over-represented in the aggressive-upon-relapse CMS3 subtype. Moreover, high KRAS ctDNA levels are predictive of a poor CRC outcome [34]. Interestingly, we found that miR-425-3p and miR-141-3p, and possibly miR-15b-5p, are over-represented in the KRAS-mutated "undruggable" subset. Fortunately, several new agents are now being developed for this subset, and it will be of interest to determine how miRNA dys-regulation intertwines with KRAS-dependent constitutive activation of the EGFR pathway [65]. Our findings prioritize future experiments of combined KRAS/miRNA targeting in a KRAS-mutated background.
Concerning the possible association of the 9-miR signature with other mutations, our cohort of CRC patients are not informative for NRAS and BRAF mutations (Table S2). A larger and informative cohort of patients will be necessary. In addition, the miRNA association with other markers (such as Ki-67 and CD34) should be considered, in order to determine the impact of miRNA expression on cell proliferation in early CRC stages.

Study Design and Ethics
This study enrolled 35 patients diagnosed as CRC patients at IRCCS Regina Elena National Cancer Institute, Rome, Italy. Demographic and clinical features are shown in Table S2. All subjects (including those belonging to the control group) signed a written informed consent and did not take any drugs at blood sampling. This study was approved by the IRCCS Regina Elena National Cancer Institute Ethical Review Board (authorization ID #CEC/541/15). Enrollment of tumor-free donors was designed in order to approach the average age and the sex distribution of the CRC cohort (described in Table S2). The age was 53.6 and the ratio M/F was 3/3.

Plasma Preparation
Blood (30 mL) was drawn in BD Vacutainer K 2 EDTA tubes and processed within 1h. Plasma was isolated by two successive rounds of centrifugation at 4 • C (2000× g for 20 min, and 16,000× g for 10 (min), and stored at −80 • C in single-use 2 mL aliquots until extraction [61]. These plasma preparations were expected to include exosomes, but to exclude large macrovesicles.

RNA Isolation
Total RNA, including miRNAs, was isolated starting from 150 µL of plasma using miRNeasy Serum/Plasma Kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions [66]. Plasma was treated to disrupt exosomes and denature miRNA-binding proteins in 5 volumes of QIAzol Lysis Reagent (Qiagen, Hilden, Germany). In total, 400 amoles of cel-miR-39-3p (Thermo Fisher Scientific, Walthman, MA, USA were added to control miRNA recovery efficiency. After the purification, total RNA was eluted in a final volume of 18 µL.

Next Generation Sequencing (RNA-Seq)
NGS analysis was performed at the Laboratory for Technologies of Advanced Therapies (LTTA) of Ferrara University. SmallRNA libraries were prepared from total RNA using the TruSeq ® Small RNA Library Prep Kit v2 (Illumina, San Diego, CA, USA, RS-200-0012/24/36/48), according to manufacturer's indications. Briefly, 35 ng of purified RNA were linked to RNA 3 and 5 adapters, retrotranscribed and amplified using Illumina primers containing unique indexes for each sample. Libraries were quantified through the Agilent Bioanalyzer by using the High Sensitivity DNA kit (Agilent, Santa Clara, CA, USA, 5067-4626) and equal amounts were pooled and submitted to size-selection in order to keep only fragments between 130-160 bp. After ethanol precipitation, library pool was quantified as stated above, denatured and diluted to 1.8 pM before being sequenced using the Illumina NextSeq500 platform and NextSeq ® 500/550 High Output Kit v2 (75 cycles) (Illumina, San Diego, CA, USA, FC-404-2005). Raw data have been demultiplexed automatically by the Illumina BaseSpace Sequence Hub (https: //basespace.illumina.com/home/index) and converted to FASTQ format. After quality checking, they were evaluated using FastQC tool (https://www.bioinformatics. babraham.ac.uk/projects/fastqc/); adapters sequences were trimmed by Cutadapt (http://cutadapt.readthedocs.io/en/stable/index.html). In this step, sequences shorter than 10 nucleotides were removed. Reads mapping was performed using the STAR algorithm (https://www.ncbi.nlm.nih.gov/pubmed/23104886), with the human microRNAs sequences from the miRBase 22 (http://www.mirbase.org/) as a reference. Raw mapped reads enumeration was performed using the htseq-count script from the HTSeq tools (http://www-huber. embl.de/HTSeq/doc/overview.html). Raw counts were normalized using DESeq2 bioconductor package (http://bioconductor.org/packages/release/bioc/html/DESeq2.html). The optimal read depth to analyze the miRNA transcriptome of plasma was determined at 10 million reads per sample [67,68]. NGS raw-data of the 9-miRNA signature are reported in Table S4.  Table S5) and a variable volume (depending on its levels in plasma) of cDNA. Droplets were automatically generated using Automated droplets generator (AutoDG) (Bio-Rad, Hercules, CA, USA) and the generated emulsion was amplified using the following amplification program: 95 • C for 10 min, 95 • C for 15 s and 60 • C for 1 min. Steps two and three were repeated for 40 cycles and the ramp rate was set at 2.5 • C/sec. A final step at 98 • C for 10 min to inactivate the polymerase enzyme activity was added. Amplified droplets were analyzed for the fluorescence content using QX200 Droplet Digital PCR system. Data analysis was performed using QuantaSoft software, version 1.7.4 (Bio-Rad, Hercules, CA, USA) [69][70][71]. The RT-ddPCR assay was chosen in consideration of its efficiency, the quantification of absolute amounts of analyzed molecules and the possibility to identify changes concerning low numbers of miRNA molecules, as expected using plasma preparation.

Statistical Analysis
In figures, the median is used to describe the miRNA expression levels, and comparisons between groups were made by using the Mann-Whitney test. Spearman's correlation coefficient was used to detect association between miRNA plasma levels.

Conclusions
To our knowledge this is the first report, with the only noted exception of miR-483-5p [52], that provides evidence for the discrimination ability (CRC vs. tumor-free donors) of the nine miRNAs making up our signature. Overall, this signature discriminates, with a high level of sensitivity and specificity, CRC patients at an early disease stage-i.e., at surgery, representing a first step toward early molecular diagnosis. In addition, these findings contribute to an area (KRAS-mutated CRC) of unmet medical need under intense investigation. KRAS/miRNA co-targeting may alleviate the shortage of targeted therapies in this aggressive CRC subset.
In addition to the obvious impact of this study on non-invasive CRC diagnostics, our data are of interest for the development of miRNA therapeutics using anti-miRNA molecules against up-regulated miRNAs (such as miR-15b-5p, miR-584-5p and miR-425-3p) or pre-miRNA mimicking the biological functions of down-regulated miRNAs. In this respect we have recently described an antisense molecule inhibiting miR-15b-5p and inducing apoptosis on the CRC cell line HT-29 [72].
Author Contributions: A.F., R.G. and P.G. conceived and designed the experiments; J.G., C.P., M.A., P.R. and E.G. performed the experiments; F.C. and S.Z. were in charge of patient recruitment; E.P. was in charge of FFPE evaluation, processing and of clinical data collection. A.F., C.S., R.G. and P.G. analyzed the data; R.G., A.F. and P.G. wrote the paper. All authors read and approved the final manuscript.
Funding: This work was supported by the Associazione Italiana per la Ricerca sul Cancro (AIRC) (IG #13575 to RG; IG #19052 to PG; Nuvenia Fellowship ID#19503 to MA). This study was also supported by the European Union (EU) Horizon 2020 Research and Innovation Programme (GA #633937, project ULTRAsensitive PLAsmonic devices for early CAncer Diagnosis (ULTRAPLACAD)) and by the Interuniversity Consortium for the Biotechnology, Italy.