Prognostic Impact of let-7e MicroRNA and Its Target Genes in Localized High-Risk Intestinal GIST: A Spanish Group for Research on Sarcoma (GEIS) Study

Simple Summary For intestinal localized high-risk gastrointestinal stromal tumors (GIST) patients’ new molecular biomarkers are urgently needed for a more accurate prognosis. In our study, miRNA profiling analyses was planned to explore new molecular biomarkers with potential prognostic role in this clinical context. Our data, revealed that 39 microRNAs (miRNAs) were significantly deregulated when comparing patients with disease relapsed versus non-relapsed cases. The underexpression of a specific miRNA let-7e and the overexpression of 4 of its target genes (ACVR1B, CASP3, COL3A1 and COL5A2) correlated significantly with worse relapse-free survival. Overall, our results suggest that miRNA profiling is a potential molecular tool useful for a more accurate prognosis for intestinal localized high-risk GIST patients. Abstract MicroRNAs (miRNAs) are small non-coding RNAs that negatively regulate gene expression at the post-transcriptional level, and they have been described as being associated with tumor prognosis. Here, miRNA profiling was planned to explore new molecular prognostic biomarkers in localized intestinal high-risk GIST. Paraffin tumor blocks of 14 and 86 patients were used in the discovery and expansion sets, respectively. GeneChip miRNA v3.0 was employed to identify the miRNAs differentially expressed between relapsed and non-relapsed patient samples, which were validated in the expansion set, by qRT-PCR. RT2 Profiler PCR Array was used for the screening of let-7e targets. Expression levels were correlated with relapse-free survival and overall survival. In the discovery set, 39 miRNAs were significantly deregulated, let-7e and miR-550 being the most underexpressed and overexpressed miRNAs in the relapsed group, respectively. In the expansion set, the underexpression of let-7e or the overexpression of 4 of its target genes (ACVR1B, CASP3, COL3A1, and COL5A2) were statistically associated with worse relapse-free survival. The expression of let-7e and 4 of its target genes are potential prognostic biomarkers in high-risk localized intestinal GIST. The expression of these genes is a potential molecular tool useful for a more accurate prognosis in this subset of GIST patients.


Introduction
Gastrointestinal stromal tumors (GISTs) are the most common mesenchymal neoplasms of the gastrointestinal tract. GISTs represent a morphological and biological continuum from incidentally discovered, <10 mm nodules to large sarcomas [1]. GISTs represent a paradigmatic solid-tumor model for targeted therapy with 3 tyrosine-kinase inhibitors (TKI) registered for first [2], second [3] and third [4] lines in advanced disease and adjuvant Imatinib for high-risk localized GIST [5]. Despite this enormous therapeutic investigation on molecular targeted therapy in this entity, the risk classification for localized GIST patients still relies on clinical-pathological features such as mitotic count, size tumor location, and tumor rupture [6].
Risk classification of GISTs entails critical information in order to make decisions around adjuvant systemic treatment for these patients. Sometimes, overcounting or undercounting just one mitotic figure can imply 3 years of adjuvant treatment with imatinib. Moreover, those cases for which neoadjuvant treatment is recommended usually have insufficient tumor tissue in the core-biopsy for an adequate mitotic count, making it difficult to establish the risk. In view of the above and in the context of a tumor model for targeted therapies, molecular prognostic variables might be instrumental for a more accurate prognosis delimitation.
Few studies have explored the prognostic relevance of genomic alterations, suggesting the potential relevance of KIT and PDGFRA genotyping in prognosis in localized GIST. More specifically, several authors have pointed out the prognostic role of deletion type mutations involving 557 and/or 558 codons of exon 11 in KIT [7,8]. Two authors have demonstrated in larger series that it is feasible to incorporate genotype information in risk classification since patients harboring mutations involving 557 and/or 558 had significantly more risk of recurrence in the intermediate subgroup [9,10]. Nevertheless, the prognostic impact of these mutations is restricted only in gastric location. For intestinal localized GIST, no molecular biomarker has been shown to be a convincing prognostic factor. Hence, there is an unmet need for molecular biomarkers for intestinal localized high-risk GIST.
MicroRNAs (miRNAs) encompass a family of more than 2500 small non-coding RNA that negatively regulate gene expression at post-transcriptional level. Based on the stability that they exhibit in formalin-fixed paraffin-embedded (FFPE) blocks [11], miRNAs were found appropriate to be explored in old archival paraffin blocks of GIST, before the adjuvant imatinib time period. With the aim of exploring new molecular biomarkers with a potential prognostic role in localized GIST, a miRNA profiling analysis was planned in the context of intestinal localized high-risk GIST patients.

Discovery Set
Fourteen patients, corresponding to cases of high-risk intestinal GIST fitting the inclusion criteria, were analyzed for the miRNA screening. There were 8 cases with relapse and 6 without tumor relapse. The median size was 11 cm (5-25 cm) and the median mitotic count was 12 mitoses per 50 hpf. The demographics of this series is shown in Table 1. The microRNA expression profiling resulted in 39 miRNAs significantly (p ≤ 0.05; FC ≥ 2) deregulated between relapsed and non-relapsed groups. The two main discriminating miRNA between both groups were lethal-7e (let-7e; underexpressed, FC = −1163.99, p < 0.0001) and miR-550 (overexpressed, FC = 204.173, p < 0.0001), when comparing relapsed with non-relapsed patients. Other relevant down-regulated miRNAs in relapsed cases were miR-17 (FC = −861.879, p < 0.0001), miR-195 (FC = −785.334, p < 0.0001) and miR-143 (FC = −627.526, p < 0.0001), while miR-1184 (FC = 184.493, p < 0.0001) was up-regulated. However, all those miRNAs presented a lower FC in absolute terms than the selected ones. Within the relapsing cases, the three with the worst prognosis (so-called relapse out group) clustered together ( Figure 1).

Expansion Set
A subset of 86 localized high-risk intestinal GIST patients, meeting the inclusion criteria, were selected as the expansion set. With a median follow-up of 117 months, 58 patients had a recurrence event (67.4%). Median size and mitoses of the primary tumor were 10 cm and 10/50 hpf, respectively. The actuarial 5-year RFS was 33% in this validation series. The most relevant demographics are depicted in Table 1. The majority of cases were diagnosed with GIST from 1995 to 2000. The univariate analyses for RFS showed statistically significant differences for mitotic count (≤10 vs. >10 mitoses per 50 hpf), tumor location (small intestine vs colorectal), and let-7e expression (levels higher or lower than the ROC cutoff). The median RFS was 98.2 (95% CI 0-218.6) and 20 (95% CI 13.5-26.5) months for cases ≤10 and >10 mitoses per 50 hpf, respectively, with p-value < 0.001. It was 33.4 (95% CI 23.7-43.1) and 13.4 (95% CI 1.8-25) months for small intestine and colorectal cases respectively with a pvalue of 0.001, and 41.1 (95% CI 18.2-64) vs. 25 (95% CI 18.2-64) months for cases with let-7e expression levels above and below the ROC cutoff, respectively, with a p-value of 0.012. The four selected target genes of let-7e ( Figure 2 and Table S1) showed prognostic relevance in the univariate analysis, following the ROC cutoff.

Expansion Set
A subset of 86 localized high-risk intestinal GIST patients, meeting the inclusion criteria, were selected as the expansion set. With a median follow-up of 117 months, 58 patients had a recurrence event (67.4%). Median size and mitoses of the primary tumor were 10 cm and 10/50 hpf, respectively. The actuarial 5-year RFS was 33% in this validation series. The most relevant demographics are depicted in Table 1. The majority of cases were diagnosed with GIST from 1995 to 2000. The univariate analyses for RFS showed statistically significant differences for mitotic count (≤10 vs. >10 mitoses per 50 hpf), tumor location (small intestine vs colorectal), and let-7e expression (levels higher or lower than the ROC cutoff). The median RFS was 98.2 (95% CI 0-218.6) and 20 (95% CI 13.5-26.5) months for cases ≤10 and >10 mitoses per 50 hpf, respectively, with p-value < 0.001. It was 33.4 (95% CI 23.7-43.1) and 13.4 (95% CI 1.8-25) months for small intestine and colorectal cases respectively with a p-value of 0.001, and 41.1 (95% CI 18.2-64) vs. 25 (95% CI 18.2-64) months for cases with let-7e expression levels above and below the ROC cutoff, respectively, with a p-value of 0.012. The four selected target genes of let-7e ( Figure 2 and Table S1) showed prognostic relevance in the univariate analysis, following the ROC cutoff.  The prognostic value of both, let-7e and miR-550 was validated by qRT-PCR, confirming its downregulation (FC = 0.6294) and up-regulation (FC = 18.0421), respectively ( Figure S1).

Expansion Set
A subset of 86 localized high-risk intestinal GIST patients, meeting the inclusion criteria, were selected as the expansion set. With a median follow-up of 117 months, 58 patients had a recurrence event (67.4%). Median size and mitoses of the primary tumor were 10 cm and 10/50 hpf, respectively. The actuarial 5-year RFS was 33% in this validation series. The most relevant demographics are depicted in Table 1. The majority of cases were diagnosed with GIST from 1995 to 2000. The univariate analyses for RFS showed statistically significant differences for mitotic count (≤10 vs. >10 mitoses per 50 hpf), tumor location (small intestine vs colorectal), and let-7e expression (levels higher or lower than the ROC cutoff). The median RFS was 98.2 (95% CI 0-218.6) and 20 (95% CI 13.5-26.5) months for cases ≤10 and >10 mitoses per 50 hpf, respectively, with p-value < 0.001. It was 33.4 (95% CI 23.7-43.1) and 13.4 (95% CI 1.8-25) months for small intestine and colorectal cases respectively with a pvalue of 0.001, and 41.1 (95% CI 18.2-64) vs. 25 (95% CI 18.2-64) months for cases with let-7e expression levels above and below the ROC cutoff, respectively, with a p-value of 0.012. The four selected target genes of let-7e ( Figure 2 and Table S1) showed prognostic relevance in the univariate analysis, following the ROC cutoff. Thus, the RFS for values above and below the ROC cutoff were as follows: for ACVR1B gene, 28.9 (95% CI 16.8-41.1) months and 43.6 (95% CI 8.1-79.1) months with a p-value of 0.023; for the COL3A1 gene, 21.3 (95% CI 7.5-35.2) months and 39.5 (95% CI 26.9-52.2) months with p-value of 0.012; for COL5A2 gene, 14.1 (95% CI 2.4-25.7) months and 35.5 (95% CI 23.9-47.1) months, with p-value of 0.003 and for CASP3 gene, 26.5 (95% CI 16.1-37) months and 39.5 (95% CI 4-75.1) months with p-value 0.018. The complete information of univariate analysis is represented in Table 2. Additionally, in order to replicate more precisely the discovery set and to select a more homogeneous population, the analysis was replicated specifically in the 75 small intestine cases. The same prognostic factors in the whole series exhibited a statistically significant difference for RFS in the small intestine series, showing greater differences between prognostic groups (Table 2 and Figure 3). The prognostic role of miR-550 was not confirmed in the expansion set ( Figure S2), neither in the whole series nor in the small intestine subset. In the whole series, cases with let-7e overexpression and low expression of each one of the let-7e target genes had significantly better RFS ( Figure S3). In the multivariate analysis, mitotic count higher than 10 mitoses per 50 hpf (HR 3.5; 95% CI 1.7-7.1; p = 0.001) and CASP3 mRNA expression above the ROC cutoff (HR 2.5; 95% CI 1.2-4.9; p = 0.010) showed independent worse prognosis for RFS. Likely, in the small intestine series these factors were the only independent prognostic variables: mitotic count higher than 10 mitoses per 50 hpf (HR 3.2; 95% CI 1.5-6.7; p = 0.002) and CASP3 mRNA expression above the ROC cutoff (HR 2.6; 95% CI 1.3-5.3; p = 0.008). If let-7e target genes are not included, then let-7e expression (HR 0.4; 95% CI 0.2-0.8; p = 0.005) and mitotic count (HR 2.5; 95% CI 1.3-5.1; p = 0.01) remained as independent prognostic predictors. The mitotic count did not correlate with let-7e expression (p = 0.16).
Only size showed a prognostic role in the univariate analysis for overall survival from the time of diagnosis. Nevertheless, the fact that imatinib has been available since 2002 and some cases had recurrences earlier than this date could induce bias in the overall survival analysis, confirming the use of relapse events as an optimal surrogate for survival in this series. Besides that, a subset of 77 patients with metastatic progression (58 recurred patients from this localized series along with 19 intestinal GIST with metastatic presentation) was considered for a sub-analysis. Among those, there were 56 cases that received imatinib. The univariate analysis screened let-7e and gene expression of let-7e targets as prognostic factors for progression-free survival (PFS) from the time of imatinib initiation. Patients with let-7e expression above and below the ROC cutoff value had progression-free survival of 69.8 (95% CI, 51.3-88.3) and 25.4 months (95% CI, 21.1-29.6), respectively, with a p-value of 0.098. Among the analyzed target genes of let-7e, ACVR1B, CASP3, and COL5A2 showed the same trend observed in the relapse-free survival analysis, a shorter imatinib PFS for values above the ROC cutoff value (Table S2). These differences did not reach statistical significance, probably in relation to the lower number of cases, since only in 33 cases was the RNA expression of these target genes available for the subset with advanced disease treated with imatinib.

Discussion
In this high-risk series of localized, completely resected intestinal GIST and imatinib adjuvant naïve, let-7e was the most significant under-expressed miRNA in the worse prognostic subset of the discovery set. Its prognostic value was validated in the expansion series, showing an independent prognostic value for RFS. let-7 constitutes a family of miRNA with 13 members encoding 9 mature miRNA that control developmental timing and differentiation [12]. Even when let-7 has been described as acting as an oncogene [13], in the vast majority of cases let-7 members also act as tumor suppressor genes [14,15]. Interestingly, let-7 is not only found to be decreased in a plethora of tumors, but the underexpression of several members of this family have been associated with poor tumor prognosis [16,17]. Concerning the mechanisms of action linked with prognosis, let-7 has been involved in several signaling pathways, which could explain its prognostic implication. Suppression of oncogene HMGA2 [18], inhibition of stemness phenotype [19,20], reduction of phosphorylation of carcinogenic signaling as AKT [21], influence in immunomodulation [22], suppression of proliferation [23,24] and control of invasion/migration [25,26] are among the different mechanisms by which let-7 can modulate tumor prognosis.
The screening for putative mRNA targets of let-7e, revealed that 4 genes (ACVR1B, CASP3, COL3A1, and COL5A2) had a significant prognostic impact on RFS, in both the discovery and expansion sets. Overexpression of each of these genes was related to underexpression of let-7e and with a worse risk of RFS in this intestinal localized GIST series. To the best of our knowledge, this is the first time that let-7e and these putative target genes have been related to prognosis in GIST.
Specifically, let-7e has been related to migration and invasion in some tumors. In breast cancer, let-7e negatively regulates Myeloid Zinc Finger 1 (MZF1), a transcription factor that ultimately upregulates lysosomal cathepsins B and L in the context of ErbB2 positive breast cancer [27]. In papillary thyroid carcinoma, let-7e inhibits migration and invasion through the downregulation of HMGB1 [28]. Similarly, two of the significantly dysregulated let-7e target genes in our study, COL3A1 and COL5A2, have relevant roles in migration and invasion, as has been described in some tumors. In bladder carcinoma, COL3A1 has been implicated in tumor invasion by regulating the MAPK signaling pathway [29], and in nasopharyngeal carcinoma, COL3A1 was identified as the only target of miR-29b in relation to migration and invasion [30]. Moreover, overexpression of COL5A2 has been related to worse prognosis with potential association with invasion and dissemination in bladder carcinoma [31], or in osteosarcoma, as an effector of NKX2-2 [32], among others.
Overexpression of ACVR1B, a key receptor of bone morphogenetic proteins (BMPs) and an important regulator of the BMP/Wnt signaling pathway and, therefore, for cancer stem cells, was associated with clinically aggressive and poor survival in hepatocellular carcinoma [33]. The implication of ACVR1B and BMP in cancer has exhibited a dual function according to different cancer cell types or different BMP ligands, promoting or preventing critical cancer hallmarks such as stemness, migration, or proliferation [34]. Protein expression of Caspase-3, either the cleaved fraction or total protein, was associated with poor prognosis in patients with moderately-differentiated carcinoma of the tongue [35]. The expression of CASP3 was observed to be elevated in some tumors [36] but decreased in others [37], maybe indicating that its expression and its prognostic influence could be largely influenced by tumor and microenvironment context. It could be speculated that caspase-3 overexpression is an epiphenomenon, at least in tumors with substantial apoptotic activity, since dying cells can release mitogens to promote the so-called apoptosis-induced proliferation [38]. However, pro-apoptotic proteins, such as BAX, are downregulated in GIST, while anti-apoptotic proteins, such as BCL2 members and inhibitor of apoptosis (IAP) proteins, are commonly upregulated in GIST [39]. Therefore, it seems that it is not easy to harmonize the finding that CASP3 overexpression, which is related to worse RFS in localized intestinal GIST, with the inhibition of apoptosis seen in GIST. It could be assumed that CASP3 failed to become activated, in the GIST context, to participate in the programmed cell death. Nevertheless, in other tumors, the expression level of cleaved caspase 3 did not correlate with spontaneous apoptosis [40], and the overexpression of cleaved caspase 3, not only of the procaspase-3, showed worse disease-free survival [35] Thus, despite the fact that dysregulation of apoptosis is detected in GIST by different means [41,42], it is still not possible to clarify the fine-tuning among different regulators of apoptosis to properly explain the apparent discordance between higher CASP3 expression and worse prognosis or apoptosis inhibition in the context of localized intestinal GIST patients. Some miRNAs have been associated with apoptosis dysregulation in GIST, with MIR221/MIR222 having been related to oncogenic GIST development in two of these studies [43,44]. Our group had reported that over-expression of MIR221/MIR222 was related to worse survival prognosis in a subset of kit negative GIST patients [45].
A weakness of this study, in part owing to the old archived samples, is that genotyping was not feasible in up to 14% of the series. The presence of mutations involving codons 557 and/or 558 in exon 11 of the gene KIT, did not exhibit prognostic impact in this series. This is in line with a large study that demonstrated these mutations had a prognostic impact only in gastric GISTs [9]. The absence of a functional validation of these results in GIST in vitro experiments is another limitation of this study.
Profiling of miRNAs was preferred over other types of profiling, such as mRNA, due to the higher stability shown by miRNAs between paraffin-embedded blocks and fresh tumor-sample analysis. In our series, archived paraffin blocks were used from cases not receiving adjuvant imatinib, meaning that the majority of them were previous to 2005.

Study Design, Patients, and Samples
Patients with the following inclusion criteria were selected from the GEIS registry: localized intestinal GIST, classified in accordance with Miettinen criteria as high-risk, having not received neoor adjuvant imatinib, absence of second primary tumor, and confirmation of no tumor abdominal rupture. Procedures were performed in accordance with guidelines established by the hospital's Ethics Committee and the study was approved by Institutional Review Boards of the university hospitals Virgen Macarena-Virgen del Rocío, on 31 May 2016.

RNA Isolation and Microarray Analysis
For the miRNA screening, 14 cases corresponding to patients with high-risk intestinal GIST, which met the inclusion criteria, were analyzed. For the screening of miRNA that discriminates between samples of relapsed and non-relapsed high-risk intestinal GIST, a GeneChip miRNA v3.0 (Affymetrix; Santa Clara, CA, USA) was used. One representative formalin-fixed and paraffin-embedded (FFPE) block was identified from each case, and 5 to 10-µm thick sections were obtained for RNA extraction using the miRNeasy FFPE ® kit (Qiagen; Hilden, Germany), according to the manufacturer´s instructions.
Quality and related size of total and small RNA were assessed by the microfluidics-based platform Agilent 2100 Bioanalyzer (Agilent; Santa Clara, CA, USA) with two chips: Agilent RNA 6000 Nano Kit for total RNA and Agilent Small RNA kit for low molecular weight RNA. In the first approach, the RIN score ranged from 1.90 to 2.50, while the other kit showed RIN values between 2.40 and 2.60. Electropherograms were visualized with the Agilent 2100 Expert software, including data collection, peak detection, and interpretation of the different profiles. All samples showed similar integrity profiles. FFPE extracted RNA has low integrity, ranging from 200 to 600 bp, but it is acceptable for non-coding RNA studies due to their small size.
For microarray analysis, 600 ng of total RNA was labeled using a FlashTag HSR RNA labeling kit (Genisphere; Hatfield, PA, USA) and hybridized to the GeneChip miRNA 3.0 Array (Affymetrix), following the manufacturer's instructions. This version contains 19,913 probesets, including 5818 human premature miRNAs, cajal body associated RNA (scaRNA), small nuclear organizers (snoRNA), and mature miRNA. The GeneChip®Scanner 3000 7G System and reagents from Affymetrix were used to hybridize, wash, stain, and scan the arrays. The hypothesis contrasts were made with an ANOVA test and adjusted p-value using the false discovery rate adjustment set to FDR <0.5 was used as significant. Normalization and statistical analysis were performed with Partek Genomic Suite 6.6 software (Partek; St. Louis, MO, USA), and technical validation of differentially expressed miRNAs was conducted using qRT-PCR.

Expansion Set
Selection criteria for the expansion set were the same as described above: intestinal GIST, classified in accordance to Miettinen criteria as high-risk, having not received neo-or adjuvant imatinib, absence of second primary tumor and confirmation of no abdominal tumor rupture. Patients had to be followed-up with scheduled CT scans by physicians of the GEIS network. Sections were obtained for RNA extraction using the RecoverAll Total RNA Extraction kit (Ambion; Austin, TX, USA). The expression of miRNAs was determined by means of qRT-PCR, using specific Taq-Man probes. miRNAs expression levels were categorized as above or below median values. Kaplan-Meier and log-rank tests were used for time-to-event variables with RFS, being the clinical endpoint.

Quantitative RT-PCR for miRNA Quantification
A biological validation was carried out on the expansion set (n = 86). For these cases, one representative block was selected and 3 × 20-µm-thick sections were cut. RNA was isolated using the RecoverAll™ Total Nucleic Acid Isolation Kit for FFPE (Ambion), according to the manufacturer's instructions. RNA concentration was measured in a NanoDrop-1000 spectrophotometer (ThermoFisher Scientific, Waltham, MA, USA) absorbance 260/280 and 260/230 ratios were measured to assess the quality of the samples.
Reverse transcription was performed from 200 ng of total RNA using the MicroRNA Reverse Transcription Kit ® (Applied Biosystems; Foster City, CA, USA), according to the manufacturer´s instructions.

Screening of mRNA Targets of let-7e
The screening for putative mRNA targets of let-7e was performed with a specific RT2 Profiler PCR Array PAHS-6008Y (Qiagen) according to the manufacturer´s instructions. miRNAs were selected by the manufacturer (Qiagen) according to biological or bioinformatic evidence. Besides, mRNA targets have been characterized as let-7e targets in public databases such as miR-DB using its own algorithm [46,47]. Each array contains 84 target genes plus 5 housekeeping genes. Briefly, an initial reverse transcription reaction was performed. The quantitative real-time PCR interrogating 84 genes previously validated as human targets of let-7 was used (Appendix A). This screening was carried out in ten samples previously used in the biological validation of the differentially-expressed miRNA expression, of which 4 correspond to relapsing samples. For this analysis, samples with optimum quality were chosen from among the biological validation set.

Quantitative RT-PCR for the Validation of mRNA Targets
The expression levels of the genes selected in the mRNA target screening of let-7 were measured by means of qRT-PCR using the following TaqMan RNA assays (Applied Biosystems): ACVR1B (Hs00244715_m1), CASP3 (Hs00234387_m1), COL3A1 (Hs00943809_m1), and COL5A2 (Hs00893878_m1). While B2M (Beta-2-microglobulin; Hs99999907_m1) and GAPDH (Hs00266705_g1) were used as housekeeping genes for data normalization. A 7500 Fast instrument (Applied Biosystems) was used. After data normalization, the relative expression of the genes was evaluated, applying the 2 −∆∆Ct method for the Fold Change (FC) calculation. Figure 4 outlines all the above-mentioned steps in search of biomarkers in this study.
Cancers 2020, 12, x 11 of 18 mRNA targets have been characterized as let-7e targets in public databases such as miR-DB using its own algorithm [46,47]. Each array contains 84 target genes plus 5 housekeeping genes. Briefly, an initial reverse transcription reaction was performed. The quantitative real-time PCR interrogating 84 genes previously validated as human targets of let-7 was used (Appendix A). This screening was carried out in ten samples previously used in the biological validation of the differentially-expressed miRNA expression, of which 4 correspond to relapsing samples. For this analysis, samples with optimum quality were chosen from among the biological validation set.

Quantitative RT-PCR for the Validation of mRNA Targets
The expression levels of the genes selected in the mRNA target screening of let-7 were measured by means of qRT-PCR using the following TaqMan RNA assays (Applied Biosystems): ACVR1B (Hs00244715_m1), CASP3 (Hs00234387_m1), COL3A1 (Hs00943809_m1), and COL5A2 (Hs00893878_m1). While B2M (Beta-2-microglobulin; Hs99999907_m1) and GAPDH (Hs00266705_g1) were used as housekeeping genes for data normalization. A 7500 Fast instrument (Applied Biosystems) was used. After data normalization, the relative expression of the genes was evaluated, applying the 2 −ΔΔCt method for the Fold Change (FC) calculation. Figure 4 outlines all the above-mentioned steps in search of biomarkers in this study. Among the 86 cases, 50 were sequenced using Sanger sequencing and the rest of the cases with NGS. DNA was isolated from 3-to 5-µm sections of FFPE tissues. After deparaffinization, the tumor tissue was processed with the QIAamp DNA Investigator Kit (Qiagen) according to the manufacturer´s instructions. Intronic primers were used to amplify exons 9, 11, 13 [48] and 17 [49] of KIT and exons 12 and 18 of PDGFRA [50] by PCR, as previously described in Martin et al. [8]. PCR products were bi-directionally sequenced in F and R, with the specific intronic primers used in the amplification diluted to a work concentration of 3.2 µM. Sequencing was performed in an ABI 3130xl Among the 86 cases, 50 were sequenced using Sanger sequencing and the rest of the cases with NGS. DNA was isolated from 3-to 5-µm sections of FFPE tissues. After deparaffinization, the tumor tissue was processed with the QIAamp DNA Investigator Kit (Qiagen) according to the manufacturer´s instructions. Intronic primers were used to amplify exons 9, 11, 13 [48] and 17 [49] of KIT and exons 12 and 18 of PDGFRA [50] by PCR, as previously described in Martin et al. [8]. PCR products were bi-directionally sequenced in F and R, with the specific intronic primers used in the amplification diluted to a work concentration of 3.2 µM. Sequencing was performed in an ABI 3130xl sequencer using the BigDye Terminator v3.1 (Applied Biosystems).
KIT and PDGFRA genotyping, by NGS, was performed using the GIST MASTR kit, following manufacturer's instructions (Multiplicom; Niel, Belgium). The assay generates a library of specific amplicons in two rounds of PCR followed by purification using Agencourt AMPure XP (Beckman Coulter, Beverly, MA, USA) and quantitation with PicoGreen, as described in Feliubadalo et al. [51].
Aligned sequences were visualized with Integrative Genomic Viewer Software 2.4.8 [52], and Variant calling was analyzed using Variant Studio 2.2 (Illumina; San Diego, CA, USA).

Statistical Analysis
Data analyses were performed with R v3.3.1 using R studio 0.99.486 and SPSS version 26 (IBM; Chicago, IL, USA). For categorical variables, frequency and percentages were calculated. For gene expression categorization, we calculate the optimal cutoff using ROC curves for their impact on progression and death. For time-to-event variables (i.e., relapse-free survival (RFS) and overall survival (OS)), we used Kaplan-Meier plots with a log-rank test to compare groups. For multivariate survival analysis, we implemented multivariate analysis with the variables that appeared to be significant in the univariate analysis, and the test was carried out according to the Cox proportional-hazard regression model with a backward variable selection. Linear regression was performed to correlate quantitative variables. All p-values reported were two-sided, and statistical significance was defined at p < 0.05.

Data Availability
Raw data from microarray experiments are available in Gene Expression Omnibus (GEO) from NCBI: GSE156715156715.

Conclusions
In conclusion, our analysis from genomic profiling of miRNAs, in a homogeneous high-risk population of localized intestinal GIST, indicated the potential prognostic relevance in RFS for let-7e and 4 of its target genes: ACVR1B, CASP3, COL3A1, and COL5A2. This knowledge opens new options for prognostic biomarkers and potential therapeutic targets. The mechanistic link underlying their prognostic role is currently being investigated in our laboratory.
Supplementary Materials: The following are available online at http://www.mdpi.com/2072-6694/12/10/2979/s1, Table S1: Results of the mRNA differentially expressed in the target screening series. Table S2: Univariate analysis of metastatic cases treated with Imatinib (n = 56). Figure S1: Target miRNA expression in technical validation; Figure S2: Target miRNA expression in biological validation; Figure S3: Relapse-free survival according to let-7e. Funding: The project was funded by the Instituto de Salud Carlos III (ISCIII (Carlos III Health Institute))-Fondo Europeo de Desarrollo Regional; FEDER (European Regional Development Fund (ERDF)), through a public competitive call (project reference PI15/01254; principal investigator Javier Martín Broto) and by the Consejería de la Salud-Junta de Andalucia-(Health Council-Regional Government of Andalusia)-Proyectos de investigación coordinados 2016, through a competitive public call (project reference PC-0537-2016-0537; principal investigator Nadia Hindi).