We examined the miRNA expression pattern in a large set of clinically and molecularly well-characterized tissue samples. Several key observations were made. First, unsupervised PCA and cluster analysis demonstrated that normal colon tissue, adenomatous tissue, pMMR1-derived carcinomas and dMMR1-derived carcinomas were all clearly separated, supporting the presence of unique molecular differences between these groups of tissue. Although distinctions between normal/tumor and pMMR/dMMR tumors have been previously reported 
, this is the first report in which a set of well-defined tubulovillous adenomas have been tested for global expression differences and shown to be distinct from normal colon and both pMMR- and dMMR-derived tumor tissue. The few reports on miRNAs in colon adenomas have largely been restricted to the analysis of only selected targets 
Second, probe-by-probe comparisons in various group comparison analyses identified a set of changes in miRNA levels that are consistent with a normal to adenoma and adenoma to carcinoma sequence. A comparison of normal tissue to adenomas, normal tissue to pMMR carcinomas and normal tissue to dMMR carcinomas identified 31, 31 and 28 miRNAs that were differentially expressed with a 2-fold change or higher, respectively. Of interest, the majority of miRNAs that were differentially expressed between normal and adenoma were also differentially expressed in both the pMMR and dMMR tissue groups. The miRNAs showing the greatest difference in expression between normal and adenoma that were also similarly changed in the tumor groups include miR-135b (6.89-fold), miR-31 (4.76-fold), miR-137 (-10-fold), miR-1 (-4.64-fold), miR-9 (-4.43-fold). Furthermore, of the 31 miRNAs differentially expressed between normal tissue and adenoma (≥2-fold change and p-value ≤6.8e-5), 11 up- and 20 down-regulated, 26 (84%) were also differentially expressed with a fold change of ≥1.41 in the normal/pMMR1 comparison while this was the case for 27 (87%) in the normal/dMMR1 comparison (Table S1)
. Importantly, all but one of these miRNAs was differentially expressed in the same direction, the single exception being miR-552 for the normal/dMMR1 comparison. Thus, the primary difference observed between the adenoma and carcinoma comparisons for many of the miRNAs appears to be a difference in the magnitude of the fold change (Table S1)
. In addition to these similarities, however, significant differences were noted. Several of the miRNAs identified were differentially expressed in the normal-carcinoma comparisons but not statistically different in the normal-adenoma comparison, for example miR-375, miR-196b, miR-153, miR-147 and miR-642 (Table S1)
. Note that the fold difference in miRNA expression as measured by the Illumina array is relative and not absolute. Determination of the absolute miRNA expression differences would require the use of more quantitative methods. As we have previously described the accuracy of this platform 
, additional studies to confirm miRNA expression differences were not performed for this report.
These data fit well with a multi-hit model of tumorigenesis. In this model, some initiating events are required to transition from normal to adenoma, while additional events are required to transition from adenoma to carcinoma. The substantial overlap for those miRNAs that are both statistically significant and show large fold changes between the normal-adenoma and the normal-carcinoma comparison suggests that many, if not most, of these miRNA changes are acquired early and persist throughout the later stages of malignant transformation. It is important to note, however, that these data should be interpreted with caution. The measured differences in these experiments reflect the average change observed for all cells present in the tissue of interest. Given the presence of substantial cellular heterogeneity, specific cell-based studies (such as in situ hybridization) will be required to distinguish those changes originating from the neoplastic cells compared to those from stromal or inflammatory cells. Thus, some of the early and persistent changes detected in these analyses may be due to changes in non-tumor related cellular processes. Additionally, the magnitude of expression differences is also dependent on the extent of the cellular heterogeneity, that is, the ratio of neoplastic to non-neoplastic cells.
Of the nine miRNA targets that demonstrated the largest fold changes, six were found in the normal to adenoma comparison ()
. Five of these showed similar expression differences in the two carcinoma groups. Of these six, four have not previously been implicated in adenoma formation (miR-31, miR-1, miR-9 and miR-99a) and, thus, represent novel findings. Of interest, miR-9 has been implicated in the c-myc
, an oncogenic pathway well characterized in CC. A great deal of literature exists for miR-31, demonstrating that this miRNA regulates a number of essential signaling pathways in mammalian cells and has been implicated in several aspects of tumorigenesis, including metastatic progression and tumor cell growth 
. miR-1 is abundantly expressed in normal skeletal muscle and is implicated in muscle differentiation 
. In tumor conditions, miR-1 was found to be down-regulated leading to the deregulation of genes associated with myogenesis 
Two of the miRNAs demonstrating the largest fold changes for the normal to adenoma comparison, miR-137 (decreased expression) and miR-135b (increased expression), have previously been reported to be important early events in colon carcinogenesis. For miR-137, Balaguer et al. 
demonstrated the specific involvement of this miRNA in both colon adenomas and carcinomas. They demonstrated that the decreased expression is due to abnormal hypermethylation and that transfection of this miRNA in CC cell lines significantly inhibited cell proliferation. This study, along with others in oral cancer and in glioblastoma 
, strongly implicates a tumor suppressor model for miR-137. In another study, Nagel et al. 
found that miR-135a and 135b were up-regulated in both colon adenomas and carcinomas, consistent with results from our study. In addition, they showed that miR-135 targets the 3′ untranslated region of APC
, suppresses its expression, and induces downstream Wnt signaling. The APC
gene has long been recognized as a key tumor suppressor in sporadic and hereditary CC 
. In addition to the adenomas, it is important to note that miR-137, miR-135 and the novel miRNAs miR-31, miR-1, miR-9 were all found to be significantly differentially expressed in both pMMR and dMMR tumors, suggesting that these alterations are biologically involved in these two very different types of tumors. Of interest, we have recently shown that miR-183, one of the miRNAs that is highly expressed in both colon adenomas and carcinomas, negatively regulated EGR1
expression, which in turn affects the expression of PTEN
. Further, miR-183 is implicated in tumor cell migration by the negative regulation of both EGR1
and PTEN 
Chr 14q32 contains one of the most miRNA rich regions in the human genome with over 25 miRNAs organized into at least five clusters 
. Results from our current study demonstrated significant involvement of this miRNA cluster in CC development, with decreased expression in adenomas and intermediate levels (between normal and adenoma) in the carcinoma subgroups ()
. The corresponding orthologous miRNAs in mice are maternally imprinted and are controlled by imprinting control regions present upstream of these miRNA clusters 
. Previous reports have shown that miR-127 present in this region is epigenetically regulated and can be modulated by chromatin-modifying drugs 
. Further, 14q32 miRNAs are also down-regulated in several solid tumors 
including osteosarcoma 
. In a recent study comparing mice iPS cells with embryonic stem cells, the complete miRNA locus was down regulated due to hypermethylation in the imprinting control regions 
. Based on these previous studies, it is very likely that the transition from normal colon to adenoma is characterized by epigenetic alterations that lead to the down-regulation of the 14q32 miRNAs in colon adenomas.
As previously published 
, our studies also demonstrate that tumors with defective DNA mismatch repair differ with respect to their miRNA expression profile from those with proficient DNA mismatch repair. Although these two tumor groups were separated by both PCA and cluster analysis, only a few miRNAs demonstrated statistically significant expression differences (≥2-fold) between the two (miR-31, miR-552, miR-592, miR-224). It is important to note, however, that many of the significant miRNAs identified among the three main two-group comparisons between normal-adenoma, normal-pMMR and normal-dMMR (Table S1
) were common to each other. Collectively, these observations indicate that although the clinically relevant pMMR and dMMR tumor subtypes differ with respect to their global miRNA expression patterns, the specific changes observed for these two tumor types are also quite similar to each other. As noted above, this similarity may reflect critical common tumor-specific processes, or they may also reflect local non-tumor-related cellular functions.
There are several noteworthy negative findings in this study. Among the pMMR1 group of tumors, for example, no separation of groups was observed by unsupervised PCA and cluster analysis for stage, gender, age of onset or between the MSS and MSI-L groups. These observations are consistent with the differential expression analyses in which there were few or no statistically significant differences between the comparisons. An analysis across tumors for each of the two dMMR subtypes (dMMR1 and dMMR2) also showed no discernible differences. All dMMR tumors, regardless of their origin (germline or epigenetic), demonstrated similar global miRNA expression patterns. This is the first systematic comparison of these two different subgroups of MMR tumors. Overall, these data suggested that the underlying molecular characteristics among those cases within both the pMMR group (young, old, Stages I-IV, MSS, MSI-L) and within the dMMR group (different MMR genes, different mechanisms of gene inactivation) are more similar to each other than expected. Although there may be a greater degree of heterogeneity at the level of gene mutations and among genes involved in particular pathways, the overall miRNA profiles appear to be fairly homogeneous. Alternatively, more subtle expression differences that have significant biologic effects may still be present but not easily distinguishable with the methods used in this study. Of note, however, the cluster analysis did suggest the possible presence of two sub-populations within the dMMR1 group of tumors, even though these were restricted to cases having epigenetic inactivation of MLH1 (). Additional studies will be required to confirm this latter intriguing observation.
Tumors with pMMR are characterized by the presence of widespread chromosomal gains and loss, and these changes have been detected by a variety of techniques such as Allelic Imbalance and array CGH studies 
. Tumors with dMMR on the other hand, demonstrate few of these changes and overall tend to be near-diploid 
. The nine most significant differentially expressed miRNAs identified in this study ()
all map to regions commonly found to have gains or losses in CC (loss of 1p, 2p, 5q, 9p, 15q, 18q and 21q and gain of 1q, 7q and 20q) 
. Chromosomal alterations as measured by array CGH 
for a subset of the cases could not explain the expression differences observed for any of the six miRNAs examined (data not shown), although the data for those miRNAs that map to multiple sites (mir-9, 1q, 5q, 15q; and miR-1, 18q and 20q) are more difficult to interpret. Overall, these data suggest that copy number differences are not likely to be responsible for the expression differences observed for the six miRNAs examined. For at least one of these, mir-137, methylation appears to be the primary mechanism leading to abnormal expression 
In summary, this is the first systematic analysis of global miRNA changes in colon adenomas along with several well-defined sub-groups of colon adenocarcinomas. The data presented provides an expanded view of miRNA changes that occur in the process of carcinogenesis. We have identified several miRNAs (miR-31, miR-1, miR-9, miR-99a, miR-137 and miR-135b) that show significant differential expression in adenomas compared to normal colon tissue, with several of these linked to critical pathways previously identified for CC, including APC/WNT signaling and cMYC. The finding of several novel miRNAs provide the opportunity to identify associations with known CC pathways or the identification of novel pathways and mechanisms that might be important in the transition from normal to adenoma and from adenoma to carcinoma. We also provide evidence that the miRNA changes detected in the early stages of disease are important in both pMMR and dMMR tumors. This data suggested the involvement of common biologic pathways in both types of tumors, in spite of the presence of numerous molecular differences between them, including differences at the miRNA level. Finally, we also demonstrate a high degree of similarity between a number of tumor subgroups, again highlighting the involvement of common biologic pathways.