RNA from 12 brain tumor samples with some variation in their histological diagnoses was sufficient to detect a set of miRNA-mRNA pairs whose fluctuations are highly correlated. The merit of this statistical detection method is that it avoids non-physiological conditions required to alter gene expression. The number of such high correlation pairs was significantly greater than expected in the random case (). A weak positive correlative effect due to distance between the pairs could be detected () as well as a weak negative correlative effect due to direct miRNA targeting of mRNAs (,). However, among the very high correlation pairs only a small number bore a predicted target relationship (11 of 72, ) or were located on the same chromosome (7 of 72). For the remaining high correlation pairs the data does not imply any direction of casualty—is the correlation due to an effect of the miRNA on the mRNA or vice versa? Key to this approach was the endogeneous variations among the samples. In other words, the correlations might stem from individual differences and are not tumor specific. By restricting the samples to gliomas rather than including healthy controls, these data in general cannot directly address carcinogenesis. The interpretation of the correlations is limited to miRNA/mRNA relationships solely within our sample set.
Validation of the correlations was obtained by over-expressing or suppressing the implicated miRNA and measuring the mRNAs predicted to correlate. These experiments frequently matched the prediction and suggested that in these cases, the correlation was due to an effect of the miRNA on the mRNA (). As expected, some pairs were not validated. Among the reasons for failure to validate are: (a) some of the high correlation pairs are false positives; (b) the U251 cells do not exactly replicate the tumors; (c) in some cases more than one miRNA contributes to the effect; (d) the direction of causality may pass from mRNA to miRNA; (e) direct causality does not underlie the relationship, rather the high correlation is the consequence that both the miRNA and the mRNA have a true biological relationship with a third element. Nevertheless, the statistical approach to endogenous fluctuations can reveal in vivo effects of miRNAs on the transcriptional profile.
Global effects of miRNAs upon the transcriptional profile and tissue specificity of mRNA expression have been reported 
. However, these studies have generally focused on miRNA targets which show more modest effects compared to the non-target effects described here. Furthermore, in one of the cited studies 
the over-expression of a miRNA by transfection can create non-physiological changes upon the mRNA target field. Because direct miRNA targeting is unlikely in the absence of a seed region, as observed in most of the high correlation pairs, other explanations for how the high correlation pairs might arise were schematized (). Among the possible explanations for high correlation pairing is the binding of the miRNA to the promoter region of the paired gene with the result that transcription increases in those sets with positive correlations. A non-coding RNA binding to a promoter has been described 
, albeit in this case suppression occurred. Also possible is that the mRNA drives the effect on the miRNA.
Mechanisms through which high miRNA-mRNA correlation pairs could operate.
Validation experiments allowed us to infer a tumor suppression pathway involving miR-181c
were among the endogenously highly correlated transcripts. We determined that miR-181c
leads to an increase in PCAF
protein levels consistent with the known positive interactions between PCAF
(). At least two possible mechanisms could explain this result. The increase in p53
protein levels could drive the increase in PCAF 
. Alternatively, the increase in p53
could occur through PCAF
, which was recently shown to stabilize p53
by inactivating MDM2
, an E3 ubiquitin ligase that negatively controls p53
. Concluding, the positive correlation of miR-181c
with these two tumor suppressor genes can potentially explain the inhibitory effect of miR-181c
on cell growth and the more subtle increase in apoptosis observed in the glioma cells (). Interestingly, miR-181c
has been reported to be low in a panel of high-grade glioblastomas tumors and glioblastomas cell lines 
and inhibition of another member of the miR-181
, in A549 lung carcinoma cells induced cell growth 
. Lower levels of miR-181c
in glioblastomas could indirectly decrease the transcript levels of the positively correlated PCAF
, resulting in decreased activation of major tumor suppressive pathways in glioblastoma cells.