Recent reports propose that a change in 3′UTR length by means of APA is a coordinated mechanism for altering expression of many genes during T cell activation (Sandberg et al., 2008
), neuronal activation (Flavell et al., 2008
) or embryonic development (Ji et al., 2009
). For a better understanding of gene regulation, it will be important to identify additional cellular, developmental and disease states that lead to increased use of proximal APA signals. We observed shorter 3′UTRs in cancer cell lines compared with non-transformed cell lines, despite similar proliferation rates of the transformed and non-transformed lines, thereby linking 3′UTR shortening with oncogenic transformation even more than with cellular proliferation.
Many proto-oncogenes play roles in proliferation and differentiation of normal cells and must be highly regulated to prevent malignant transformation. In normal cells, the full-length, annotated transcripts were expressed and presumably responsible for normal proto-oncogene function, but in cancer cells, substantially more shorter isoforms were also expressed, which typically differed from the full-length mRNAs only in the length of their 3′UTRs. Our reporter assays and immunoblots revealed that the APA-mediated 3′UTR shortening can have striking functional consequences in the cancer cell lines, with the shorter mRNA isoforms typically producing ten-times more protein. These results suggested that the cancer-associated shortening of 3′UTRs could activate oncogenes, thereby reinforcing the transformed state. In support of this hypothesis, we found that the shorter IMP-1 isoform promoted oncogenic transformation far more efficiently than did the longer one.
The oncogenic potential of IMP-1
was previously demonstrated using a transgenic mouse model with IMP-1
under the promoter of whey acidic protein, which is induced in mammary epithelial cells in pregnant and lactating female mice (Tessier et al., 2004
). The construct overexpressing IMP-1
in these mice contained only the first 288 nt of the 6.3 kb annotated 3′UTR. Our results, which showed both that cancer cells endogenously express a shorter isoform with a 369-nt 3′UTR (with the orthologous mouse isoform estimated to have a 336-nt 3′UTR) and that this shorter isoform is the one that was oncogenic, support the in vivo
relevance of the previous transgenic model.
Part of the IMP-1
oncogene activation observed in our soft-agar assays was explained by escape of the shorter isoform from miRNA-mediated repression. Since the discovery of miRNAs in mammalian cells, much has been learned about their roles in tumorigenesis. Some miRNAs, including the miR-17~92 cluster are amplified in human cancers and act as oncogenes when overexpressed in mice (Ota et al., 2004
; He et al., 2005
). Others, including miR-15/16, miR-34 and the let-7
miRNAs, are deleted or downregulated in cancers and are reported to act as tumor-suppressor genes (Calin et al., 2002
; Johnson et al., 2005
; Mayr et al., 2007
; Yu et al., 2007
; He et al., 2007
). These miRNAs playing tumor-suppressor roles can do so only when their targets retain the cognate regulatory sites. Thus, another important mechanism for oncogenic transformation is the loss of miRNA sites in the mRNAs of oncogenes. This loss can occur through genetic aberrations, such as the translocations that abrogate let-7
repression of the HMGA2
oncogene (Mayr et al., 2007
; Lee and Dutta, 2007
). Our results uncovered an epigenetic mechanism (using the broader sense of the word “epigenetic”) that achieved the same effect: oncogenes can escape miRNA-mediated repression through 3′UTR shortening due to the APA prevalent in cancer cells. Our data suggests that this epigenetic mechanism is more prevalent than is the genetic one because cancer cells had the shortest UTRs as indicated by their low TLI (a transcriptome-wide index), and APA generated the different transcripts in over a third of the expressed genes examined in detail.
Escape from miRNA-mediated repression explained only part of the upregulation conferred by APA. In addition to miRNAs, AU-rich or GU-rich elements can specify repression (Chen and Shyu, 1995
; Vlasova et al., 2008
), but a search of the tested 3′UTRs did not reveal these elements. Thus, our results implicated additional, as-yet-undefined regulatory elements in these 3′UTRs. Interestingly, the net effect of these elements was never activating—whenever an effect was observed, it was repressive. Although activating regulatory elements undoubtedly occur in some 3′UTRs and the net effect of all RNA elements might be activating in some 3′UTRs or cell types, our results support the idea that regulatory phenomena acting on 3′UTRs are generally repressive. This conclusion holds despite the bias in selecting for study 3′UTRs with multiple sites to coexpressed miRNAs, because even after all those sites were mutated, we still observed only repression. The overall repressive character of 3′UTRs could help explain why nearly all successful transgenic mouse models of tumorigenesis overexpress oncogenes missing large segments of their annotated 3′UTRs (Ruther et al., 1987
; Fedele et al., 2002
; Primrose and Twyman, 2006
). As with IMP-1
, our results indicate that these models expressing shorter versions of the mRNA are more biologically relevant than might have been anticipated when they were first generated.
One of the most interesting open questions regarding APA in cancer cells is, what mechanism underlies the recognition and increased utilization of proximal polyadenylation signals? In the sequence surrounding the proximal polyadenylation sites we never found point mutations that would have changed the strength of the polyadenylation signal, with the caveat that by 3′ RACE we investigated only the sequences upstream of the cleavage sites. Although mutations downstream of the cleavage sites cannot be excluded, we hypothesize that differential expression of trans
-acting factors explains the use of proximal polyadenylation sites in cancer cells. Factors that might influence the choice of polyadenylation signal include limiting components of the polyadenylation machinery, RNA-binding proteins that bind in the vicinity of the proximal signal and influence recognition by the polyadenylation machinery (Takagaki et al., 1996
; Martincic et al., 1998
; Veraldi et al., 2001
; Lutz, 2008
; Wang et al., 2008
), and perhaps factors that influence transcriptional elongation rate(Kornblihtt, 2005
). To begin to identify such factors, we examined published array data comparing breast cancer cells with a breast epithelial cell line, MCF10A (Hoeflich et al., 2009
). A survey of the constitutive components of the polyadenylation machinery and other candidates from the literature revealed several that were significantly upregulated in the cancer lines (Figure S8
). These included the mRNAs of cleavage and polyadenylation specificity factor 1 (CPSF1) and cleavage stimulation factor 2 (CSTF2), which recognize the poly(A) signal and accessory sequences including the downstream G/U-rich sequence, respectively, raising the intriguing possibility that an increase of these factors might help increase utilization of sub-optimal proximal poly(A) signals in cancer cells.
We imagine a complex scenario in which some trans-acting factors act globally, some act tissue specifically, and some act gene specifically, with the combinatorial expression of all the different trans-acting factors determining the probability of using each proximal polyadenylation signal. The observation of higher amounts of shorter mRNAs in cancer cells compared with normal cells, with no examples of the reverse for any of the genes and cell types studied, suggested a role for globally acting factors. That some cell lines showed high amounts of shorter transcripts for all genes investigated further implicated the role of globally acting factors. Nonetheless, the differences observed for different genes in different cell types suggested a role for additional factors acting more specifically. Such complexity could explain the differential impact of different oncogenes in different tissues. Indeed, some of the genes for which we did not observe alternative mRNAs are known oncogenes in tissues not included in our panel of cell lines (ARID3B in neuroblastomas, MYB and PLAG1 in hematological malignancies). Perhaps shorter mRNA transcripts might be found in the tissues where these genes have oncogenic effects. Moreover, the prospect of some factors acting more specifically opens the possibility for exceptions to the trend of shorter isoforms expressed preferentially in cancer cells. Combinatorial use of tissue-specific and gene-specific trans-acting factors could for some genes (most intriguingly, for tumor-suppressor genes) lead to higher amounts of shorter mRNAs in normal cells rather than in cancer cells.
The prevalence of APA in cancer lines brought to the fore the question of what influence it might have on oncogenes, and our results for IMP-1
supported the idea that APA was activating. However, APA creates shorter, less repressed isoforms of more than just known oncogenes, as illustrated for DICER1
. Because some of these genes likely act in opposition to oncogenes, the net functional significance of APA in cancer cells is unknown, and in principle could even be tumor suppressive. During both normal development and cancer development there is often a dichotomy of cell proliferation and differentiation (Derynck and Wagner, 1995
). The association of APA with cell proliferation suggests that it might represent a coordinated gene-expression program that antagonizes differentiation during normal development. Accordingly, we propose that cancer cells coopt and exaggerate this proliferation/de-differentiation program with the net effect of enhanced tumorigenesis.
Our observations that the shorter mRNAs were found in transformed cells and that expression of the shorter mRNA of IMP-1 (and presumably other oncogenes) can lead to transformation suggests a APA-mediated feed-forward loop in cancer that might lead to a more aggressive phenotype. This proposal is in agreement with the report on mantle cell lymphomas, in which the patients that have shorter Cyclin D1
3′UTRs have the worst prognosis (Rosenwald et al., 2003
; Wiestner et al., 2007
). However, when this process of generating shorter mRNAs would start and whether APA might play a role in early tumorigenesis is unclear. In general, shorter mRNAs were not observed in non-transformed cell lines, which usually already have one hit towards cancer but are not yet fully transformed. Perhaps if an early lesion activates a signaling pathway that is able to change the expression of a key trans
-acting factor, then APA would contribute to early steps in oncogenesis.
Oncogenes are reported to be overexpressed in human tumors much more frequently than genetic lesions are detected at these loci. Our results could explain some of this discrepancy, because overexpression of oncogenes due to APA-mediated shortening of mRNA transcripts was a widespread phenomenon in the cancer cells investigated. The epigenetic nature of this mechanism for oncogene activation suggests that it can be reversed, perhaps providing a new strategy for cancer treatment.