Downloaded from rnajournal.cshlp.org on September 30, 2021 - Published by Cold Spring Harbor Laboratory Press

Targeting of TGFβ signature and its essential component CTGF by miR-18 correlates with improved survival in glioblastoma

JAMIE L. FOX,1,2 MICHAEL DEWS,1 ANDY J. MINN,2,3,4 and ANDREI THOMAS-TIKHONENKO1,2,4 1Division of Pathobiology, Department of Pathology & Laboratory Medicine, Center for Childhood Cancer Research, Children’s Hospital of Philadelphia, Philadelphia, Pennsylvania 19104, USA 2Cancer Biology Graduate Program, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, Pennsylvania 19104, USA 3Abramson Family Cancer Research Institute and Department of Radiation Oncology, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, Pennsylvania 19104, USA

ABSTRACT The miR-17∼92 cluster is thought to be an oncogene, yet its expression is low in glioblastoma multiforme (GBM) cell lines. This could allow unfettered expression of miR-17∼92 target such as connective tissue growth factor (CTGF; or CCN2), which is known to contribute to GBM pathogenesis. Indeed, microRNA-18a (but not other miR-17∼92 members) has a functional site in the CTGF 3′ UTR, and its forced reexpression sharply reduces CTGF and mRNA levels. Interestingly, it also reduces the levels of CTGF primary transcript. The unexpected effects of miR-18a on CTGF transcription are mediated in part by direct targeting of Smad3 and ensuing weakening of TGFβ signaling. Having defined the TGFβ signature in GBM cells, we demonstrate a significant anti-correlation between miR-18 and TGFβ signaling in primary GBM samples from The Cancer Genome Atlas. Most importantly, high levels of miR-18 combined with low levels of the TGFβ metagene correlate with prolonged patient survival. Thus, low expression of the miR-17∼92 cluster, and specifically miR-18a, could significantly contribute to GBM pathogenesis. Keywords: miR-17∼92 cluster; miR-18a; TGFβ; CTGF; CCN2; glioblastoma

INTRODUCTION encoded by paralogous clusters (e.g., miR-18b on chromo- some X). The microRNA cluster miR-17∼92 has been well-studied in Previously, we used publicly available microarray data from human . It exerts oncogenic effects in multiple systems human cancer cell lines, known as the Wooster data set such as B-cell lymphoma (Ota et al. 2004; He et al. 2005; (Greshock et al. 2010), to look for patterns of expression of O’Donnell et al. 2005), adenocarcinoma (Hayashita et al. 2005; Dews et al. 2006), neuroblastoma (Fontana et al. 2008; MIR17HG (Dews et al. 2010). Despite its purportedly onco- Chayka et al. 2009; Mestdagh et al. 2010), and retinoblastoma genic role, we observed a wide range of MIR17HG expression (Conkrite et al. 2011). Processed from the MIR17HG primary across tumor types, with particularly low expression in glio- transcript, this cluster encodes six unique mature microRNA blastoma multiforme (GBM). We hypothesized that certain ∼ sequences: miR-17-5p, -18a, -19a, -20a, -19b, and -92a (for re- cancer types may express relatively low levels of miR-17 92 view, see Mendell 2008). Altogether they contain four unique in order to alleviate repression of target genes important “seed sequences,” which are located at the 5′ end of the in those specific malignancies. To corroborate this hypothe- microRNA (positions 2-8 or 2-9) and are generally thought sis, we searched for genes whose expression anti-correlated to be responsible for target recognition (for review, see with that of MIR17HG. According to this analysis, the genes Bartel 2009). miR-17-5p and -20a share the same seed se- with the first- and second-ranked negative correlation were quence, as do miR-19a and -19b. miR-18a and -92a seed se- THBS1 (encoding -1, or TSP1) and connec- quences are unique, although each has related microRNAs tive tissue growth factor (CTGF) (Dews et al. 2010). While the role of TSP1 in cancer has been well recognized, the contribu- tion of CTGF to neoplastic growth remains unclear. 4Corresponding authors CTGF (also known as CCN2) is a member of the CCN fam- E-mail [email protected] ily of secreted (for review, see de Winter et al. 2008). E-mail [email protected] Article published online ahead of print. Article and publication date are at The role of CTGF in cancer is somewhat controversial. De- http://www.rnajournal.org/cgi/doi/10.1261/rna.036467.112. pending on tumor type and model system, CTGF has been

RNA (2013), 19:177–190. Published by Cold Spring Harbor Laboratory Press. Copyright © 2013 RNA Society. 177 Downloaded from rnajournal.cshlp.org on September 30, 2021 - Published by Cold Spring Harbor Laboratory Press

Fox et al. shown to be anti-migratory and anti-metastatic (Chang et al. miR-17∼92 cluster. Indeed, TargetScan (Lewis et al. 2005) 2004), pro-metastatic (Kang et al. 2003, 2005), positively cor- predicts two partially overlapping sites for miR-18 and related with poor patient survival (Sala-Torra et al. 2007), and miR-19 (Fig. 1A, top) in the CTGF 3′ UTR. Although others either pro-angiogenic (Hall-Glenn et al. 2012) or anti-angio- have reported functional interaction between miR-18 and the genic (Chang et al. 2006). Even within the same study, CTGF CTGF 3′ UTR (Ohgawara et al. 2009; Ernst et al. 2010), no exhibited pro-angiogenic effects on its own, but inhibited mutagenesis studies have confirmed the specific binding VEGF-dependent (Inoki et al. 2002). site. Additionally, although it has been suggested that miR- In GBM, several reports have suggested important roles for 19 targets CTGF (Ernst et al. 2010; van Almen et al. 2011), CTGF. CTGF is overexpressed in ∼58% of primary GBM to the best of our knowledge, there are no published data compared with normal brain tissue, and its expression is asso- demonstrating direct binding of miR-19 to the 3′ UTR. ciated with disease progression and poor patient survival (Xie To determine whether miR-18a and/or miR-19 directly and et al. 2004). Forced overexpression of CTGF in GBM cell lines specifically bind the reported sites, 100-bp fragments of the has been shown to increase in vitro cell migration (Demuth CTGF 3′ UTR were cloned downstream from a Renilla et al. 2008), proliferation, and resistance to tumor drugs, as luciferase cassette to serve as a surrogate 3′ UTR. These frag- well as increase tumor size and vascularization in vivo (Yin ments contained the wild-type sequence (“WT”), a mutation et al. 2010). Because CTGF expression is both important for in the region unique to the miR-18 (“18mut”) or miR-19 GBM malignancy and significantly anti-correlated with that (“19mut”) seed, or a mutation in the region common to of miR-17∼92 in human cancers, we postulated that GBM the two seeds (“18 + 19mut”) (Fig. 1A, bottom). These plas- may maintain low levels of miR-17∼92 to allow unfettered ex- mids and cognate microRNA mimics were cotransfected pression of CTGF. Indeed, according to TargetScan, a web- into the human cell line DLD1 rendered hypomorphic for based algorithm that predicts microRNA targets (Lewis et al. Dicer and thus expressing low basal levels of all microRNAs 2005), and several published reports (Dews et al. 2006; (Cummins et al. 2006). As observed previously (Sundaram Ohgawara et al. 2009; Ernst et al. 2010; van Almen et al. et al. 2011), miR-17∼92 mimics did not have a significant ef- 2011), miR-17∼92 could target CTGF directly via miR-18a, fect on luciferase activity in the absence of any portion of the -19a, and -19b. However, currentdata do not provide evidence CTGF 3′ UTR (Fig. 1B, “Vector”). Similarly, miR-19a, -19b, of a direct mechanism of regulation by miR-19 or validate the or the combination of the two did not inhibit luciferase activ- specific miR-18a binding site. Additionally, previous data ity in the presence of the CTGF WT 3′ UTR or any of the from our laboratory and collaborators have also linked miR- mutants (Fig. 1B; Supplemental Fig. 1A). In contrast, miR- 17∼92 to the TGFβ pathway (Dews et al. 2010; Mestdagh 18a specifically inhibited luciferase activity in the presence et al. 2010), an established transcriptional activator of CTGF of the CTGF WT 3′ UTR, indicating that miR-18a but not expression (Grotendorst 1997; Abreu et al. 2002; Leask and miR-19 is able to directly interact with this sequence. This in- Abraham 2004; Perbal 2004; Baxter and Twigg 2009). teraction was only marginally affected by mutations in the In this study, we demonstrate that, of all miR-17∼92 com- miR-19 seed. In contrast, mutation of the region specific to ponents, only miR-18a affects endogenous CTGF protein ex- the miR-18 seed or the region common to the two seeds re- pression in GBM. This mechanism has both direct and duced the degree of repression, indicating direct functional indirect components, the latter involving transcriptional reg- interaction at this site. ulation of CTGF by TGFβ via targeting of Smad3 by miR- While a luciferase sensor assay indicates the ability of a giv- 18a. This discovery prompted us to evaluate the relationship en microRNA and 3′-UTR sequence to interact, it does not between miR-18, TGFβ signaling, and patient outcome in confirm that this interaction actually affects the endogenous primary GBM samples. We found a negative correlation be- target in a given cell type. To determine which member(s) tween miR-18 expression and the TGFβ signature and a pos- of the miR-17∼92 cluster inhibits endogenous CTGF expres- itive correlation between miR-18 expression and patient sion, we used established human GBM cell lines. As anticipat- survival. These data reveal a unique axis of regulation, where- ed, in all cell lines tested, the endogenous miR-17∼92 levels by miR-17∼92 expression may be maintained at low levels in were just at the detection levels, making loss-of-function GBM to allow expression of CTGF and other TGFβ-induced experiments problematic. Thus, A172 GBM cells, profiled genes, and therefore contribute to GBM pathogenesis. in the Wooster data set, were transfected with microRNA mimics corresponding to each individual member of the clus- ter. Because of very low basal levels, we were unable to mea- RESULTS sure the degree of overexpression reliably. However, while miR-17-5p and miR-20a exhibited inconsistent effects, only miR-18a is the miR-17∼92 component that regulates miR-18a consistently and strongly inhibited CTGF protein CTGF in glioblastoma expression (Fig. 1C). This selectivity validated the use of the The observed anti-correlation between MIR17HG and CTGF microRNA mimic approach. RNA levels (Dews et al. 2010) may have resulted from regu- When miR-18a mimic was transfected into A172 cells, there lation of CTGF by one or more individual members of the was also a profound (∼80%) inhibitory effect on CTGF

178 RNA, Vol. 19, No. 2 Downloaded from rnajournal.cshlp.org on September 30, 2021 - Published by Cold Spring Harbor Laboratory Press

miR-18 targets TGFβ signature in glioblastoma

25 ABControl mimic 1.5-fold miR-18 seed miR-18a mimic ** 1.5-fold common seed 20 miR-19b mimic ** 2.7-fold 2.2-fold miR-18a: 3’ GAUAGACGUGAUCUACGUGGAAU 5’ ** ** 15 CTGF 3’UTR: 5’...A C C AAAAGUUACAUGUUUGCACCUUU ...3’ miR-19a: 3’ A G U CAAAACGUAUCUAAACGUGU 5’ miR-19b: 3’ A G U CAAAACGUACCUAAACGUGU 5’ 10

miR-19 seed 5 Relative luciferase activity WT: TTTGCACCTT miR-18 seed mutant: TTTGCACTCC 0 miR-19 seed mutant: CCCGCACCTT Vector WT 18mut 19mut 18+19mut Common seed mutant: TTTATGCCTT psiCHECK2 construct

D 1.8 1.6 1.4 C 1.2 1 Mimic: ControlmiR-17-5p miR-18a miR-19a miR-19b miR-20a miR-92 0.8 0.6 actin 0.4 0.2 ** CTGF 0

1.00 0.33 0.09 0.81 0.70 0.48 1.13 expression CTGF/ Relative mRNA Control miR-92 miR-18a miR-19a miR-19b miR-20a miR-17-5p Actin Transfected Mimic (20 nM)

1.2 Control mimic EF HCT116A172 HCT116A172 miR-18a mimic 1.0 ** 0.8

0.6

1.5 kb 0.4 ** 1 kb ** ** ** ** 0.2 500 bp Relative mRNA expression CTGF/actin Relative mRNA

0.0 GAPDH CTGF A172 LN18 SF268 U118 U251 U373

FIGURE 1. miR-18a is the miR-17∼92 component that regulates CTGF in glioblastoma. (A, top) Alignment of the CTGF 3′ UTR with miR-18a and miR-19 sequences, with predicted seed sequences indicated. (Bottom) Partial sequences of luciferase sensors with underlines indicating mutated nu- cleotides. (B) Dual luciferase assay performed in DLD-1 Dicer hypomorph cells cotransfected with indicated luciferase constructs and control, miR- 18a, or miR-19b mimic. (C) Representative Western blot of CTGF and actin (loading control) from A172 human GBM cell line transfected with indicated microRNA mimics 48 h prior to harvesting. Numbers below the CTGF band indicate CTGF band intensity relative to that of the actin band. (D) qPCR for CTGF mRNA in A172 cells transfected with the indicated mimics for 24–48 h. (E)3′-RACE products specific for GAPDH or CTGF 3′ UTR from HCT116 or A172 cells. (F) qPCR for CTGF from human GBM cell lines transfected with control or miR-18a mimic for 24 h. Error bars represent the standard deviation of three or more independent experiments; (∗∗) P < 0.01 by Student’s t-test. mRNA (Fig. 1D). In contrast, miR-19 did not significantly re- prevented binding of miR-19. To test this possible scenario, duce CTGF mRNA expression. To obtain a positive control the CTGF 3′ UTR was amplified from A172 total cDNA using for miR-19 mimic functionality, we also measured expres- 3′-RACE. As controls, we used HCT116 cells and the sion of thrombospondin-1, a known target for both miR- GAPDH 3′ UTR. All four 3′-RACE reactions yielded single 18 and miR-19 (Dews et al. 2006, 2010). miR-19a and -19b bands of appropriate sizes (Fig. 1E). Direct sequencing of (as well as -18a) mimic reduced thrombospondin mRNA lev- the A172 CTGF product verified that its 3′ UTR was full els (Supplemental Fig. 1B), confirming their functionality. length, with no mutations to the reported miR-19 site as re- Nevertheless, a mutation in the CTGF 3′ UTR could have ported in GenBank.

www.rnajournal.org 179 Downloaded from rnajournal.cshlp.org on September 30, 2021 - Published by Cold Spring Harbor Laboratory Press

Fox et al.

To confirm that miR-18a regulation of CTGF is not limit- Nevertheless, CTGF regulation by miR-18a has a clear tran- ed to A172 cells, we transfected miR-18a mimic into a panel scriptional component. Interestingly, similar results were ob- of six human GBM cell lines. Although basal expression of tained when measuring thrombospondin hnRNA levels CTGF mRNA, and the degree to which miR-18a inhibited (Supplemental Fig. 1E), suggesting that miR-18a may regu- it, varied between cell lines, miR-18a significantly inhibited late thrombospondin and CTGF transcription in a similar CTGF mRNA levels in each case (Fig. 1F), suggesting that fashion. this microRNA is rate-limiting for CTGF expression in GBM. Thus, of the individual members of the miR-17∼92 miR-18a regulates CTGF transcription in part via cluster, only miR-18, and not miR-19a or -19b, is a bona regulation of TGFβ signaling fide regulator of CTGF expression. Because CTGF is known to be transcriptionally induced by the TGFβ pathway, it is possible that miR-18a regulates miR-18a significantly inhibits CTGF transcription CTGF transcription by inhibiting TGFβ signaling. Since all To elucidate the mechanism(s) whereby miR-18a regulates previous experiments in this study were performed in the ab- CTGF, we compared its effects on mRNA and protein levels. sence of exogenous ligand, this proposed mechanism would The observed 80%–90% inhibition of CTGF mRNA expres- require autocrine TGFβ signaling in A172 cells. To determine sion by miR-18a was equal in magnitude to inhibition of the if A172 cells have such signaling, expression of TGFβ receptor protein (Fig. 2A). This was surprising because direct regula- II (TGFBR2), a required component of the pathway (Fig. tion by microRNAs is traditionally thought to have strong ef- 3A), was inhibited using siRNA. Knockdown of TGFBR2 fects on protein but modest effects on mRNA (Karginov et al. and abrogation of TGFβ signaling (indicated by lack of down- 2007; Baek et al. 2008; Selbach et al. 2008). The latter are stream Smad3 phosphorylation) were confirmed by Western commonly explained by RNA depolyadenylation and ensu- blotting (Fig. 3B). Notably, in the presence of siRNA target- ing destabilization (Wu and Belasco 2005; Giraldez et al. ing TGFBR2, both induction by TGFβ1 and basal expression 2006). To examine whether this is the case, A172 cells were of CTGF were reduced, confirming that A172 cells have au- treated with Actinomycin D to prevent new transcription, tocrine TGFβ signaling. Then, to determine to what extent and CTGF mRNA levels were measured by qPCR. In contrast miR-18a regulation of CTGF is dependent on TGFβ signal- to short-lived (half-life ∼1.5 h) c-Jun and c-Myc transcripts, ing, A172 cells with TGFBR2 knockdown were transfected CTGF mRNA was relatively stable, with a half-life of ∼5–6h with miR-18a mimic. While basal expression of CTGF was (Fig. 2B). Then the cells were simultaneously treated with still decreased in the TGFBR2 knockdown cells, the effective- both Actinomycin D and control or miR-18a mimic for 8 ness of miR-18a against CTGF expression was diminished h, at which point miR-18a already exerts its effects on compared with that seen in control siRNA-treated cells mRNA levels (Supplemental Fig. 1C). Surprisingly, the (Fig. 3C, top and bottom; see bracket). mRNA half-life was only marginally affected by the miR- To determine whether miR-18a is the only member of the 18a mimic (Fig. 2C), indicating that another mechanism of miR-17∼92 cluster that can overcome CTGF induction by regulation could be in place. TGFβ, A172 cells were transfected with microRNA mimics One alternative explanation for the profound down-regu- corresponding to individual members of the miR-17∼92 lation of CTGF mRNA by miR-18a is indirect transcriptional cluster, then stimulated with TGFβ1. Again, only miR-18a repression, which would affect the CTGF primary transcript, strongly inhibited CTGF protein expression, both in the ab- also known as the heterogeneous nuclear RNA (hnRNA). To sence and presence of exogenous TGFβ1 (Fig. 3D). Thus, measure CTGF hnRNA expression, qPCR primers were de- miR-18a effectively inhibits TGFβ-induced CTGF expression. signed corresponding to two separate exon–intron junctions of the ctgf (Supplemental Fig. 1D). These primers were miR-18a directly targets human Smad3 validated using treatment with Actinomycin D, which almost completely obliterated CTGF hnRNA while preserving ∼20% We next asked which component(s) of the TGFβ signaling of CTGF mRNA (Fig. 2D). As an additional validation, stim- pathway is/are susceptible to repression by miR-18a. To this ulation of A172 cells with the TGFβ1 cytokine, a well-estab- end, A172 cells were transfected with miR-18a mimic and lished inducer of CTGF transcription, resulted in sharply treated with TGFβ1, and Western blotting was performed increased CTGF hnRNA levels (Fig. 2E). We then used this for each principal component of the TGFβ pathway shown CTGF hnRNA qPCR assay to measure the effects of miR- in Figure 3A. Although miR-18a surprisingly increased 18a on CTGF transcription. Transfection of miR-18a mimic Smad2 and TGFBR2 levels, it had no inhibitory effect on into A172 cells decreased CTGF mRNA levels by 80%, but any components of the pathway except Smad3 protein (Fig. CTGF hnRNA levels were also sharply decreased by nearly 3E). Inhibition of Smad3 by miR-18a was especially pro- 60% (Fig. 2F). The 20% differential between the two effects nounced in the presence of exogenous TGFβ1 and resulted may reflect changes in mRNA stability that were unseen at in a decrease in phosphorylated Smad3, which is a function- 8 h of miR-18a treatment (Fig. 2C) but appear within 24 h. al readout for TGFβ signaling. To determine if this effect is

180 RNA, Vol. 19, No. 2 Downloaded from rnajournal.cshlp.org on September 30, 2021 - Published by Cold Spring Harbor Laboratory Press

miR-18 targets TGFβ signature in glioblastoma

1.8 1.6 ABCTGF mRNA c-Jun 1.6 CTGF protein 1.4 c-Myc 1.4 1.2 1.2 CTGF 1 1 0.8 0.8 0.6 0.4 0.6 ** ** 0.2 0.4 0

Relative expression CTGF/actin 0.2

Control miR-92 expression relative to control mRNA 0 miR-18a miR-19a miR-19b miR-20a miR-17-5p 0 1 2 3 4 5 6 7 8 Transfected Mimic (20 nM) Actinomycin D treatment (hrs)

CD1.2 1.2 Control mimic Control miR-18a mimic 1 Expon.(Control mimic) 1 ActD Expon.(miR-18a mimic) 0.8 0.8

0.6 0.6

ActD/control 0.4 0.4 0.2

Relative CTGF mRNA expression Relative CTGF mRNA 0.2 0 RNA expression relative to actin RNA 0 1 2 3 4 5 6 7 8 0 Hours after treatment CTGF mRNA CTGF hnRNA Probe #1 CTGF hnRNA Probe #2

1.2 E 4 F Control mimic Vehicle miR-18a mimic 3.5 TGF- 1 3 0.8 2.5 Effect on transcription 2 0.6

1.5 Effect on ** ** 0.4 mRNA 1 stability

0.5 0.2 ** RNA expression relative to actin RNA RNA fold change relative to vehicle RNA 0 0 CTGF mRNA CTGF hnRNA CTGF hnRNA CTGF mRNA CTGF hnRNA Probe #1 CTGF hnRNA Probe #2 Probe #1 Probe #2

FIGURE 2. miR-18a significantly inhibits CTGF transcription. (A) qPCR and quantitated Western blot analysis for CTGF in A172 cells transfected with indicated mimics for 24–48 h. (B) qPCR for indicated genes in A172 cells treated with Actinomycin D relative to untreated samples. (Vertical dashed lines) Approximate half-life. (C) qPCR for CTGF in A172 cells treated with Actinomycin D and simultaneously transfected with control or miR-18a mimic. (Vertical dashed lines) Approximate half-life. (D) qPCR for CTGF mRNA or hnRNA from A172 cells treated with Actinomycin D for 8h.(E) qPCR for CTGF mRNA or hnRNA from A172 cells treated with vehicle or TGFβ1 for 24 h. (F) qPCR for CTGF mRNA or hnRNA from A172 cells transfected with control or miR-18a mimic for 24 h. Error bars represent the standard deviation of three or more independent experiments; (∗∗) P < 0.01 by Student’s t-test.

unique to miR-18a, microRNA mimics corresponding to According to TargetScan and RNA22, another algorithm each member of the miR-17∼92 cluster were transfected indi- for microRNA target prediction (Rigoutsos et al. 2006), the vidually into A172 cells, which were then treated with TGFβ1. Smad3 3′ UTR contains four putative miR-18 sites (Fig. Similar to experiments examining CTGF expression, only 3F), suggesting that regulation of Smad3 by miR-18a could miR-18a inhibited expression of both total and phosphorylat- be direct. Therefore, fragments of the Smad3 3′ UTR contain- ed Smad3 (Supplemental Fig. 1F). Taken together, these re- ing the wild-type sequence (“WT”) or mutated versions sults indicate that miR-18a might regulate human Smad3, (“Mut”) of each of the putative miR-18 sites were cloned either directly or indirectly. downstream from a Renilla luciferase cassette as described

www.rnajournal.org 181 Downloaded from rnajournal.cshlp.org on September 30, 2021 - Published by Cold Spring Harbor Laboratory Press

Fox et al.

A BC TGF 1 siRNA Control TGFBR2 Control siRNA TGFBR2 siRNA TGFBR2 TGFBR1 Mimic Ctl miR-18a Ctl miR-18a

P P TGF 1 - + - + Smad3 Smad2 TGFBR2 CTGF

Smad4 Smad4 Actin TGFBR2

CTGF pSmad3 Actin Smad3 CTGF Actin 1 TGF - 0.9 dependent 0.8 effect of 0.7 miR-18a Vehicle TGF 1 0.6 D 0.5 0.4 Mimic: mimic 0.3 Control miR-17-5pmiR-18a miR-19a miR-19b miR-20a miR-92 ControlmiR-17-5p miR-18a miR-19a miR-19b miR-20a miR-92 0.2 0.1 CTGF 0 Expression relative to control Control miR-18a Control miR-18a mimic mimic mimic mimic Control siRNA | TGFBR2 siRNA 1.00 0.81 0.11 1.29 1.23 1.24 1.54 2.22 3.25 0.18 3.84 2.96 3.39 4.62 Actin

Control mimic miR-18a mimic E 1.6 TGF 1 (hrs) 0 0.5 8 0 0.5 8 G WT Smad2 1.4 Mut

pSmad2 1.2

Smad3 1 **

pSmad3 0.8

Smad4 0.6 ** TGFBR1 0.4 ** TGFBR2 0.2 Luciferase activity relative to control mimic CTGF 0 Empty vector CTGF Smad3 #1 Smad3 #2 Smad3 #3 Smad3 #4 Actin

F

Wild type constructs: 4658 miR-18a: 3’ GAUAGACGUGAUCUACGUGGAAU 5’ 1 392-398 1793-1799 2520-2526 2860-2866 Site #1: 5’ CACCTTG 3’ Site #2: 5’ GCACCTTG 3’ Site #3: 5’ CACCTTA 3’ Site #1: Site #2: Site #3: Site #4: Site #4: 5’ GCACCTT 3’ CACCTTG GCACCTT CACCTTA GCACCTT

FIGURE 3. miR-18a regulates CTGF transcription in part via direct regulation of human Smad3. (A) Simplified diagram of Smad-dependent TGFβ signaling pathway. (B) Western blot analysis of A172 cells transfected with nontargeting siRNA or siRNA targeting TGFBR2 for 24 h, then treated with vehicle or 5 ng/mL TGFβ1 for 3 h. (C, top) Western blot of lysates from A172 cells transfected with nontargeting siRNA or siRNA targeting TGFBR2 for 24 h, then transfected with control or miR-18a mimic for 24 h. (Bottom) Quantitation of CTGF Western blot above; levels of CTGF protein are normalized to the control mimic in each cell type. The bracket indicates contribution of TGFβ signaling to miR-18a repression of CTGF. (D) Representative CTGF Western blot from A172 cells transfected with 20 nM indicated microRNA mimics for 48 h, followed by 8 h of stimulation with vehicle or 5 ng/mL TGFβ1. The numbers below the CTGF band indicate CTGF band intensity relative to that of the actin band. (E) Western blot analysis of A172 cells transfected with 20 nM control or miR-18a mimic for 24 h, then treated with vehicle or 5 ng/mL TGFβ1 for the indicated times. (F, left) Alignment of predicted miR-18 sites in Smad3 3′ UTR and miR-18a sequence. (Right) Distribution of miR-18 sites within Smad3 3′ UTR; (black boxes) 100-bp fragments used in luciferase sensor constructs. (G) Dual luciferase assay performed in DLD-1 Dicer hypomorph cells cotransfected with indicated luciferase sensors and miR-18a mimic. Values were normalized to luciferase activity from cells simultaneously cotrans- fected with indicated luciferase sensors and control mimic. Error bars represent the standard deviation of two independent experiments; (∗∗) P < 0.01 WT versus Mut sequence by Student’s t-test.

182 RNA, Vol. 19, No. 2 Downloaded from rnajournal.cshlp.org on September 30, 2021 - Published by Cold Spring Harbor Laboratory Press

miR-18 targets TGFβ signature in glioblastoma above for the CTGF 3′ UTR. Upon cotransfection of these ture using gene set enrichment analysis. Because expression plasmids and microRNA mimics, miR-18a significantly values for miR-18a and -18b were highly correlated (correla- (∼50% reduction) decreased luciferase activity in the pres- tion coefficient of 0.79, P < 0.001), these values were averaged ence of the site #2 WT sequence, but had no effect when for further analyses. Consistent with the metagene approach, this sequence was mutated (Fig. 3G), indicating that miR- this analysis revealed that the signature as a whole was signifi- 18a is able to bind Smad3 3′ UTR directly at this site. miR- cantly suppressed in GBM tumors that concomitantly ex- 18a also had a small effect on luciferase activity in the pres- pressed high levels of miR-18 (Fig. 4B). ence of the site #4 WT sequence; however, this decrease To explore the potential disease relevance of our findings, was ∼10%, making it a minor contributor to Smad3 down- we investigated the clinical importance of the miR-18/TGFβ regulation. Taken together, these data suggest that miR-18a axis. We hypothesized that miR-18 may repress key TGFβ-in- could be a negative regulator of TGFβ signaling via direct in- duced genes, in addition to CTGF, that contribute to tumor- hibition of Smad3 in GBM. igenesis in GBM. We noted that nine genes induced by TGFβ and significantly anti-correlated with miR-18 expression have been previously implicated in GBM pathogenesis (Fig. miR-18 expression inversely correlates 4C). An additional 26 genes have been implicated in other with the TGFβ signature and influences forms of cancer (Table 1). survival in human GBM Consistent with the importance of the miR-18/TGFβ axis, We wanted to determine if miR-18 could dampen TGFβ Kaplan-Meier survival analysis revealed that GBM patients signaling by testing its effects on TGFβ-induced genes. with high expression of miR-18 and low expression of the However, sets of genes induced by TGFβ (referred to as TGFβ metagene had improved overall survival compared TGFβ signatures) differ significantly between cell types (for with patients with low miR-18 or high TGFβ metagene ex- review, see Massague 2008). To establish the TGFβ signature pression (Fig. 4D). When CTGF expression levels were sub- in a human GBM cell line, we treated A172 cells with vehicle stituted for the TGFβ metagene, the same trend emerged: or TGFβ1 and profiled changes in using Patients with high expression of miR-18 and low expression microarray analysis. From this analysis, 211 genes showed in- of CTGF had better survival than the low miR-18/high CTGF creased expression after TGFβ1 treatment with at least a 1.5- cohort (Supplemental Fig. 1H). This trend, however, was not fold change and an adjusted P-value <0.01. These genes were statistically significant (P = 0.143), attesting to the impor- used to define the GBM TGFβ signature (Supplemental tance of repression by miR-18 of other TGFβ signature genes. Table). Altogether, these data suggest a mechanism whereby GBMs To examine the association between the TGFβ signature maintain their neoplastic phenotype by retaining low levels and miR-18 expression in primary human GBM, we used of the miR-17∼92 cluster and specifically miR-18a, which data from The Cancer Gene Atlas GBM study (Cancer would otherwise inhibit expression of CTGF and other key Genome Atlas Research Network 2008; Verhaak et al. TGFβ signature genes (Fig. 4E). 2010). For each tumor, expression of the TGFβ signature was summarized into a single value, or metagene, by averaging DISCUSSION the expression of the 149 genes common between the above- established signature and the microarray platform used in the It is generally accepted that the miR-17∼92 cluster functions TCGA. This TGFβ metagene was then used to determine as an oncogene in its own right (Ota et al. 2004; Hayashita which microRNAs negatively correlated with the TGFβ signa- et al. 2005; He et al. 2005; Petrocca et al. 2008) or as a key ture in primary GBM samples. Using a false discovery rate of downstream effector of the Myc family of oncoproteins, 10%, 98 out of 535 unique microRNAs examined were found such as c- and N-Myc (O’Donnell et al. 2005; Dews et al. to be anti-correlated with the GBM TGFβ metagene. Both 2006; Chayka et al. 2009). miR-17∼92 contributes to various miR-18a and -18b, which has the same seed sequence as aspects of cancer initiation and progression (Mendell 2008). miR-18a but is expressed from a separate transcript, ranked For example, our laboratory has demonstrated that miR- among the most negatively correlated microRNAs, with 17∼92 augments tumor angiogenesis in the KRAS-driven co- each appearing in the top 7% (Fig. 4A). There was also a sig- lon carcinoma model (Dews et al. 2006), although it possess- nificant anti-correlation between the TGFβ metagene and es a cell-intrinsic anti-angiogenic function in endothelial cells miR-19a and -19b, which may reflect coordinated expression (Doebele et al. 2010). In the colon cancer model, induction of of the entire MIR17HG transcript. Among the different mo- angiogenesis is accompanied by down-regulation of several lecular subtypes of GBM (Verhaak et al. 2010), the strongest anti-angiogenic members of the TSR superfamily, including negative association between miR-18a and -18b with the thrombospondin-1 and clusterin (Dews et al. 2010). The TGFβ metagene was observed in the proneural subtype (P role of thrombospondin-1 in anti-angiogenesis is well estab- = 0.006 and P = 0.025, respectively) (Supplemental Fig. 1G). lished (see below). Additionally, we have observed that resto- As a complementary analysis, we also examined the rela- ration of clusterin expression in miR-17∼92-overexpressing tionship between miR-18 and all 149 genes in the TGFβ signa- cells blocks angiogenesis and overall tumor growth in colon

www.rnajournal.org 183 Downloaded from rnajournal.cshlp.org on September 30, 2021 - Published by Cold Spring Harbor Laboratory Press

Fox et al.

FIGURE 4. miR-18 expression inversely correlates with the TGFβ signature and influences survival in human GBM. (A) Correlation between TGFβ metagene and miR expression. (Blue) The miRs that are negatively or positively correlated at a false discovery rate of 0.1; (red) miR-18a and miR-18b. (B) Gene set enrichment analysis (GSA) demonstrating inverse correlation between TGFβ signature genes and miR-18 expression. The heatmap shows the expression of the 149 genes in columns with each tumor sample in rows. (Orange) High expression; (blue) low. For each gene, the gene score from the GSA is indicated in the plot above the heatmap, with negative scores indicating negative association with miR-18 expression. The expression of miR-18 for each sample is shown in the graph to the right of the heatmap. (C) Negative correlation between select TGFβ target genes and average miR- 18 expression levels. The correlation coefficients are indicated; all P-values are <0.05. (D) Kaplan-Meier survival curves for GBM patients stratified by high and low expression of miR-18 and the TGFβ metagene (TGFbMeta). The P-value shown is from a comparison between miR-18(hi)/TGFbMeta (lo) versus the other groups. (E) Proposed model wherein miR-18a inhibits CTGF via direct inhibition of mRNA (and possibly inhibition of trans- lation), as well as via Smad3-dependent CTGF transcription.

184 RNA, Vol. 19, No. 2 Downloaded from rnajournal.cshlp.org on September 30, 2021 - Published by Cold Spring Harbor Laboratory Press

miR-18 targets TGFβ signature in glioblastoma

TABLE 1. TGFβ-induced cancer genes anti-correlated with miR-18a expression in TCGA

Genes implicated in GBM Oncogenic function PubMed ID

FAP Anti-apoptotic 16187021 FGF2 Pro-proliferative 19340397 Tumorigenic 8559302 INHBA Overexpressed compared with normal brain, pro-proliferative, activates p-Smad2/3 20926007 PDPN Overexpressed, pro-proliferative, pro-migratory 22394497 PFKFB3 Overexpressed 17805487 PPARGC1A Amplified 19304958 RFX2 May contribute to GBM stem cell maintenance 22415835 THBS1 Implicated in cell motility 11745436 Expression correlates with GBM malignancy 11132930 UNC5B Pro-proliferative 22496621

Genes implicated in other cancers Oncogenic function PubMed ID AMIGO2 Required for gastric adenocarcinoma tumorigenicity in mouse studies 15107827 BHLHE40 Anti-apoptotic in oral cancer 22572381 COL1A1 Associated with invasive phenotype in HCC 22681909 Associated with chemoresistance in ovarian cancer 22249249 COL8A1 Up-regulated in papillary thyroid carcinoma 19725780 COL8A2 Up-regulated in papillary thyroid carcinoma 19725780 CSF1R Sustains ERK activation and proliferation in breast cancer cell lines 22096574 Pro-proliferative in renal cell carcinoma 22052465 DIXDC1 Pro-proliferative and tumorigenic in colon cancer 19572978 FHL2 Overexpressed and associated with poor prognosis in many types of cancer 17352216 16378916 17145880 15161045 FOXJ1 Pro-proliferative and associated with poor prognosis in HCC 22488567 GADD45B Associated with chemoresistance 12162804 GLS Induced by c-Myc 19033189 IER3 Required for pancreatic transformation and associated with poor prognosis in PDAC 22565310 Overexpressed in ovarian cancer 22095100 JUNB Pro-migratory and pro-invasive via TGFβ in HCC 22747445 Pro-proliferative and pro-invasive in gastric cancer 22252121 Overexpressed in lymphoma 12393503 KIAA1199 Pro-proliferative in colon cancer 21772334 Overexpressed and correlated with poor prognosis and lymph node metastasis in gastric 19434458 cancer LIMK2 Tumorigenic in breast carcinoma 22492986 Tumorigenic, pro-metastatic, and pro-angiogenic in pancreatic cancer 20047470 LTBP2 Overexpressed in PDAC 21755970 Pro-migratory in melanoma 12716902 MPL Established oncogene in leukemias 8393355 NCF2 Associated with increased mortality in renal cell carcinoma 21944129 NRP2 Pro-tumorigenic in prostate cancer 22777769 Pro-metastatic in papillary thyroid carcinoma 21880798 Pro-tumorigenic in colorectal cancer via TGFβ 21747928 Associated with lymph node metastasis and poor prognosis in breast cancer 19580679 Pro-proliferative, pro-migratory, and pro-angiogenic in gastric cancer 19409892 PALLD Overexpressed and pro-migratory in pancreatic cancer 17194196 PFTK1 Overexpressed and correlated with poor prognosis and chemoresistance in esophageal 22333595 squamous cell carcinoma Pro-invasive in HCC 21577206 PLAU Overexpressed in PDAC 22487470 Associated with resistance to irradiation in esophageal cancer 15365572 PTPRB Overexpressed in gastric cancer 16338072 RASGRP1 Overexpression in mouse model sufficient to cause T-ALL 22116551 Overexpression in mouse skin leads to 17210708 TRIB1 Oncogenic in leukemia 17227832 Expression correlates with poor prognosis in ovarian cancer 17971902 ZNF267 Overexpressed, pro-proliferative, and pro-migratory in HCC 21840307

www.rnajournal.org 185 Downloaded from rnajournal.cshlp.org on September 30, 2021 - Published by Cold Spring Harbor Laboratory Press

Fox et al. cancer (Dews et al. 2006) and neuroblastoma (Chayka et al. in patients with TGFβ metagene-high tumors (blue curve in 2009). While TSR proteins in solid tumors are targeted by Fig. 4D). This finding underscores the importance of the both miR-18 and miR-19, in B-lymphoid malignancies, the TGFβ pathway as an miR-18 effector in GBM. In addition key oncogenic component of miR-17∼92 is miR-19 (Mu to CTGF, several genes comprising the TGFβ signature are et al. 2009; Olive et al. 2009), acting primarily through the likely to contribute to GBM pathogenesis. PTEN pathway. Nine TGFβ target genes that significantly anti-correlate Because glioblastomas are highly angiogenic tumors sus- with miR-18 expression in primary GBM have been linked ceptible to TSR mimetics (for review, see Chung et al. 2007) to GBM malignancy (Fig. 4C); 26 others have been described and are known to inactivate PTEN (Li et al. 1997), one might in the literature to have possible or established roles in other predict that miR-17∼92 would be expressed at high levels in forms of cancer (Table 1). For example, FGF2, INHBA, GBM. However, we have demonstrated that this is not the PDPN, and UNC5B are associated with proliferation of case; in fact, MIR17HG expression was lower in GBM com- GBM cells (see Table 1), and their reactivation following pared with other tumor types (Dews et al. 2010). To explain miR-18a down-regulation might contribute to tumor expan- this paradox, we proposed that elevated levels of miR- sion. Additionally, INHBA, which encodes one of the sub- 17∼92 members other than miR-19 could compromise ex- units of Activin A, a TGFβ superfamily ligand (Deli et al. pression of genes that may be essential for GBM pathogenesis. 2008), has been shown to activate Smad2/3 phosphorylation A similar scenario has been described for other miRs; for ex- in GBM (see Table 1). These data implicate a possible posi- ample, miR-222 is oncogenic in many cancers by virtue of tar- tive-feedback loop between TGFβ and activin signaling path- geting p27 cdk inhibitor (Galardi et al. 2007; le Sage et al. ways, which could play an important role in GBM 2007) but suppresses the development of erythroleukemia pathogenesis. because of its propensity to target c-KIT (Felli et al. 2005). Finally, THBS1, which encodes the protein thrombospon- In GBM, such a key gene could be CTGF, which is involved din-1, is also suppressed by miR-18a (Sundaram et al. 2011), in GBM pathogenesis via its effects on cell migration and its expression negatively correlates with that of miR-18 in (Demuth et al. 2008), proliferation, angiogenesis, and multi- primary GBM (Fig. 4C; Table 1). Thrombospondin has been drug resistance (Yin et al. 2010). Importantly, CTGF expres- shown by many laboratories to have anti-angiogenic proper- sion is known to be associated with GBM progression and ties in various model systems (Good et al. 1990; Weinstat- poor patient survival (Xie et al. 2004). Saslow et al. 1994; Jimenez et al. 2000; Zaslavsky et al. 2010; Here we demonstrate that miR-18a, a member of the miR- Sundaram et al. 2011). However, in GBM, its expression cor- 17∼92 cluster, regulates CTGF expression by interacting relates with malignancy of primary tumors (Kawataki 2000), directly with its 3′ UTR. miR-18a-dependent regulation of and repression of thrombospondin decreases migration of CTGF expression results in 80%–90% decreases of both the GBM cell lines (Amagasaki 2001), suggesting that, like protein and mRNA and significantly inhibits CTGF hnRNA CTGF, thrombospondin may have contradictory roles in levels; therefore, we postulated the existence of a transcrip- cancer depending on tumor type. Additionally, thrombo- tional component. Indeed, we observed that miR-18a effi- spondin can promote TGFβ signaling itself (Crawford et al. ciently inhibits induction of CTGF by TGFβ, which is an 1998), generating another possible positive-feedback loop important pathophysiological CTGF regulator (Grotendorst in GBM with low levels of miR-17∼92. 1997; Abreu et al. 2002; Leask and Abraham 2004; Perbal In general, TGFβ signaling plays a complex role in cancer 2004; Baxter and Twigg 2009). Mechanistically, we discovered and normal cell growth (for review, see Bierie and Moses that miR-18a modulates TGFβ signaling through direct inhi- 2006). For example, we and our collaborators have shown bition of Smad3, which in this cell type appears to be a dom- that in neuroblastoma, NMYC-dependent inhibition of inant mechanism, in addition to the previously described TGFβ correlates with increased in vitro proliferation and in targeting of TGFBR2 by miR-17-5p and -20a (Volinia et al. vivo engraftment (Mestdagh et al. 2010). The TGFβ pathway 2006). In fact, a recent high-profile study argued that miR- is also a frequent target of loss-of-function mutations in co- 20a is tumor suppressive in GBM because of its inhibitory ef- lon cancer (Cancer Genome Atlas Research Network 2008). fects on TGFBR2 (Chen et al. 2012) but did not investigate its In contrast, aberrant activation of the TGFβ pathway has effects on survival of GBM patients. been linked to metastasis and angiogenesis in breast cancer Consistent with the existence of these multiple axes of reg- (Kang et al. 2005). Similar mechanisms might dominate in ulation, miR-18 expression anti-correlates with that of the GBM, where TGFβ pathway activation was reported to corre- TGFβ metagene in primary human GBM. In the proneural late with cell migration (Demuth et al. 2008) and radiation GBM subtype, specifically, expression of both miR-18a and resistance (Zhang et al. 2011). Furthermore, its down-regula- its paralog, -18b, significantly anti-correlate with that of the tion by miR-17∼92 sets the stage for a highly coordinated role TGFβ metagene. Most importantly, high levels of miR-18 reversal, whereby TGFβ signaling not only switches from a combined with low levels of the TGFβ metagene (red curve major tumor suppressor in colon cancer to neoplastic growth in Fig. 4D) correlate with prolonged patient survival. Of promoter in GBM, but also causes miR-17∼92 to follow the note, high levels of miR-18 do not confer survival advantage opposite pattern.

186 RNA, Vol. 19, No. 2 Downloaded from rnajournal.cshlp.org on September 30, 2021 - Published by Cold Spring Harbor Laboratory Press

miR-18 targets TGFβ signature in glioblastoma

MATERIALS AND METHODS prepared with oligo(dT) primer using the High Capacity cDNA RT kit (Life Technologies). 3′ UTRs were amplified according to the ′ Cell line maintenance and treatments manufacturers’ instructions using the 3 -RACE System for Rapid Amplification of cDNA Ends (Invitrogen) with the following oligo- A172, LN18, SF268, U118, U251, and U373 cell lines (provided by nucleotides: CTGF: ATGGCATGAAGCCAGAGAGT; GAPDH: Drs. Jay Dorsey and M. Celeste Simon, University of Pennsylvania) ACCGGGAAGCCCATCACC. Products were separated on a 1% aga- were maintained in DMEM supplemented with 10% fetal bovine se- rose gel. A single band was excised from the A172 CTGF sample and rum and antibiotics. DLD1 cells bearing a hypomorphic mutation in submitted for sequencing with the primer used to generate the specif- Dicer (Cummins et al. 2006) were maintained in McCoy’s supple- ic 3′ UTR (“CTGF” above) and the following oligonucleotide: CTGF, mented with 10% fetal bovine serum and antibiotics. MicroRNA 3RACE SEQ: GGAGGCATCAGTGTGTCCTTGGC. mimics were purchased from Dharmacon and transfected at a final concentration of 20 nM using Lipofectamine 2000 (Invitrogen) for DLD1 Dicer-hypomorphs or HiPerfect (QIAGEN) for GBM cell lines. Short interfering RNA (siRNA) against the human type II Microarray analysis of TGFβ target genes TGFβ receptor was purchased as a Smart Pool from Dharmacon RNA was harvested from duplicate cultures of A172 cells treated and transfected at a final concentration of 5 nM into A172 cells us- with vehicle or 5 ng/mL TGFβ1 for 6 h using TRI reagent (Sigma- ing HiPerfect (QIAGEN). TGFβ1 ligand (R&D Systems) and Aldrich). Total RNA (50 ng) from each sample was submitted Actinomycin D (Sigma-Aldrich) were used at final concentrations to the Functional Genomics Core at the University of Pennsylvania of 5 ng/mL and 5 μg/mL, respectively. Perelman School of Medicine for labeling and hybridization. Ampli- fied cDNA was prepared using the WT-Ovation Pico Amplification Western blotting System (NuGEN Technologies). Amplified cDNA (2 µg) was directly labeled using the BioPrime Array CGH Genomic Labeling Western blotting was performed as previously described (Dews et al. System (Invitrogen) with Cy3- and Cy5-labeled nucleotides (GE 2010) with the addition of the following primary antibodies: Amersham Biosciences). Dye switching was used with the replicates TGFBR1 (Santa Cruz Biotechnology) and phosphorylated Smad2 to eliminate variations introduced by dye bias. Labeled samples were (Cell Signaling). Quantitation by Odyssey Infrared Imager (LI- hybridized overnight to the Agilent Human GE 4× 44K v2 COR Biosciences) was performed as previously described (Dews Microarray. Arrays were washed and then scanned with the model et al. 2010) with the addition of CTGF primary antibody (Santa G2565B Agilent DNA microarray Scanner (Agilent Technologies). Cruz Biotechnology), incubated at 1:1000, and IRDye 800CW don- Median intensities of each element on the array were captured key anti-goat secondary antibody (LI-COR Biosciences), incubated with Agilent Feature Extraction version 9.53 (Agilent Technologies). at 1:10,000. Quality-control diagnostic plots were prepared for each array, and those failing to exhibit high-quality hybridizations were excluded Luciferase reporter constructs and sensor assays from further analysis. The data were normalized by the print tip loess method using the LIMMA (linear models for microarray Luciferase reporter plasmids were constructed and luciferase sensor data) package in R. For statistical analysis, genes were called differ- assays were performed as previously described (Dews et al. 2010). entially expressed using the Significance Analysis of Microarrays See the Supplemental Materials and Methods for sequences. (SAM) one class response package with a false discovery rate (FDR) of 20%. Genes marked as absent, i.e., with expression levels near background, were omitted. Quantitative real-time PCR analysis Total RNAs were isolated using TRI reagent (Sigma-Aldrich). cDNAs were prepared with random hexamers for mRNA or β TaqMan microRNA Assays for microRNA, using the High Analysis of microRNAs correlated with TGF Capacity cDNA RT kit (Life Technologies). For mRNA, cDNA am- signature plification reactions were performed using the PowerSYBR Green To define the TGFβ signature, genes with an adjusted P-value of PCR master mix (Life Technologies). c-Jun and c-Myc were mea- <0.01 and a fold-change of at least 1.5 were selected from the above sured using QIAGEN QuantiTect primer assays, and Tsp primers microarray analysis, resulting in 211 genes. To create a single value were obtained from the Harvard Primer Bank; see the Supplemental that represents the average expression of the TGFβ signature (TGFβ Materials and Methods for the remaining primer sequences. metagene) using primary human GBM microarray data from MicroRNA expression was analyzed and normalized to that of TCGA, genes that were shared between the Agilent Human GE 4× RNU6B using TaqMan microRNA Assays and the TaqMan Gene ex- 44K v2 Microarray and Affymetrix U133A platform were identified pression Master Mix (Life Technologies). Quantitative PCR reac- by gene symbol and probe sets mapping to the same gene symbol tions were performed on an Applied Biosystems Viia7 machine were averaged, resulting in 149 genes. For each sample, a value for and analyzed with Viia7 RUO software (Life Technologies). the TGFβ metagene was calculated by averaging the median centered and normalized expression data for the 149 genes. For TCGA 3′-RACE microRNA expression data, probes that mapped to the same microRNA were averaged. Then, SAM with a 90th percentile FDR Total RNA from untreated exponentially growing A172 or HCT116 of 0.1 was used to determine microRNAs associated with the cells was isolated using TRI reagent (Sigma-Aldrich). cDNAs were TGFβ metagene.

www.rnajournal.org 187 Downloaded from rnajournal.cshlp.org on September 30, 2021 - Published by Cold Spring Harbor Laboratory Press

Fox et al.

Gene set enrichment analysis Bartel DP. 2009. MicroRNAs: Target recognition and regulatory func- tions. Cell 136: 215–233. For gene set enrichment analysis, we used Gene Set Analysis (GSA) Baxter RC, Twigg SM. 2009. Actions of IGF binding proteins and re- implementation. Here, the average expression value of miR-18a and lated proteins in adipose tissue. Trends Endocrinol Metab 20: 499– miR-18b for each sample was used as a quantitative response vari- 505. Bierie B, Moses HL. 2006. Tumour microenvironment: TGFβ: The mo- able, and a gene set list was constructed using the 149 mapped genes – β lecular Jekyll and Hyde of cancer. Nat Rev Cancer 6: 506 520. in the TGF signature and 99 randomly selected gene sets of the Cancer Genome Atlas Research Network. 2008. Comprehensive geno- same size. Restandardization was performed using the entire expres- mic characterization defines human glioblastoma genes and core sion data, and the maxmean statistic was used. The significance of pathways. Nature 455: 1061–1068. the enrichment for the TGFβ signature was calculated based on Chang CC, Shih JY, Jeng YM, Su JL, Lin BZ, Chen ST, Chau YP, 1000 permutations. Yang PC, Kuo ML. 2004. Connective tissue growth factor and its role in lung adenocarcinoma invasion and metastasis. J Natl Cancer Inst 96: 364–375. Chang CC, Lin MT, Lin BR, Jeng YM, Chen ST, Chu CY, Chen RJ, Kaplan-Meier survival analysis Chang KJ, Yang PC, Kuo ML. 2006. Effect of connective tissue α β growth factor on hypoxia-inducible factor 1 degradation and tu- The association of miR-18 and TGF signature or CTGF expression mor angiogenesis. J Natl Cancer Inst 98: 984–995. with clinical outcome for TCGA GBM samples was estimated by Chayka O, Corvetta D, Dews M, Caccamo AE, Piotrowska I, Santilli G, Kaplan-Meier survival analysis. Each patient was stratified based Gibson S, Sebire NJ, Himoudi N, Hogarty MD, et al. 2009. Clusterin, on miR-18, which represented the average expression of miR-18a a haploinsufficient tumor suppressor gene in neuroblastomas. J Natl and miR-18b, and the value of either the TGFβ metagene or Cancer Inst 101: 663–677. CTGF expression. For both variables, a mean cut-point was used Chen AJ, Paik JH, Zhang H, Shukla SA, Mortensen R, Hu J, Ying H, Hu B, Hurt J, Farny N, et al. 2012. STAR RNA-binding protein to define low versus high expression. The four groups resulting Quaking suppresses cancer via stabilization of specific miRNA. from stratification using these two binary factors were then com- Genes Dev 26: 1459–1472. pared for differences in overall survival. Significance was calculated Chung EY, Dews M, Maity A, Thomas-Tikhonenko A. 2007. Aiding using the log-rank test. and ABT’ing treatment for glioblastoma. Cancer Biol Ther 6: 802– 804. Conkrite K, Sundby M, Mukai S, Thomson JM, Mu D, Hammond SM, ∼ SUPPLEMENTAL MATERIAL MacPherson D. 2011. miR-17 92 cooperates with RB pathway mu- tations to promote retinoblastoma. Genes Dev 25: 1734–1745. Supplemental material is available for this article. Crawford SE, Stellmach V, Murphy-Ullrich JE, Ribeiro SM, Lawler J, Hynes RO, Boivin GP, Bouck N. 1998. Thrombospondin-1 is a ma- jor activator of TGF-β1 in vivo. Cell 93: 1159–1170. Cummins JM, He Y, Leary RJ, Pagliarini R, Diaz LA Jr, Sjoblom T, ACKNOWLEDGMENTS Barad O, Bentwich Z, Szafranska AE, Labourier E, et al. 2006. The – We thank Dr. Elena Sotillo and Dr. Prema Sundaram (Children’s colorectal microRNAome. Proc Natl Acad Sci 103: 3687 3692. ′ Deli A, Kreidl E, Santifaller S, Trotter B, Seir K, Berger W, Schulte- Hospital of Philadelphia) for their advice concerning 3 -RACE and Hermann R, Rodgarkia-Dara C, Grusch M. 2008. Activins and acti- hnRNA analysis, respectively. We are grateful to Dr. Jonathan vin antagonists in hepatocellular carcinoma. World J Gastroenterol Schug and Alan Fox (University of Pennsylvania) for their help 14: 1699–1709. with performing and interpreting microarray analysis. Current de Winter P, Leoni P, Abraham D. 2008. Connective tissue growth fac- members of our laboratory, in particular Dr. Elena Sotillo, Dr. tor: Structure–function relationships of a mosaic, multifunctional – James Psathas, and Rebecca Rivard, are acknowledged for many protein. Growth Factors 26: 80 91. Demuth T, Rennert JL, Hoelzinger DB, Reavie LB, Nakada M, Beaudry stimulating discussions. J.L.F. thanks her thesis committee members, C, Nakada S, Anderson EM, Henrichs AN, McDonough WS, et al. Dr. Nancy Speck, Dr. Mitchell Weiss, Dr. Xianxin Hua, and Dr. 2008. Glioma cells on the run—the migratory transcriptome of 10 Warren Pear (University of Pennsylvania) for guidance and encour- human glioma cell lines. BMC Genomics 9: 54. agement. This work was supported by NIH grant R01 CA122334 (to Dews M, Homayouni A, Yu D, Murphy D, Sevignani C, Wentzel E, A.T.T.) and a Cancer Research Institute Immunobiology training Furth EE, Lee WM, Enders GH, Mendell JT, et al. 2006. grant (to J.L.F.). The Functional Genomics Core at the University Augmentation of tumor angiogenesis by a Myc-activated – of Pennsylvania Diabetes Research Center (DRC) was supported microRNA cluster. Nat Genet 38: 1060 1065. Dews M, Fox JL, Hultine S, Sundaram P, Wang W, Liu YY, Furth E, by MIH grant P30-DK19525. Enders GH, El-Deiry W, Schelter JM, et al. 2010. The myc-miR- 17∼92 axis blunts TGFβ signaling and production of multiple Received September 20, 2012; accepted November 12, 2012. TGFβ-dependent antiangiogenic factors. Cancer Res 70: 8233–8246. Doebele C, Bonauer A, Fischer A, Scholz A, Reiss Y, Urbich C, Hofmann WK, Zeiher AM, Dimmeler S. 2010. Members of the REFERENCES microRNA-17-92 cluster exhibit a cell-intrinsic antiangiogenic func- tion in endothelial cells. Blood 115: 4944–4950. Abreu JG, Ketpura NI, Reversade B, De Robertis EM. 2002. Connective- Ernst A, Campos B, Meier J, Devens F, Liesenberg F, Wolter M, tissue growth factor (CTGF) modulates cell signalling by BMP and Reifenberger G, Herold-Mende C, Lichter P, Radlwimmer B. 2010. TGF-β. Nat Cell Biol 4: 599–604. De-repression of CTGF via the miR-17-92 cluster upon differentia- Amagasaki K, Sasaki A, Kato G, Maeda S, Nukui H, Naganuma H. 2001. tion of human glioblastoma spheroid cultures. Oncogene 29: Antisense-mediated reduction in thrombospondin-1 expression re- 3411–3422. duces cell motility in malignant glioma cells. Int J Cancer 94: 508–512. Felli N, Fontana L, Pelosi E, Botta R, Bonci D, Facchiano F, Liuzzi F, Baek D, Villen J, Shin C, Camargo FD, Gygi SP, Bartel DP. 2008. The Lulli V, Morsilli O, Santoro S, et al. 2005. MicroRNAs 221 and impact of microRNAs on protein output. Nature 455: 64–71. 222 inhibit normal erythropoiesis and erythroleukemic cell growth

188 RNA, Vol. 19, No. 2 Downloaded from rnajournal.cshlp.org on September 30, 2021 - Published by Cold Spring Harbor Laboratory Press

miR-18 targets TGFβ signature in glioblastoma

via kit receptor down-modulation. Proc Natl Acad Sci 102: 18081– Li J, Yen C, Liaw D, Podsypanina K, Bose S, Wang SI, Puc J, Miliaresis C, 18086. Rodgers L, McCombie R, et al. 1997. PTEN, a putative protein tyro- Fontana L, Fiori ME, Albini S, Cifaldi L, Giovinazzi S, Forloni M, sine phosphatase gene mutated in human brain, breast, and prostate Boldrini R, Donfrancesco A, Federici V, Giacomini P, et al. 2008. cancer. Science 275: 1943–1947. Antagomir-17-5p abolishes the growth of therapy-resistant neuro- Massague J. 2008. TGFβ in cancer. Cell 134: 215–230. blastoma through p21 and BIM. PLoS One 3: e2236. Mendell JT. 2008. miRiad roles for the miR-17-92 cluster in develop- Galardi S, Mercatelli N, Giorda E, Massalini S, Frajese GV, Ciafre SA, ment and disease. Cell 133: 217–222. Farace MG. 2007. miR-221 and miR-222 expression affects the pro- Mestdagh P, Boström AK, Impens F, Fredlund E, Van Peer G, De liferation potential of human prostate carcinoma cell lines by target- Antonellis P, von Stedingk SK, Ghesquière B, Schulte S, Dews M, ing p27Kip1. J Biol Chem 282: 23716–23724. et al. 2010. The miR-17-92 microRNA cluster regulates multiple Giraldez AJ, Mishima Y, Rihel J, Grocock RJ, Van DS, Inoue K, components of the TGF-β pathway in neuroblastoma. Mol Cell 40: Enright AJ, Schier AF. 2006. Zebrafish MiR-430 promotes deadeny- 762–773. lation and clearance of maternal mRNAs. Science 312: 75–79. Mu P, Han YC, Betel D, Yao E, Squatrito M, Ogrodowski P, de Good DJ, Polverini PJ, Rastinejad F, Le Beau MM, Lemons RS, Stanchina E, D’Andrea A, Sander C, Ventura A. 2009. Genetic dis- Frazier WA, Bouck NP. 1990. A tumor suppressor-dependent inhib- section of the miR-17∼92 cluster of microRNAs in Myc-induced itor of angiogenesis is immunologically and functionally indistin- B-cell lymphomas. Genes Dev 23: 2806–2811. guishable from a fragment of thrombospondin. Proc Natl Acad Sci O’Donnell KA, Wentzel EA, Zeller KI, Dang CV, Mendell JT. 2005. c- 87: 6624–6628. Myc-regulated microRNAs modulate E2F1 expression. Nature 435: Greshock J, Bachman KE, Degenhardt YY, Jing J, Wen YH, Eastman S, 839–843. McNeil E, Moy C, Wegrzyn R, Auger K, et al. 2010. Molecular target Ohgawara T, Kubota S, Kawaki H, Kondo S, Eguchi T, Kurio N, class is predictive of in vitro response profile. Cancer Res 70: Aoyama E, Sasaki A, Takigawa M. 2009. Regulation of chondrocytic 3677–3686. phenotype by micro RNA 18a: Involvement of Ccn2/Ctgf as a major Grotendorst GR. 1997. Connective tissue growth factor: A mediator target gene. FEBS Lett 583: 1006–1010. of TGF-β action on fibroblasts. Cytokine Growth Factor Rev 8: Olive V, Bennett MJ, Walker JC, Ma C, Jiang I, Cordon-Cardo C, Li QJ, 171–179. Lowe SW, Hannon GJ, He L. 2009. miR-19 is a key oncogenic com- Hall-Glenn F, De Young RA, Huang BL, van Handel B, Hofmann JJ, ponent of mir-17-92. Genes Dev 23: 2839–2849. Chen TT, Choi A, Ong JR, Benya PD, Mikkola H, et al. 2012. Ota A, Tagawa H, Karnan S, Tsuzuki S, Karpas A, Kira S, Yoshida Y, CCN2/connective tissue growth factor is essential for pericyte adhe- Seto M. 2004. Identification and characterization of a novel gene, sion and endothelial basement membrane formation during angio- C13orf25, as a target for 13q31-q32 amplification in malignant lym- genesis. PLoS One 7: e30562. phoma. Cancer Res 64: 3087–3095. Hayashita Y, Osada H, Tatematsu Y, Yamada H, Yanagisawa K, Perbal B. 2004. CCN proteins: Multifunctional signalling regulators. Tomida S, Yatabe Y, Kawahara K, Sekido Y, Takahashi T. 2005. A Lancet 363: 62–64. polycistronic microRNA cluster, miR-17-92, is overexpressed in hu- Petrocca F, Visone R, Onelli MR, Shah MH, Nicoloso MS, de Martino I, man lung cancers and enhances cell proliferation. Cancer Res 65: Iliopoulos D, Pilozzi E, Liu CG, Negrini M, et al. 2008. E2F1-regu- 9628–9632. lated microRNAs impair TGFβ-dependent cell-cycle arrest and apo- He L, Thomson JM, Hemann MT, Hernando-Monge E, Mu D, ptosis in gastric cancer. Cancer Cell 13: 272–286. Goodson S, Powers S, Cordon-Cardo C, Lowe SW, Hannon GJ, et Rigoutsos I, Huynh T, Miranda K, Tsirigos A, McHardy A, Platt D. 2006. al. 2005. A microRNA polycistron as a potential human oncogene. Short blocks from the noncoding parts of the have Nature 435: 828–833. instances within nearly all known genes and relate to biological pro- Inoki I, Shiomi T, Hashimoto G, Enomoto H, Nakamura H, Makino K, cesses. Proc Natl Acad Sci 103: 6605–6610. Ikeda E, Takata S, Kobayashi K, Okada Y. 2002. Connective tissue Sala-Torra O, Gundacker HM, Stirewalt DL, Ladne PA, Pogosova- growth factor binds vascular endothelial growth factor (VEGF) Agadjanyan EL, Slovak ML, Willman CL, Heimfeld S, Boldt DH, and inhibits VEGF-induced angiogenesis. FASEB J 16: 219–221. Radich JP. 2007. Connective tissue growth factor (CTGF) expression Jimenez B, Volpert OV, Crawford SE, Febbraio M, Silverstein RL, and outcome in adult patients with acute lymphoblastic leukemia. Bouck N. 2000. Signals leading to apoptosis-dependent inhibition Blood 109: 3080–3083. of neovascularization by thrombospondin-1. Nat Med 6: 41–48. Selbach M, Schwanhausser B, Thierfelder N, Fang Z, Khanin R, Kang Y, Siegel PM, Shu W, Drobnjak M, Kakonen SM, Cordon- Rajewsky N. 2008. Widespread changes in protein synthesis induced Cardo C, Guise TA, Massague J. 2003. A multigenic program medi- by microRNAs. Nature 455: 58–63. ating breast cancer metastasis to bone. Cancer Cell 3: 537–549. Sundaram P, Hultine S, Smith LM, Dews M, Fox JL, Biyashev D, Kang Y, He W, Tulley S, Gupta GP, Serganova I, Chen CR, Manova- Schelter JM, Huang Q, Cleary MA, Volpert OV, et al. 2011. p53-re- Todorova K, Blasberg R, Gerald WL, Massague J. 2005. Breast cancer sponsive miR-194 inhibits thrombospondin-1 and promotes angio- bone metastasis mediated by the Smad tumor suppressor pathway. genesis in colon cancers. Cancer Res 71: 7490–7501. Proc Natl Acad Sci 102: 13909–13914. van Almen GC, Verhesen W, van Leeuwen RE, van de Vrie M, Karginov FV, Conaco C, Xuan Z, Schmidt BH, Parker JS, Mandel G, Eurlings C, Schellings MW, Swinnen M, Cleutjens JP, van Hannon GJ. 2007. A biochemical approach to identifying Zandvoort MA, Heymans S, et al. 2011. MicroRNA-18 and microRNA targets. Proc Natl Acad Sci 104: 19291–19296. microRNA-19 regulate CTGF and TSP-1 expression in age-related Kawataki T, Naganuma H, Sasaki A, Yoshikawa H, Tasaka K, Nukui H. heart failure. Aging Cell 10: 769–779. 2000. Correlation of thrombospondin-1 and transforming growth Verhaak RG, Hoadley KA, Purdom E, Wang V, Qi Y, Wilkerson MD, factor-β expression with malignancy of glioma. Neuropathology 20: Miller CR, Ding L, Golub T, Mesirov JP, et al. 2010. Integrated geno- 161–169. mic analysis identifies clinically relevant subtypes of glioblastoma le Sage C, Nagel R, Egan DA, Schrier M, Mesman E, Mangiola A, characterized by abnormalities in PDGFRA, IDH1, EGFR, and Anile C, Maira G, Mercatelli N, Ciafre SA, et al. 2007. Regulation NF1. Cancer Cell 17: 98–110. of the p27Kip1 tumor suppressor by miR-221 and miR-222 promotes Volinia S, Calin GA, Liu CG, Ambs S, Cimmino A, Petrocca F, Visone R, cancer cell proliferation. EMBO J 26: 3699–3708. Iorio M, Roldo C, Ferracin M, et al. 2006. A microRNA expression Leask A, Abraham DJ. 2004. TGF-β signaling and the fibrotic response. signature of human solid tumors defines cancer gene targets. Proc FASEB J 18: 816–827. Natl Acad Sci 103: 2257–2261. Lewis BP, Burge CB, Bartel DP. 2005. Conserved seed pairing, often Weinstat-Saslow DL, Zabrenetzky VS, VanHoutte K, Frazier WA, flanked by adenosines, indicates that thousands of human genes Roberts DD, Steeg PS. 1994. Transfection of thrombospondin 1 are microRNA targets. Cell 120: 15–20. complementary DNA into a human breast carcinoma cell line

www.rnajournal.org 189 Downloaded from rnajournal.cshlp.org on September 30, 2021 - Published by Cold Spring Harbor Laboratory Press

Fox et al.

reduces primary tumor growth, metastatic potential, and angiogen- associated with oncogenic activities and drug resistance in glioblas- esis. Cancer Res 54: 6504–6511. toma multiforme. Int J Cancer 127: 2257–2267. Wu L, Belasco JG. 2005. Micro-RNA regulation of the mammalian lin- Zaslavsky A, Baek KH, Lynch RC, Short S, Grillo J, Folkman J, 28 gene during neuronal differentiation of embryonal carcinoma Italiano JE Jr, Ryeom S. 2010. Platelet-derived thrombospondin-1 cells. Mol Cell Biol 25: 9198–9208. is a critical negative regulator and potential biomarker of angiogen- Xie D, Yin D, Wang HJ, Liu GT, Elashoff R, Black K, Koeffler HP. 2004. esis. Blood 115: 4605–4613. Levels of expression of CYR61 and CTGF are prognostic for tumor Zhang M, Kleber S, Rohrich M, Timke C, Han N, Tuettenberg J, Martin- progression and survival of individuals with gliomas. Clin Cancer Res Villalba A, Debus J, Peschke P, Wirkner U, et al. 2011. Blockade of 10: 2072–2081. TGF-β signaling by the TGFβR-I kinase inhibitor LY2109761 en- Yin D, Chen W, O’Kelly J, Lu D, Ham M, Doan NB, Xie D, Wang C, hances radiation response and prolongs survival in glioblastoma. Vadgama J, Said JW, et al. 2010. Connective tissue growth factor Cancer Res 71: 7155–7167.

190 RNA, Vol. 19, No. 2 Downloaded from rnajournal.cshlp.org on September 30, 2021 - Published by Cold Spring Harbor Laboratory Press

Targeting of TGFβ signature and its essential component CTGF by miR-18 correlates with improved survival in glioblastoma

Jamie L. Fox, Michael Dews, Andy J. Minn, et al.

RNA 2013 19: 177-190 originally published online December 18, 2012 Access the most recent version at doi:10.1261/rna.036467.112

Supplemental http://rnajournal.cshlp.org/content/suppl/2012/12/03/rna.036467.112.DC1 Material

References This article cites 65 articles, 26 of which can be accessed free at: http://rnajournal.cshlp.org/content/19/2/177.full.html#ref-list-1

License

Email Alerting Receive free email alerts when new articles cite this article - sign up in the box at the Service top right corner of the article or click here.

To subscribe to RNA go to: http://rnajournal.cshlp.org/subscriptions

Copyright © 2013 RNA Society