Manipulation of PK-M mutually exclusive alternative splicing by antisense oligonucleotides

Alternative splicing of the pyruvate kinase M gene involves a choice between mutually exclusive exons 9 and 10. Use of exon 10 to generate the M2 isoform is crucial for aerobic glycolysis (the Warburg effect) and tumour growth. We previously demonstrated that splicing enhancer elements that activate exon 10 are mainly found in exon 10 itself, and deleting or mutating these elements increases the inclusion of exon 9 in cancer cells. To systematically search for new enhancer elements in exon 10 and develop an effective pharmacological method to force a switch from PK-M2 to PK-M1, we carried out an antisense oligonucleotide (ASO) screen. We found potent ASOs that target a novel enhancer in exon 10 and strongly switch the splicing of endogenous PK-M transcripts to include exon 9. We further show that the ASO-mediated switch in alternative splicing leads to apoptosis in glioblastoma cell lines, and this is caused by the downregulation of PK-M2, and not by the upregulation of PK-M1. These data highlight the potential of ASO-mediated inhibition of PK-M2 splicing as therapy for cancer.


Summary
Alternative splicing of the pyruvate kinase M gene involves a choice between mutually exclusive exons 9 and 10. Use of exon 10 to generate the M2 isoform is crucial for aerobic glycolysis (the Warburg effect) and tumour growth. We previously demonstrated that splicing enhancer elements that activate exon 10 are mainly found in exon 10 itself, and deleting or mutating these elements increases the inclusion of exon 9 in cancer cells. To systematically search for new enhancer elements in exon 10 and develop an effective pharmacological method to force a switch from PK-M2 to PK-M1, we carried out an antisense oligonucleotide (ASO) screen. We found potent ASOs that target a novel enhancer in exon 10 and strongly switch the splicing of endogenous PK-M transcripts to include exon 9. We further show that the ASO-mediated switch in alternative splicing leads to apoptosis in glioblastoma cell lines, and this is caused by the downregulation of PK-M2, and not by the upregulation of PK-M1. These data highlight the potential of ASO-mediated inhibition of PK-M2 splicing as therapy for cancer.

Introduction
Cancer cells preferentially use the glycolytic metabolic pathway with lactate generation, even under normal oxygen conditions [1]. This metabolic feature is termed the Warburg effect. Expression of the type II isoform of the pyruvate kinase M gene (PKM2, referred to here as PK-M) has been shown to mediate this effect, and to facilitate the proliferation of cancer cells in vivo [2].
The PK-M gene consists of 12 exons; exons 9 and 10 are alternatively spliced in a mutually exclusive fashion to give rise to M1 and M2 isoforms, respectively [3]. PK-M catalyses the final step in glycolysis to generate pyruvate and ATP from phosphoenolpyruvate and ADP [4]. Exons 9 and 10 each encode a 56-amino acid segment that confers distinctive properties to the respective PK-M isozymes. PK-M1 is constitutively active, whereas PK-M2 is allosterically regulated by fructose-1,6-bisphosphate levels and interaction with tyrosine-phosphorylated signalling proteins [5].
Consistent with the correlation between proliferation and PK-M2 expression, PK-M2 is highly expressed in embryonic tissues and in a broad range of cancer cells, whereas PK-M1 is predominantly expressed in terminally differentiated tissues [2,6]. In particular, the mammalian target of rapamycin pathway, which is a central mediator of cellular growth and proliferation, transcriptionally induces PK-M expression through the transcription factor hypoxia inducible factor-1a (HIF-1a) [7].
Paradoxically, downregulation of PK-M2 kinase activity is required for cancer cell growth. As PK-M2 is a rate-limiting enzyme in glycolysis, inhibition of PK-M activity decreases carbon flux through the catabolic glycolytic pathway, allowing upstream intermediates to be shunted to anabolic pathways, and thereby facilitating proliferation [2] and detoxification of reactive oxygen species (ROS) [8]. Apart from allosteric regulation, PK-M2 kinase activity can be suppressed in other ways, including growth-signalling-mediated inhibition through the binding of phosphorylated tyrosine proteins to its allosteric pocket [5], direct phosphorylation at Y105 [9], acetylation at K305 [10] and oxidation at C358 via ROS [8]. The multitude of avenues leading to PK-M2 inhibition underlies the importance of PK-M2 regulation and glycolysis in tumorigenesis.
PK-M2 has also been demonstrated to have critical functions beyond its kinase activity. In particular, PK-M2 can translocate into the nucleus and serve as a transcriptional co-activator for HIF-1a [11] and b-catenin [12] in mediating transactivation of targets important for tumour growth and proliferation. In spite of the importance of PK-M2 in tumorigenesis, little is known regarding alternative splicing of the PK-M pre-mRNA that predominantly generates the PK-M2 isoform in all cancer cells.
We and others previously showed that exon 9 inclusion, which generates the PK-M1 isoform, is actively repressed in cancer cells via the well-characterized PTB/nPTB and hnRNPA1/A2 splicing repressors [6,13]. We further showed that critical cis-elements that mediate the predominant PK-M2 splicing pattern in cancer cells are located in the exons, and we identified a potent exon-10 exonic splicing enhancer (ESE) that recruits the splicing activator SRSF3, promoting use of this exon [14].
Because of the importance of PK-M2 as a potential cancer drug target, and in order to systematically search for new splicing elements that mediate the PK-M2 dominant splicing patterns, we conducted an antisense oligonucleotide (ASO) screen targeting endogenous PK-M exon 10 to discover ASOs that switch the expression of the PK-M2 isoform to the PK-M1 isoform in cancer cells.
ASOs have generally been used to downregulate mRNA expression by homing to target mRNAs via Watson-Crick base pairing and inducing ribonuclease H (RNase H)-mediated degradation of the RNA [15]. Certain chemical modifications have been used to generate a class of nuclease-resistant ASOs with high affinity for their RNA targets and which do not operate through the RNase H mechanism. ASOs have been engineered with a wide range of chemical modifications, such as 2 0 -Omethyl (2 0 OMe), 2-O-methoxyethyl (MOE), peptide nucleic acid, locked nucleic acid and phosphorodiamidate morpholino. These ASOs work by direct sequence-specific annealing to the target transcript to block either splicing cis-elements, or ribosome recruitment so as to inhibit translation [16,17].
Our splicing ASO screen uncovered potent oligonucleotides that increase PK-M1 inclusion in cancer cell lines. We show that the effective ASOs principally target a new ESE in exon 10, and transfection of these ASOs into glioblastoma cell lines induces apoptosis. We demonstrate that this phenotype is caused by the ASO-mediated downregulation of PK-M2 expression, highlighting the therapeutic potential of the PK-M pre-mRNA as an ASO drug target.

3.
1. An exon 10-focused antisense oligonucleotide screen uncovers potent oligonucleotides that increase exon 9 inclusion and exon 10 skipping in cell lines We previously reported that the critical cis-elements involved in the activation of exon 10 in proliferating cells are located within the exon [14]. To pinpoint the locations of these elements, we performed a systematic ASO walk along the entire length of PK-M exon 10. These ASOs have a phosphorothioate backbone and MOE at the 2 0 ribose position, which allows them to bind to RNA targets with high affinity, while conferring resistance to both endogenous nucleases and to cleavage of the target RNA by RNase H [18,19]. The overlapping 15-mer ASOs were designed to cover the entire 167 nt exon 10 in 5 nt steps.
To examine the effects of individual ASOs on endogenous PK-M transcripts, we transfected each ASO into HEK-293 cells at a nominal final concentration of 30 nM, and analysed the splicing of the endogenous PK-M transcripts by radioactive RT-PCR, 48 h after transfection (figure 1a). Some exon 10 ASOs strongly increased the proportion of PK-M1 mRNA, with a concurrent increase in the amount of doubleskipped mRNA ( joining of exons 8 and 11) and a decrease in PK-M2 mRNA, suggesting that these ASOs target functional ESEs in exon 10. The two most potent ASOs that decreased the proportion of PK-M2 mRNA were 10W 45 -59 (figure 1c, lane 10) and 10W 140 -154 (see the electronic supplementary material, figure S1b, lane 10). ASO 10W 140 -154 targets the previously characterized exon 10 SRSF3 motif [14], whereas ASO 10W 45 -59 targets a non-overlapping 15 nt region in the middle of exon 10 (termed the 10W region).

Antisense oligonucleotide microwalks centred on the 10W exonic splicing enhancer region
To find optimal ASOs that target the exon 10 region defined by ASO 10W 45 -59 , we performed an ASO microwalk (figure 1b,d).
Overlapping 15 nt ASOs targeting this region were designed in 1 nt steps. A total of 12 ASOs were synthesized for this region, and transfected into HEK-293 cells at a nominal concentration of 60 nM. ASO 10M 46 -60 was the most potent in increasing endogenous PK-M1 mRNA (figure 1d, lane 16) and decreasing PK-M2 mRNA abundance. We also performed a second microwalk centred on the region defined by ASO 10W 140 -154 , and found that the ASO 10MS 139 -153 optimally abrogated the SRSF3-dependent ESE in exon 10 (see the electronic supplementary material, figure S1c, lane 10).
3.3. Antisense oligonucleotides 10W 45 -59 and 10M 46 -60 target a novel activation region of PK-M exon 10 Because of the strong effect of ASOs 10W 45 -59 and 10M 46 -60 on PK-M splicing, we characterized the 15 nt 10W candidate ESE region in detail. To map the enhancer elements present in the 10W 45 -59 region, we took advantage of the high sequence identity between exons 9 and 10, and individually duplicated the entire 10W region of exon 10, or the first 7 nt (F7) or last 8 nt (B7) of the 10W region, into their corresponding exon 9 locations (figure 2a  rsob.royalsocietypublishing.org Open Biol 2: 120133   Figure 2. Characterizing the 10WASO target region. (a) Scheme of method used to duplicate the exon 10 10W region into exon 9 in a minigene. The minigene comprises the same genomic region as indicated in figure 1a. To amplify minigene transcripts, we used a primer annealing to a vector-specific sequence upstream of the genomic insert, pcDNAF [14]. Minigene mutant names are indicated below. The indicated exon 9 (green) nucleotides at the top were mutated to the corresponding exon 10 (red) sequences on the right. The 10W minigene duplicates the entire exon 10 10W region into exon 9; the 10F minigene duplicates the first 8 nt of 10W45-59; and the 10B minigene duplicates the last 7 nt of 10W 45  Standard deviations for ( f ) are 1.2%, 1.6% and 1.5%; for (g) they are 0.4%, 2.0%, 3.9%; and for (h) they are 0.2%, 0.6% and 0.6%, corresponding to control, 10W 45 -59 and 10M 46 -60 ASOs, respectively; n ¼ 3.
rsob.royalsocietypublishing.org Open Biol 2: 120133 wild-type minigene, any strong ESEs comprised by the candidate regions should lead to an increase in PK-M1 mRNAs expressed from the minigene. Duplication of the B7 region, but not the F7 and 10W region, led to increased exon 9 inclusion (figure 2b, compare lanes 2 and 4 with lane 6). This result suggests that the 8 nt B7 motif is a bona fide exon 10 ESE. Interestingly, duplication of F7 alone led to a further decrease in exon 9 inclusion, compared with the wild-type minigene (figure 1c, lanes 1 and 2), suggesting that F7 is an exon 10 splicing silencer that also represses the inclusion of exon 9 when placed in the context of the 10 W minigene.
3.4. Antisense oligonucleotides 10W 45 -59 and 10M 46 -60 interfere with exon 10 definition To characterize the mechanism of action of ASOs 10W 45 -59 and 10M 46 -60 on the inclusion of exon 9 and skipping of exon 10, we co-transfected these ASOs with the previously characterized PK-M wild-type or duplicated exon 10 minigenes ( figure 2c,d). The wild-type minigene comprises the flanking exons 8 and 11, and the complete genomic region between both exons, whereas the duplication construct has exon 10 cleanly replacing exon 9.
As expected, both ASOs 10W 45 -59 and 10M 46 -60 switched the splicing of the minigene transcript by simultaneously increasing the amount of M1 mRNA and decreasing the amount of M2 mRNA expressed from the wild-type minigene (figure 2c). However, ASO 10W 45 -59 increased exon 9 inclusion to a greater extent than 10M 46 -60 , although the latter decreased exon 10 inclusion to a greater extent, resulting in higher levels of double-skipped (Skp) transcripts (figure 2c, compare lanes 4 and 6).
Co-transfection of ASOs 10W 45 -59 or 10M 46 -60 with the exon 10 duplication minigene interfered with the inclusion of exon 10, leading to a large increase in double-skipped species (figure 2d). ASO 10M 46 -60 was especially potent, nearly converting all the mRNA to the Skp isoform (figure 2d, compare lanes 4 and 6). These results suggest that both ASO 10W 45 -59 and ASO 10M 46 -60 interfere with the activation of exon 10.
3.5. The 10W region of exon 10 is the critical binding site for antisense oligonucleotides 10W 45 -59 and 10M 46 -60 Alignment of the 10W region with the PK-M exons 8-11 genomic region revealed a highly homologous region in intron 9 (figure 2e). To weigh the relative contributions of the exon 10 and intron 9 complementary regions for the effect of ASOs 10W 45 -59 and 10M 46 -60 on PK-M splicing, we made minigene mutations that eliminated the presumptive target sites in exon 10, intron 9 or both, and determined the effect of the ASOs on splicing of the mutant-minigene transcripts. Three mutants were generated (figure 2e): (i) we mutated the exon 10 10W region by duplicating the corresponding exon 9 region (termed the d10W construct); (ii) we introduced a 15 nt deletion in intron 9 that removed the homologous target region (termed the dInt9 construct); and (iii) we introduced both mutations in the same minigene (termed the dM construct).
There was a slight decrease in baseline minigene PK-M2 mRNA expressed from the d10W minigene (figure 2f, lane 2), suggesting that the duplicated exon 9 region contains weak repressor elements. This was not the case for the dInt9 construct, suggesting that this region alone does not have a major effect in dictating M1/M2 ratios.
The loss of the 10W binding site largely abrogated exon 9 inclusion and exon 10 skipping promoted by ASOs 10W 45 -59 and 10M 46 -60 ( figure 2f, lanes 4 and 6). By contrast, removal of the intron 9 homologous region did not block the effect of ASOs 10W 45 -59 and 10M 46 -60 on splicing (figure 2g, lanes 4 and 6). However, when both binding sites were removed, the effect of ASOs 10W 45 -59 and 10M 46 -60 was completely abrogated ( figure 2h, lanes 4 and 6). The above results indicate that ASOs 10W 45 -59 and 10M 46 -60 largely mediate exon 9 inclusion through the 10W complementary region in exon 10, though a small effect is attributable to cross-hybridization with the homologous region in intron 9.
3.6. Exon 10 antisense oligonucleotides switch PK-M RNA and protein expression in glioblastoma cells in a dose-dependent manner To objectively compare the performance of ASOs targeting the 10W region versus those targeting the previously characterized SRSF3 region, we did side-by-side ASO transfections at different doses, with final concentrations of 30, 60 or 90 nM (figure 3a,c) in the glioblastoma cell lines A172 (figure 3a) and U87-MG (figure 3c). We chose to use glioblastoma cells because: (i) a characteristic splicing switch from PK-M1 to PK-M2 occurs during gliomagenesis [6,20]; and (ii) glioblastoma cell lines have a higher basal level of PK-M1 mRNA, which we expected to facilitate the ASO-mediated PK-M splicing switch [6]. As expected, there was a dose-dependent increase in exon 9 inclusion and exon 10 skipping in these cell lines, with ASOs 10W 45 -59 and 10M 46 -60 performing better than ASO 10MS 139 -153 (figure 3a,c, compare lanes 3-14 with lanes [15][16][17][18]. Consistent with the minigene experiments, ASO 10W 45 -59 gave rise to more PK-M1 mRNA (figure 3a,c, lanes 4-8), whereas 10M 46 -60 gave rise to more double-skipped mRNA (figure 3a,c, lanes 9-14) and a larger decrease in PK-M2 mRNA. In addition, there was generally a larger switch in PK-M alternative splicing in these cells, compared with HEK-293 cells.
Using isoform-specific antibodies, we measured the amount of PK-M1 and PK-M2 proteins in the same lysates as above (figure 3b,d). As expected, PK-M1 and PK-M2 protein isoform levels closely mirrored their mRNA levels. There was detectable PK-M1 protein after transfection of each of the three ASOs, but ASO 10M 46 -60 resulted in the greatest decrease in PK-M2 levels. Downregulation of PK-M2 protein was also observed by indirect immunofluorescence staining with the PK-M2 antibody when either ASO 10W 45 -59 or 10M 46 -60 , but not the control ASO, was transfected into A172 or U87-MG cell lines (see the electronic supplementary material, figure S2).  4a). As expected, the proportion of Annexin-V-positive cells increased in an ASO dose-dependent manner, indicating that ASO-mediated switching of PK-M splicing induces apoptosis in these cell lines. ASO 10M 46 -60 was the most potent in inducing apoptosis among the three ASOs tested.

Antisense oligonucleotide-mediated apoptosis in glioblastoma cells is caused by the downregulation of endogenous PK-M2 expression
ASO transfection elicited a simultaneous increase in PK-M1 expression and a decrease in PK-M2 expression. To pinpoint which of these changes is responsible for inducing apoptosis in glioblastoma cells, we generated stable cell lines that express human PK-M1 cDNA in a doxycyclineinducible manner (figure 4c), or PK-M2 cDNA constitutively (figure 4e). To investigate the role of PK-M1 in ASO-mediated apoptosis, we added doxycyline to the PK-M1 inducible cells for three days, and then treated them with ASO 10W 45 -59 , 10M 46 -60 or control ASO. There was a similar increase in the number of Annexin-V-positive cells in the cells that did or did not overexpress PK-M1, suggesting that PK-M1 induction did not cause apoptosis in these cells (figure 4d).
To investigate the role of PK-M2 downregulation in apoptosis, we overexpressed PK-M2 in U87-MG and A172 cells, and then treated them with the maximum dose of ASO -P -P -P -P -P -P -P -P -P -P -P -P -P -P -P -P -P -P enz: ASO:  expression, but not the induction of PK-M1 expression, leads to apoptosis in these cell lines.

Discussion
We performed a systematic exon 10 ASO screen, and developed potent ASOs that increase the inclusion of exon 9 and skipping of exon 10 in transfected cell lines. We found that these ASOs switch PK-M splicing especially effectively in glioblastoma cells, and they induce apoptosis. Finally, we showed that it is the downregulation of PK-M2 expression that causes this apoptosis. Glioblastomas exhibit a high rate of glycolysis, even under normal oxygen conditions [6]. Consistent with this phenomenon, the splicing profile of glioblastoma cell lines, as with other cancer cell lines, is predominantly PK-M2. Because PK-M2 is a rate-limiting glycolytic enzyme, targeting    rsob.royalsocietypublishing.org Open Biol 2: 120133 this isoform might inhibit the entire glycolytic pathway, leading to apoptosis, as previously observed by glucose withdrawal in glioblastoma cells in vitro [21]. The replacement of PK-M2 in tumour cells with PK-M1 also leads to a shunting of glycolytic end-products into the mitochondria [5], phenocopying the apoptotic phenotype displayed by glioblastoma cells after treatment with dichloroacetic acid [22,23]. In addition, because PK-M2 also functions as a co-activator of HIF-1 [11] and b-catenin [12] transactivation, reducing the level of PK-M2, as opposed to just inhibiting its kinase function, might also interfere with anti-apoptotic and pro-proliferative functions [24] mediated by these factors. How does blocking the activation of the PK-M exon 10 10W region result in activation of exon 9? This may be explained in terms of 3 0 splice site (ss) competition, such that the loss of exon 10 definition leads to more efficient recognition of the exon 9 3 0 ss by the spliceosome. This was previously observed upon knockdown of SRSF3 [14]-a recently discovered exon 10 activator-leading to a large increase in PK-M1 mRNA levels in HEK-293 cells. Although here we have successfully targeted a novel exon 10 ESE motif, we do not presently know which splicing factor(s) actually binds to this 10W ESE. In addition, even though the highly homologous 10W-like region in intron 9 does not by itself play a strong role in exon 10 activation and/or exon 9 repression, rescue experiments indicate that it works in tandem with the 10W ESE in exon 10. It will be of interest to identify the factors responsible for recognition of the exon 10 10W ESE and of the related element in intron 9, so as to elucidate the exact mechanism of exon 10 activation mediated by these elements. However, due to the complicated mechanism controlling mutually exclusive exon use in this gene, we cannot rule out the potential influence of RNA secondary and/or tertiary structures in mediating exon 10 activation. Consequently, apart from or in conjunction with disrupting RNA-protein interactions, these ASOs might also work by disrupting putative RNA higher-order structure present in exon 10.
Additionally, the 10W region appears to comprise adjacent silencer (10F) and enhancer (10B) regions. Although it is clear from the minigene-duplication experiments that the silencing effect of 10F is masked by 10B in the native context of exon 10, but is dominant in the context of exon 9, the mechanism responsible for this context-specific repression remains unknown.
Although certain exon 10 ASOs increase the expression of PK-M1 protein isoform and decrease that of PK-M2 protein isoform, we cannot rule out that these ASOs might partially disrupt PK-M2 mRNA export and translation by remaining hybridized to the spliced mRNA. Although previous experiments targeting SMN2 pre-mRNA argued against the influence of analogous ASOs on mRNA transport and translation [25], further work will be necessary to establish whether the PK-M ASOs act solely at the level of modulating alternative splicing.
Although siRNA-mediated knockdown of PK-M2 in the NCI 60 panel of cell lines has also been reported to induce apoptosis [26], that strategy might be less effective, considering that the onset of apoptosis was greatly delayed (144 h versus 36 h in the present study). In addition, because glioblastoma cells can freely take up certain chemically modified ASOs in vivo [27] without the aid of delivery agents, ASO-mediated PK-M switching could potentially be a particularly effective therapeutic strategy.
Although the ASOs we have screened are potent enough to switch the PK-M splicing profile of glioblastoma cells from PK-M1 to PK-M2 in vitro, the effects of these ASOs on cells and tissues that predominantly express PK-M1 are presently unknown. This will be difficult to test in vitro, because all cell lines assayed to date predominantly include exon 10 and express the PK-M2 isoform, regardless of their transformation status [26] (data not shown). However, given our successful experience with intracerebroventricular administration of ASOs in rodents and non-human primates [28][29][30][31], it should be possible in the future to assess the safety of PK-M2 reduction in normal CNS tissues, as well as to evaluate the efficacy of these ASOs in mouse glioblastoma models.

Cells and transfections
HEK-293, U87-MG and A172 cells were obtained from ATCC and grown in DMEM supplemented with 10 per cent (v/v) FBS, penicillin and streptomycin, at 378C and 5 per cent CO 2 . 5 mg of minigene plasmid per 10 cm dish, or 1 mg per 6-cm dish, was transiently transfected using Lipofectamine 2000 (Invitrogen). Total RNA from transfected cells was harvested after 48 h. For ASO transfections, 2 Â 10 6 cells were first plated in 6 cm dishes, and then transfections were performed using an ASO : Lipofectamine 2000 ratio of 20 pmoles : 1 ml. For Annexin V assays and RT-PCR experiments in glioblastoma cells, total RNA and protein from transfected cells were harvested after 36 h.

Oligonucleotide synthesis
2 0 -O-methoxyethyl-modified oligonucleotides with a phosphorothioate backbone were synthesized on an Applied Biosystems 380B automated DNA synthesizer (Applied Biosystems, Foster City, CA) and purified as described [32]. The oligonucleotides were dissolved in water. The sequence of the control ASO is 5 0 -TCATTTGCTTCATACAGG-3 0 .

Retrovirus transduction
To generate cell lines that overexpress human PK-M1 isoform in a doxycyline-dependent manner, A172 cells were first infected with MSCV-rtTA-hygro virus, and selected in hygromycin for two weeks. Human PK-M1 cDNA was amplified from A172 cells transfected with 10M 46 -60 ASO using the following primer pair: hPKT7cDNAF (5 0 -GGGGAACTCGA GATGGCTTCTAGGATGGCATCGATGACAGGTGGCCAAC AGATGGGCATGTCGAAGCCCCATAGTGAAGCCG-3 0 ) and rsob.royalsocietypublishing.org Open Biol 2: 120133 hPKT7cDNAR (5 0 -GGGGAAGAATTCTCACGGCACAGGAA CAACACGCATG-3 0 ), with Phusion high-fidelity DNA polymerase (Finnzymes). The resulting amplicon containing the T7 tag was gel-purified and cloned between the EcoRI and XhoI sites of the retroviral TtiGP plasmid. A172-rTA cells were then infected with TtiGP-PKM1 virus. To make cell lines constitutively overexpressing human PK-M2, PK-M2 cDNA from A172 cells was amplified using the same primers and cloned between the EcoRI and XhoI sites of the retroviral PIG plasmid. A172 and U87-MG cells were then infected with the PIG-PKM2 virus. All infected cells were then selected with 100 mg ml 21 puromycin for 3 days. All plasmids were sequenced to confirm their identities.
To generate the dInt9 mutant, two fragments were generated from the wild-type minigene construct, using the following primer pairs: FEcoRV and PKMdelB12R (5 0 -TGC CCTGCCATGACCTCCCAGACGAGAAGAGGCTCTGTGCC CAG-3 0 ) and PKMdelB125 (5 0 -ACAGAGCCTCTTCT CGTCT GGGAGGTCATGGCAGGGCAG-3 0 ). To generate the dM double mutant, the same two fragments were generated from the d10W minigene. Both fragments were then gelpurified and subjected to a second OE PCR using FEcoRV and RKpnI. All generated fragments were then cloned between the EcoRV and KpnI sites of the modified wild-type minigene plasmid.

RT-PCR
Total RNA (2 mg) was extracted from cell lines using Trizol (Invitrogen). Contaminating DNA was removed with DNAase I (Promega). Reverse transcription was carried out using ImPromp-II reverse transcriptase (Promega). Semiquantitative PCR using Amplitaq polymerase (Applied Biosystems) was performed by including [a-32 P]-dCTP in the reactions. The human-specific primer sets used to amplify endogenous transcripts anneal to PK-M exons 8 and 11, and their sequences are: hPKMF, 5 0 -AGAAACAGCCAAAGGG GACT-3 0 ; hPKMR, 5 0 -CATTCATGGCAAAGTTCACC-3 0 . To amplify minigene-specific transcripts, the forward primer was replaced with a primer annealing to the pcDNA3.1(þ) vector, pcDNAF (5 0 -TAATACGACTCACTATAGGG-3 0 ). After 28 amplification cycles for minigene-derived transcripts, and 27 cycles for endogenous transcripts, the reactions were divided into two aliquots for digestion with PstI (New England Biolabs) or no digestion. The products were analysed on a 5 per cent native polyacrylamide gel, visualized by autoradiography and quantified on a Typhoon 9410 phosphorimager (GE Healthcare) using MULTI GAUGE software v. 2.3. The percentage M1 mRNA in endogenous transcripts was calculated using the GC-content-normalized intensities of the top undigested band (M1, A), the bottom two digested bands (M2, B1 B2) in the PstI-digest lanes and the double-skipped species (D), if detectable. The percentage M1 mRNA from minigene transcripts was calculated using the GC-content-normalized intensities of the top undigested band (a, M1) and other higher-mobility digested bands corresponding to M2 and its variant species (b-g, as described above) in the PstI-digest lanes. All the PCR products were gel-purified, cloned and sequenced to verify their identities.

Immunofluorescence
Cells were first transfected with ASOs as above, and then plated onto four-well culture slides (BD Biosciences) 24 h post-transfection. At 36 h post-transfection, cells were washed with PBS and fixed with 3.7 per cent formaldehyde in PBS for 20 min. Cells were then permeabilized in 0.1 per cent Triton X-100 in PBS for 10 min after washing in PBS, blocked for 20 min in blocking buffer (1% goat serum in PBS) and then incubated overnight with rabbit monoclonal anti-PKM2 antibody (Cell Signaling Technology). After washing three times with PBS, the cells were then incubated for 1 h in blocking buffer containing Alexa Fluor 594 conjugated goat anti-rabbit secondary antibody (Molecular Probes/Invitrogen). Cells were washed with PBS, and then culture slides were disassembled and mounted with Prolong Gold mounting solution containing DAPI (Molecular Probes/Invitrogen). Cells were analysed using a Zeiss Axioplan1 upright fluorescent microscope.