Exon-centric regulation of pyruvate kinase M
alternative splicing via mutually exclusive exons
Zhenxun Wang1,2, Deblina Chatterjee1,3, Hyun Yong Jeon1,3, Martin Akerman1,
Matthew G. Vander Heiden4, Lewis C. Cantley5, and Adrian R. Krainer1,*
1Cold Spring Harbor Laboratory, PO Box 100, Cold Spring Harbor, NY 11724, USA
2Watson School of Biological Sciences, Cold Spring Harbor, NY 11724, USA
3Graduate Program in Molecular and Cellular Biology, Stony Brook University, Stony Brook, NY 11794, USA
4Koch Institute for Integrative Cancer Research at MIT, Cambridge, MA 02139, USA
5Division of Signal Transduction, Beth Israel Deaconess Medical Center and Department of Systems Biology, Harvard Medical School, Boston, MA 02115, USA
* Correspondence to: Adrian R. Krainer, E-mail: firstname.lastname@example.org
Alternative splicing of the pyruvate kinase M gene (PK-M) can generate the M2 isoform and promote aerobic glycolysis and tumor
growth. However, the cancer-specific alternative splicing regulation of PK-M is not completely understood. Here, we demonstrate
that PK-M is regulated by reciprocal effects on the mutually exclusive exons 9 and 10, such that exon 9 is repressed and exon
10 is activated in cancer cells. Strikingly, exonic, rather than intronic, cis-elements are key determinants of PK-M splicing
isoform ratios. Using a systematic sub-exonic duplication approach, we identify a potent exonic splicing enhancer in exon 10,
which differs from its homologous counterpart in exon 9 by only two nucleotides. We identify SRSF3 as one of the cognate
factors, and show that this serine/arginine-rich protein activates exon 10 and mediates changes in glucose metabolism. These find-
ings provide mechanistic insights into the complex regulation of alternative splicing of a key regulator of the Warburg effect, and
also have implications for other genes with a similar pattern of alternative splicing.
Keywords: alternative splicing, cancer metabolism, pyruvate kinase, SRSF3
Cancer cells exhibit a metabolic phenotype termed aerobic gly-
colysis, or the Warburg effect, characterized by increased glycoly-
sis with lactate generation, regardless of oxygen availability
(Vander Heiden et al., 2009). Expression of the type II isoform
of the pyruvate kinase M gene (PKM2, referred to here as
PK-M) mediates this metabolic phenotype, and confers a prolif-
erative advantage to tumor cells in vivo (Christofk et al., 2008a).
Pyruvate kinase (PK) catalyzes the final step in glycolysis, gen-
erating pyruvate and ATP from phosphoenolpyruvate and ADP
(Dombrauckas et al., 2005). The exons 9 and 10 of the PK-M
gene can each encode a 56-amino-acid segment, and be alterna-
tively spliced in a mutually exclusive (ME) fashion to give rise to
M1 and M2 isoforms, respectively (Noguchi et al., 1986). PK-M1 is
constitutively active and predominantly expressed in terminally
differentiated tissues (Christofk et al., 2008a; Clower et al.,
2010). PK-M2 is expressed in cancer cells, as well as in fetal
and undifferentiated adult tissues, and is allosterically regulated
by fructose-1,6-bisphosphate (FBP) and can interact with
tyrosine-phosphorylated signaling proteins (Christofk et al.,
2008a, b). The growth signal-mediated inhibition of PK-M2
activity contributes to cancer cell growth by decreasing carbon
flux through the catabolic glycolytic pathway, allowing accumu-
lated upstream intermediates to be shunted to anabolic pathways
to facilitate cell proliferation (Hitosugi et al., 2009).
and Ranganathan, 2009). Several mechanisms involved in MEexon
selectionhave been described,but how theyare coordinatelyregu-
are often homologous, indicating an exon-duplication origin
(Letunic et al., 2002). The exons 9 and 10 of PK-M are ME exons
whose ME splicing mechanism might be novel, as the length
(401 bp) and sequence of intron 9 rule out steric interference that
could prevent double splicing due to the spacing of the branch
site and the 5′splice site (5′ss) (Smith and Nadal-Ginard, 1989).
cancer and proliferating cells, and also implicated two pairs of
splicing-repressor paralogs—PTB/nPTB and hnRNPA1/A2—in
exon 9 repression (Clower et al., 2010; David et al., 2010). It
remains unclear whether additional repressors block exon 9, and
9 repression? Moreover, though hnRNPA1/A2 and PTB appear to
bind in the intronic regions flanking exon 9 (David et al., 2010), it
remains unclear where the critical cis-elements responsible for the
Received July 11, 2011. Revised August 9, 2011. Accepted August 21, 2011.
#The Author (2011). Published by Oxford University Press on behalf of Journal of
Molecular Cell Biology, IBCB, SIBS, CAS. All rights reserved.
Published online November 1, 2011
Journal of Molecular Cell Biology (2012), 4, 79–87
PK-M2 splicing pattern in proliferating cells are distributed, i.e. are
they present in the exons, the introns, or both?
To address these questions, we constructed a minigene that
recapitulates the splicing-regulatory features of endogenous
PK-M. We demonstrate that exon 10 is activated in cancer cells
independently of exon 9 repression. We further show that
exonic, but not intronic, cis-elements, are the key determinants
of PK-M alternative splicing. Using a sub-exonic duplication strat-
egy, we further mapped an exonic splicing enhancer (ESE) in exon
10, and found that SRSF3 (formerly SRp20), an oncogenic
member of the serine/arginine-rich (SR) protein family of splicing
activators, is its cognate binding factor. SRSF3 knockdown in
cancer cells rescues PK-M1 expression and decreases lactate pro-
duction and cellular proliferation.
PK-M minigene recapitulates alternative splicing
of the endogenous gene
The ME exons 9 and 10 of PK-M are identical in length, and
highly homologous at the nucleotide and amino acid sequence
levels (Figure 1A). To analyze the mechanism of ME splicing of
PK-M pre-mRNA, and identify splicing cis-elements, we generated
a minigene transcribed from a CMV promoter (Figure 1B). The
minigene consists of the genomic region encompassing exons 9
and 10, introns 8, 9, and 10, and proximal portions of the flanking
exons 8 and 11. Detection and analysis of minigene-derived tran-
scripts utilized an RT–PCR and restriction-digestion strategy, as
described for endogenous PK-M transcripts (Figure 1B) (Clower
et al., 2010). To characterize all possible minigene-derived
species after cell transfection, we selectively amplified them
from total cDNA using a forward primer specific for upstream
vector sequence, and a reverse primer annealing to constitutive
exon 11 (Figure 1B).
Figure 1C shows the profiles of endogenous and minigene-
derived transcripts in HEK-293 cells (similar data not shown for
HeLa, A172, and SKNBE cells). The minigene predominantly
expressed the exon 10-included PK-M2 isoform, paralleling
endogenous PK-M2/M1 splicing ratios. Two additional, minor
minigene RNAs not found in endogenous PK-M transcripts were
also reproducibly detected: an RNA lacking both exons 9 and 10,
Figure 1 Detection of endogenous and minigene-specific PK-M spliced isoforms. (A) Nucleotide (top) and amino acid (bottom) sequence align-
ments of ME exons 9 (M1) and 10 (M2). Identical nucleotides are shown by vertical dashes. Identical and similar amino acids are highlighted in
red and yellow, respectively. The distinctive phosphotyrosine-binding residue and the FBP-binding pocket of PK-M2 are indicated. The percen-
tages of nucleotide and amino acid identity are shown. (B) Diagram of the human PK-M minigene. The minigene comprises the intact introns 8,
9, and 10, the intact alternative exons 9 and 10, and portions of the flanking constitutive exons 8 and 11. The numbers above each exon and
intron show the length in nucleotides. A vector-specific forward primer (dashed arrow) and a reverse primer annealing to exon 11 were used to
amplify minigene-derived transcripts; to amplify endogenous transcripts in untransfected cells, a forward primer annealing to exon 8 (solid
arrow) was used instead. To distinguish between exon 9-included (M1 isoform) and exon 10-included (M2 isoform) transcripts, cDNA amplicons
were cleaved with NcoI (N), PstI (P) or both (NP). Note the additional NcoI site at the 5′end of the PK-M minigene, which is absent from the
endogenous gene. (C) Radioactive RT–PCR and restriction digest of endogenous and minigene-derived PK-M transcripts in HEK-293 cells. RNA
was isolated from untransfected cells or 48 h after transfection of the plasmid minigene. Asterisks (*) indicate restriction fragments correspond-
ing to unspliced PK-M pre-mRNA. cDNAs and fragments fromendogenous mRNAs are indicatedon the left in uppercase font; those derived from
minigene-specific transcripts are shown on the right in lowercase bold font. The most important bands are indicated in blue font. The bands
correspond to: uncut M1 fragment (A, 398 nt; a, 481 nt); uncut M2 fragment (B, 398 nt; b, 481 nt); NcoI-cleaved M1 3′fragment (A1, 248 nt; a1,
248 nt); NcoI-cleaved M1 5′fragment (A2, 144 nt; a2, 150 nt); PstI-cleaved M2 5′fragment (B2, 185 nt; b2, 268 nt); PstI-cleaved M2 3′fragment
(B1, 213 nt; b3, 213 nt). Minigene M2 cDNA is cleaved by NcoI, giving rise to b1 (404 nt). Two additional species are observed from minigene-
specific transcripts:a spliced mRNA that skips both exons9 and10(d, 314nt) andis cleavedbyNcoI (d1, 237 nt), and anM2 transcript thatuses
a cryptic 3′splice site 115 nt upstream of the authentic 3′ss of exon 10 (c, 596 nt), and whose 3′fragment after PstI digestion is identical to b3.
Due to its lowabundance, the corresponding 5′end afterPstI digestion can only be detected with certain mutant minigenes (see Figure2A). The
5′end of the PstI-cleaved M2 fragment, b2, is additionally cleaved by NcoI to a shorter fragment (b4, 194 nt) plus a 74-nt fragment that runs off
the gel. SeeSupplementary Figure S1 fora detailed description of these M2-variant species. The numbers belowthe gel indicate the percentage
of exon 9-included transcripts (%M1); standard deviations (SD) are ≤0.2% (n ≥ 3).
Journal of Molecular Cell BiologyWang et al.
species d) and a variant PK-M2 RNA spliced via a cryptic 3′ss
upstream of the authentic 3′ss of exon 10 (Figure 1C and
Supplementary Figure S1; exon 10 3′ss cryptic species c).
Blocking exon 10 inclusion does not fully rescue exon 9 inclusion
Given the predominant PK-M2 splicing of endogenous and
minigene transcripts in proliferating cells, we tested whether
blocking exon 10 inclusion by inactivating or weakening its
splice sites might force a corresponding increase in exon 9
inclusion.The 5′ss or 3′ss of exon 10were inactivated by mutating
the invariant G residues at the intron borders, and the 3′ss was
separately weakened by transversions within its upstream poly-
pyrimidine tract (PPT) (Figure 2A).
Mutating the exon 10 5′ss resulted in a larger mRNA with an
extended exon 10 (Figure 2A and Supplementary Figure S1,
lanes 5–8; exon 10 5′ss cryptic species e). This mRNA resulted
from splicing viaacryptic 5′ss105 ntdownstreamof the authentic
5′ss of exon 10. There was no significant increase in the double-
skipped RNA species d. Thus, exon 10 definition was essentially
preserved, even though the normal 5′ss was inactivated.
In contrast, mutating the exon 10 3′ss largely abrogated exon
10 definition, giving a large increase in double-skipped tran-
scripts (Figure 2A and Supplementary Figure S1, lanes 9–12;
species d). A small increase in the exon 10 cryptic 3′ss species
c was also observed. These results suggest that the exon 10
3′ss is essential for exon 10 definition. Similarly, weakening the
exon 10 PPT led to some loss of exon 10 definition, with large
increases in double-skipped RNA and in the use of the upstream
cryptic 3′ss (Figure 2A, lanes 13–16; species c and d). This result
implies that exon 10 inclusion is an active process, such that the
loss of exon 10 activation results in non-productive splicing of the
With all three minigene mutants, there was only marginal
recovery of exon 9 inclusion (Figure 2A, lanes 5–16). Therefore,
exon 9 inclusion in proliferating cells is repressed independently
of exon 10 splicing.
Strengthening the splice sites of exon 9 leads to aberrant PK-M
Because blocking exon 10 splicing did not significantly rescue
M1 splicing in transformed cells, we tested whether directly
strengthening the exon 9 splice sites might do so. We therefore
mutated the exon 9 5′ss to the consensus 5′ss sequence
(Figure 2B, lanes 5–8), and strengthened the exon 9 3′ss by
increasing thepyrimidine content
(Figure 2B, lanes 9–12).
As expected, strengthening either the exon 9 5′ss or 3′ss led to
an increase in M1 splicing. However, a new exon 9 plus exon 10
doubly-included mRNA species was also observed in both cases
(Figure 2B and Supplementary Figure S1; double-spliced species
f). Mutating the 5′ss to the consensus resulted in a large increase
in the amount of M1 RNA and in high levels of the double-spliced
mRNA (Figure 2B, lanes 5–7), whereas slightly increasing the
3′ss strength gave somewhat lower levels of these two mRNAs
(lanes 9–11). The concurrent appearance of the double-spliced
mRNA suggests that exon 10 definition—and by extension, exon
10 activation—is largely independent of exon 9 splicing.
Figure 2 Effects of splice-site relative strengths on inclusion of exon 9. Mutant minigenes were analyzed by transient transfection into HEK-293
cells, followed by radioactive RT–PCR and restriction digests, as in Figure 1C. (A) Mutations that inactivate or weaken the splice sites (ss) of
exon 10. The mutated 5′ss, 3′ss, or PPT nucleotides are indicated in light blue. Numbers indicate the position of the mutated nucleotide, either
upstream (+), or downstream(2) of the exon.Bands markedwith asterisks (*) arefragmentsof PK-M pre-mRNA, as in Figure1C. Bands already
described in Figure 1C are indicated on the left. New and/or important bands are indicated on the right, and the key bands are labeled in blue
font (see also Supplementary Figure S1). The 5′ss mutation gives rise to a new M2 variant (e, 586 nt) derived from use of a cryptic 5′ss 105 nt
downstream of the authentic 5′ss; this band is cleaved by NcoI (e1, 509 nt), and PstI digestion generates a longer 3′fragment (e2, 318 nt). The
increasein the M2 3′ss cryptic variant (c, as described inFigure1C)upon mutation of the PPTor3′ss allows detection of the 5′PstI fragment (c3,
383 nt). The %M1 inclusion is indicated at the bottom; SD: 0.2% (WT); 0.3% (exon 10 5′ss mut); 2% (exon 10 3′ss mut); and 2% (exon 10 Py?)
(n ≥ 3). (B) Mutations that strengthen the splice sites of exon 9. The mutated 5′ss and PPT nucleotides are indicated in light blue. Uppercase
and lowercase letters indicate exonic and intronic sequences, respectively. Bands are labeled as in A. A new exon 9–exon 10 doubly included
mRNA is indicated on the right (f, 648 nt). This band is sensitive to both PstI (f1, 435 nt) and NcoI (f2, 415 nt) digestion. See also Supplementary
Figure S1. The %M1 inclusion is indicated at the bottom; SD: 0.2% (WT), 4% (exon 9 5′ss con), and 3% (exon 9 3′ss Py?) (n ≥ 3).
Exon-centric regulation of PK-M alternative splicingJournal of Molecular Cell Biology
Exonic splicing silencers and enhancers are key determinants of
PK-M splicing ratios
To prioritize PK-M regions for cis-element analysis, we next
asked whether intronic or exonic cis-elements play critical roles
in activating exon 10 and/or repressing exon 9. First, we dupli-
cated exon 10 in place of exon 9 in the minigene (Figure 3,
lanes 9–12). If exon 9 repression depends on cis-elements
present in intron 8 or 9, there should be inefficient use of the
upstream copy of exon 10, because it would be under the influ-
ence of these repressive elements, and the pattern should be
similar to that of the wild-type minigene. Instead, we observed
the striking appearance of a doubly-included exon 10 mRNA
species (Figure 3, lane 9; exon 10 doubly-included species g)
indicating that the upstream exon 10 was still activated,
regardless of its position. This finding strongly suggests that
enhancer elements involved in exon 10 definition are present in
the exon itself.
Similarly, we duplicated exon 9 in place of exon 10 in the mini-
gene (Figure 3, lanes 5–8). If exon 10 splicing is normally acti-
vated through flanking cis-elements in intron 9 or 10, there
should be a strong increase in exon-9-included transcripts, com-
pared with the wild-type minigene, because the downstream
copy of exon 9 would now be under the influence of such
elements. However, there was no such increase from the exon-9-
duplicated minigene transcripts (Figure 3, lanes 5–8). This finding
suggests that splicing-silencing elements involved in repressing
exon 9 are located in the exon itself.
Whenthepositionsofexons 9and 10wereswapped,leavingall
the introns unchanged (Figure 3, lanes 13–16), the M1 and M2
isoform ratio was similar to that of the wild-type minigene,
although therewasa decrease in M1abundance, and an expected
decrease in the use of the cryptic 3′ss upstream of the original
exon 10 (i.e. in intron 9), because this 3′ss is now juxtaposed
with the repressed exon 9 3′ss. This finding indicates that exon
10, when moved to exon 9, is spliced as efficiently as in its
original location. This striking result suggests that exonic
cis-elements involved in PK-M splicing are sufficient to activate
exon 10 and repress exon 9, independently of their respective
positions along the gene and the influence of their flanking
To confirm these results, and to determine the role of the splice
sites in ME exon use, we swapped the 3′ss or 5′ss of both exons.
Computational analysis (Yeo and Burge, 2004) suggested that the
3′ss of exon 9 is weaker than those of exons 10 and 11
(Supplementary Figure S2). Exon 9 inclusion was therefore
expected to increase when its 3′ss was replaced by the stronger
one from exon 10. Instead, the M1 isoform abundance decreased,
suggesting that proper definition of exon 9 requires its native,
albeit weaker, 3′ss, for contextual reasons (Supplementary
Figure S2, lanes 3 and 11). There were no significant changes
when the 5′ss were swapped (lanes 3 and 7). These observations
again indicate that exonic, rather intronic, cis-elements are the
key determinants of PK-M splicing ratios.
A strong ESE in exon 10 is necessary and sufficient for activation
of the exon
We systematically searched for critical ESEs in exon 10. Taking
advantage of the high nucleotide-sequence identity between
exons 9 and 10, and their identical lengths (Figure 1A), we dupli-
cated 15–30 nt stretches of exon 10 into the corresponding
location in exon 9, to find sub-exonic regions that are sufficient
to activate exon 9 inclusion (Figure 4A).
The last 30 nt, but not the last 15 nt of exon 10 strongly
increased exon 9 inclusion, suggesting that a strong ESE is
present within, or overlaps with, the penultimate 15 nt of exon
10 (Figure 4B, compare lanes 4–6 with lanes 7–9). We then ana-
lyzed the entire 30-nt stretch using SFmap (Akerman et al., 2009;
Paz et al., 2010), a method to predict splicing-regulatory motifs.
This analysis yielded a near-consensus, conserved SRSF3 motif
(Schaal and Maniatis, 1999) within the penultimate 15-nt
segment (Supplementary Figure S3). To determine whether this
SRSF3 motif alone can account for the observed M1 splicing acti-
vation, weduplicated the 7-nt SRSF3 motif from exon 10 into exon
9, by mutating the only two nucleotides that differ between exons
9 and 10 within this heptamer. Remarkably, duplicating only the
SRSF3 motif, but not the flanking 8-nt or the final 15-nt region,
activated exon 9 to a similar extent as that achieved by duplicat-
ing the entire 30-nt region (Figure 4B, compare lanes 10–12 with
lanes 7–9 and 13–15). We conclude that the SRSF3 motif is an
actual exon 10 ESE.
We then reciprocally abrogated the motif in exon 10 by
mutating itinto thecorresponding
(Figure 4C). This resulted in decreased exon 10 splicing, and
a large increase in double-skipped RNA (species d), together
with a concomitant increase in the M1 isoform (species a).
Taken together, these data suggest that the SRSF3 motif is a
bona fide ESE that is both necessary and sufficient for exon 10
Figure 3 Effects of exonic cis-elements on PK-M alternative splicing.
Mutant minigenes were analyzed by transient transfection into
HEK-293 cells, followed by radioactive RT–PCR and restriction
digests, as in Figure 1C. Minigenes were constructed with clean dupli-
cations or swaps of exons 9 and 10, as shown at the top. Bands
marked with black asterisks (*) or labeled on the left are as described
in Figure 1C. Key bands are labeled in blue font. The band marked
with a red asterisk indicates a PCR artifact. A new exon 10–exon 10
doubly-included mRNA expressed from the exon 10 Dup minigene
is indicated on the right (g, 648 nt). It generates unique fragments
upon NcoI (g1, 571 nt) and PstI (g2, 167 nt) digestion. See also
Supplementary Figure S1.
Journal of Molecular Cell BiologyWang et al.
SRSF3 binds specifically to the exon 10 ESE
The mapped ESE resembles a consensus SRSF3 motif, so we
used RNA-affinity pulldowns to ask whether SRSF3 indeed
binds to this sequence. Synthetic 24-nt RNAs were covalently
linked via their 3′ends to agarose beads, incubated with HeLa
nuclear extract under splicing conditions, and washed at two
different salt concentrations. Bound proteins were then eluted
and analyzed for SRSF3 binding by immunoblotting (Figure 5).
We initially compared RNA sequences 6–29-nt upstream from
the last 3′nucleotide in both exons (Figure 5B). As a control,
we mutated the exon 10 SRSF3 motif to the consensus
(AUCGUCC to CUCGUCC). Coomassie-blue staining and mass
spectrometry analysis revealed a different composition of
bound proteins between exon 9 and exon 10 RNAs (Figure 5A,
lane 2 vs lane 3). We observed similar patterns under more strin-
gent washing conditions (300 mM instead of 150 mM KCl; data
not shown). More importantly, SRSF3 bound strongly to the
exon 10 RNA. Because hnRNPA1 bound to all RNAs tested, we
usedit asaninternalloading control. Immunoblotting
unambiguously revealed that SRSF3 bound only to the exon 10
and consensus SRSF3 RNAs, but not to the exon 9 RNA, at both
salt concentrations (Figure 5B).
To pinpoint the precise location of SRSF3 binding in exon 10,
we mutated the SRSF3 motif to the corresponding exon 9
sequence (Figure 5C). This gave a large decrease in SRSF3
binding (Figure 5C), indicating that the motif is necessary for
strong SRSF3 binding to these short exon 10 RNA fragments.
SRSF3 is necessary for exon 10 inclusion
Although abrogating the SRSF3 motif in exon 10 led to greater
exon 9 inclusion (Figure 4D), we could not rule out the presence
of functional cis-elements in the corresponding exon 9 sequence
that might influence the observed PK-M1 inclusion ratio. To better
assess the effect of SRSF3 on exon 10 inclusion, we tested
whether knocking down SRSF3 in the context of both the wild-type
minigene and the duplicated SRSF3 motif minigene mutant (10 SR
minigene; Figure 4B) would enhance exon 9 inclusion, as exon 10
inclusion would be expected to decrease. Indeed, knocking down
SRSF3 using two different siRNAs (Figure 5D) increased M1
Figure 4 Mapping an ESE in exon 10. (A) Method used to map an ESE within the last 30 nt of exon 10. The indicated exon 9 (green) nucleotides
were mutated to corresponding exon 10 (red) sequences. The last 30 nt of exon 10, when moved to exon 9, activated inclusion of exon 9. An
SRSF3 SELEX motif (Schaal and Maniatis, 1999) identified by SFmap (Akerman et al., 2009) is shown by a black rectangle. Exon 10 candidate
regions that were duplicated into exon 9 are indicated at the bottom, and the construct names (red) are given on the left. (B) The SRSF3 motif is
the ESE in the last 30 nt of exon 10. The constructs from (A) are indicated at the top. Labeled bands are as in Figure 1C. The %M1 inclusion is
shown at the bottom (n ≥ 3). Duplication of the exon 10 SRSF3 motif into exon 9 was sufficient to rescue exon 9 inclusion. The %M1 inclusion
is indicated at the bottom; SD: 0.2% (WT), 0.1% (10 15), 5% (10 30), 4% (10 SR), 1% (10 5′) (n ¼ 3). (C) Replacing the SRSF3 motif in exon10 by
duplicating exon 9 sequences. The indicated exon 10 (red) nucleotides were mutated to the corresponding exon 9 (green) sequences. Exon 9
regions that were duplicated into exon 10 are indicated below, and the construct names (green) are given on the left. (D) Inactivating the SRSF3
motif in exon 10 causes skipping of exon 10. Candidate exon 9 regions duplicated into exon 10 are indicated at the top. Labeled bands are as in
Figure 1C. %M1-included and %double-skipped (%Skp) transcripts are indicated below. Inactivating the SRSF3 motif in exon 10 is sufficient to
increase the abundance of the double-skipped RNA species. Transcript-level changes are indicated at the bottom; SD: 0.3% (WT), 0.3% (9 30),
1% (9 1530), 2% (9 SR) (n ¼ 3).
Exon-centric regulation of PK-M alternative splicingJournal of Molecular Cell Biology
inclusion and the double-skipped RNA with the wild-type minigene
SR minigene; Figure 4D). We also observed a strong decrease in
with the 10 SR minigene (Figure 5E, lanes 7–12). Thus, SRSF3 is
necessary for exon 10 inclusion.
SRSF3 affects endogenous PK-M splicing
We next tested whether SRSF3 could affect endogenous PK-M
alternative splicing; as expected, knockdown and overexpression
gave reciprocal effects. SRSF3 knockdown in HEK-293 cells
increased PK-M1 mRNA to 12 folds, which was also reflected at
nificant (David et al., 2010), but this could have reflected weaker
knockdown and/or negative selection resulting from the use of a
stably transfected shRNA. As a reciprocal experiment, we overex-
pressed SRSF3 in the glioblastoma cell line A172 (Supplementary
Figure S4B), which expresses relatively high levels of PK-M1
(Clower et al., 2010) and low levels of SRSF3 (Supplementary
Figure S4D). As expected, SRSF3 overexpression promoted an
increase in M2, and a 5-fold decrease in M1 (Figure 6C).
SRSF3 activates endogenous PK-M exon 10
To determine whether the effect of SRSF3 overexpression
(Figure 6C) reflects increased exon 10 activation and/or exon 9
repression, we first rescued M1 inclusion by knocking down
hnRNPA1/A2 and PTB in HEK-293 cells, and then we overex-
pressed SRSF3 and assessed its effects on endogenous PK-M
transcripts (Figure 6D and Supplementary Figure S4A). SRSF3
overexpression partially restored the level of PK-M2 mRNA
(Figure 6D), suggesting that SRSF3 only activates exon 10.
Indeed, SRSF3 overexpression did not affect splicing of the dupli-
cated exon 9 minigene, and SRSF3 knockdown in the context of
the 9 SR minigene had no significant effect on PK-M1/M2 splicing
ratios, further suggesting that SRSF3 does not directly affect exon
9 splicing (Supplementary Figure S5).
SRSF3 is necessary for aerobic glycolysis and cellular proliferation
PK-M isoform ratios influence aerobic glycolysis (Christofk
et al., 2008a), and therefore, we determined the effect of SRSF3
knockdown on this process, as assayed by the extent of cellular
lactate production. SRSF3 knockdown in HEK-293 cells resulted
in a significant 2-fold decrease in lactate production (Figure 6E).
Because the Warburg effect also strongly influences cellular pro-
liferation (Christofk et al., 2008a), we next assayed the effect of
SRSF3 knockdown oncell
(MTT) assay was performed over the course of 7 days in
decreased the proliferation rate. Thus, SRSF3 promotes cellular
proliferation and aerobic glycolysis at least in part by influencing
PK-M isoform ratios. Moreover, because SRSF3 knockdown
Figure 5 SRSF3 binds to the motif in exon 10 and is necessary for exon 10 inclusion. (A) The indicated RNAs covalently linked to agarose beads
were incubated with HeLa cell nuclear extract under splicing conditions, and the beads werewashed three times with buffer containing 150 mM
KCl. Bound proteins were eluted with SDS and analyzed by SDS–PAGE and Coomassie-blue staining. The mobilities of size markers (M) are
indicated. Prominent bands were excised and analyzed by mass spectrometry; the identified proteins are indicated on the right. (B and C)
Sequences of synthetic RNA oligonucleotides used for affinity pulldowns and western-blotting analysis. Descriptive names of RNAs are indi-
cated on the left. The SRSF3 motifs are enclosed by rectangles. Underlined nucleotides indicate the differences between the SRSF3 motif
and the corresponding sequences in exon 9. Western blots of eluted proteins are shown on the right of each set of sequences. The RNAs
and wash conditions are indicated at the top. Antibodies against SRSF3 and hnRNPA1 were used. HeLa nuclear extract was used as a positive
control; hnRNPA1 was used as a loading control, as it binds to these short RNAs to similar extents. (D and E) The indicated SRSF3 siRNA or
control luciferasesiRNAwas co-transfected into HEK-293 cells with thewild-type or10 SR minigene. Knockdownof SRSF3 was verified by immu-
noblotting, as showninD. Minigene-specific transcript-levelchanges for PK-M1 andPK-M2 areshown in E. siRNAs used areindicatedatthe top.
Labeled bands areas inFigure1C. Asterisks indicate PstI-cleaved pre-mRNA.%M1-included, %M2-included, and%double-skipped (%Skp) tran-
scripts are indicated below, with the following SD: 0.3% (WT/luc), 1.2% (WT/SR3#1), 1.5% (WT/SR3#2), 5.7% (10 SR/luc), 3.5% (10 SR/
SR3#1), and 2.7% (10 SR/SR3#2) (n ¼ 4). P-values (Student’s t-test) comparing %M2 from luc siRNA to that from SR3#1 and SR3#2
siRNAs co-transfected with the 10 SR minigene were 0.01 and 0.02, respectively.
Journal of Molecular Cell Biology Wang et al.
mainly increased PK-M1 protein (Figure 6B), this suggests that
the increase in PK-M1, rather than a decrease in PK-M2, is respon-
sible for the effects on metabolism and proliferation.
We have demonstrated that PK-M ME splicing involves a two-
component circuit: exon 9 is repressed and exon 10 is activated
in proliferating cells, and these two effects are essentially inde-
pendent (Supplementary Figure S6).
By duplicating and swapping ME exons in a PK-M minigene, we
showed that the key cis-elements controlling PK-M alternative
splicing are located within the ME exons themselves. This is the
first demonstration of exon-centric alternative splicing regulation
via ME exons. As a first proof of this principle, we mapped a bona
fide SRSF3 ESE in exon 10 that provedsufficient to activate exon 9
splicing in cancer cells when placed in this exon. Although double
inclusion of exon 10 was not the major product from the exon
10-duplication minigene (Figure 3), this cannot be due to repres-
sion via the flanking intron 8 and 9 regions, because exon 10 was
spliced efficiently when it was cleanly swapped with exon
9. Instead, we speculate that intronic elements are likely involved
in the ME exon selection properties of PK-M.
Remarkably, the SRSF3 ESE motif in exon 10 differs from the
corresponding exon 9 sequence by only two nucleotides, which
are wobble bases in the respective codons (Supplementary
Figure S3A and B). The first wobble base is highly conserved,
perhaps reflecting the importance of this nucleotide in mediat-
ing SRSF3 recruitment and functionality. The corresponding
two nucleotides in exon 9 are also conserved, suggesting
selection against the creation of an exon 9 SRSF3 activation
motif (Supplementary Figure S3B). However, we cannot rule
out the existence of a corresponding exonic splicing silencer
element in exon 9. The use of two wobble nucleotides to
code for a key splicing signal illustrates the impact of
sequence changes that would be translationally neutral,
except that they drastically affect the structure of the resulting
protein by changing alternative splicing of the entire exon
(Cartegni et al., 2002).
Although mutating just two nucleotides in exon 10 to the cor-
responding exon 9 sequence (9 SR minigene) gave rise to more
double-skipped and PK-M1 mRNAs, this was not the case when
the surrounding exon 9 sequence was also duplicated (91530
and 930 minigenes). This probably reflects the presence of
additional cis-elements in exon 9, such as putative ESEs that
Figure 6 SRSF3 affects endogenous levels of PK-M1/M2, aerobic glycolysis, and cellular proliferation. (A and B) The indicated SRSF3 siRNA or
control luciferase siRNA was transfected into HEK-293 cells. (A) Transcript-level changes for PK-M1 and PK-M2 are shown. %M1 is indicated at
the bottom, with the following SD: ≤ 0.2% (luc); 0.3 % (SRSF3) (n ¼ 3). (B) Changes at the protein level, as seen in a representative blot. (C)
T7-tagged SRSF3 cDNAwas transfected in increasing amounts into A172 cells. Cells were harvested after 48 h. The PK-M mRNA level was deter-
mined by RT–PCR. SD ≤1% in all cases (n ¼ 5). (D) hnRNPA1/A2/PTB siRNAs were co-transfected into HEK-293 cells, followed by transfection
of SRSF3 cDNA 24 h later. Cellswereharvested 36 hafter the second transfection. Transcript-level changes are indicated atthe bottom, with the
following SD: 1% (A1/A2/PTB siRNA); 3% (A1/A2/PTB siRNA + SRSF3) (n ¼ 5). P-value (Student’s t-test) comparing A1/A2/PTB siRNA with
A1/A2/PTB + SRSF3 is0.002.(E) The indicated SRSF3 orluciferase siRNAwastransfected intoHEK-293 cells. Lactateproductionwasmeasured
48 h after transfection. Error bars represent SD (n ¼ 3). *P , 0.05 (Student’s t-test). (F) Analysis of cell proliferation by the MTT assay. HEK-293
cells were transfected with luciferase or SRSF3 siRNA and grown on 96-well plates. Error bars represent the standard error of the mean. Day 1 is
the next day after seeding.
Exon-centric regulation of PK-M alternative splicing Journal of Molecular Cell Biology
may compensate for the loss of exon definition upon mutation of
the SRSF3 motif.
Whereas the 5′ss of exons 9 and 10 do not playa dominant role
in exon selection—considering that the M1 and M2 mRNA levels
do not change upon swapping these 5′ss (Supplementary Figure
S2B)—mutational analysis indicates that the 3′ss are necessary
for definition of their respective exons, and they compete with
each other. Exon definition, and ultimately, proper ME exon selec-
tion in the PK-M gene, is apparently dependent on the outcome of
competition between the alternative 3′ss. We speculate that the
recovery of endogenous PK-M1 transcripts upon SRSF3 knock-
down may reflect this competition mechanism, resulting in a
loss of exon 10 definition, and allowing the basal splicing machin-
ery to be recruited to the exon 9 3′ss. Loss of exon 10 definition is
also supported by the increase in double-skipped RNA from the
wild-type minigene upon SRSF3 knockdown. However, it is
unclear why the extent of PK-M1 inclusion is weaker in minigene
than in endogenous RNA, leading to the accumulation of unpro-
ductively spliced double-skipped RNA.
We were surprised to find no rescue of exon 9 inclusion when
its 5′ss was swapped with that of exon 10—despite a report
that hnRNPA1 represses exon 9 inclusion by binding to the
exon 9 5′ss (David et al., 2010)—as this swap presumably
removed the repressive hnRNPA1 binding site; perhaps this lack
of rescue reflects contextual effects. However, our results
confirm and extend an earlier study that duplicated the exon 10
5′ss in a heterologous minigene reporter, and found no change
in PK-M splicing (Takenaka et al., 1996). Moreover, in the
context of intact pre-mRNAs, it is not known how well hnRNPA1
binding to a motif that is part of a 5′ss can compete with
binding of spliceosomal components, such as U1 and U6
snRNPs. Given that hnRNPA1 does affect exon 9 inclusion in
vivo (Clower et al., 2010; David et al., 2010), hnRNPA1-induced
exon 9 repression could occur either indirectly—through regu-
lation of a splicing factor that in turn regulates PK-M alternative
splicing—or through additional cis-elements located elsewhere
on the PK-M pre-mRNA.
The change in endogenous PK-M1 upon SRSF3 knockdown was
roughly comparable to the effects of knocking down the known
repressors of exon 9, hnRNPA1/A2 and PTB (Clower et al.,
2010; David et al., 2010). However, knocking down these
factors did not completely rescue exon 9 inclusion, and as in
other systems, we anticipate the existence of additional activa-
tors of exon 10 and/or repressors of exon 9 that contribute to
maintaining exon 10 definition in proliferating cells. With
respect to the enhancer region we identified in exon 10, knock-
RNA-affinity chromatography—hnRNPK and RBM3—did not
change PK-M splicing ratios (data not shown).
Consistent with its expected ability to facilitate cellular prolifer-
ation byaltering glycolytic metabolism, SRSF3 is overexpressedin
ovarian cancers (He et al., 2004) and cervical cancer cell lines,
whereas in normal cervical tissue, its expression is restricted to
the basal proliferating layers (Jia et al., 2009). SRSF3 is also a
target of the oncogenic b-catenin/TCF-4 pathway in colorectal
cancer cells (Goncalves et al., 2008). Moreover, SRSF3 overex-
pression is sufficient for transformation of immortal mouse
fibroblasts (Jia et al., 2010), indicating that similar to its
paralog, SRSF1 (Karni et al., 2008), SRSF3 is an oncoprotein.
Given the multitude of potential SRSF3 downstream target
genes, we expect that SRSF3-mediated tumorigenesis reflects
splicing changes in multiple effector genes, including PK-M.
Important unanswered questions include: what are the
additional factors that govern exon 9 and 10 use in tumor cells?
Can the PK-M2 isoform be completely switched to the PK-M1
isoform by manipulating splicing factor levels in tumor cells?
How is exon 9 preferentially selected in quiescent cells? Are
tion in differentiated cells? Answers to these questions could
Materials and methods
Cells and transfections
HeLa and HEK-293 cells were grown and transiently transfected
as described (Clower et al., 2010). Total RNA was harvested
after 36 h.
Two siRNAs targeting human SFRS3 (Sigma Genosys) have the
(SR3#1) and 5′-CGUAGUCGAUCUAGGUCAA-3′(SR3#2). siRNAs
against hnRNPA1, hnRNPA2, and PTB were used as described
(Cartegni et al., 2006; Clower et al., 2010). HEK-293 cells (1 ×
106) in six-well plates were transfected with 200 pmol of siRNA
duplex using Lipofectamine 2000 (Invitrogen). Cells were
harvested 48 h later.
Immunoblotting was carried out as described (Clower et al.,
2010). Primary antibodies were: b-tubulin (Genscript rAb,
1:5000), hnRNPA1 (mAb UP1-55, culture supernatant) (Hua
et al., 2008), SRSF3 (Zymed mAb, 1:1000), PK-M2 (rabbit,
1:2000), and PK-M1 (rabbit, 1:2000) (Christofk et al., 2008a, b).
Secondary antibodies were goat anti-mouse or anti-rabbit HRP
conjugates (1:20000; Bio-Rad).
DNA oligonucleotides were obtained from Sigma Genosys. The
PK-M2 minigene was constructed by amplifying a 6.4-kb PK-M
exon 8–11 fragment from human genomic DNA (Promega)
using Phusion High-Fidelity DNA Polymerase (Finnzymes) and
primers PKMminigeneF (5′-GGGGAAAGATCTGCCACCATGGGAGA
AACAGCCAAAGGGGAC-3′) and PKMminigeneR (5′-GGGGAACTCGA
GCTAGACATTCATGGCAAAGTTCACC-3′). The product was cloned
between the BamHI and XhoI sites of pcDNA3.1(+) (Invitrogen).
See Supplementary experimental procedures for the generation
of mutant, exonic and sub-exonic duplication, exon-swap, and
Radioactive RT–PCR was carried out as described (Clower
et al., 2010). The human-specific primer sets used to amplify
endogenous transcripts anneal to PK-M exons 8 and 11, and
their sequences are: hPKMF: 5′-AGAAACAGCCAAAGGGGACT-3′;
hPKMR: 5′-CATTCATGGCAAAGTTCACC-3′. To amplify minigene-
specific transcripts, the forward primer was replaced with a
primer annealing tothepcDNA3.1(+)
5′-TAATACGACTCACTATAGGG-3′. After 26 amplification cycles for
Journal of Molecular Cell Biology Wang et al.
minigene-derived transcripts, and 24 cycles for endogenous tran- Download full-text
scripts, the reactions were divided into four aliquots for digestion
with NcoI, PstI (New England Biolabs), both, or neither. The pro-
ducts were analyzed on a 5% native polyacrylamide gel, visual-
ized by autoradiography, and quantified on a FLA-5100
phosphoimager (Fuji Medical Systems) using Multi Gauge soft-
ware Version 2.3. All the PCR products were gel-purified,
cloned, and sequenced to verify their identities.
RNA-affinity chromatography and mass spectrometry
et al., 1999; Hua et al., 2008). After the final wash, the beads were
resuspended in 4× Laemmli buffer and boiled for 5 min to elute
minent bands were excised for in-gel trypsin digestion and analysis
by MALDI-TOF mass spectrometry. Spectra were analyzed with the
MASCOT search engine.
For measurements of lactate secretion, cells were transfected
with siRNA in six-well plates. Twenty-four hours later, the cells
were replated (three replicates per condition) at subconfluent
density (25000 cells/well) in 12-well dishes, and after 24 h, the
cells were switched to serum-free medium without phenol red for
20 min, and lactatesecretedintothe mediumwasmeasured, intri-
plicateforeachsample,usinga LactateAssayKit II(BiovisionInc.).
Optical density readings at 450 nm were averaged foreach sample
replicate set, then averaged for each condition replicate set, and
finally normalizedto the cellnumber measuredfrom parallelwells.
HEK-293cells(3 × 106)weretransfectedwith400 pmoloflucifer-
ase control or SRSF3 siRNA in a 6-cm dish. After 24 h, cells were
seeded into 96-well plates (2500 cells/well). The next day, and
every 2 days thereafter, MTT (Sigma) was added to fresh medium
at a final concentration of 0.5 mg/ml. Cells were incubated at
378C for 4 h and then solubilized with DMSO. Optical density at
560 nm was then determined for each well and averaged.
Supplementary material is available at Journal of Molecular Cell
We thank Xavier Roca and Yimin Hua (Cold Spring Harbor
Laboratory), and Cynthia Clower (Harvard Medical School) for
Z.W. is supported by a fellowship from the Agency for Science,
Technology and Research, Singapore. This work was supported
by grant I1-A34 from the Starr Cancer Consortium to A.R.K.,
L.C.C., and M.G.V.H.
Conflict of interest: none declared.
Akerman, M., David-Eden, H., Pinter, R.Y., et al. (2009). A computational
approach for genome-wide mapping of splicing factor binding sites.
Genome Biol. 10, R30.
Caputi, M., Mayeda, A., Krainer, A.R., et al. (1999). hnRNP A/B proteins are
required for inhibition of HIV-1 pre-mRNA splicing. EMBO J. 18, 4060–4067.
Cartegni, L., Chew, S.L., and Krainer, A.R. (2002). Listening to silence and
understanding nonsense: exonic mutations that affect splicing. Nat. Rev.
Genet. 3, 285–298.
Cartegni, L., Hastings, M., Calarco, J.A., et al. (2006). Determinants of exon 7
splicing in the spinal muscular atrophy genes, SMN1 and SMN2.
Am. J. Hum. Genet. 78, 63–77.
Chacko, E., and Ranganathan, S. (2009). Comprehensive splicing graph analy-
sis of alternative splicing patterns in chicken, compared to human and
mouse. BMC Genomics 10(Suppl 1), S5.
Christofk, H., Vander Heiden, M., Harris, M.H., et al. (2008a). The M2 splice
isoform of pyruvate kinase is important for cancer metabolism and
tumour growth. Nature 452, 230–233.
Christofk, H., Vander Heiden, M., Wu, N., et al. (2008b). Pyruvate kinase M2 is
a phosphotyrosine-binding protein. Nature 452, 181–186.
Clower, C.V., Chatterjee, D., Wang, Z., et al. (2010). The alternative splicing
repressors hnRNP A1/A2 and PTB influence pyruvate kinase isoform
expression and cell metabolism. Proc. Natl Acad. Sci. USA 107, 1894–1899.
deregulate pyruvate kinase mRNA splicing in cancer. Nature 463, 364–368.
Dombrauckas, J.D., Santarsiero, B.D., and Mesecar, A.D. (2005). Structural
basis for tumor pyruvate kinase M2 allosteric regulation and catalysis.
Biochemistry 44, 9417–9429.
Goncalves, V., Matos, P., and Jordan, P. (2008). The beta-catenin/TCF4
pathway modifies alternative splicing through modulation of SRp20
expression. RNA 14, 2538–2549.
He, X., Ee, P.L., Coon, J.S., et al. (2004). Alternative splicing of the multidrug
resistance protein 1/ATP binding cassette transporter subfamily gene in
ovarian cancer creates functional splice variants and is associated with
increased expression of the splicing factors PTB and SRp20. Clin. Cancer
Res. 10, 4652–4660.
Hitosugi, T., Kang, S., vander Heiden, M.G., et al. (2009). Tyrosine phosphoryl-
ation inhibits PKM2 to promote the Warburg effect and tumor growth. Sci.
Signal. 2, ra73.
Hua, Y., Vickers, T.A., Okunola, H.L., et al. (2008). Antisense masking of an
hnRNP A1/A2 intronic splicing silencer corrects SMN2 splicing in transgenic
mice. Am. J. Hum. Genet. 82, 834–848.
Jia, R., Liu, X., Tao, M., et al. (2009). Control of the papillomavirus early-to-late
switch by differentially expressed SRp20. J. Virol. 83, 167–180.
Jia, R., Li, C., McCoy, J.P., et al. (2010). SRp20 is a proto-oncogene critical for
cell proliferation and tumor induction and maintenance. Int. J. Biol. Sci. 6,
Karni, R., Hippo, Y., Lowe, S.W., et al. (2008). The splicing-factor oncoprotein
SF2/ASF activates mTORC1. Proc. Natl Acad. Sci. USA 105, 15323–15327.
Letunic, I., Copley, R.R., and Bork, P. (2002). Common exon duplication in
animals and its role in alternative splicing. Hum. Mol. Genet. 11,
Noguchi, T., Inoue, H., and Tanaka, T.(1986). The M1- and M2-type isozymes of
rat pyruvate kinase are produced from the same gene by alternative RNA
splicing. J. Biol. Chem. 261, 13807–13812.
Paz, I., Akerman, M., Dror, I., et al. (2010). SFmap: a web server for motif analy-
sis and prediction of splicing factor binding sites. Nucleic Acids Res.
Schaal, T.D., and Maniatis, T. (1999). Selection and characterization of
pre-mRNA splicing enhancers: identification of novel SR protein-specific
enhancer sequences. Mol. Cell. Biol. 19, 1705–1719.
Smith, C.W. (2005). Alternative splicing—when two’s a crowd. Cell 123, 1–3.
Smith, C.W., and Nadal-Ginard, B. (1989). Mutually exclusive splicing of alpha-
tropomyosin exons enforced by an unusual lariat branch point location:
implications for constitutive splicing. Cell 56, 749–758.
Takenaka, M., Yamada, K., Lu, T., et al. (1996). Alternative splicing of the pyr-
uvate kinase M gene in a minigene system. Eur. J. Biochem. 235, 366–371.
Vander Heiden, M., Cantley, L., and Thompson, C.B. (2009). Understanding the
Warburg effect: the metabolic requirements of cell proliferation. Science
Yeo, G., and Burge, C.B. (2004). Maximum entropy modeling of short sequence
Exon-centric regulation of PK-M alternative splicingJournal of Molecular Cell Biology