An internal ribosome entry site in the coding region of tyrosyl-DNA phosphodiesterase 2 drives alternative translation start

Tyrosyl-DNA phosphodiesterase 2 (TDP2) is a multifunctional protein that has been implicated in a myriad of cellular pathways. Although most well-known for its phosphodiesterase activity removing stalled topoisomerase 2 from DNA, TDP2 has also been shown to interact with both survival and apoptotic mitogen-activated protein kinase (MAPK) signaling cascades. Moreover, it facilitates enterovirus replication and has been genetically linked to neurological disorders ranging from Parkinson's disease to dyslexia. To accurately evaluate TDP2 as a therapeutic target, we need to understand how TDP2 performs such a wide diversity of functions. Here, we use cancer cell lines modified with CRISPR/Cas9 or stably-expressed TDP2-targeted shRNA and transfection of various TDP2 mutants to show that its expression is regulated at the translational level via an internal ribosome entry site (IRES) that initiates translation at codon 54, the second in-frame methionine of the TDP2 coding sequence. We observed that this IRES drives expression of a shorter, N-terminally truncated isoform of TDP2, ΔN-TDP2, which omits a nuclear localization sequence. Additionally, we noted that ΔN-TDP2 retains phosphodiesterase activity and is protective against etoposide-induced cell death, but co-immunoprecipitates with fewer high-molecular-weight ubiquitinated peptide species, suggesting partial loss-of-function of TDP2's ubiquitin-association domain. In summary, our findings suggest the existence of an IRES in the 5′ coding sequence of TDP2 that translationally regulates expression of an N-terminally truncated, cytoplasmic isoform of TDP2. These results shed light on the regulation of this multifunctional protein and may inform the design of therapies targeting TDP2 and associated pathways.

ized function is its ability to remove stalled topoisomerase 2 (TOP2) 3 moieties from the ends of DNA by cleaving the bond between the active site tyrosine of TOP2 and the 5Ј-phosphate of DNA, allowing religation of DNA and preventing the formation of double-stranded breaks (8). Tyrosyl-DNA phosphodiesterase 2 (TDP2) phosphodiesterase activity is not restricted to protein-DNA bonds, but can also act on protein-RNA junctions. This was documented when TDP2 was discovered to be synonymous with Vpg unlinkase, which removes the protein Vpg from the 5Ј end of incoming viral RNA during enterovirus infections (9).
In cancer, TDP2 has been suggested to be both pro-and antiproliferative. TDP2 is overexpressed in up to 80% of nonsmall cell lung carcinoma cell lines (NSCLC), and its knockdown is correlated with decreased proliferation in cell culture, reduced tumor xenograft size, and decreased phosphorylation of the mitogen-activated protein kinase (MAPK) pathway components, RAF-1, MEK-1/2, and ERK-1/2, whereas overexpression of TDP2 by lentivirus correlates with the opposite phenotype (6). Furthermore, increased cytoplasmic immunohistochemical staining of TDP2 has been correlated to NSCLC disease stage in patient samples (6). Conversely, in osteosarcoma cell lines, such as U2OS and Saos-2, which normally express low levels of TDP2, overexpressing TDP2 causes decreased proliferation in culture, reduced colony formation, increased caspase activation, and G 2 /M phase cell cycle stalling (7).
Internal ribosome entry sites are structured regions of RNA that are capable of directly recruiting the eukaryotic ribosome to initiate translation in a cap-independent manner. First characterized in picornaviruses, the utilization of viral internal ribosome entry site (IRESs), in combination with their inhibition of cap-dependent host translation, allows the virus to effectively redirect the translation machinery toward synthesis of their own proteins (10). Recently, IRESes have also been controver-sially reported in select cellular mRNAs, several of which encode isoforms of stress-response proteins whose functions are essential during situations that otherwise suppress cap-dependent translation, such as p53 after DNA damage (11,12) and HSP70 following heat shock (13).
Here we present our findings on the regulation of TDP2 expression at the translational level. A novel IRES in the 5Ј coding sequence (CDS) of TDP2 produces an N-terminally truncated isoform (⌬N-TDP2), corresponding to residues 54 -362, which lacks its N-terminal nuclear localization sequence (NLS) and part of its previously reported ubiquitinassociation (UBA) domain. This ⌬N isoform is localized diffusely throughout the cell, in contrast to full-length TDP2 (FL-TDP2), which is predominantly localized to the nucleus. ⌬N-TDP2 retains its phosphodiesterase activity both in vitro and in stable cell lines, and has slightly higher affinity for TNF receptor-associated factor 3 (TRAF3), but greatly reduced binding to high-molecular-weight ubiquitinated species. These results expand our understand of the regulation of a protein that can mediate and participate in a diversity of cellular processes, and aid in the rational design of therapies targeted toward TDP2 and its associated pathways.
We then replicated the experiment using CRISPR/Cas9 to knockout TDP2 in A549 and U2OS cells, and also found that A549 cells exhibited a growth defect without TDP2, whereas U2OS cells did not. We additionally tested CRISPR/Cas9 clones that appeared heterozygous for TDP2 knockout by immunoblot and found a positive correlation between TDP2 expression and proliferation in A549 cells but not U2OS cells (Fig. 1, D and  E). Although our sample size is limited, our data suggests that expression of the shorter isoform of TDP2 is correlated with decreased proliferation upon TDP2 depletion, and confirms previously published data that the effects of TDP2 knockdown are cell-line dependent (6,7).

The shorter isoform lacks exon 1 but not exon 2 sequences of TDP2
To identify the composition of the shorter TDP2 isoform, we immunoprecipitated both species from A549 cells, separated the isoforms by SDS-PAGE ( Fig. 2A), and subjected the bands from a corresponding silver-stained gel to MS coverage analysis. The resulting identified tryptic peptides from the upper band mapped to the entirety of TDP2, whereas the lower band lacked peptide sequences mapping to exon 1 (Fig. 2B), which encodes codons 1-55.
We then tested the ability of different Cas9 sgRNAs to disrupt TDP2 expression. 293T and A549 cells both express the shorter isoform of TDP2 in addition to FL-TDP2 (Fig. 2C), although A549 cells express a higher ratio of 43-kDa TDP2 to FL-TDP2. When Cas9 was targeted to exon 2 of TDP2 ( 182 ctccaccggaggctcgaagt 201 ), all of the clones that lost expression of FL-TDP2 also lost expression of the shorter isoform (Fig. 2C), however, when Cas9 was targeted to exon 1 of TDP2 ( 4 gagttggggagttgcctgga 23 ), the majority of clones lost only FL-TDP2 expression (Fig. 2D). This is consistent with the 43-kDa TDP2 isoform lacking sequences from exon 1 and therefore being independent of frameshifts introduced to exon 1.

RNA-seq datasets and qPCR do not suggest the shorter TDP2 isoform is a product of an alternative mRNA transcript
To investigate the possibility of the 43-kDa TDP2 isoform being a product of a second mRNA species lacking exon 1, we downloaded and mapped previously published RNA-seq datasets of A549 and U2OS whole cell mRNA from NCBI GEO (14). We analyzed two U2OS samples from separate studies and two A549 samples also from separate studies (15)(16)(17)(18) for reads mapping to the TDP2 locus (Fig. S1). There were no observable differences in the number of reads mapping to exon 1 of TDP2 in the A549 samples when compared with the U2OS samples, despite A549 cells expressing significantly more 43-kDa TDP2 compared with U2OS cells. Sashimi plots mapping exon-exon junction reads also did not suggest exon 1 skipping or the inclusion of alternative exons in any dataset analyzed (Fig. S1A).
Additionally, we designed multiple primers targeting exon 1 and downstream exons of TDP2 (Fig. S1B) and utilized these in qPCR analysis of U2OS and A549 whole RNA-derived cDNA. ⌬⌬C t calculations were modified to reflect the ratio of exon 1 to downstream exons in A549 versus U2OS No decrease was observed in the ratio of exon 1 normalized to any downstream exon in the A549 samples compared with the U2OS samples (Fig. S1, C and D), making it unlikely that there are any major alternative mRNA species of TDP2 lacking exon 1 in A549 cells.

The shorter isoform of TDP2 is expressed from FL-TDP2 cDNA but loses N-terminal tags
Transient expression of WT FL-TDP2 cDNA in 293T cells produces both isoforms (Fig. 3, A and B), as previously observed (6). We utilized this expression pattern to test for mutations that might disrupt expression of the shorter isoform by cloning mutant TDP2 sequences into either an N-terminal 3ϫ FLAG tag construct (pCMV3ϫFLAG Fig. 3A) or a C-terminal His 6 tag construct (pcDNAv5HisA Fig. 3B), and assaying for changes in expression of the shorter isoform. Of note, samples introduced with N-terminally 3ϫ FLAG-tagged FL-TDP2 cDNA (Fig. 3A,

Characterization of an IRES in TDP2
FL) generated a separate band above endogenous FL-TDP2, corresponding to FL-TDP2 tagged with 3ϫ FLAG, but no separate band above the shorter isoform. Instead, the band intensity of the 43-kDa band increased, suggesting that the shorter isoform was still expressed from the cDNA but without the tag, consistent with the shorter isoform being an N-terminally truncated product. When tagged at the C terminus, both isoforms being translated from the introduced cDNA retained their tags (Fig. 3B, FL) and four distinct TDP2 bands were produced. This led us to name the shorter band, ⌬N-TDP2.

Expression of ⌬N-TDP2 depends on the second in-frame methionine codon, Met 54
The second in-frame methionine of TDP2 is codon 54. Omission of the first 53 amino acids would be consistent with the size difference between the isoforms seen by immunoblot, so we hypothesized that the shorter isoform might be an alternatively translated product beginning from the second methionine. Silent mutations changing the Kozak sequence, which aids ribosomal recognition of a translation-initiating codon, flanking Met 54 had no effect on the increase in band intensity of ⌬N-TDP2 (Fig. 3A, G-1A M54); however, M54A (Fig. 3A, M54A) or M54L (Fig. 3, A and B, M54L) mutations abolished expression of ⌬N-TDP2 from the introduced cDNA constructs, indicating that expression of ⌬N-TDP2 depended on the second in-frame AUG codon. Directly omitting codons 1-53 of the C-terminally His 6 -tagged TDP2 also produced a ⌬N-TDP2 product running at the same size by immunoblot as that being translated from full-length TDP2 cDNA (Fig. 3,  A and B, ⌬N).

Characterization of an IRES in TDP2
To extend these findings, we generated a 293T M54L-TDP2 cell line using Cas9 and the sgRNA targeted toward exon 1 of TDP2 co-transfected with a segment of donor DNA containing the M54L mutation (Fig. 3C). The resulting cell line exclusively expressed FL-TDP2 (Fig. 3D). PCR amplification and sequencing of the TDP2 locus from genomic DNA using primers 97 bp upstream and 111 bp downstream of the introduced donor DNA (Fig. 3C, black arrows) returned a single, clean sequence containing the expected M54L mutation (Fig. S2), although the decrease in expression of FL-TDP2 seen in whole cell lysates suggested that a subset of TDP2 alleles were also rendered incapable of producing either isoform by Cas9.

Translation of ⌬N-TDP2 from FL-TDP2 cDNA is independent of ribosomal scanning from the 5-end of mRNA
To test whether ⌬N-TDP2 is initiated from the second inframe methionine due to readthrough of the Met 1 AUG during ribosomal scanning from the 5Ј cap, we introduced either three out-of-frame AUG codons flanked by Kozak residues in the ϩ1 reading frame relative to the Met 1 AUG (Fig. 4, 3ϫ ATG) (19)) immediately upstream of the Met 1 start codon in the pcDNAv5HisA backbone. Expression of this construct in 293T cells demonstrated that 3ϫ ATG selectively eliminates expression of FL-TDP2, whereas the 30-bp HP and 34-bp HP selectively decrease expression of FL-TDP2 (Fig. 4), suggesting that expression of ⌬N-TDP2 is not dependent on ribosome scanning from the 5Ј mRNA end.
A similar analysis was carried out in vitro using T7 RNA polymerase transcription of various FL-TDP2 cDNA clones with a His tag at the C terminus coupled to translation in mRNA-dependent rabbit reticulocyte lysate (Fig. S3). In this system, we could be certain that all the mRNAs were fulllength, because they were transcribed from the T7 RNA polymerase promoter in the FL-TDP2 cDNA constructs. Consistent with the results obtained from cell transfections, we found that the FL cDNA clone generated both FL-TDP2 and ⌬N-TDP2, whereas the M54L cDNA clone produced only FL-TDP2, and the 3ϫ ATG TDP2 clone only ⌬N-TDP2. These in vitro results make it very unlikely that the ⌬N-TDP2 expressed from the FL-TDP2 cDNA is a product of a cryptic promoter or splicing causing omission of the Met 1 codon.

The 5 CDS of TDP2 has internal ribosome entry site activity
Secondary structure prediction using RNAfold (20) and RNA structure (21) (Fig. 5A) and ⌬G values predicted by mFold (22) (Table S1) suggested that the 5Ј CDS of TDP2 between the first and second methionine codons falls in a highly structured region, so we hypothesized that translation of ⌬N-TDP2 may be driven by an IRES. We used a bicistronic Renilla and firefly luciferase construct, Figure 2. The shorter TDP2 isoform lacks sequences mapping to exon 1 of TDP2. A, both isoforms of TDP2 were immunoprecipitated from A549 cells using either no antibody (Ϫ) or a polyclonal rabbit antibody (ϩ), and input, flow-through (FT), and elution (IP) samples were immunoblotted with TDP2 4-2c. B, the two major antibody-reactive bands were excised from a corresponding silver-stained gel and digested for MS coverage analysis. Highlighted regions indicate tryptic peptides detected. C, TDP2 and tubulin co-immunoblots of lysates from clones derived from 293T (top) and A549 (bottom) cells transfected with Cas9 and a sgRNA targeting exon 2 ( 182 ctccaccggaggctcgaagt 201 ) of TDP2. * ϭ clones that lost expression of both isoforms of TDP2. D, TDP2 and tubulin coimmunoblots of lysates from clones derived from 293T (left) and A549 (right) cells transfected with Cas9 and a sgRNA targeting exon 1 ( 4 gagttggggagttgcctgga 23 ) of TDP2. MW, molecular weight marker.

Characterization of an IRES in TDP2
in which the 5Ј CDS of TDP2 was inserted between the Renilla and firefly luciferase coding regions to test for IRES activity. Renilla luciferase (RL) is driven by a CMV promoter, followed by a linker region and a downstream firefly luciferase (FFL). The expression of FFL measures the ability of the linker region to recruit ribosomes to initiate translation. We used a construct with no linker region as a negative control (pRL-FFL) and the hepatitis C virus IRES in the linker region as a positive control (pRL-HCV-FFL) (both were gifts from the lab of Dr. Richard Lloyd (23)). The 5Ј-most (pRL-TDP2 2-100-FFL) and an internal (pRL-TDP2-101-200-FFL) 297-nucleotide, 99-codon segment of the TDP2 CDS were cloned into the linker region to test for IRES activity (Fig. 5B). Dual luciferase assays and the resulting ratio of FFL to RL expression suggested that the 5Ј-most 297 nucleotides of TDP2's CDS possessed potential IRES activity (Fig. 5C). The ratio of FFL to RL expressed from the pRL-HCV-FFL construct was consistent with previously published results using the same 293T cell line (23). A and B, mutational analysis by overexpression of the TDP2 CDS and the second in-frame methionine codon using (A) an N-terminal 3ϫ FLAG tag construct and B, a C-terminal His 6 tag construct. Constructs include empty vector (ev), WT full-length TDP2 (FL), TDP2 omitting codons 1-53 (⌬N), TDP2 with its N-terminal nuclear localization signal mutated (mNLS), silent mutation of the guanine in the Ϫ1 position relative to the second in-frame methionine codon to adenosine (G-1A M54), TDP2 with the first methionine codon mutated to leucine (M1L), TDP2 with the second in-frame methionine mutated to either alanine (M54A) or leucine (M54L), and two catalytically inactive mutants (E152A and D262A). C, diagram of donor DNA containing the M54L mutation used in conjunction with exon 1-targeted Cas9 to modify 293T cells to exclusively express FL-TDP2. Black arrows indicate primers used to amplify genomic region for verification sequencing (Fig. S2). D, immunoblots of whole cell lysates from unmodified 293T cells and a 293T clone genetically modified by Cas9 to harbor the TDP2 M54L mutation. MW, molecular weight marker. pcDNAv5HisA-FL-TDP2 was modified to include sequences that disrupt ribosomal scanning upstream of the Met 1 start codon and tested for expression of FL-and ⌬N-TDP2. Constructs include adding three out-of-frame ATG codons upstream of Met 1 (3xATG) and introducing a 30-(30 bp HP) or 34-bp (34 bp HP) hairpin as previously described (19). 293T cells were transfected with the indicated construct and harvested for immunoblot. MW, molecular weight marker.

Characterization of an IRES in TDP2
We verified the integrity of the bicistronic mRNA being generated from the pRL-FFL construct by RT-PCR using the primers denoted in Fig. 5B (black arrows) on cDNA generated from 293T cells transiently transfected with the constructs. We confirmed that pRL-TDP2-2-100-FFL generated only one major band of the expected intact size (Fig. S4A), suggesting that there was no aberrant splicing of the CMV promoter to FFL. Additionally, to rule out the presence of a cryptic promoter in the linker region, 293T cells were co-transfected with the indicated pRL-FFL construct and an shRNA against the upstream RL (shRL) (Fig. S4B). If a cryptic promoter were present and able to generate monocistronic RNA products containing FFL, we would have expected shRL to preferentially repress expression of only RL. Both RL and FFL were proportionally suppressed in each construct, suggesting that the pRL-TDP2-2-100-FFL construct produced an intact bicistronic mRNA.
These data suggest that a potential IRES in the 5Ј CDS of TDP2 may drive translation initiation at the second in-frame methionine codon, Met 54 , to produce ⌬N-TDP2 from transientlytransfected FL-TDP2 cDNA. As a side note, the ⌬N-TDP2 protein being expressed off of constructs directly omitting the first 53 codons (Fig. 3, A and B, ⌬N) is likely a product of canonical translation initiation, as omission of codons 1-53 also leaves out half of the tested, putative IRES. This would also explain the lower expression levels seen of ⌬N-TDP2 protein off of FL-TDP2 cDNA versus ⌬N-TDP2 cDNA by immunoblot because canonical cap-dependent translation is generally significantly more efficient than IRESmediated translation.

Silent mutations to the 5 CDS of TDP2 correspondingly alter expression of the downstream ORF in different constructs
Because many previously reported cellular sequences that displayed positive IRES activity have now been contended to be artifacts of the bicistronic constructs used to detect them (23), we tested various silent mutations in the 5Ј CDS of TDP2 (Fig.   S5A, arrows, Table S2) in both the FL-TDP2 His-tagged cDNA construct (Fig. S5, B and C) and the bicistronic pRL-TDP2 2-100-FFL construct (Fig. S5D) to look for consistency in response of the downstream open reading frame (ORF).
The silent mutations were introduced into either the construct containing the full cDNA of TDP2 tagged with C-terminal His 6 and assayed for ⌬N-TDP2 expression relative to FL-TDP2 expression by immunoblot (Fig. S5B) and the band intensities quantified (Fig. S5C), or to the pRL-TDP2 2-100-FFL construct and assayed by dual-luciferase for FFL expression relative to RL expression (Fig. S5D). Mutation sets 2, 2.1, and 4 increased, whereas mutation set 1 decreased the ability of the IRES to initiate translation of both ⌬N-TDP2 from the TDP2 cDNA and FFL from the bicistronic RL-FFL construct, mutation set 2.2 produced no obvious change in expression of either downstream ORF, and mutation sets 3 and 5 decreased expression of ⌬N-TDP2 but made little difference in FFL expression.
The observation that these silent mutations can produce similar expression changes of a downstream ORF in two separate constructs is consistent with the hypothesis that this section of the TDP2 mRNA can regulate translation initiation. This supports the theory that the 5Ј CDS of TDP2 contains a putative IRES and that expression of FFL from the bicistronic luciferase vector is not simply an artifact of that particular construct.

⌬N-TDP2 lacks the N-terminal nuclear localization sequence in TDP2
While investigating the functional differences between FLand ⌬N-TDP2, we found a NLS in the N terminus of TDP2 using cNLS mapper (24 -26) that had been independently described (27). The pattern of the core basic NLS residues ( 23 KKRR 26 ) followed by the second in-frame methionine in TDP2 is conserved from humans through mice (Fig. S6A).

Characterization of an IRES in TDP2
To verify the functionality of the NLS in TDP2, we mutated the 23 KKRR 26 NLS motif to 23 AAAA 26 to generate mNLS-TDP2. FL-TDP2, ⌬N-TDP2, and mNLS-TDP2 were tagged with N-terminal eGFP and imaged by fluorescence microscopy (Fig. S6B). Although FL-TDP2 was predominantly nuclear, ⌬N-TDP2 and mNLS-TDP2 were localized diffusely throughout the cytoplasm and nucleus. The incomplete exclusion of ⌬N-TDP2 and mNLS-TDP2 from the nucleus is expected because both proteins are smaller than the nuclear pore diffusion limit of ϳ60 kDa.

⌬N-TDP2 is catalytically active but less protective against etoposide than nuclear TDP2
Because the catalytic domain of TDP2 is in its C terminus (residues 114 -358 (28)), we expected ⌬N-TDP2 to retain its phosphodiesterase activity. To test this, we purified GSTtagged TDP2 from bacteria and measured the ability of the different isoforms to cleave p-nitrophenyl-thymidine-5Ј-phosphate (T5PNP), as previously described (29). FL-, ⌬N-, and mNLS-TDP2 were all able to cleave T5PNP, whereas the catalytically inactive mutants, E152A and D262A, were not (Fig. 6A).
When treated with the indicated concentrations of the TOP2 poison and DNA-damaging drug, etoposide, for 72 h, FL-, ⌬N-, mNLS-, and M54L-TDP2 expression conveyed a survival advantage over the knockout MEFs and MEFs re-expressing the catalytically inactive mutant, E152A, of TDP2 (Fig. 6, C and D).
Of the cell lines re-expressing catalytically-active TDP2, consistent with their localizations, predominantly-nuclear M54L-TDP2 exhibited slightly higher tolerance to etoposide (EC 50 ϭ 70.4 nM) than ⌬N-TDP2 (EC 50 ϭ 60.3 nM), whereas mNLS-TDP2, which likely diffuses less readily into the nucleus compared with ⌬N-TDP2 due to a higher molecular weight, was the most sensitive to etoposide (EC 50 ϭ 45.9 nM).

⌬N-TDP2 and mNLS-TDP2 have increased binding affinity for TRAF3 but reduced binding to ubiquitinated high-molecularweight species
TDP2 has been previously reported to interact with ERK3 (31), TAK1 (2), and TRAF3, TRAF5, and TRAF6 (1). We investigated whether ⌬N-TDP2 had a different binding profile to these interacting proteins by transiently co-expressing FL-TDP2-6xHis with these genes cloned into FLAG-tagged constructs in 293T cells and immunoprecipitated with anti-FLAG beads. All previously reported interactors were able to co-immunoprecipitate (IP) TDP2 (Fig. 7A). Because TRAF3 appeared to be the strongest interactor by our IP conditions, we performed the reverse IP by co-expressing FLAG-tagged FL-TDP2, ⌬N-TDP2, mNLS-TDP2, M54L-TDP2, and E152A-TDP2 with

Characterization of an IRES in TDP2
His-tagged TRAF3 (Fig. 7B) and immunoprecipitating for FLAG. ⌬N-TDP2 and mNLS-TDP2 displayed a slight but consistent increase in the amount of TRAF3 pulled down compared with FL-TDP2 (13 and 28% increase, respectively) or M54L-TDP2. Because TRAF3 is primarily cytoplasmic, this result would be consistent with increased co-localization of the ⌬N-TDP2 form of TDP2 with TRAF3 in the cytoplasm leading to increased protein-protein interaction.
We also co-blotted the IPs with an anti-ubiquitin antibody (Fig. 7B, bottom left and bottom right). Met 54 falls 28 amino acids into the UBA domain of TDP2, and, as expected, ⌬N-TDP2 had significantly reduced binding to the high-molecular-weight ubiquitinated species, noted by the reduction in ubiquitin smears in the IP samples compared with FL-TDP2 or M54L-TDP2.

Discussion
Here we have presented data supporting the existence of a potential IRES in the 5Ј coding sequence of TDP2. Our results suggest that the IRES drives expression of a shorter, N-terminally truncated isoform of TDP2, ⌬N-TDP2, from an alternative translation initiation start site, the second in-frame methionine (Met 54 ). Mutation of Met 54 causes loss of ⌬N-TDP2 expression both from transiently transfected cDNA and in 293T cells as well as in an in vitro translation system. ⌬N-TDP2 differs from FL-TDP2 in that it lacks an N-terminal NLS, causing it to remain diffuse throughout the cell, whereas FL-TDP2 is predominantly nuclear. Furthermore, whereas, ⌬N-TDP2 retains catalytic activity and conveys protection against etoposide, its relocalization may give it different preferences for certain TDP2 interacting proteins, including ubiquitinated species (Fig. 8).
Recently, Huang and colleagues (32) described characterization of a shorter isoform of TDP2 present in select NSCLC cell lines, specifically A549 cells, and concluded that it was a product of an alternative transcription initiation event generating a new mRNA species lacking exons 1 and 2, whereas appending a

Characterization of an IRES in TDP2
novel first exon containing a mitochondrial targeting sequence (MTS-TDP2) (32). Their isoform and the IRES isoform we describe are 304 and 309 residues in length, respectively. The negligible size difference would make it difficult to distinguish the two possible shorter proteins by SDS-PAGE separation and immunoblotting. However, our MS data strongly suggest that the shorter isoform present in A549 cells contains peptides mapping to exon 2 of TDP2, and our CRISPR/Cas9-modified knockout cell lines also demonstrate that expression of the shorter isoform depends on exon 2 sequences remaining inframe. Analysis of RNA-seq datasets from A549 and U2OS cells also did not reveal a significant number of reads mapping to alternative exon 1 by Huang et al. (32), despite the shorter isoform being the dominant TDP2 protein species expressed in A549 cells. In addition, we have shown that expression of the shorter product depends on the Met 54 codon both in 293T cells and from transfected FL-TDP2 cDNA, because mutation of the Met 54 codon is sufficient to abrogate ⌬N-TDP2 expression. It is possible that the shorter isoform detected by immunoblot is a mixture of MTS-TDP2 and ⌬N-TDP2, or that certain cell lines express one or the other predominantly. Regardless, our data demonstrating that ⌬N-TDP2 is expressed from FL-TDP2 cDNA strongly suggest the existence of an IRES in the coding sequence of TDP2, as it is highly unlikely that MTS-TDP2 can be expressed from FL-TDP2 cDNA by splicing in an alternative first exon.
We also do not rule out the possibility that the Met 54 codon can be recognized by ribosomes via mechanisms other than internal recruitment, such as ribosomal shunting past the Met 1 codon. Our in vitro and cellular expression experiments with hairpins and decoy AUG codons inserted upstream of Met 1 argue that ⌬N-TDP2 expression is independent of ribosomal scanning from the 5Ј mRNA end, and our dual-luciferase experiments suggest that internal ribosome recruitment contributes to expression of at least a portion of the ⌬N-TDP2 detected: the pRL-TDP2 2-100-FFL construct produced a consistent 1.7-fold increase in FFL expression compared with a similar construct with no linker. It remains to be tested whether this increase varies between cell lines. However, for transient transfection of the cDNA and for cell lines that heavily overex-press ⌬N-TDP2, ribosomal shunting facilitating translation initiation at Met 54 could also account for a portion of the ⌬N-TDP2 seen. To further test how much translation of the shorter isoform is entirely cap-independent versus facilitated by mechanisms such as ribosomal shunting, future experiments should test expression of the different isoforms from uncapped, FL-TDP2 mRNA.
IRES elements are generally associated with expression of stress-response proteins, as a mechanism for cells to maintain expression of key survival genes even when cap-dependent translation is inhibited, such as expression of p53 following DNA damage (11,12). TDP2 is well documented as a DNAdamage preventing phosphodiesterase, and has been reported to interact with both growth and apoptotic pathways (3)(4)(5)(6)33) in different contexts. In NSCLC, TDP2 facilitates proliferation and tumor growth (6), whereas overexpression of TDP2 in osteosarcoma cell lines such as U2OS and Saos-2 correlate with decreased proliferation and increased apoptosis (7). A key question to ask would be how TDP2 can switch between these opposing functions. Here we have presented data suggesting that a subset of NSCLC that respond negatively to TDP2 knockdown by shRNA concomitantly express ⌬N-TDP2, a shorter, cytoplasmic isoform of TDP2 by use of an IRES in the 5Ј CDS of TDP2 mRNA. Combining this observation with previously reported data that increased cytoplasmic TDP2 staining positively correlates with the disease stage in NSCLC patient samples (6) leads to the speculation that ⌬N-TDP2 is potentially oncogenic and facilitates proliferation in certain cell lines, whereas FL-TDP2 does not. However, our analyses thus far have not revealed a distinct function for ⌬N-TDP2 compared with FL-TDP2 that would convey a proliferation advantage. The next step will be to identify conditions under which the utilization of the TDP2 IRES can be induced, and determine how much the nuclear-cytoplasmic redistribution contributes to the regulation of TDP2's potential growth-inducing and growth-inhibiting functions.
We have taken preliminary steps in looking for differences in protein interactors between FL-TDP2 and ⌬N-TDP2. We found a very modest preference for ⌬N-TDP2 to interact with TRAF3 when compared with FL-TDP2 or M54L-TDP2. This result would be expected as TRAF3 is cytoplasmically located, and should therefore be in closer proximity to cytoplasmic TDP2. Interestingly, however, TDP2 has been previously shown to inhibit NF-B activation by stimulation with CD40L (1), which activates NF-B through the noncanonical pathway dependent on TRAF3 degradation and subsequent NIK stabilization (34). We are performing ongoing experiments to determine whether the two isoforms can differently affect responses to CD40L.

Characterization of an IRES in TDP2
cin and 1ϫ NEAA and ATCC recommended levels of sodium pyruvate. All transient transfections were done using 1:3 g of DNA:l 1 mg/ml of PEI in OptiMEM.

Stable cell line generation
shRNAs were from the Mission shRNA library (Sigma). Lentivirus particles were generated using 293T cells and infectious media used with 8 g/ml of Polybrene for subsequent infection followed by puromycin selection (1 g/ml). TDP2 ⌬ex1-3/⌬ex1-3 MEFs re-expressing TDP2 variants were generated by cloning the corresponding TDP2 isoform into pCAG-GFP and generating retrovirus for infection as previously described (35), followed by FACS to isolate GFPϩ cells. Cas9 guide RNAs were designed using CRISPR and px458 constructs and cell lines were generated as previously described (36). Donor DNA used to introduce the M54L mutation is documented in Fig. 3C. Clones were screened using the MyTaq Extract-PCR kit (Bioline) and products digested to check for loss of HaeIII digestion indicating incorporation of donor DNA.

Proliferation and survival assays
Cells were seeded into 96-well plates at 3,000 cells/well and allowed to adhere overnight before being switched to 2% FBS/ DMEM for proliferation assays or 10% FBS/DMEM supplemented with the indicated concentrations of etoposide. Etoposide (VP16, Selleckchem S1225) stocks were prepared in DMSO and diluted to indicated concentrations. Crystal violet staining was performed as previously described (37).

Mass spectrometry
Silver-stained gel slices were first destained and then reduced and alkylated with 10 mM tris(2-carboxyethyl)phosphine hydrochloride (Roche Applied Science) and 55 mM iodoacetamide (Sigma), respectively. In-gel digestion was performed overnight with trypsin (Promega) according to the manufacturer's specifications. The protein digests were pressure-loaded onto 250-m inner diameter fused silica capillary (Polymicro Technologies) columns with a Kasil frit packed with 3 cm of 5 m C18 resin (Phenomenex). After desalting, each loading column was connected to a 100-m inner diameter fused silica capillary (Polymicro Technologies) analytical column with a 5-m pulled-tip, packed with 12 cm of 5-m C18 resin (Phenomenex).
Each split column was placed in-line with a 1200 quaternary HPLC pump (Agilent Technologies) and the eluted peptides were electrosprayed directly into an LTQ Orbitrap XL mass spectrometer (Thermo Scientific). The buffer solutions used were 5% acetonitrile, 0.1% formic acid (buffer A) and 80% acetonitrile, 0.1% formic acid (buffer B). The 120-min elution gradient had the following profile: 10% buffer B beginning at 10 min to 40% buffer B at 70 min, 60% buffer B at 90 min, and then 100% buffer B at 100 min continuing to 110 min. A cycle consisted of one full scan mass spectrum (400 -1600 m/z) in the Orbitrap at 60,000 resolution followed by five data-dependent collision-induced dissociation (CID) MS/MS spectra in the LTQ. Charge state screening was enabled and unassigned charge states and charge state 1 were rejected. Dynamic exclusion was enabled with a repeat count of 1, a repeat duration of 30 s, an exclusion list size of 500, and an exclusion duration of 180 s. Dynamic exclusion early expiration was enabled with an expiration count of 3 and an expiration signal-to-noise ratio of 3. Application of mass spectrometer scan functions and HPLC solvent gradients were controlled by the Xcalibur data system (Thermo Scientific).
MS/MS spectra were extracted using RawXtract (version 1.9.9.2) (38). The Integrated Proteomics Pipeline (IP2) (Integrated Proteomics Applications) was used to generate filter peptide-spectra matches. MS/MS spectra were searched with the ProLuCID (version 1.3.5) algorithm (39) against a human UniProt protein database downloaded on March 03, 2014 that had been supplemented with common contaminants and concatenated to a decoy database in which the sequence for each entry in the original database was reversed (40). A total of 177,652 protein entries were searched. Precursor mass tolerance was 50 ppm and fragment mass tolerance was 600 ppm. For protein identifications, the ProLuCID search was performed using no enzyme specificity and static modification of cysteine due to carboxyamidomethylation (57.02146). ProLu-CID search results were assembled and filtered using the DTA-Select (version 2.1.3) algorithm (41), requiring at least partial enzyme specificity (cleavage C-terminal to Arg or Lys residue) and a minimum of two peptides per protein identification. The number of missed cleavages was not specified. The protein identification false positive rate was kept below 1% and all peptide-spectra matches had less than 10 ppm mass error. DTASelect assesses the validity of peptide-spectra matches using the

Characterization of an IRES in TDP2
cross-correlation score (XCorr) and normalized difference in cross-correlation scores (␦CN). The search results are grouped by charge state and tryptic status and each subgroup is analyzed by discriminant analysis based on a nonparametric fit of the distribution of forward and reversed matches.

Fluorescence imaging
U2OS cells were plated onto glass coverslips and transiently transfected with the indicated eGFP constructs as described above. 24 -48 h after transfection, cells were washed with PBS and fixed in 4% formaldehyde in PBS for 20 min. Slides were washed once in PBS and twice in TBS then blocked and permeabilized in TBS ϩ 3% FBS ϩ 0.1% Triton X-100 for 10 min then washed twice again TBS. Cells were then stained with 1 g/ml of Hoechst 33342 (Invitrogen H3570) in TBS for 10 min, washed 3 times with TBS, and mounted onto slides with Pro-Long Gold Antifade (Invitrogen P36930). Slides were imaged using a Zeiss 710 confocal microscope.

RNA preparation and qPCR
RNA was isolated from cells and cells transfected with the indicated constructs using the NucleoSpin RNA extraction kit (Macherey-Nagel 740955). All purified RNA products were subjected to an additional DNase treatment followed by isopropyl alcohol precipitation and 70% ethanol wash to ensure removal of genomic contamination. Reverse transcription was done using SuperScript III (Invitrogen 18080093). qPCR was conducted using SYBR Green (Thermo Science). Control wells with cDNA preparations lacking addition of the reverse transcriptase enzyme were included for all samples to ensure no genomic DNA contamination.

In vitro transcription and translation
In vitro transcription and translation was done using the TNT Quick Coupled T7 Transcription/Translation System (Promega L1171). The resulting protein solution was diluted with 50 mM Tris-HCl, pH 8, ϩ 150 mM NaCl ϩ 0.5% Nonidet P-40 ϩ 10 mM imidazole supplemented with 1 mM EDTA, 1 mM Na 2 VO 3 , 10 mM NaF, 10 mM ␤-glycerophosphate, 1 g/ml of pepstatin, 1 g/ml of aprotinin, and 20 g/ml of leupeptin to a final volume of 1 ml. 50 l of TALON Metal Affinity Resin (Takara 635501) was added and samples were incubated for 2 h at 4°C with rotation. Beads were washed 3 times with 50 mM Tris-HCl, pH 8, ϩ 150 mM NaCl ϩ 0.5% Nonidet P-40 ϩ 10 mM imidazole then boiled in 50 l of 1ϫ SDS buffer.

Immunoprecipitation and pulldowns
For MS coverage analysis, A549 cells were lysed by sonication in 50 mM Tris-HCl, pH 8, ϩ 400 mM NaCl supplemented with 1 g/ml of pepstatin, 1 g/ml of aprotinin, and 20 g/ml of leupeptin. Cell debris was cleared by centrifugation and supernatants were precleared with protein A beads before incubating with either no antibody or H300 rabbit polyclonal ␣TDP2 antibody followed by protein A bead capture. Beads were washed 4 times in lysis buffer and eluted by boiling in 1ϫ SDS buffer. Eluate was run out on a 10% bis-acrylamide gel and silver stained.
FLAG-tagged protein constructs were transiently co-transfected with the construct of interest into 293T cells. 24 h posttransfection, cells were switched to media containing 0.1 M bortezomib for 16 h, after which cells were rinsed in PBS and scraped off into 1.1 ml of 50 mM Tris-HCl, pH 8, ϩ 150 mM NaCl ϩ 0.5% CHAPS supplemented with 1 mM EDTA, 1 mM EGTA, 1 mM Na 2 VO 3 , 10 mM NaF, 10 mM ␤-glycerophosphate, 1 g/ml of pepstatin, 1 g/ml of aprotinin, and 20 g/ml of leupeptin. Cells were allowed to swell on ice for 15 min and disrupted by trituration by pipetting up and down slowly three times with a P1000 pipette tip. Cell debris was pelleted in a microcentrifuge. Part of the resulting supernatant was taken as an input sample, and 30 l of ␣FLAG M2 Affinity Gel bead slurry (Sigma A2220) added to the remaining and incubated at 4°C for 2 h with rotation. Beads were washed 4 times with lysis buffer and eluted by boiling in 100 l of 1ϫ SDS buffer.
His pulldowns were conducted by lysing cells and adding 10 mM imidazole with 25-50 l of TALON Metal Affinity Resin (Takara 635501). Samples were incubated at 4°C for 2 h with rotation. Beads were washed 3-4 times with lysis buffer ϩ 10 mM imidazole and eluted by boiling in 1ϫ SDS buffer.