Hairpin-end conformation of adeno-associated virus genome determines interactions with DNA-repair pathways.

The palindromic terminal repeats (TRs) of adeno-associated virus (AAV) form DNA hairpins (HPs) are essential for replication and for priming the conversion of single-stranded virion DNA to double strand. In recombinant AAV (rAAV) gene-delivery vectors, they are targets for the DNA-repair pathways leading to circularization, concatemerization and, infrequently, chromosomal integration. We investigated the effect of the TR HP on recombination by comparing specific DNA substrates transfected into wild-type and DNA-repair-deficient cells. DNA molecules with the TR sequences constrained in the T-shaped HP conformation at one or both ends were subject to a loss of gene expression, which was partially relieved in ataxia telangiectasia mutated (ATM(-/-)) cells. The ATM-dependent effect was mediated by transcriptional silencing of a subset of HP-containing molecules in cis rather than a loss of DNA, and was dependent on the specific T-shaped structure of the HP and not the primary sequence. DNA molecules with simple U-shaped HP ends were unaffected by ATM-dependent silencing. The silenced molecules remained in a linear conformation, in contrast to the expressed molecules, which were circularized. In the absence of ATM activity, this subset remained linear but was actively expressed. DNA molecules with the TR sequence in the open duplex conformation, or without TR sequences, were unaffected by ATM mutation and were predominantly converted to circular forms. A separate HP-specific effect in normal cells resulted in a loss of DNA substrate in the nucleus and was ATM independent. These results suggest that the presence of the HP structure on rAAV vector genomes subjects them to specific, and sometimes unproductive, DNA-repair/recombination pathways.

INTRODUCTION

One of the defining features of the adeno-associated virus (AAV) genome is the palindromic, GC-rich terminal repeat (TR) at each end of the genome. These stable T-shaped hairpins play a vital role in AAV biology, serving as replication origins and priming sites for conversion of single-stranded virion DNA to double-stranded DNA (dsDNA) templates for gene expression. They are the only essential cis acting components of recombinant AAV (rAAV) gene delivery vectors. The TR at the 3’ end of the virion DNA serves as primer for DNA synthesis, and remains constrained in the hairpin conformation in the duplex molecule. The TR at the original 5’ end of the genome is thought to be displaced by the elongating replication fork to form a palindromic dsDNA end. Because the TRs form the ends of the linear genome, they are also targets for DNA recombination, typically circularization and concatemerization, but infrequently also chromosomal integration.

The potential for genotoxicity stemming from rAAV vector DNA integration raises the question as to whether the hairpin structures formed by the TRs are especially recombinogenic compared to other forms of DNA ends.[1, 2] Nakai et al compared integration of DNA molecules with and without AAV TR sequences by hydrodynamic transfection into mouse liver cells and found little difference between them.[3] While this suggested that the TR palindromes did not promote integration, the DNA molecules used in this experiment were not constrained in the hairpin structural conformation like those of rAAV vector genomes.

The TR ends of the linear AAV genome are recognized by the host cell as DNA double-strand breaks (DSB), and recombination events are mediated by any of several DSB repair pathways. Previous studies have demonstrated the participation of both non-homologous end-joining (NHEJ) and homologous recombination (HR) in the circularization and concatemerization of rAAV genomes. The NHEJ pathway depends on the activity of DNA-PKCS, a PI3-like kinase that is critical for rAAV recombination in non-dividing cells such as myocytes, though it appears to be less important in hepatocytes and dividing cells.[4-8] In contrast, HR is the dominant recombination pathway in S-phase, and is orchestrated through the actions of another PI3-like kinase protein, ATM.[9] A third member of the PI3-like kinase family, ATR (ATM-and rad3-related), is more specifically involved in repairing DNA lesions involving stalled replication forks or significant regions of ssDNA sequence.

We have recently reported that rather than promoting recombination of rAAV ends, wt ATM activity leads to silencing of gene expression from a large fraction of rAAV genomes.[10] Further, these molecules remain linear rather than recombining to form circles. In the absence of ATM, these genomes are expressed normally, but remain linear. This suggests that they are committed to a different pathway or compartment from the genomes that are normally expressed and circularized. Similarly, ATR activity led to a smaller degree of silencing of conventional single-strand rAAV vectors (ssAAV), but not self-complementary AAV (scAAV) vectors, which do not expose single-stranded DNA in the nucleus.[11, 12]

In this study, we test the effect of the TR hairpin structure on recombination by transfecting cells with linear DNA substrates having covalently closed hairpin TRs, or fully duplex molecules with or without the palindromic TR sequences. Transfected linear DNA can be circularized by both NHEJ and HR pathways.[13] We find that a single AAV TR in the hairpin conformation in cis is necessary and sufficient to mediate the previously observed ATM-dependent silencing of gene expression from rAAV genomes. This effect is independent of the primary sequence, but requires the specific T-shaped secondary structure of the terminal repeat. There is also a separate hairpin-dependent, ATM-independent, loss of substrate DNA molecules.

RESULTS

Terminal repeat structure affects the fate of transfected DNA

In order to determine whether the AAV TR hairpin structures are more likely to undergo recombination than other forms of DNA ends, we constructed 3 different DNA molecules that contained the TR sequences either in a covalently closed hairpin conformation (TRHP), or an open DNA duplex conformation (TR+), or containing no TR sequences (TR-) (. The plain linear molecules were made simply by cutting the vector sequences out of plasmid constructs and filling-in to produce blunt ends, while the covalently closed molecules were made using a previously described procedure for creating no-end AAV DNA substrate.[14] This molecule, containing TR hairpins at both ends, has previously been used to reconstitute AAV Rep-dependent DNA replication in vitro and yielded all of the predicted products for rolling hairpin replication.[15]

To measure circularization, two different GFP vector configurations were used: one with the intact GFP cassette with an intron in the middle of the coding region (, and a second with a previously described circularization-dependent (CD) arrangement (.[16] This had the left and right halves of GFP (split through the intron) at the ends of the DNA molecules, such that expression would be possible only after circularization brought the two ends together. This allowed the frequency of recombination between the different end-structures to be measured by comparing expression from the intact construct, which is unaffected by recombination, and the CD construct, which is dependent on recombination for expression.

Substrate DNAs were transfected using polyethylenimine (PEI), which requires endocytosis and acidification of early endosomes to release the DNA into the cytoplasm prior diffusion into the nucleus.[17, 18] The transfections were set up with supercoiled plasmid DNA (pSP72) as carrier and supercoiled red fluorescent protein-coding plasmid (pDSRed2-C1) to normalize transfection efficiency. The molar ratios of the three molecules were first titrated using supercoiled GFP-expressing plasmid, so as to deliver approximately 1-2 molecules of fluorescent protein gene per transfected cell nucleus; i.e., approximately half of the transfected cells expressed only green or only red fluorescent protein, and half expressed both. This allowed accurate quantization of DNA fate since a GFP signal would be the product of a single-molecular event. Because the carrier and RFP plasmids were supercoiled, they were unlikely to interact with the experimental GFP-DNA ends, or to induce DNA damage signaling.

After transfection into Hela cells, the DNA with TRs in the closed hairpin conformation yielded a lower frequency of GFP expression than DNA with either open duplex TRs, or no TR sequences (). This was the case whether they were associated with the intact GFP, or the CD-GFP gene. This surprising result suggested that the presence of the hairpin structure leads to degradation or silencing.

AAV TR in hairpin conformation induces an ATM-dependent decrease in gene expression from transfected DNA substrates

We previously described an ATM-dependent loss of functional rAAV genomes in wild type (WT) fibroblasts compared to ATM-/- fibroblasts transduced either with ssAAV or scAAV vectors.[10] A dsDNA molecule with a TR in the hairpin conformation is an intermediate in the fate of both ssAAV and scAAV genomes, respectively generated by second strand DNA synthesis or by self-complementary base-pairing. Therefore, we investigated whether the decrease in GFP expression from transfected dsDNA molecules with TRHP ends was ATM-dependent.

Wild-type transformed fibroblasts (GM00637 J), and well characterized transformed ATM-/- fibroblasts (GM05849 E),[19] were transfected with the three DNA molecules containing different end structures, carrying either the GFP or GFP-CD expression cassette (). While the two open end molecules (TR+ and TR-) were unaffected by the absence of ATM, the number of cells expressing GFP from the hairpin molecules was increased by 2-fold in ATM-/- cells compared to WT cells. This was similar to our previously reported effect of ATM on transduction with rAAV vectors, though less pronounced (2.6- and 6.6-fold increases for ssAAV and scAAV vector transducution, respectively). The transfection-based experiment suggested that the ATM-dependent decrease in gene expression requires the presence of the AAV terminal repeats specifically in the hairpin conformation.

While expression from the intact GFP TRHP substrate increased by 2-fold in ATM-/- cells compared to WT, there was no corresponding increase from the GFP-CD molecules with hairpin ends, suggesting that the TRHP substrates that became available for expression in the absence of ATM do not circularize. This also parallels the results from our viral transduction-based studies, where the number of cells expressing intact GFP increased in the absence of ATM, while the number that expressed GFP from a circularized genome remained the same. In contrast to the HP-containing molecules, ATM-deficiency caused a small increase in the circularization of substrates with either TR+ or TR- open-blunt ends, suggesting that ATM is not required for processing these types of ends.

Interestingly, while gene expression in ATM-/- cells transfected with TRHP substrates increased compared to WT, it did not reach the same levels as cells transfected with TR+ or TR- substrates, suggesting that there is an additional TRHP-dependent loss of gene expression that is independent from the ATM pathway (see below).

A single AAV TR in the hairpin conformation is necessary and sufficient for loss of gene expression

The expected product of conversion of an AAV genome into dsDNA would be a linear molecule with a closed hairpin TR at one end and an open duplex at the other. To determine whether this structure is subject to the same ATM-dependent silencing as observed with our double-hairpin TRHP substrate, we cut off one or both hairpins by restriction enzyme digestion prior to transfection into WT and ATM-/- fibroblasts (.

While cutting off one hairpin from the substrate had no impact on the reduced number of GFP positive cells, removing both TRs restored its behavior to exactly that of the unmodified DNA fragment that contained no TR sequences (TR-). An increase in the number of GFP positive cells upon removal of both hairpins was observed in both WT and ATM-/- cells, though the effect was significantly greater in WT cells. This indicated that the presence of one AAV TR in the hairpin conformation was sufficient for the substrate to interact with an ATM-dependent pathway, leading to a significant loss of expression. Further, the additional increase in GFP positive cells in the absence of ATM upon removal of the hairpin ends suggested that there was a second, ATM-independent loss of functional DNA substrate when it was associated with a TR in the hairpin conformation.

T-shaped TR hairpin conformation, but not primary sequence, induces ATM-dependent decrease in gene expression

To determine how the secondary structure and/or primary sequence of the TR contribute to the ATM-dependent effect on gene expression, we designed a series of palindromic oligonucleotides with modified primary sequence, but preserving the secondary structure. These hairpin oligonucleotides were ligated onto both ends of linear TR- DNA molecules, generating covalently closed GFP-expressing DNA substrates. Substrate wtTRHP (B+C) contained the T-shaped HP formed by the wild-type B and C palindromes but lacked the A palindrome portion (). Substrate revTRHP contained the T-shaped TRs with the sequence of the wild-type AAV TR reversed 5’ to 3’ (not the reverse complement). Substrate TRHP(AT) contains T-shaped TRs with a portion of the GC base-pairs substituted with AT such that it is no longer a GC-rich sequence. Substrate Simple TRHP(AT) contains TRs with an AT-rIch sequence that fold into a simple U-shaped hairpin secondary structure.

These substrates, and the previously described TR+, TR- and TRHP, were transfected into HeLa cells in presence or absence of an ATM inhibitor which we previously showed to have a similar effect on rAAV transduction as ATM-deficient cells (.[10] While the two open end molecules (TR+ and TR-) were unaffected by the presence of the ATM inhibitor, the number of cells expressing GFP from the T-shaped hairpin molecules [TRHP, wtTRHP (B+C), revTRHP and TRHP(AT)] was increased by 2-fold in the presence of the ATM inhibitor. Each of the four T-shaped hairpin molecules showed similar levels of expression, regardless the different primary sequence of their TRs, suggesting that it is the specific secondary structure that contributes to the ATM-dependent loss of gene expression.

Interestingly, while there is not an ATM effect on expression from the Simple TRHP(AT) molecule, expression is reduced compared to TR+ and TR- molecules, suggesting that the ATM-independent effect operates on any hairpin structure.

TRHP-dependent loss of gene expression requires HP in cis, not in trans

The observed deficiency of gene expression from DNA substrates with TR ends in the hairpin conformation suggested that the molecules were recognized through both ATM-dependent and ATM-independent mechanisms, and silenced, degraded, or failed to reach the nucleus (see following section for ATM-independent loss of HP genomes in the nucleus). However, it remained possible that the TR hairpin structures triggered activation of DNA damage signaling pathways leading to a general decrease in cellular gene expression. This could lead to a similar decrease in the number of cells scored as GFP positive after transfection.

In order to distinguish between the possibilities of the TRHP affecting gene expression in trans, versus strictly cis effects, we tested the effect of the TRHP structure on GFP expression from a separate co-transfected molecule. Human embryonic kidney 293 cells were co-transfected with the GFP-expressing TR+ substrate (open duplex ends) and a second substrate carrying an unrelated transgene flanked with TR sequences in either the open DNA duplex conformation (PD-L1-TR+) or the covalently closed hairpin conformation (PD-L1-TRHP) ( A greater amount of the TRHP or TR+ PD-L1 substrate was transfected with the GFP TR+ reporter (5:1molar ratio) to ensure that any cell transfected with the GFP reporter also received the hairpin test molecule. The PD-L1 gene cassettes contained a liver-specific promoter LSP that was not active in the transfected cells in order to preclude transcriptional quenching effects.

There was no significant difference in the number of cells expressing GFP from the transfected GFP-TR+ reporter in the presence or absence of co-transfected PD-L1 molecules, whether they contained open duplex or covalently closed hairpin ends. This indicated that the TR hairpin effect on gene expression operated only in cis, without significant global effects on cellular viability or transgene expression. These results point to either transcriptional silencing or degradation of DNA molecules with TRHP ends, or possibly both.

The ATM-dependent decrease in TRHP-linked gene expression is due to transcriptional silencing while the ATM-independent decrease is due to loss of substrate molecules

We have recently reported that an ATM-dependent loss of expression from transduced rAAV genomes is due to silencing, rather than loss of vector genomes.[10] In order to determine whether the same mechanism was responsible for the deficiency in gene expression from the transfected DNA molecules with TRHP ends, we used quantitative real-time PCR (qPCR) to measure the number of DNA substrate molecules within the nuclei of transfected cells. WT and ATM-/- fibroblasts were transfected with the three intact GFP-substrates (TR+, TR- and TRHP) and harvested at 24 h post-transfection. The cells were lysed and low molecular weight DNA was extracted from nuclear and cytoplasmic fractions for qPCR analysis. Because the covalently closed hairpin DNA molecules are resistant to PCR amplification (data not shown), all DNA samples were cut with Mlu I prior to qPCR to remove the hairpin ends. Transfection efficiency was normalized by co-transfecting supercoiled pCMVβ plasmid with each of the DNA substrates and quantifying using a separate primer/probe set specific for the LacZ gene.

In both WT and ATM-/- cells, the copy number of TRHP substrate in the nucleus decreased by 4-fold compared to either TR+ or TR- substrates (. Importantly, there was no difference in TRHP copy number between the WT and ATM-/- cells. This suggests that the ATM-dependent decrease in gene expression from TRHP DNA molecules is due to silencing, similar to transduction with rAAV vectors, while the ATM-independent mechanism is due to loss of nuclear DNA molecules.

ATM-dependent TRHP silencing is saturated at high dose transfection

We used a competition-based experiment to determine the effect of higher doses of transfected TR molecules on ATM-dependent silencing of HP-containing reporter. The doses of the GFP reporter DNA molecules were kept constant to facilitate quantitation of GFP-expressing cells, while competitor DNA was increased (. The competitor for each GFP-expressing DNA substrate contained the LSP-PD-L1 transgene, as above, and the same TR conformation as the reporter molecule against which it competed. Because the HP effect was measured as GFP positive cells per GFP copy number, the ATM-independent loss of HP molecules was not a factor in this assay. We began to observe saturation of the ATM-dependent HP silencing effect when the competitor reached 54-fold excess over the reporter molecule (162 ng/well and 3ng/well, respectively), at which point approximately half of the TRHP molecules were still silenced. Since only 1-2 reporter molecules reach the nucleus at 3 ng/well of reporter DNA, we estimated that at least 25-50 HP-containing molecules were subject to silencing in the presence of 54-fold excess competitor. When the competitor increased to 162-fold excess, no loss of TRHP reporter expression was observed, suggesting that the silencing mechanisms had reached saturation.

Time-course of ATM-dependent effect on rAAV expression

To determine whether the ATM-dependent decrease in gene expression was stable over time, we assessed RNA transcript levels after transducing WT fibroblasts with a self-complementary rAAV vector carrying a CMV-GFP expression cassette (scAAV-GFP) in the presence or absence of an ATM inhibitor (. Viral vector infection was used in this experiment to achieve a more synchronous delivery of HP DNA molecules. Cells were harvested at different time points during the first 48 hours post-infection, and GFP transcripts levels were quantified by RT-PCR, comparing each time point to background from uninfected cells, and using GAPDH as an internal standard.

The GFP transcript levels were greater in cells treated with the ATM inhibitor than in untreated cells at each time point. The difference reached statistical significance at 6 hours post-infection, and continued to increase to 5.3-fold at 24 hours and 7.5-fold at 48 hours. The slight decrease in transcript levels detected in untreated, but not ATM inhibitor treated cells after 12 hours post infection is likely due to the slower doubling time observed in inhibitor-treated cells, which reduces the dilution and loss of episomal DNA molecules as the cells divide.

DISCUSSION

To directly assess whether the AAV TR hairpin structure is specifically recombinogenic, we tested three different types of defined DNA substrate molecules by transfection into normal and recombination-deficient cell lines. These substrates contained the TR sequences in either a simple duplex or a covalently closed hairpin conformation, or contained blunt DNA ends with no TR sequences. Their fates in the transfected cells revealed important differences in the way these structures are recognized and processed. Most strikingly, the presence of the TR in the hairpin conformation resulted in the loss of functional DNA substrate, as determined by expression of a GFP cassette. This occurred whether there was a hairpin present at both ends of the DNA molecule, or at only one end, suggesting that the presence of the hairpin structure causes the DNA to be targeted for silencing, degradation, or altered trafficking. It is unlikely that the observed effect on gene expression is caused by topological constraints in the covalently-closed DNA molecule with a hairpin at each end, because the same decrease in gene expression is observed from molecules where one of the hairpins has been removed to leave one end open. Importantly, this would be the predicted structure for a rAAV vector genome that has undergone second-strand synthesis, or an scAAV genome that has been released from its capsid and folded into a dsDNA conformation.

The loss of functional genomes with one or two TRs in the hairpin conformation is partially relieved in cells lacking ATM activity. This is consistent with the effects of ATM mutation or inhibition on transduction from both single-strand rAAV and scAAV vectors, where transduction is increased in ATM-/- cells relative to wild type.[10] Despite the increased transduction from rAAV vectors in ATM-/- cells, as measured by GFP expression, the amount of vector DNA in the nucleus, including both linear and circular forms, does not change between the mutant and WT cells. This suggests a silencing effect from interaction with ATM or associated factors, which closely parallels the behavior of transfected DNA with hairpin ends, where the excess of actively expressing genomes in ATM-/- cells remain linear. Thus, in normal cells, DNA molecules with one or two hairpin ends can either interact with the ATM-dependent pathway and remain linear and silent, or they become circularized and expressed, presumably through interaction with a different set of recombination factors. In the absence of ATM, the same subset of hairpin molecules goes through the pathway that leads to circularization and expression, and the remainder that would have interacted with the ATM-dependent pathway are able to express from linear molecules. The time course analysis shows that the ATM-dependent silencing begins as soon as 3 hours post-infection and does not change over the subsequent 48 hours, suggesting that interaction with ATM is an early event, and that viral genomes are not released from silencing through at least the first 48 hours after infection.

We have not as yet determined the direct mechanism for ATM-initiated silencing of AAV hairpin-containing molecules, chromatin modifications are an attractive possibility. It has been observed that ATM-mediated phosphorylation of histone H2AX is associated with reduced transcription in regions undergoing DSB repair.[20] Also, ubiquitylation of H2A correlates with transcription repression, and E3 ubiquitin ligases RNF8 and RNF168 affect ubiquitylation of H2A on damaged chromosomes in an ATM-dependent manner.[21-24] Future work employing immunoprecipitation of specific chromatin components complexed with rAAV vectors, or DNA substrates with the TR in different conformations, may reveal the mechanism at the molecular level.

While the ATM-dependent effect on gene expression is clearly dependent on the hairpin end structure, there is a second TR hairpin-dependent effect that is not related to ATM activity, and results in the loss of DNA molecules, as measured by qPCR. This may be mediated by a specific DNA recombination/repair pathway that recognizes the hairpin structure and leads to extensive DNA degradation. Alternatively, the loss may be part of a host cell defense mechanism recognizing hairpin DNA ends, leading to altered trafficking to the nucleus. Since we are transfecting naked DNA, this could occur either in the nucleus or during transport through the cytoplasm. We do not know whether this effect is specific to transfected DNA because parvovirus vectors cannot be replicated and packaged without hairpin ends. Further studies to identify the mechanism of hairpin-dependent loss of genomes may reveal whether the viral genomes are also affected.

Considering the loss and silencing of DNA molecules with hairpin ends, evaluating the direct effect of the hairpin on the frequency of recombination is complex. However, it is clear that the majority of transfected molecules without hairpin ends express GFP from circularized molecules. Thus, it is probably not accurate to suggest that the AAV hairpin end is inherently recombinogenic, because simple double-strand DNA ends are at least equally prone to undergo recombination. Rather, recombination involving AAV hairpin ends is mechanistically different from that of simple DNA ends, with multiple pathways leading to different fates.

We had previously published a study evaluating the efficiency of recombination between the hairpin end versus open end in scAAV genomes, using the formation of intermolecular concatemers as a model.[16] These experiments indicated that recombination between two hairpins was twice as frequent as recombination between two open ends, and recombination between an open and a hairpin end was intermediate in frequency. As in the present study, the assay was based on reconstitution and expression of a split GFP gene. Considering the present results, these concatemerization events probably occurred within the fraction of genomes that do not interact with ATM, and ultimately are expressed and undergo recombination. This suggests that within this compartment, the hairpin end may be considered recombinogenic, in that it is more likely to join with another DNA end.

Notably, a fourth DNA substrate was planned for this study, but could not be made in our hands. This would have been a dsDNA molecule with two open ITRs at each end, and the TR of each strand in the hairpin conformation. The TRs of AAV are often depicted schematically in this conformation, and it is a critical intermediate step in the replication scheme which allows priming for the next round of DNA synthesis. We could not find partial denaturation or annealing conditions that would allow the ITRs to remain in the hairpin conformation, instead finding that they always reannealed completely to form simple duplex ends. The conformation was determined by susceptibility to digestion with the restriction enzyme Ahd I at a site in the apex of the B palindrome (data not shown). This suggests that the double open hairpin TR may be formed only transiently, or must be maintained in the context of host or viral proteins. Further, the blunt-ended, fully duplex TR may best represent the actual conformation of the open end of a duplex AAV genome under physiological conditions.

MATERIALS AND METHODS

Cell lines

The SV40-transformed fibroblast cell lines from a normal individual (WT cells, GM00637 J) and from an ataxia telangiectasia (AT) homozygous patient (ATM-/- cells, GM05849 E) were purchased from Coriell Institute (Camden, NJ). WT and ATM-/- cells were grown in MEM Eagle with Earle's BSS (LONZA BioWhittaker) supplemented with 10% FBS and 1X GlutaMAX (GIBCO).

HeLa cells were grown in DMEM (GIBCO) supplemented with 10% FBS. Human embryonic kidney 293 cells were grown in DMEM (GBCO) supplemented with 10% cosmic calf serum (HyClone). All cell lines were cultured as monolayer cultures at 37°C in 5% CO2 humidified incubator.

DNA substrates

TR+, TR- and TRHP substrates were excised from plasmids carrying either the intact (scAAV-GFP) or circularization-dependent (scAAV-GFP-CD) vectors described previously.[16] TR+ substrate was excised by Pvu II digestion from both plasmids. TR- substrate was excised by Mlu I digestion from scAAV-GFP plasmid, or Bgl II and Sal I digestion from scAAV-GFP-CD plasmid, followed by fill-in of 5’ overhangs with Klenow. TRHP has been described previously.[14] Briefly, it was excised by Pvu II digestion from both plasmids and incubated for 6 min with Exonuclease III to expose single-stranded 5’ termini, which fold into the hairpin conformation. Subsequently, T4 DNA Polymerase and T4 DNA Ligase were used to repair the gaps and covalently close the ends of the DNA molecules. Finally, the preparation was digested with excess Exonuclease III to remove any molecules containing nicks or gaps (covalently closed hairpin DNA molecules remained resistant). A single hairpin was removed from intact GFP-expressing TRHP by Kpn I digestion (TRHP/-), and both hairpins removed by digestion with Mlu I (~TR-).

To generate the four covalently closed DNA substrates with modified TR sequence or structure, the following palindromic oligonucleotides were synthesized: wtTRHP (B+C): 5’-CGCGAGGCCGGGCGACCAAAGGTCGCCCGACGCCCGGGCTTTGCCCGGGCGGCCT; revTRHP: 5’-CGCGTCCGGCGGGCCCGTTTCGGGCCCGCAGCCCGCTGGAAACCAGCGGGCCGGA; TRHP(AT): 5’-CGCGAGGCGAACATAGTACAACTATGTTCATTATGTATGTCTCATACATAAGCCT; and Simple TRHP(AT): 5’-CGCGAGGCGAACATAGTACAACTATGTTCGCCT. Each oligonucleotide was boiled and cooled to room temperature, phosphorylated, and ligated to the scAAV-GFP plasmid which was previously digested with MluI to excise the intact GFP cassette without the TR ends. The molar ratio for ligation was 10:1 oligo to plasmid molecule. The ligation reaction was subsequently digested with Exonuclease III to eliminate any unligated DNA, and covalently closed substrate of approximately 2200bp was purified by gel electrophoresis.

DNA molecules used as competitors (PD-L1-TR+, PD-L1-TR-, and PD-L1-TRHP) were excised from a plasmid (p-trsLSP-PD-L1) in which the expression cassette of the scAAV vector contains a mouse PD-L1 transgene [25] controlled by a liver-specific promoter (LSP). p-trsLSP-PD-L1 was derived from the previously described p-trsLSP-GFP [26], in which the eGFP-coding sequence was excised and replaced by the mouse PD-L1 gene. PD-L1-TR+ substrate was excised by Pvu II digestion. PD-L1-TR- substrate was excised by BssH II digestion, followed by a fill-in of 5’ overhangs with Klenow. PD-L1-TRHP was excised by Pvu II digestion, and covalently closed ends were generated as described above. All DNA substrates were gel purified from low melting point agarose (using long wavelength (365 nm) UV illumination to avoid inducing UV- damage repair pathways in recipient cells) and concentrations were normalized by comparing aliquots on ethidium bromide stained gels (Kodac Gel Logic Imaging system).

Transfection assays

Cells were seeded on glass cover slips pre-coated with Collagen Type I in 24-well plates and transfected using polyethylenimine (PEI) (Polysciences, ~25,000 MW). Precipitates were prepared by mixing DNA in media at 55 ug /ml followed by PEI (PEI:DNA ratio at 3.6:1), and incubated 10 minutes at room temperature. The standard DNA transfection in triplicate comprised supercoiled pSP72 (Promega) as carrier (1.6 μg/well), supercoiled pDSRed2-C1 (Clonetech) (17 ng/well), and linear DNA substrate (17 ng/well). Precipitates were added to cells in 24-well plates at 30-35 ul per well. HeLa cells were seeded at 30% confluence on day 1, co-transfected on day 2, and fixed on day 3 with 4% paraformaldehyde in phosphate-buffered saline. Cells on coverslips were mounted on slides and GFP and RFP positive cells counted under a fluorescent microscope. To assay for the ATM effect in Hela cells, cells were seeded as described above and treated with 10 uM ATM inhibitor (KU55933, Tocris) or 10 uM DMSO on day 1. Treatment was maintained until cells were harvested on day 3.

WT and ATM-/- cells were co-transfected with pSP72 (1.6ug/well), pDSRed2-C1 (12.5ng/well) and linear DNA substrate (12.5ng/well) and assayed as described above.

To assay for cis versus trans effects on substrate expression, 293 cells were seeded on day 1 at 30% confluence in 24-well plates. On day 2, cells were co-transfected with pSP72 (1.6ug/well), supercoiled DsRed-express (3ng/well), and either linear GFP-TR+ DNA substrate (3ng/well) or GFP-TRHP DNA substrate (3ng/well), plus either PDL1-TRHP (15ng/well) or PDL1-TR+ DNA substrates (15ng/well). At 24 hr post-transfection, cells were fixed assayed as described above.

For the competition assay, 293 cells were seeded as described above, and co-transfected with supercoiled pCMVβ (3 ng/well) to normalize for transfection efficiency, plus constant amounts of GFP-expressing TR+, TR- or TRHP DNA substrate (3ng/well), decreasing amounts of pSP72 carrier, and increasing amounts of PD-L1-TR+, PD-L1-TR-, or PD-L1-TRHP competitor, respectively. Competitors were added at 0, 18, 54, 162 or 486 ng/well, while pSP72 was added at 1.6, 1.58, 1.55, 1.44, or 1.11 ug/well, respectively. Precipitates were prepared as described above, and were added to cells at 34-40 ul per well. 24 h after transfection GFP positive cells were counted under a fluorescent microscope without fixing. To evaluate GFP expression per GFP substrate molecule, nuclear fractions were isolated as described previously (Wang, Zhu et al. 2006), and low molecular weight DNA was separated by Hirt extraction and digested with MluI to cut off the TR sequences. Copy numbers were quantified using Absolute Blue QPCR ROX Mix on an Applied Biosystems Prism 7000 Sequence Detector System. Data were analyzed by absolute quantification, and pCMVβ copy numbers were used to normalize for transfection efficiency. The primer-probe sequence specific for CMVβ amplification was: forward primer, 5’CCCGTATTTCGCGTAAGGAA; reverse primer, 5’GTTGATGTCATGTAGCCAAATCG (300nM final concentration); and probe, 5’FAM CTTTTACTTTTTTATCATGGGAGCC TAMRA3’ (100nM final concentration). Substrate DNAs carrying GFP-expression cassette were quantified using the following primer-probe sequences: forward primer, 5’AAGCAGCACGACTTCTTCAA; reverse primer, 5’TCGTCCTTGAAGAAGATGGT (300nM final concentration); and probe, 5’ FAM CCATGCCCGAAGGCTACGTC TAMRA 3’ (100nM final concentration).

To evaluate the effect of TR conformation and ATM mutation on DNA substrate copy number, WT and ATM-/- fibroblasts were seeded in 6-well plates at 30% confluence on day 1. On day 2, cells were co-transfected with pSP72 (9.6 ug/well), supercoiled pCMVβ (Clonetech) plasmid (75ng/well) and TR+, TR- or TRHP DNA substrate (75ng/well). 24hs after transfection, cells were harvested and copy number was quantified by Real Time PCR as described above.

Time course assay for ATM-dependent silencing

WT fibroblasts were seeded on 6-well plates on day 1 at 40% confluence and treated with 10 uM ATM inhibitor (KU55933, Tocris) or 10 uM DMSO. Treatment was maintained until cells were harvested at different time points after infection. On day 2, the cells were infected with a rAAV vector previously described (scAAV-GFP),[10] at a multiplicity of infection (MOI) of 10 HeLa C12 infectious unit per cell. Medium with virus was replaced with fresh medium at 3 h post-infection. Cells were harvested at 0, 3, 6, 12, 24, 36, and 48 h postinfection and total RNA was isolated using the SV total RNA isolation System (Promega). Reverse transcription of mRNA was carried out in 40 ul final volume from 1ug total RNA using random-hexamer primers and the Verso cDNA synthesis Kit (Thermo Scientific). Reverse-transcribed RNA was amplified with Absolute Blue QPCR SYBR Green ROX Mix (Thermo Scientific) plus 0.07 μM of gene-specific forward and reverse primers in 25 ul final volume on an Applied Biosystems 7000 Real-Time PCR System. Data were analysed by relative quantification using the 2 (-ΔΔCt) method. GFP transcript abundance was quantified in infected cells relative to the uninfected cells harvested at time 0, using GAPDH transcripts as internal control (GFP primers: forward: AAGCAGCACGACTTCTTCAA; reverse: GTAGTTGTACTCCAGCTTGT; GAPDH primers: forward: GGAGTCAACGGATTTGGTCG; reverse: GGAATCATATTGGAACATGTAAACC).

Statistical analysis

All statistical analyses were performed using Student's T-Test (GraphPad Prism 5 software). Error bars indicate mean with standard deviation.