Induction of the alternative lengthening of telomeres pathway by trapping of proteins on DNA

Abstract Telomere maintenance is a hallmark of malignant cells and allows cancers to divide indefinitely. In some cancers, this is achieved through the alternative lengthening of telomeres (ALT) pathway. Whilst loss of ATRX is a near universal feature of ALT-cancers, it is insufficient in isolation. As such, other cellular events must be necessary - but the exact nature of the secondary events has remained elusive. Here, we report that trapping of proteins (such as TOP1, TOP2A and PARP1) on DNA leads to ALT induction in cells lacking ATRX. We demonstrate that protein-trapping chemotherapeutic agents, such as etoposide, camptothecin and talazoparib, induce ALT markers specifically in ATRX-null cells. Further, we show that treatment with G4-stabilising drugs cause an increase in trapped TOP2A levels which leads to ALT induction in ATRX-null cells. This process is MUS81-endonuclease and break-induced replication dependent, suggesting that protein trapping leads to replication fork stalling, with these forks being aberrantly processed in the absence of ATRX. Finally, we show ALT-positive cells harbour a higher load of genome-wide trapped proteins, such as TOP1, and knockdown of TOP1 reduced ALT activity. Taken together, these findings suggest that protein trapping is a fundamental driving force behind ALT-biology in ATRX-deficient malignancies.


INTRODUCTION
Due to the inherent inability of DN A pol ymerases to replicate the distal ends of linear chr omosomes, chr omosomal DNA is progressi v ely shortened with each round of cell di vision. To circumv ent potentially detrimental effects on genome stability and loss of genetic information, the genome has de v eloped specialised nucleoprotein structures, telomer es, which ar e comprised of many kilobases (kb) of a tandem repeat sequence, TTAGGG, culminating in a 3' protrusion of single stranded G-rich DNA of 50-400 nucleotides. Telomeric sequence is bound by a specialised protein complex, denoted Shelterin, which comprises the proteins TRF1, TRF2, POT1, TIN1, TPP1 and RAP1 ( 1 ).
Telomeres range from 3 to 12 kb in humans and progressi v ely shorten by about 200 bp per cell division due to the end-replication problem. Once telomeres reach a critical length, termed the Hayflick limit, they elicit DNA damage checkpoint activation, leading to cellular senescence or telomere-induced apoptosis.
One hallmark of cancer cells is their ability to circumvent telomere shortening via a telomere maintenance mechanism (TMM). In the majority of cancers, this is achie v ed through upregulation of telomerase, a specialised ribonucleoprotein that acts to progressi v ely add telomeric repeats to the end of chromosomes. More recently it has emerged that a subset of cancers maintain their telomere length through a telomerase independent TMM, known as the alternati v e lengthening of telomeres (AL T) pathway. The AL T pathway is particularly prevalent in cancers of mesenchymal origin (such as osteosarcoma), and se v er al cancers of the centr al nervous system, such as glioblastoma ( 2 ). ALT in human cancer cells is generally considered to be a form of aberrant telomer e r ecombina tion and conserva ti v e DNA synthesis, known as break-induced replication (BIR), occurring during both the G2 and mitosis phases of the cell cycle (3)(4)(5). A three-protein axis has been implicated in facilitating ALTmediated telomere synthesis; POLD3, PCNA and RAD52 (3)(4)(5). A second RAD52-independent pathway has also recently been reported, suggesting that ALT is in fact a bifurca ted pa thway, with similarities to telomerase independent telomere maintenance pathways originally described in budding yeast ( 6 , 7 ).
Further work in human cells has suggested that replication stress arising at telomeres can potentiate the ALT pathway ( 4 ). Indeed, owing to their repetiti v e nature, telomeres are thought to be inherently difficult sequences to replicate and telomeres have been shown to phenotypically resemble common fragile sites due to the overt fragility they exhibit under conditions of r eplication str ess ( 8 ). This is likely in large part due to the propensity of the G-rich repetiti v e telomeric sequence to adopt non-canonical DNA secondary structures, including the G-quadruplex (G4) conformation and R-loops; three-stranded nucleic acid structures consisting of an RN A:DN A hybrid and a displaced piece of single stranded DNA ( 9 ). In line with this notion, abro gation of FANCM activity, w hich has known roles in R-loop resolution and fork stabilisation (10)(11)(12), or depletion of RNase H, which can degrade DN A:RN A hybrids, have both recently been shown to potentiate markers of the ALT pathway (12)(13)(14).
Almost all ALT+ cancer cells exhibit loss of the ATRX gene and / or its interaction partner DAXX (15)(16)(17), howe v er, the precise role of ATRX / DAXX loss in the initiation and maintenance of ALT remains unclear. Previous work by our lab and others has demonstrated that ectopic expression of ATRX leads to a DAXX-dependent suppression of the ALT pathway ( 18 , 19 ). Howe v er, depletion or knockout of ATRX in telomerase positi v e or primary cell lines is generally insufficient to induce markers of the ALT pathway (18)(19)(20)(21), with the notable exception of a minority of glioma cell lines (22)(23)(24). ALT has also been induced in cells where further proteins, along with ATRX, have also been knocked down; for example, p53, TERT and the histone demethylase KDM4B ( 25 ). Ther e must, ther efor e, be other genetic, epi-genetic or cellular e v ents r equir ed for ALT induction that act in concert with ATRX loss -but, to da te, wha t these e v ents are remains unclear.
ATRX is a multifunctional protein, involved in various critical cellular and genetic processes. ATRX is a chromatin remodelling factor of the Snf2 family which, together with the histone chaperone DAXX, facilitates the incorporation of the histone variant H3.3 into defined genomic sites, such as pericentric and telomeric chromatin (26)(27)(28). ATRX has various roles in the maintenance of genome stability, including regulation of non-canonical DNA secondary structures, such as G4s and R-loops, with ATRX null cells displaying increases in both of these structures (29)(30)(31)(32). ATRX also has multiple reported roles in DNA replication, including the pre v ention of replication fork stalling, potentiation of fork restart and the pre v ention of e xcessi v e nucleolytic degradation of stalled forks ( 20 , 31 , 33-35 ). Additionall y, ATRX facilitates DN A double strand break (DSB) repair, through both homologous recombination (HR) and non-homologous end joining (NHEJ) pathway (36)(37)(38)(39). It is clear that the loss of one or more of these roles could be instrumental to the induction of the ALT pathway -but the exact nature of the perturbation causing the 'second factor' remains undefined.
In this work, we explore the role of protein trapping in ATRX-deficient cells and demonstrate that formation of DNA-protein covalent complexes (DPCCs) and / or accumulation of proteins on DNA is a fundamental driving force in both natural and artificial models of ALT-cancer. We hypothesise that these DPCCs cause replication fork collapse in the absence of ATRX, which is the substrate for subsequent BIR and ALT activity.

Cell lines and cell culture conditions
All cell lines were obtained from ATCC with the exception of HeLa LT which was a gift from Roderick O'Sullivan (Uni v ersity of Pittsburgh, USA), as described in ( 21 ). All cells were cultured at 37 • C in 5% CO 2 in standard Dulbecco's modified Eagle's medium (DMEM) media supplemented with 10% foetal calf serum, 1% L -glutamine and 1% PenStrep (all Gibco). Cells were split e v ery 2-4 days to maintain at a confluency no greater than 90% using 0.05% trypsin / EDTA (Gibco).

CRISPR-Cas9 knockout
CRISPR-Cas9 knockout of ATRX was performed using a modified pSpCas9(bb)-2A-GFP tagged Cas9 vector containing sgRNAs targeting exon 16 of ATRX (TOP: 5'-caccGTCCAATAA CAA CCA ∧ AGT-3', BOTTOM: 5'-aaacACT ∧ T GGTT GTTATT GGAC-3') with an expected cut site at lysine 1536. Cells were sorted on a BD FAC-SAria Fusion cell sorter based on GFP expression 24 h after transfection into single wells and grown into clones. Knockout was determined by western blotting. DAXX knockouts were performed using a commercially available DAXX Cas9 plasmid (Santa Cruz, sc-400686-knockout-2) and were FACS sorted based on GFP expression. Transfected cells were selected using 0.4 g / ml puromycin and Nucleic Acids Research, 2023, Vol. 51, No. 13 6511 then sorted into single wells to obtain clones. Combinatorial knockouts were made through sequential knockout of genes.

T r eatment of cells with genotoxic agents
Cells were treated with these genotoxic agents at the following doses for 24-72 h prior to downstream analysis with the ALT assays used in this stud y: pyridosta tin / PDS (Sigma Negati v e controls with the drug diluent (water, PBS or DMSO) but without drug were included with each assay.

Western blotting
To pr epar e w hole cell l ysates, about 2 × 10 6 cells were lysed in 200 l ice-cold IgePal lysis buffer (50 mM Tris-HCl pH 7.4, 100 mM NaCl, 1 mM MgCl 2 , 10% glycerol, 5 mM NaF, 0.2% IgePal-CA630) containing Pierce protease inhibitor tablet (Thermo Fisher Scientific) for 45 min on ice. Insoluble components were removed by centrifugation and protein lysate concentration was calculated by NanoDrop. At least 10 g of lysate was loaded per lane into precast 4-12% bis-tris gels with MOPS running buffer (both Thermo Fisher Scientific). Membranes were transferred at 4 • C overnight onto 0.45 m pore size PVDF membranes (Millipor e) pr e-activated in 100% methanol using NuPage transfer buffer (Thermo Fisher Scientific) supplemented with 10% methanol. The following day, membranes were rinsed briefly in PBST (0.1% tween in PBS) and then blocked for 1 hour at room temperature in 5% milk in PBST. Primary antibod y incuba tions then followed in 2.5% milk in PBST. Membranes were washed three times for 10 min with PBST before incubation with HRP-conjugated secondary antibodies diluted in 2.5% milk in PBST for 1 h at room temper ature. Membr anes were once again washed three times for 10 min with PBST and membranes were then de v eloped using SignalFire ECL reagent (Cell Signaling) onto X-ray films (Amersham).
To pr epar e subcellular fractions of nuclear soluble and chromatin bound fractions, 1 million HeLa LT ATRX 1 cells wer e tr eated with the panel of PARPi drugs for 48 h and then cells were collected. For the fractionation, a subcellular protein fractionation kit (78840, Thermo Fisher Scientific) was used according to manufacturer's instructions. Immunoblotting was carried out as above. Signal intensity was measured using ImageJ software.

shRNA knockdowns
shRNA experiments were performed using the following commerciall y available MISSION shRN A plasmid DN A: MISSION shRNA TOP1 (NM 003286-3990) and MIS-SION shRNA PARP1 (NM 001618-7930) (both Sigma) that were packaged into lentiviruses in house. Puromycin kill curves were performed to establish the minimum lethal dose for untransduced cells at 48h timepoint. Cells (30,000) were seeded into a 24-well plate and, the f ollowing da y, 250 l of the lentivirus mix (containing 35 l of lentivirally packaged shRNA, 215 l of DMEM and 0.3 l of polybrene) was added. Se v enty-two h after transduction, cells were held under selection with puromycin (3 g / ml) for up to 8 days before being harvested or fixed for downstream assays. Western blots were performed to check efficiency of knockdown. A negati v e control, using a scrambled shRNA sequence, was included with each assay.

Immunofluorescence / ImmunoFISH
Immunofluorescence experiments were performed with 50 000 cells seeded onto 13 mm #00 thickness glass coverslips in 24-well tissue culture plates. Cells were briefly washed once in PBS , pre-permea bilised in 0.5% triton X-100 (Sigma) in PBS for 1 minute on ice and then fixed with 4% paraformaldehyde (Thermo Fisher Scientific; 16% diluted to 4% in PBS) for 20 min at room temperatur e. Fix ed cells were then washed three times in PBS for 5 min and then permeabilised with 0.5% triton X-100 on ice for 6 min. The cells were washed a further three times in PBS for 5 min each and then blocked for 1 h in blocking solution (1% BSA in PBS). After blocking, samples were incubated for at least 1 h with primary antibodies diluted in blocking buffer. Samples were washed four times with PBST for 5 min and then incubated for 1 h with fluorescently labelled secondary antibodies raised in the appropriate species diluted in blocking buffer. Following three more 5 minute washes with PBST, the samples were mounted in VectaShield containing DAPI and visualised using a DeltaVision widefield microscope with either 60 × or 100 × objecti v es. Images were processed using Fiji ImageJ, and downstream analysis of co-localisation and / or foci number and intensity was performed on CellProfiler (41)(42)(43). Cells from at least two independent biological replicates were analysed for each experimental condition, with a minimum of 100 cells per repeat. The following primary and secondary antibodies were used with the indicated dilutions: mouse anti-PML (Santa Cruz, sc-966, 1:300), rabbit anti-TRF2 (Novus Biologicals, NB110-57130, 1:500), mouse anti-RPA32 (Abcam, ab2175, 1:500), rabbit anti-ATRX (Abcam, ab97508, 1:500), mouse anti-pATM[S1981] (Santa Cruz, sc-47739, 1:250), mouse anti-BLM (Santa Cruz, sc-365753, 1:300), mouse anti-Top1cc (Merck, MABE1084, 1:200), goat anti-rabbit Alexa Fluor 568 (Life Technologies, A11036, 1:3000), goat anti-mouse Alexa Fluor 488 (Life Technologies, A11029, 1:3000). For Top1cc immunofluorescence, pre-permeabilisation time was increased to 5 min and permeabilisation time to 20 min.
ImmunoFISH experiments were carried out as described above up until the PBST washes after secondary antibody incubation. Following these, samples were post-fixed with 4% paraformaldehyde (as above) for 10 min at room tempera ture. If dena tured telomeres were required (all experiments except RPA-ssTel analysis), 3.5 N HCl was added to the coverslips at room temperature for 13 min and the reaction was then quenched by washing twice for 5 min with ice cold PBS. Samples were then washed twice with PBST for 5 min and once with 2 × SSC buffer (0. dispensed on parafilm. The f ollowing da y, coverslips were returned to 24-well dishes and washed three times for 10 min each with 2 × SSC at 37 • C, followed by two 5 min washed in PBST and one 5 min wash in PBS. Coverslips were then mounted with DAPI and imaged as above.

Metaphase spreads and telo-FISH
Cells were seeded out into 6 cm dishes and treated with PDS f or 24 h. Kary omax colcemid (50 l of 10 g / ml, Gibco) was added to the cells for 1-4 h to arrest cells at the metaphase stage of mitosis. Cells were then trypsinised and collected. Pre-warmed hypotonic solution (75 mM KCl) was then added to the cells and incubated for 15 min at 37 • C. Cells were spun down and fixati v e (3:1 methanol:glacial acetic acid) was gently added dropwise to the pellet to a final volume of 10 ml. Cells were incubated at -20 • C for 30 min, spun down, and fixati v e was replaced twice. Cells were then resuspended in an appropriate volume of fresh fixati v e and cells were dropped onto warmed glass slides. Spr eads wer e hardened for 5 days at room temperature and then Telo-FISH was carried out as above. Fragile telomer es wer e scor ed w hen m ultiple telomeric signals were seen that were spatially separated from chromatid ends, as in ( 8 ).

C-circle assay
Genomic DNA was extracted from pellets of up to 1 × 10 6 cells using the PureLink genomic DNA extraction kit (Thermo Fisher Scientific) and quantified using a Nan-oDrop. 30 ng of genomic DNA was amplified in a 20 l rolling circle amplification reaction containing 7.5 U of 29 polymerase (New England Biolabs), 0.1% Tween-20, 200 g / ml BSA (New England Biolabs) and 1 mM each of dTTP, dGTP and dATP (all New England Biolabs) diluted in nuclease-free water. The r eactions wer e incubated in a PCR cycler (Bio-Rad T100 Thermal Cycler) for 8 h at 30 • C followed by 20 min at 65 • C. A negati v e control (without 29 polymerase) and a positi v e control (U2OS gDNA) was included in each run. Amplified samples were then diluted with 180 l 2 × SSC buffer and transferred onto a Zeta-Probe membrane (Bio-Rad) using a slot blot filtration manifold (Bio-Rad BIO-DOT, 48-well). The membranes were dried for 15 min at room temperature and then UV crosslinked using the 'Auto Crosslink' setting on a Stratalinker 2000 (UV-A). The membrane was pre-hybridised with 10 ml DIG Easy Hyb (Roche) for 20 min at room temperature on a roller and then incubated in a 37 • C hybridisation oven for 2 h with a 3'DIG-labelled [CCCTAA] 5 probe diluted in 10 ml DIG Easy Hyb to a final concentration of 40 nM. Membranes were then washed twice for 5 min in MS wash buffer (0.1 M maleic acid, 3 M NaCl, 0.3% Tween-20, pH 7.5) and b locked in MS b locking buffer (1% milk and 1% BSA in 0.1 M maleic acid, 3 M NaCl, pH 7.5) for 30 min at room temperature. Anti-DIG-AP Fab fragments (Roche) were added to the MS blocking buffer (1:20 000) and the membranes were incubated for 30 min at room temperature. The membranes were washed three times for 15 min with MS wash buffer and then incubated with 2 ml of CDP-Star chemiluminescent substrate solution (Roche) for X-ray film detection. Intensities on membrane images were quantified using ImageJ software. Unless otherwise stated, normalised CC le v els (A.U) were expr essed r elative to a U2OS r efer ence sample (30 ng) from the same membrane. A gDNA dilution series (4:2:1) was loaded onto each membrane to assess assa y linearity. Assa ys were performed on at least two independent DNA samples for each test condition.

Terminal restriction fragment (TRF) assay
Genomic DNA ( > 2 g) was digested overnight at 37 • C with HinfI and RsaI restriction endonucleases (NEB). Following digestion, samples were run on 0.8% agarose gels in 1 × TAE at 60 V overnight. Gels were denatured using 1 M NaOH and then neutr alised with neutr alisa tion buf fer (0.5 M Tris-HCl, 1.5 M NaCl) and blotted onto Zeta-Probe membrane by upward capillary transfer overnight. Blots were then probed with a DIG-tagged telomere probe as described abov e. The b lots were analysed and quantified using TeloMetric software ( 44 ).

Monochrome multiplex qPCR (mm-qPCR)
MM-qPCR was carried out as previously reported with minor alterations ( 45 ). Primers were bought through Life Technologies using standard desalted oligonucleotides. Primer sets used were GlobinF (CGGCGGCGGGCGGC-GCGGGCTGGGCG GCTTCATCCACGTTCACCTTG), GlobinR (GCCCGGCCCGCCGCGCCCGTCCCGCCG-GAGGAGAAGTCTGCCGTT), hTeloG (A CA CTAA-GGTTT GGGTTT GGGTTT GGGTTT GGGTTAGT GT) and hTeloC (TGTTAGGTA TCCCTA TCCCTA TCC CTA TCCCTA TCCCTAACA). Fi v e concentrations of r efer ence genomic DNA purified from HeLa LT were pr epar ed by 3-fold serial dilution (from 150 ng to 1.85 ng) to generate standar d curv es for relati v e quantitation of T / S ratios. For each sample, 20 ng of genomic DNA was mixed with 0.75 × PowerUp SYBR Green Master Mix (Thermo Fisher Scientific), the rele vant forwar d and re v erse primers (300 nM), and water to a final volume of 20 l per well and analysed using a Thermo Fisher QuantStudio 3 qPCR machine with the following cycle conditions: dena tura tion for 15 min a t 95 • C , followed by two cycles of 15 s at 94 • C / 15 s at 49 • C and 32 cycles of 15 s at 94 • C / 10 s at 62 • C / 15 s at 74 • C with signal acquisition and 10 s at 84 • C / 15 s at 88 • C with signal acquisition. Samples were run in triplicate, and analysis was repeated six times using independent runs.

Reverse transcription quantitative polymerase chain reaction (RT-qPCR)
RNA was extracted from cells using the PureLink RNA mini kit (Thermo Fisher Scientific). 1 g of total RNA was re v erse transcribed to cDNA using 200 U of Super-Script IV re v erse transcriptase (Thermo Scientific), and 50 M random hexamer oligos, according to manufacturer instructions. cDNA was diluted to 200 l with nucleasefree water. For each sample, 5 l of cDNA was mixed with 0.75x PowerUp SYBR Green Master Mix (Thermo Fisher Scientific), the primers (300 nM) and water to a final volume of 20 l per well and analysed using a Thermo Fisher QuantStudio 3 qPCR machine with the following cycle conditions: 95 • C for 15 min, followed by 40 cycles of 10s at 95 • C, 20s at 60 • C and 20s at 72 • C. Melting curve analysis was performed immediately following this. Primers were bought through Life Technologies using standard desalted oligonucleotides. Primers used were hTERT F1579 (GCT GACGT GGAAGAT GAGCGT GC) hTERT R1616 (T CCT CA CGCAGA CGGT GCTCT G), 7SK F7 (GAGGGCGATCT GGCT GCGACAT) and 7SK R112 (ACAT GGAGCGGT GAGGGAGGAA). TERT expression was normalised to 7SK expression using the CT method. Data is expressed as CT comparing the untreated controls versus PDS-treated cells. Averages were calculated from two biological replicates, each run in triplicate.

RADAR (rapid approach to DNA adduct recovery) assay
RADAR assa y was perf ormed as in ( 46 ) with minor modifica tions. Approxima tely 5 × 10 5 cells wer e tr eated with PDS, CPT, ETO or CX-5461 (as above) for 1 h. After drug treatment, medium was aspirated and cells were lysed directly on the plate by adding 1 ml of lysis reagent (RLT Plus, Qiagen). Nucleic acids were recovered by adding 0.5 × volume of 100% ethanol, incubation at -20 • C for 10 min and then centrifugation at maximum speed at 4 • C for 15 min. The supernatant was removed and the pellet was washed twice with 75% ethanol, followed by 10 min centrifugation at maximum speed. The nucleic acid pellet was then dissolved in 200 l of freshly prepared 8 mM NaOH and rotated overnight at 4 • C to ensure complete solubilisation.

Quantification and statistical analysis
Statistical analysis was done using GraphPad Prism 9 (GraphP ad Softwar e Inc.) and Social Science Statistics calcula tors ( https://www.socscista tistics.com/ ). Unpaired ttests were used to compare two groups and one-way ANOVA with Welch correction was used to compare more than two groups (both parametric data). Kruskall-Wallis test was used for non-parametric unpaired data. Chisquared test was used for 2 × 2 contingency analysis. Linear r egr ession was used for correlation analysis. Sample sizes and P -values are shown in the figure legends and significance was considered as * P < 0.05. ** P < 0.01, *** P < 0.001, **** P < 0.0001. ns denotes no significance.

Stabilisation of G-quadruplex structures induces an ALTphenotype in ATRX-null cells
Owing to their highly repetiti v e nature, telomeric sequences have a high propensity to adopt non-canonical DNA secondary structures, including the G-quadruplex (G4) conformations and R-loops ( 9 ). Recent work has shown that telomeres of ALT cells are characteristically enriched in both G4 structures and R-loops and form a linked structure known as a G-loop, where a G4 and an R-loop form on opposing strands ( 47 ). Based on these findings, we investigated whether stabilisation of G4 structures can induce ALT.
The HeLa long telomere cell line (HeLa LT) is a HeLa subclone which has previously been shown to be amenable to ALT induction through depletion of ASF1 ( 21 ). Two independent ATRX CRISPR-Cas9 knockout HeLa LT clones were generated (Supplementary Figure S1A) and, consistent with previous reports, neither clone elicited an increase in any of the cardinal ALT markers, including C-circles (Supplementary Figures S1B and S1C), ALT-associated PML nuclear bodies (APBs) (Supplementary Figures S1D and E) or telomere length heterogeneity (Supplementary Figure S1F). Strikingly, howe v er, addition of a G4 stabilising ligand, pyridostatin (PDS), resulted in increased le v els of each of these key ALT markers in the two clones lacking ATRX (but not wildtype cells) -and the le v el was comparable to the ar chetypical ALT cell line, U2OS (Figur es 1 A-D). Addition of PDS also elicited an increase in telomere intensity and a decrease in total telomere number, this being consistent with increased telomere clustering and synthesis as observed in ALT (Figures 1 E and F). We also saw an increase in telomere heterogeneity ( Supplementary Figures S2A and B), an increase in telomere copy number as measured by qPCR (Supplementary Figure S2C) and repression of TERT expression (Supplementary Figure S2D). These data were strongly indicati v e of a bona fide switch to ALT-based telomere maintenance.
Another hallmark of ALT activity is increased replication stress and accumulation of DNA damage signals at telomer es, termed telomer e damage induced foci (TIFs), a potential dri v er of ALT ( 16 ). Telomeric accumulation of DNA damage markers (such as 53BP1, ␥ H2AX and ATM phosphoryla ted a t serine 1981 (pATM s1981 )) have all been used to identify TIFs ( 48 ). Imm uno-FISH anal ysis assessed co-localisation between pATM S1981 and telomeric foci, and allowed us to quantify le v els of DNA damage at telomer es. Following tr eatment with PDS, both ATRX wildtype and deficient cells showed increased le v els of telomeric pATM S1981 , howe v er, ATRX deficient cells had significantly higher TIF le v els (Supplementary Figure S3). We also measur ed telomer e fragility in our cells, finding that ATRX-null cells treated with PDS had a significantly higher percentage of fragile telomeres as compared to untreated cells ( Figures  1 G and H). It is generally considered that fragile telomeres result from aberrant processing of stalled replication forks and, as such, can be used as a measure of replication fork stalling ( 8 ).
Pre vious wor k has shown that the human ALT pathway relies on BIR-dependent mechanisms of telomere synthesis. The BIR pathway is a three-protein axis comprised of POLD3, PCNA and RAD52 (3)(4)(5). BLM helicase has also been shown to be the downstream effector in the BIR axis and is vital for APB assembly with subsequent ALT telomere synthesis ( 7 , 21 , 49 ). To confirm that the observed induction of ALT in our system was a BIR-dependent process, we knocked down BLM, POLD3 and RAD52 in our cells in untreated and PDS treated cells (Supplementary Figure  S4A). We found that knockdown of BLM and POLD3 sig-nificantly reduced the induction of C-circles in our ATRX knockout cells treated with PDS ( Figures 1 I and J). APB induction was also pre v ented when BLM was knocked down (Supplementary Figures S4B and C). RAD52 knockdown, howe v er, failed to reduce C-circle le v els (Figures 1 I and J), supporting pre vious wor k suggesting that C-circle formation is RAD52-independent ( 7 ). We also confirmed that BLM is recruited to telomeres upon PDS treatment, and this recruitment is significantly greater in ATRX knockout cells versus wildtype ( Supplementary Figures S4D and E).
ATRX is recruited to GC-rich and repetiti v e regions of the genome --such as telomer es --wher e it has been shown to have a role in unfolding of G4-structures at these genomic regions ( 30-32 , 50 ). We therefore examined the recruitment of ATRX to telomeres following PDS treatment. Imm uno-FISH anal ysis demonstrated an increased number of ATRX foci per nucleus and a strong co-localisation of ATRX protein at telomeres following exposure to the drug, supporting the previously described role for ATRX in the resolution of G4s (Supplementary Figures S4F-H). Taken together, these data confirm that following treatment with the G4-stabilising agent PDS, there is recruitment of ATRX to telomeres and, in the absence the ATRX, telomeres undergo aberrant processing via a BIR-dependent process resulting in markers of the ALT pathway.
As a small minority of ALT-cancer cells exhibit DAXX loss , rather than ATRX loss , we generated CRISPR-Cas9 mediated DAXX knockout clones in the HeLa LT cell line, both in the context of wildtype ATRX and ATRX knockout (Supplementary Figure S5A). On treatment with PDS, markers of ALT were observed in the DAXX-depleted clones as were previously seen in the ATRX-depleted clones (Supplementary Figures S5B-D). Importantly, codepletion of both ATRX and DAXX failed to confer any cumulati v e increase in cardinal ALT markers, supporting an epista tic rela tionship between the two proteins and implicating the H3.3 deposition pathway in the pre v ention of ALT.
ALT has been suggested to be associated with telomeric replicati v e stress, with an accumulation of RPA2 at telomeres upon ALT induction ( 21 , 51 ) and an increase in single stranded telomeric DNA (ssTel), as detected by nondenaturing telomeric FISH ( 52 ). Addition of PDS significantly increased detectable RPA-ssTel foci in both ATRX knockout clones, with no detectable increase in wildtype ATRX HeLa LT cells ( Supplementary Figures S5E and F). Similarly, an increase in RPA-ssTel foci was seen in the DAXX knockout clones (Supplementary Figure S5G).

T r apping of TOP2A protein leads to induction of ALT
It has been suggested that stable non-canonical DNA structures, such as G4s, promote the formation of topoisomerase 2 (TOP2)-mediated DNA breaks ( 53 ). The cytotoxic effect of G4 stabilizing drugs (such as PDS and CX-5461) has been shown to be dependent on trapping of TOP2A and presence of R-loop structures ( 54 , 55 ). This led us to consider whether the observed induction of ALT upon PDS trea tment of A TRX -null HeLa LT cells is dependent on TOP2A trapping. The ra pid a pproach to DN A adduct recovery (RADAR) assay ( 46 ) was performed and this confirmed that PDS treatment led to a significant increase in trapped TOP2A le v els on the DNA and --to a lesser extent --TOP1 (Supplementary Figures S6A and B). TOP2A was depleted using siRNA in the HeLa LT clones (Supplementary Figure  S6C), and this abrogated the le v el of C-circles and APBs observed upon PDS treatment in ATRX null cells (Figures  2 A, B and 1 D), suggesting that PDS-initiated induction of ALT was indeed dependent on TOP2A. To confirm that this result was not specific to PDS, a second G4-stabiliser (CX-5461) was used; this also induced ALT hallmarks (Supplementary Figures S6D-F) and caused increased le v els of trapped TOP2A (Supplementary Figure S6G), as previously reported ( 55 , 56 ).
To explore this pathway further, we next treated the cells with etoposide (ETO), a canonical TOP2 poison which directly stabilises the TOP2 cleavage complex (TOP2cc) on DNA. The increased trapping of TOP2A was confirmed by RADAR assay (Supplementary Figures S6A and B). Strikingly, treatment of the ATRX knockout clones with ETO elicited an increase in C-circles, APBs and RPA-ssTel foci to le v els comparab le to U2OS cells, whereas no notab le increase was observed in the ATRX wildtype cells (Figures 2 C-H). Knockdown of TOP2A decreased the effect of ETO, pre v enting the robust induction of ALT markers, ther efor e r einfor cing the r equir ement for TOP2A trapping, as opposed to loss of activity, in triggering the ALT pathway upon loss of ATRX (Figures 2 C, D).

T r apping of other proteins induces ALT in ATRX-deficient cells
DNA protein covalent complexes (DPCCs) are formed as transient intermediates in a variety of protein-DNA interactions and more than 20 different proteins have been reported to form DPCCs ( 46 ). As such, we next considered whether the induction of ALT in ATRX-null cells was restricted to trapping of TOP2A, or whether the enforced trapping of other DNA interacting proteins could elicit the same effect.
Topoisomerase I (TOP1) can be trapped onto DNA through treatment with TOP1 poisons such as camptothecin (CPT) and this effect was confirmed in our system (Supplementary Figure S6A and B). Treatment with CPT led to a strong induction of C-circles, APBs and RPA ssTel foci in ATRX-deficient cells. (Figures 3 A-F). It was noted that, in the case of APBs, there was also a significant increase in ATRX wildtype cells treated with CPT, howe v er the number of APBs were still significantly greater in the two ATRX knockout clones (Figure 3 D). The same effect was noted in DAXX knockout clones ( Supplementary Figures S7A-C). Induction of ALT was not observed when cells wer e tr eated with concurr ent knockdown of TOP1 protein and CPT, consistent with the observed effect being due to trapping of TOP1 rather than loss of catalytic activity (Figures 3 A, B and Supplementary Figures S8A and B).
PARP1, a central player in the DNA damage response, is also known to accumulate on chromatin when cells are treated with a variety of PARP inhibitors (PARPi). It should be noted, howe v er, tha t ra ther than the covalent complexes formed by TOP1 (Top1cc) and TOP2A (Top2cc), the trapping of PARP1 on DNA is non-covalent in nature. It was recently demonstrated that the persistent accumulation of PARP1 foci upon treatment with PARPi was due to rapid and continuous recruitment and exchange of PARP1 at DNA lesions ( 57 , 58 ).
Notably, the potency of different inhibitors to trap PARP1 varies markedly, with allosteric differences in trapping behaviours (Supplementary Figure S9A). PJ34 is a catalytic inhibitor of PARP1 activity that has been reported to have no trapping action ( 59 ). Conversely, talazoparib is a PARPi which has consistently demonstrated extremely strong trapping potency (59)(60)(61)(62). It has previously been reported that niraparib strongly traps PARP1, olaparib exhibits an intermediate le v el and v eliparib e xhibits no trapping potency, a pattern which does not correlate with the catalytic inhibitory properties for each drug (61)(62)(63). This dogma has, howe v er, been challenged recently, with one group demonstra ting tha t although olaparib does display tr apping tendencies, nir aparib might belong to the nontrapping class of agents ( 58 ). We considered it essential, ther efor e, to confirm the trapping potency of these agents in our cellular system.
Nuclear and chromatin protein fractions were extracted from untreated cells and those treated with a PARPi panel. This demonstra ted tha t PJ34 and veliparib did not induce trapping of PARP1, whilst olaparib induced a moderate amount of trapping (Figures 4 A, B). Niraparib and talazoparib were both found to strongly trap PARP1 on chromatin, with talazoparib having e v en greater action than niraparib (Figures 4 A, B). Importantly, all the drugs reduced PARylation le v els to a similar degree (Supplementary Figure S9B). We took advantage of these properties to further test our hypothesis that it is the trapping of proteins on DNA that dri v es the ALT pa thway. Trea tment of ATRX wildtype cells with PARP1-inhibitors failed to induce C-cir cles (Figur es 4 C, D). Strikingly, howe v er, treatment of both ATRX knockout clones with trapping PARPi led to an induction of C-circles (Figures 4 C, D and Supplementary Figures S9C and S9D). Veliparib and PJ34 (nontrapping PARPi) failed to elicit any detectable C-circles, w hilst ola parib, nira parib and talazoparib elicited a stepwise increase in C-circles (Figures 4 C, D and Supplementary Figures S9C and S9D). There was a striking correlation between trapping potency and le v el of induced C-circles ( R 2 = 0.971, P = 0.0003) (Figure 4 G). The same pattern was observed upon APB analysis (Figures 4 E, F and Supplementary Figure S9E), with a similarly striking correlation ( R 2 = 0.880, P = 0.0005) (Figure 4 H). As seen with CPT, there was a significant increase in APB le v els in ATRX wildtype cells treated with the strongest trapper, talazoparib, howe v er once again the effect in the ATRX knockout clones was significantly greater (Figure 4 E). As seen with other agents, induction of ALT was not observed when cells were tr eated with concurr ent knockdown of PARP1 protein and PARP trapping drugs, again confirming the mechanism of action as protein tr apping, r ather than loss of protein catalytic activity (Supplementary Figures S9F-H).
As the effect of protein trapping did not appear to be specific to any one protein, but rather a general effect of DPCC formation or protein trapping, we next investigated the   effect of for maldehyde. For maldehyde is a potent crosslinking agent, which forms covalent bonds between proteins and DNA. Treatment of the cells with formaldehyde had no effect on APBs in ATRX wildtype HeLa LT cells but caused a striking increase in ATRX-deplete clones, although curiously no C-circles could be amplified (Figures 5 A, B and Supplementary Figure S10A). A possib le e xplanation for this discrepancy is that formaldehyde is causing crosslinking of the GC-rich C-circle DNA, thereby prohibiting rolling circle amplification.
Taken together, this data suggested that it is the trapping of proteins on DNA that dri v es ALT in the absence of ATRX and not replicati v e stress per se . To further test this hypothesis, we treated the ATRX deficient cells with the DN A pol ymerase inhibitor aphidicolin (APH) and hydroxyurea (HU), at both low and high doses. APH has previously been shown to generate replicati v e stress at common fragile sites, including telomeres ( 8 ), but does not generate DPCCs, and hydroxyurea (HU) causes replicati v e stress through depletion of nucleosides. APH and HU treatments failed to generate a significant increase in C-circles or APBs in any HeLa LT cell clones, including the ATRX knockout clones --e v en when treated at high doses (Supplementary Figures S10B-F). Taken together, we conclude that trapping of proteins on DNA, in combination with ATRX loss, can induce the ALT pathway.

Induction of ALT by trapped proteins is dependent on MUS81-endonuclease activity
Trapping of proteins on DNA leads to e xcessi v e replicati v e stress, because when a trapped protein is encountered by the replication f ork, f ork stalling and reversal will occur unless the trapping is resolved. We saw an increase in ␥ H2AX le v els in our ATRX knockout cells treated with trapping agents (Supplementary Figure S11A and S11B), indicati v e of replication fork collapse and the generation of DSBs. MUS81 endonuclease has been shown to re v erse e xcessi v e DNA supercoiling resulting from protein trapping ( 64 ). Further, MUS81 localises to APBs in ALT cells, and depletion of MUS81 results in reduction of ALT specifictelomer e r ecombination ( 65 ). During S-phase, the MUS81-EME2 complex is responsible for the restart of stalled replication forks and telomere recombination ( 66 ). It was consider ed, ther efor e, whether the observed induction of ALTmarkers on trapping of proteins was dependent on MUS81 activity. Cells were treated with siRNA to MUS81 to knockdown expression (Supplementary Figure S11C). In contrast to drug treatment alone, when ATRX-deficient cells were co-treated with a protein trapping drug (ETO, CPT, niraparib or PDS) and siMUS81, there was minimal induction of markers of ALT (Figures 5 C-E and Supplementary Figures S11D-F).
RPA stabilisation of single-stranded DN A (ssDN A) intermediates arising from a resected DSB, has previously been shown to be critical for BIR in yeast ( 67 ). Consistent with this, following treatment of cells with PDS, there was recruitment of RPA to telomeres in ATRX-deficient (but not ATRX-wildtype) cells ( Supplementary Figures S11D  and E). Howe v er, on co-treatment with PDS and siMUS81, there was a significant reduction in RPA recruitment to the telomeres, indica ting tha t in the absence of MUS81, such ssDNA intermediates were not generated ( Supplementary  Figures S11D and E). We conclude , therefore , that the induction of ALT in the context of ATRX-loss and presence of DPCCs is dependent on MUS81 endonuclease activity and generation of DSBs by MUS81. Given the welldescribed role of MUS81 in resection of stalled replication forks (64)(65)(66), we believe that it is highly likely the observed DSBs arise from collapse of stalled replication forks.

ATRX-deficient cells have higher levels of trapped protein on DNA
Finally, we investigated whether ALT+ cell lines have increased le v els of trapped proteins. Previous proteomic work suggested that TOP1, TOP2A and PARP1 can all be detected at telomeres of the ALT+ U2OS cell line, potentially indicati v e of protein trapping at telomeres (Supplementary Figure S12A) ( 21 ). Re-analysis of this BioID data indicated that, in U2OS cells, there is a particular excess of TOP2A and TOP2B at telomeres, as compared to HeLa LT ( 21 ) (Supplementary Figure S12B). This might indica te tha t in this cell line, the predominate telomeric trapped protein is TOP2A / B, howe v er, it is difficult to draw firm conclusions as the comparator is a non-isogenic line. We next took advantage of a monoclonal antibod y tha t specificall y reco gnises TOP1cc but not free TOP1 or DNA ( 68 ). Consistent with a specificity of the antibody for TOP1cc, imm unofluorescence anal ysis showed a significant increase in the le v els of both total and telomeric TOP1cc in both ATRX wildtype and deplete HeLa LT cells upon tr eatment with CPT (Figur es 6 A-C). We note that levels of TOP1cc were generally higher in the ATRX deplete cells, and this increase was exacerbated upon CPT treatment (Figures 6 A-C). Importantl y, w hen comparing the ALT + cell line SW26 with the isogenic telomerase positi v e, ALT-cell line SW39 ( 69 ), we found that the ALT+ cell line exhibited 2.4-times higher levels of trapped TOP1 protein as compared to its ALT-isogenic counterpart; the result was strongly significant ( t = 6.22, P < 0.0001) (Figures 6 D, E). The le v el of TOP1cc in a panel of natural ALT+ cell lines was also assessed; this demonstra ted tha t ALT+ cells generally had higher genome-wide le v els of trapped TOP1 protein in comparison to ALT-cells (Figure 6 F).
It has previously been shown that knockdown of the histone chaperones ASF1a and ASF1b in HeLa LT cells leads to induction of the ALT pathway ( 21 ). We knocked down ASF1 in our HeLa LT cells (Supplementary Figure S13A) and confirmed this led to induction of the ALT pathway (Supplementary Figure S13B). Strikingly, this induction of ALT correlated with a significant increase in TOP1cc levels ( Supplementary Figures S13C and D), indicating that trapping of proteins on DNA is strongly linked to ALTinduction in these cells.
Next, the effect of protein trapping in the canonical ALT+ osteosarcoma cell line, U2OS, was explored. Crucially, depletion of TOP1 through both siRNA and shRNA (Supplementary Figure S14A) led to a significant reduction in both APBs and C-circles in U2OS ( Figure 6 G and H). We also tested another ALT+ cell line, MGBM1, howe v er in this case found no effect on C-circle le v els and only a minor   Figures S14B and S14C). This suggested that in natural ALT+ cancer cells, formation of TOP1cc is not the only natural cause of ALT activity, consistent with our earlier observations that trapping of various proteins can induce the phenotype.
To explore this further we knocked down TOP2A (Supplementary Figure S14D) and found there was a decrease in C-circle le v els in both MGBM1 and U2OS cell lines (Supplementary Figures S14E and S14F). There was a modest but statistically significant decrease in APBs observed in MGBM1, but no observed difference in U2OS cells (Supplementary Figure S14G). Finally, knockdown of PARP1 using shRNA in U2OS (Supplementary Figure S14H) led to a significant decrease in both C-circles and APBs (Supplementary Figures S14I-K). Knockdown of PARP1 was additionally attempted in MGBM1, but the cells were not viable.

DISCUSSION
DNA is always associated with various proteins as part of normal genomic processes --howe v er these frequent, close interactions carry the risk of formation of either DNApr otein covalent cr oss-links (DPCCs) or non-covalent trapping of proteins on DNA. Many enzymes --such as topoisomer ases, polymer ases, methyltr ansfer ases, glycosylases and pol y(ADP-ribose) pol ymerases --form re v ersib le covalent intermediates with DN A during catal ysis, and these intermediates might accidentally form DPCCs ( 70 ). DPCCs are ubiquitous in the genome but are bulky and can, therefor e, interfer e with vital processes such as DNA replication, transcription, chromatin remodelling, and DN A topolo gy manipulation (71)(72)(73)(74)(75)(76). DPCCs can be generated by a variety of endogenous or environmental sources, including chemotherapeutical drugs, ionizing or UV radiation and oxidati v e stress, as well as erroneous enzymatic activity of normal DNA-interacting proteins. Enhancement of protein trapping has become an important mechanism of action for chemotherapeutic agents, such as etoposide, camptothecinderi vati v es (such as irinotecan), and niraparib. It has been known for some time that ALT cancers are particularly sensiti v e to these a gents, b ut the link between the ALT pathway and DPCC formation has not previously been investigated.
Here, it was demonstrated that se v er al chemother apeutic agents were capable of inducing ALT in HeLa LT cells which lacked ATRX, but not in ATRX wildtype clones. The agents which were able to induce ALT markers in the ATRX-deplete cells wer e p yridostatin and CX-5461 (G4 stabilising agents that trap TOP2A), camptothecin (TOP1 poison), etoposide (TOP2 poison), and niraparib and talazoparib (PARP1 trappers). The same effect was observed when the ATRX-deplete clones were treated with formaldehyde, an agent which induces widespread cross-links between protein and DNA. Crucially, olaparib (a less potent PARP1 trapping agent) did not induce markers of ALT to the same extent as niraparib or talazoparib, and PJ34 and veliparib (non-trapping PARP1 inhibitors) did not induce any markers of the ALT phenotype. Further, treatment with aphidicolin and hydroxyurea failed to elicit an increase in car dinal ALT mar kers in ATRX-deplete cells. These data strongly support the concept that it is not replicati v e stress per se driving ALT in ATRX-deplete cells, but replicative stress specifically caused by trapping of proteins.
It was considered likely, therefore, that trapping of proteins --including TOP1, TOP2 and PARP1 --on DNA was a crucial step in initiation of ALT processes --and so it was conjectured that depletion of the implicated proteins should not have the same effect as poisons. It was demonstra ted tha t depletion of TOP1 and PARP1 in two naturally occurring ALT lines (U2OS and MGBM1) led to a significant decrease in APBs; the effect of depletion of TOP2 led to a small decrease in APBs in MGBM1 but not U2OS. The effect on C-circle le v els was less clear cut, with a decrease observed on silencing of PARP1 and TOP2A, but not on TOP1. It is likely, then, tha t the ef fect is not specific to one protein, but rather a generic effect of protein trapping, the magnitude of the effect possibly being related to relati v e gene e xpression le v els, temporospa tial rela tions or e v en stochastic effects. Further proteomic investigations into these questions might re v eal insights into ALT-cancer biology.
Next, we considered the role of MUS81-endonuclease, an enzyme which has been implicated in the repair of DNA in response to protein trapping, restart of stalled replication forks and telomere recombination in ALT-cells (64)(65)(66). As expected, in the absence of MUS81, protein-trapping drugs were unable to induce markers of ALT in ATRX-deficient cells. Further, ALT telomere synthesis has previously been demonstrated to be reliant on BLM and the BIR pathway; here, we demonstrated that the induction of ALT observed under conditions of protein trapping was, indeed, dependent on these pathways. This confirmed that in our artificial ATRX knockout cellular system, trapping of proteins was likely initiating ALT through the same pathway that naturally occurs in ALT + malignant cells: that is, MUS81and BLM-dependent break-induced replication. Crucially, it has been demonstra ted tha t MUS81 endonuclease has a role in re v ersal of e xcessi v e DNA supercoiling resulting fr om pr otein trapping, as well as cleavage of replication inter mediates in G2, for med as a result of DNA protein cross-links, to facilitate BIR ( 64 , 77 ). Taken together, these findings provide a cogent model of ALT-biolo gy, w here trapped proteins lead to stalling of replication fork progression, with subsequent recruitment of MUS81 and fork resection, which provides the genetic substrate for BLMmediated BIR, but only in the absence of ATRX / DAXX (Figure 6 I).
An important consideration is whether protein trapping is one potential natural driving force behind ALT, or an artificial way that ALT can be induced in cellular systems using chemothera peutic agents. Firstl y, w e show ed that knockdown of ASF1 also elicited an increase in TOP1cc and concurrent induction of ALT; this supports the conclusion tha t accumula tion of trapped proteins is intrinsic to the ALT process, rather than a coincidental e v ent. Further, we showed that natural ALT + cell lines exhibit higher baseline le v els of trapped TOP1cc relati v e to ALT-cells, including in isogenic paired AL T+ / AL T-lines (SW26 / SW39). It is unlikely, howe v er, that a single trapped protein is responsible for this phenomenon. We observed that knockdown of TOP1, TOP2A, PARP1 in two ALT cell lines -U2OS and MGBM1 -had differing magnitudes of effect, indicating cell-specific differences. Further, it was previously demonstra ted tha t TOP2cc were enriched at telomeres in ATRXdeficient cells ( 78 ). Future work to characterise the repertoire of trapped telomeric proteins across ALT cancer cells will be of great interest and help answer these questions.
Another important consideration concerns the role of ATRX in pre v enting for k collapse in the presence of protein trapping. ATRX has previously been implicated with multiple roles at the replication fork which could explain the data we present here, including roles in both the protection of stalled forks from nucleolytic degradation ( 33 ) and the restart of stalled replication forks ( 34 , 37 ). Of note, ATRX has been proposed to interact with factors known to modulate both fork restart and protection and has also been shown to interact and cooperate with FANCD2 to recruit CtIP to stalled replication forks and promote MRE11 dependent fork restart ( 37 ). ATRX itself has also been shown to interact with components of the MRN complex, raising the intriguing possibility that ATRX constitutes a component of a fork restart complex ( 20 , 34 , 37 ). Understanding how these interactions are regulated will likely gi v e important insights as to how ATRX facilitates progressi v e DNA replication in the presence of trapped proteins on DNA.
Taken together, these data strongly suggest that trapping of protein on DNA is one potential key driving force in the natural aetiology of ALT, although why ALT cells should have higher levels of trapped proteins remains to be explor ed. Inter estingly, we observed that knock out of ATRX in the HeLa LT cell line in itself elicited an increase in TOP1cc le v els. This data is consistent with pre vious findings which has shown that ATRX loss is associated with increases in R-loops, including telomeric R-loops ( 29 ) which have been linked to increases in DNA supercoiling and protein trapping. Howe v er, gi v en that these cells do not exhibit canonical markers of ALT prior to treatment with genotoxic agents, we can conclude that this small increased basal le v el of TOP1cc is not sufficient to trigger the ALT pa thway in isola tion. Within a tumour, howe v er, we must consider the heterogenous cellular population due to clonal evolution, and the effect of tumour micr oenvir onment. It is likely that the increased le v el of trapped protein in ATRXdeficient cells dri v es the initiation of ALT when in the context of other genetic, metabolic or cellular challenges that further increase the le v els of DPCCs.
Of note, the human cancers in which ALT is most prominent are also those most likely to harbour mutations in IDH1 ( 2 ). The IDH1 R132H mutation leads to the accumulation of the oncometabolite 2-hydroxygluterate (2-HG) which in turn inhibits multiple enzymes related to cytosine and histone methylation, resulting in perturbations in the expression of multiple genes. Recent research has shown that IDH1 R132H mediated loss of XRCC1 expression is key to ALT induction in ATRX deplete cells ( 79 ). Of note a novel role has recently been identified for XRCC1 in the pre v ention of trapping of PARP1 on single strand break intermediates during base excision repair ( 80 ). As such, it is tempting to specula te tha t IDH1 mutant ALT cancers may exhibit high levels of trapped PARP1 through loss of XRCC1 protein activity.
The formation of TOP1cc can be facilitated by the interaction of TOP1 with oxidised bases such as 7,8-dihydro-8-o x oguanine (8-o x oG) ( 81 ), suggesting that the accumulation of oxidati v e stress and reacti v e oxygen species (ROS) in ALT cancers might serve as a pathway to trap TOP1. Consistent with this possibility, telomerase suppression in a mouse lymphoma model has been found to lead to development of ALT+ tumours and, of note, the resultant tumours characteristically exhibited mitochondrial dysfunction and increased le v els of ROS ( 82 ). Each of these hypotheses will need to be experimentally explored in more detail to fully elucidate the mechanisms underlying the observed phenomenon.
Here, we propose a model whereby, in the absence of ATRX or DAXX, forks become unstable upon encountering a DPCC. Gi v en that many proteins have been purported to have a role in fork stability this raises the important question as to w hy ATRX a ppears to be the dominant tumour suppressor for ALT cancers? We consider the likely explanation for this phenomenon is the reported role for ATRX in the maintenance of telomere sister chromatid cohesion, thereby allowing for out of register BIR and net changes in telomere length ( 22 , 83 ). As such, this dual functionality of ATRX means loss of ATRX is uniquely dangerous for the activation of ALT.
Finally, the translational and clinical implications of these findings are of high importance. TOP1 / TOP2A and PARP poisons ar e alr eady an important component of many treatment protocols for glioblastoma, sarcoma and other ALT + tumour types. Howe v er, understanding that protein trapping induces a hyper-ALT phenotype should inform more rational drug combination therapy --for example, combining a non-trapping PARPi (such as veliparib) with a trapping agent (such as irinotecan) could be nonadditi v e at best, or antagonistic at worst. It should also inform combinations with non-trapping based drugs --for example, studies have demonstra ted tha t FANCM loss leads to a hyper-ALT phenotype, and so we could predict that combination of these agents with trapping drugs will have a synergistic effect (10)(11)(12). It is also hoped that further insight into the basic biology of ALT cancer cells could lead to the design of more novel targeted chemotherapeutic agents and, e v entually, to improv ed outcomes for individuals with ALT+ cancers.

DA T A A V AILABILITY
ImageJ scripts for z-projecting of Deltavision image files together with CellProfiler pipelines for foci quantification and co-localisation analysis of the projected images are available at https://github.com/CLYNESLAB/Image-Analysis and https://doi.org/10.5281/zenodo.7630004 . All other data is contained within the manuscript and / or supplementary files.

SUPPLEMENT ARY DA T A
Supplementary Data are available at NAR Online.