Visualising G-quadruplex DNA dynamics in live cells by fluorescence lifetime imaging microscopy

Guanine rich regions of oligonucleotides fold into quadruple-stranded structures called G-quadruplexes (G4s). Increasing evidence suggests that these G4 structures form in vivo and play a crucial role in cellular processes. However, their direct observation in live cells remains a challenge. Here we demonstrate that a fluorescent probe (DAOTA-M2) in conjunction with fluorescence lifetime imaging microscopy (FLIM) can identify G4s within nuclei of live and fixed cells. We present a FLIM-based cellular assay to study the interaction of non-fluorescent small molecules with G4s and apply it to a wide range of drug candidates. We also demonstrate that DAOTA-M2 can be used to study G4 stability in live cells. Reduction of FancJ and RTEL1 expression in mammalian cells increases the DAOTA-M2 lifetime and therefore suggests an increased number of G4s in these cells, implying that FancJ and RTEL1 play a role in resolving G4 structures in cellulo. Direct observation of G-quadruplexes (G4s) in live cells is challenging. Here the authors report a method to identify G4s within the nuclei of live and fixed cells using a fluorescent probe combined with fluorescence lifetime imaging microscopy.

G uanine-rich sequences of DNA can fold into tetrastranded helical assemblies known as G-quadruplexes (G4s). These structures are implicated in a number of essential biological processes such as telomere maintenance, transcription, translation, and replication [1][2][3][4] . A series of bioinformatical studies initially suggested that there are over 350,000 putative G4-forming sequences in the human genome 5,6 , with subsequent studies reporting an even larger number 7 . Following these predictions, sequencing studies using purified human genomic DNA revealed over 700,000 sequences that form G4 structures under in vitro conditions 8 . More recently, the prevalence of G4s in human chromatin has been investigated using an immunoprecipitation technique 9,10 . These studies showed that there are over 10,000 sequences in the human genome that can form G4 DNA structures under cellular conditions 9 . Interestingly, these G4 structures are mainly located in gene promoter regions and in 5′-untranslated regions (5′-UTR) of genes supporting the proposed hypothesis that G4 DNA is involved in a number of essential biological regulatory processes.
While the exact roles that G4 structures play in biology are still under significant scrutiny, it is commonly accepted that G4 formation can be detrimental to certain biological processes and can lead to DNA damage 11,12 . Therefore, it is not surprising that several helicases such as Pif1, RecQ, RTEL1, FancJ, and BLM have been found to unfold G4 structures in vitro 13 . While it is known that G4 DNA helicases are important in maintaining genome integrity in cells, the direct link between their in vitro G4 unwinding activity and genome instability associated with their mutations is still missing.
Considering the wide range of biological processes associated with G4s, there has been significant interest in developing tools to detect and visualise G4 DNA structures in cells. With widespread application in immunofluorescent staining, high-affinity antibodies have been developed to visualise G4 in cells [14][15][16][17][18][19] . An early antibody found to be selective against telomeric G4 showed nuclear staining in the ciliate Stylonychia lemnae 15 . Subsequent studies have reported high-affinity antibodies able to visualise G4 DNA and G4 RNA in mammalian cells by immunofluorescent staining [16][17][18] . While these elegant studies are the most direct evidence of the presence of G4s in cells, they have some potential drawbacks. The fixation process can denature the cellular DNA and induce DNA degradation 20,21 , and the high affinity of the antibodies for G4 could artificially increase the presence of G4 structures. Furthermore, antibodies are not suitable for use in live cells and hence cannot be used to study G4 dynamics in real time.
These limitations can be overcome by using small-molecule optical probes [22][23][24] . Most G4 DNA optical probes reported to date rely on a large enhancement in emission intensity upon binding G4 ('switch-on'), compared to binding duplex (ds) DNA 24 . For example, Thioflavin T (ThT) and its derivatives have shown to be switch-on probes for G4s, with up to 150-fold fluorescence enhancement vs. dsDNA (under optimal conditions) and have been used for live cell imaging of both DNA and RNA G4s [25][26][27] . However, ThT is known to bind to other biologically relevant molecules (e.g. protein aggregates 28 ) and is highly sensitive to the matrix microviscosity 29 , potentially triggering non-G4 emission. While these small-molecule probes provide a good way of studying G4s in cell-free environments, their emission intensity is concentration dependent. In the vast excess of nuclear dsDNA (and other biomolecules), distinguishing G4 foci from background emission is challenging, even with a very high switchon ability. Additionally, extracting quantitative information from emission intensity alone is difficult, since the cellular concentrations of neither the probe nor G4s are known. More complex approaches are required to gain insight in cellulo, such as the recently reported study using single-molecule fluorescence imaging for G4 detection 30 .
An alternative approach is to use a change in the fluorescence lifetime (a concentration-independent parameter) of a probe upon binding to different DNA structures [31][32][33][34][35] . For example, the fluorescence lifetime of 3,6-bis(1-methyl-2-vinyl-pyridinium) carbazole diiodide (o-BMVC) is longer when bound to G4 (ca. 2.5 ns) than non-G4 DNA sequences (ca. 1.5 ns). This difference was used to identify long lifetime foci in both the nucleus and cytoplasm of mammalian cells, however, this dye is only permeable to fixed cells and so no dynamic live cell work was possible 33,34 . More recently, a fluorescence lifetime-based probe (a tripodal cationic fluorescent molecule, NBTE) was reported that can distinguish G4 (3-4 ns) and dsDNA (2.0-2.5 ns) in live cells 35 . We previously reported that DAOTA-M2 [ Fig. 1a] has a very different fluorescence lifetime when bound to G4 structures as compared to duplex or single-stranded DNA, with good live cell permeability and low cytotoxicity 32 . Herein we show that DAOTA-M2 can be used to visualise dynamic processes involving G4s in live cells. We confirm that while the probe works in fixed cells, there are differences in G4 abundance, highlighting the importance of carrying out G4 imaging experiments in live, rather than fixed environments. We demonstrate that DAOTA-M2 responds to the expression of DNA helicases FancJ and RTEL1 which are involved in genome stability and the distribution of G4 in live cells. Finally, we present a quantitative fluorescence lifetime-based assay to visualise the interaction of small molecules (which are not fluorescent themselves) with G4 structures in live cells.

Results
Biophysical characterisation of DAOTA-M2 interacting with DNA. Previously, we have shown that DAOTA-M2 is a medium-strength DNA binder 36 , with a moderately greater affinity for G4 over dsDNA (K d = ca. 1.7 µM for dsDNA and ca. 1.0 µM for G4) 32,37 . Despite similar binding affinities, the intensity-weighted average lifetime (τ w ) upon binding is found to be dependent on the DNA topology. These values range from ca. 5-7 ns when bound to dsDNA, ca. 7-11 ns when bound to RNA and ca. 9-12 ns when bound to G4 DNA [ Fig. 1b, c] and are independent of absolute concentration [ Supplementary Fig. 1a]. Given the complex excited state decay of DAOTA-M2, we have now optimised our fitting algorithm and have chosen to use τ w as a reporter instead of τ 2 as used previously, due to a more straightforward and accurate fitting of fluorescence lifetime imaging data (vide infra) 30 .
To establish the effect of proteins, lipids, carbohydrates and biomolecules other than nucleic acids on the lifetime of DAOTA-M2, we used a highly protein-concentrated and nucleic aciddepleted Xenopus egg extract 38 . τ w increases from 2.5 ± 0.3 ns in aqueous buffer to 7.3 ± 0.3 ns in cell extract [ Fig. 1d], indicating an effect from other biomolecules on the fluorescence lifetime. Reassuringly, τ w increases to 10.8 ± 0.6 ns on addition of G4 to the extract, and decreases to 6.1 ± 0.3 ns with dsDNA, consistent with the results observed in aqueous buffered solutions [ Fig. 1c]. This result implies a higher affinity of DAOTA-M2 for DNA over other cellular biomolecules. In a nuclear environment, given the vast excess of nucleic acid, DAOTA-M2 should preferentially bind to DNA and the lifetime will give an indication of the DNA topology.
We next set out to examine whether the DAOTA-M2 lifetime can report on competitive interactions of other binders to G4 in vitro. Due to a higher affinity for quadruplexes, addition of G4 to a mixture of DAOTA-M2 and dsDNA increases the lifetime from 6.1 ± 0.2 ns [Fig. 1e, green trace] to 11.7 ± 0.4 ns [Fig. 1e, red trace].
The subsequent addition of pyridostatin (PDS), a selective G4 binder with a greater affinity for G4 than DAOTA-M2 (K d = ca. 0.5 39,40 and ca. 1.0 µM 32 , respectively) 41 , displaces DAOTA-M2 from G4 back to dsDNA, accompanied by a drop in τ w [Fig. 1f, black dots]. When the experiment is repeated using DAPI, which is a well-established dsDNA minor groove binder 42 and non-G4 binder, only a small drop in τ w is observed as DAOTA-M2 stays bound to G4 [Fig. 1f]. The dynamic equilibrium between dsDNA and G4 bound DAOTA-M2 can also be disrupted through the addition of a large excess of dsDNA into a solution of G4 and DAOTA-M2 [ Supplementary Fig. 1b].
We then investigated DAOTA-M2 displacement using Nisalphen, another DNA binder which has a higher affinity for both G4 and dsDNA than DAOTA-M2 [ Supplementary Fig. 1c] 43,44 . In this case τ w drops below the region associated with dsDNA to ca. 3.7 ns, close to that of free dye [ Fig. 1f, orange dots]. A range of metal-salphen complexes (with metal = Ni, Cu, VO, Zn) are an ideal series to study structurally related molecules with different G4 affinities, as variation in the metal centre changes the metal coordination geometry, which has a dramatic effect on G4 binding: strong (Ni, Cu), medium (VO) and weak (Zn) 45,46 . Our DAOTA-M2 lifetime-based data [ Fig. 1g] confirms a similar This data indicates that in mixtures of G4 and dsDNA, analysis of the DAOTA-M2 lifetime can accurately differentiate between distinct DNA topologies, thus, we next looked to investigate DAOTA-M2 in live cells using fluorescence lifetime imaging microscopy (FLIM).
Imaging G4s in cellulo using DAOTA-M2 and FLIM. Given the in vitro assay results [ Fig. 1] confirming that DAOTA-M2 lifetime can predict the strength of the interaction between small molecule binders and G4s, we set out to use DAOTA-M2 to study the interaction of any molecule with G4s in live cells. Incubation of U2OS cells with DAOTA-M2 (20 µM, 24 hr) results in nuclear staining [ Fig. 2a], no obvious changes to cell morphology [ Supplementary Fig. 4c], no increased DNA damage response [ Supplementary Fig. 4a, b], and some accumulation of cells in S and G2 phases of the cell cycle [ Supplementary Fig. 3a, b].
We were able to record FLIM images at high magnification, making it possible to visualise spatial lifetime distributions within the nucleus [ Fig. 2d]. Consistent with in vitro experiments, fluorescence decays were fitted to a bi-exponential decay model [ Fig. 2f] to obtain the corresponding fluorescence lifetimes (τ w ). Figure 2b, d show resulting FLIM maps; each cell nucleus displays spatially heterogenous lifetime distributions, with areas corresponding to long (12-13 ns, blue) and short (9-10 ns, red) lifetimes. Even at this high resolution, each pixel will contain lifetime information from DAOTA-M2 bound to a large number of duplex and G4 DNA sites which results in a histogram of lifetime distribution within a 10-12 ns range [ Fig. 2e  PDS (as well as many other G4 binders) is known to cause DNA damage, arrest cell growth and activate DNA damage response (DDR) pathways 47 . To establish whether the observed changes in DAOTA-M2's fluorescence lifetime in cells could be due to PDSdependent DNA damage rather than displacement of the probe from G4 structures, we carried out a control experiment with cisplatin. This compound is known to form DNA intra-strand links and activate the apoptotic pathway, but not to bind G4 DNA 48 . Encouragingly, co-incubation of cisplatin with DAOTA-M2 did not lead to a decrease in the fluorescence lifetimes recorded by FLIM [ Supplementary Fig. 8a]. We also caused DNA damage by inducing double strand breaks with 2 Gy gamma irradiation. Irradiation of cells had no effect (p = 7.4 × 10 −1 ) on the DAOTA-M2 lifetime [ Fig. S8b], confirming that these two types of DNA damage were not the cause of the lifetime decrease observed after incubation of cells with PDS.
FLIM studies using fixed cells. It is common for immunostaining or small molecule staining of G4 in cells to be performed following PFA fixation (needed for permeabilising the cellular membrane), which crosslinks proteins, increases cell rigidity and chemically alters the cell morphology. This process is known to impact nucleic acids within the cell, denaturing DNA and causing DNA damage 20,21 . However, DNAse and RNAse treatment is possible in fixed cells, and allows for examination of the effect of nuclear RNA on the DAOTA-M2 lifetime. This is an important control given the overlap in lifetimes when DAOTA-M2 is bound to G4 structures and non-G4 RNA [ Fig. 1c].
Fixing with 4% PFA results in a drop in average τ w by ca. 1 ns to 10.6 ± 1.1 ns, an interesting result as it implies changes in the number of G4s present upon fixation [ Fig. 3d]. Reassuringly, addition of PDS (10 μM, 1 hr) has the same response as in live cells, with τ w dropping to mean = 9.1 ± 1.4 ns and median = 8.5 ± 1.4 ns. In this experiment, the bi-modal distribution (formed due to limited PDS uptake by some cells) is best represented using the median. Treatment of fixed cells with RNAse (RNA digestion monitored using RNASelect TM , Supplementary Fig. 9) has no effect (p = 4.4 × 10 −1 ) on the average τ w , confirming that RNA in the cell nucleus does not affect DAOTA-M2 lifetime. DNAse treatment, however, did result in a drop in nuclear lifetime to 9.3 ± 0.9 ns. This value falls in-between that of DAOTA-M2 bound to dsDNA (median = 8.5 ± 1.4 ns) and in a mixture of dsDNA/G4 (10.6 ± 1.1 ns in fixed cells). This pattern is replicated in the Xenopus egg extract experiment described above [ Fig. 1d].
Thus, our fixed cell experiments confirm that nuclear RNA does not contribute to the high DAOTA-M2 lifetime observed in fixed cells; this data gives us confidence that RNA is unlikely to interfere with live cell experiments. Therefore, the DAOTA-M2 lifetime can be attributed to G4 DNA structure formation. At the same time our data seem to indicate that more G4s are stained by DAOTA-M2 in live rather than in fixed cells (all of which are being equally displaced by PDS), although the effect of fixation on other cellular components and its knock-on effect on DAOTA-M2 binding cannot be excluded.
Use of DAOTA-M2 to investigate helicases in live cells. We next investigated if DAOTA-M2 could report on the dynamics of G4 DNA inside live cells. We chose to disrupt the expression of  Fig. 4], which have been extensively reported, in vitro, to play a role in the resolution of G4s 49-51 , and monitor this using DAOTA-M2 in human and mouse cell lines. Cells lacking these proteins display general DNA replication stress, proposed to be the consequence of persisting G4s 52,53 . RTEL1 loss has been shown to cause sensitisation of mouse embryonic fibroblast (MEF) cells to G4 stabilising ligands, whilst FancJ loss did notsuggesting an increase in stable G4-DNA with the loss of RTEL1, but little or no increase with the loss of FancJ 54 .
To alter the general G4 content in cellulo, we used mutant MEF cells in which expression of the FancJ gene is eliminated, and the RTEL1 gene can be deleted using an adenovirus 54 . This allows for four combinations of MEF cells with FancJ and/or RTEL1 showed a modest, but significant (p = 6.6 × 10 −3 ) increase in the mean lifetime [8.0 ± 1.0 ns, Supplementary Fig. 10c].
As studying RTEL1 requires cells to be infected, we studied both uninfected MEF cells and MEF cells infected with a null adenovirus. Given we saw the potential for small variations in DAOTA-M2 lifetime after infection [ Supplementary Fig. 10c], the null infected samples were used as our basis for comparison for RTEL1 mutants [ FancJ and RTEL1 deficient MEF cells are known to present a DNA damage response 54 , therefore to control for DDR influence on DAOTA-M2 lifetime, MEF cells were exposed to 2 Gy gamma irradiation. ɣH2AX positive MEF cells did not show DAOTA-M2 lifetime variation compared to non-irradiated MEFs [Supplementary Fig. 8c]. Additionally, we noticed that DAOTA-M2 treatment on all four MEF cell lines induced an equivalent increase in ɣH2AX intensity [ Supplementary Fig. 4d, e] and a moderate accumulation of cells in S and G2 phases of the cell cycle [ Supplementary Fig. 3c, d], when compared to control MEFs treated with DMSO. Importantly, these phenotypes in the mutant MEF cell lines are similar to the one observed in WT MEFs and so are independent of the genotypes. Therefore, the recorded fluctuations in DAOTA-M2 lifetimes between mutant and WT MEF cells are indicative of an effect of the absence of the helicases FancJ and RTEL1 on G4 homoeostasis, and not on DNA damage or cell cycle.
These results demonstrate the potential for DAOTA-M2 to be used directly to monitor the role of G4 DNA in live cells without affecting cells through fixation or co-incubation.
In cellulo Fluorescence Lifetime Displacement Assay. G4s are considered as potential drug targets, and therefore it would be extremely useful to assess the ability of any given G4-targeted pharmaceutical to bind to this structure. Currently, a gold standard is to assess G4 DNA binders in vitro, with crucial information on whether they also target G4s in live cells missing. To add to our previous work using PDS as known G4 binder 32 , our next aim was to develop a FLIM based Fluorescence Lifetime Indicator Displacement Assay (FLIDA) to investigate a range of G4 binders on their ability to displace DAOTA-M2 from live cells. As our test dataset, we chose the Cu/Ni/VO/Zn-salphen complexes described in the in vitro studies (see above) since they are structurally related but have different G4 affinities [ Fig. 1g]. All cell cultures incubated with potential G4 binders showed good viability alone (at 1 μM G4 binder over 24 hr) and in combination with DAOTA-M2 at the experimental conditions used for imaging [ Supplementary Fig. 11].
In cellulo FLIDA using Ni-salphen resulted in a rapid drop in lifetime over 2 hr to 8.9 ± 0.7 ns, and after 6 hr the lifetime had  Fig. 5a and Supplementary Fig. 12]. For Ni, VO and Zn-salphens, the in vitro trend [ Fig. 1g] is replicated in cellulo [ Fig. 5], correlating with the binding affinity of each complex for G4 [43][44][45] . Cu-salphen, a strong G4 binder, did not show the magnitude of lifetime change expected from in vitro data, a possible consequence of lower cellular or nuclear uptake.
Thus, it appears that dynamic measurements enabled by FLIDA in live cells offer significant advantages in monitoring the ability of potential G4 binders to target G4 structures.

Discussion
It is becoming increasingly evident that G4 DNA structures form in vivo and play important biological roles. However, to date it has been difficult to visualise these structures in real-time and in live cells. Based on our studies herein using FLIM with the optical probe DAOTA-M2, we have been able to establish a robust displacement assay (FLIDA) to quantitatively study the interaction of any molecule with G4 DNA in vitro and in live cells. The advantage of our displacement assay as compared to others previously reported, is that it is based on fluorescence lifetime rather than fluorescence intensity, which means the analysis is not dependent on concentrations and hence can be easily applied to a cellular environment. To achieve this, it was necessary to first confirm that other biomolecules present in cells would not interfere with the ability of DAOTA-M2 to differentiate different DNA topologies. The fluorescence lifetime of our probe in nucleic acid-depleted egg extracts supplemented with G4 or dsDNA are consistent with those in aqueous buffer. Additionally, we have shown that RNA does not contribute to the DAOTA-M2 lifetime, through the digestion of RNA in fixed cells. This was of importance given the overlap in in vitro lifetimes when DAOTA-M2 binds to non-G4 RNA, and G4 structures. In live cells, the average nuclear lifetime is ca. 11.0 ns, which drops to ca. 8.5 ns on displacement using PDS (10 µM, 24 h), a highly specific G4 binder. Given our extensive characterisation in buffered aqueous solutions, in extract and in fixed cells, this apparent drop in nuclear lifetime was assigned to the displacement of DAOTA-M2 from G4. In general, the lifetime values in cellulo are slightly higher than those from buffered and extract solutions. We attribute this difference to the change in physical and experimental conditions between bulk measurements in vitro and FLIM decays in cellulo (see experimental section). What is more important than the absolute lifetime value (which can be affected by physical parameters such as refractive index 59,60 ), is the relative lifetime change between a control set of cells and those where the G4 structures have been disrupted.
To establish the validity of the cellular FLIDA assays, we carried out experiments with a series of structurally related G4 binders (i.e. the metal-salphens, with M = Cu, Ni, VO, Zn). As indicated above, there is a clear structure-activity relationship between the in vitro binding affinities and the live cell FLIDA experiments. As a further negative control, we carried out FLIDA in live cells with DAPI, which localised in the nucleus (and bound to DNA) but did not change the nuclear τ w of DAOTA-M2, consistent with DAPI not displacing the probe from G4 DNA.
Interestingly, cell fixation using PFA decreased the DAOTA-M2 lifetime which implies a difference in the nuclear distribution of G4 structures upon fixation (with a reduction in number of G4 sites available for binding with DAOTA-M2), a result that has not been observed previously. The FLIDA could be repeated in fixed cells using PDS, with displacement occurring to the same lifetime value as in live cells.
Having established how DAOTA-M2 interacts with G4 DNA using known G4 binders, we next designed experiments in which the nuclear G4 dynamics could be disturbed, without the need for co-incubation with strong G4 ligands. To achieve this, we reduced helicase enzyme expression in live cells through known protocols that knock-down FancJ in U2OS and knock-out FancJ and/or RTEL1, in MEF cells, respectively. Reducing the expression of known G4 helicases increases the number and/or stability of G4s, as the DAOTA-M2 lifetime increases. Studies using FancJ deficient human cells (U2OS and Hela) have reported that exposure to G4 ligands elicits a response in phenotypic markers (e.g. proliferation) when compared to wild type cells 50 , however, no such sensitivity was observed in FancJ deficient MEF cells 54 . We observe a modest lifetime increases when FancJ expression is reduced (0.5 and 0.3 ns, for U2OS and MEF cells, respectively). This increase is smaller than the decrease in lifetime after PDS displacement (ca. 2.5 ns), which is expected given that this helicase transiently targets specific regions of the genome and probably acts only on a proportion of the cellular G4 content inside each cell. Although the lifetime increase in MEF cells on loss of FancJ is small (0.3 ns), a larger effect is observed in MEF cells where RTEL1 has been deleted (0.6 ns). This is consistent with studies showing that RTEL1 loss led to sensitisation by G4stabilising ligands, which was not observed with FANCJ loss. Taken together, these findings are in line with the suggestion that RTEL1 targets and resolves more efficiently genomic G4-forming sites than FancJ 54 .
We also noticed that MEF cells have a lower baseline average nuclear lifetime compared to U2OS (ca. 8 ns and ca. 11 ns, respectively). Embryonic MEF cells contain different euchromatin and heterochromatin organisation compared to U2OS cells 61,62 . While it is possible that the differences in chromosome packing between different cell lines have a direct effect on the lifetime of DAOTA-M2, we confirmed that high-intensity signal (indicating staining of chromatin) by DAOTA-M2 in the nucleus of both U2OS and MEF cells does not correlate with DAOTA-M2 lifetime ( Supplementary Fig. 5). Additionally, as the PDS G4 displacement works in both live and fixed cells (Fig. 3), this lifetime change is unlikely to be only due to chromatin rearrangement.
Altering DNA helicase expression has multiple cellular and molecular effects on the genome, including transcriptome and RNA trafficking 51,63,64 , the possibility of forming ssDNA, doublestrand breaks and affecting RNA molecules. Treatment of U2OS cells with cisplatin showed a slight increase in nuclear lifetime, indicating that intra-strand DNA adducts and inter-strand crosslinks can result in a shift in DAOTA-M2 lifetime (Supplementary Fig. 8a). However, dsDNA breaks induced by 2 Gy gamma irradiation had no effect on the DAOTA-M2 lifetime (Figures S8b, c). In all MEF cells used in this study, incubation with DAOTA-M2 resulted in the same DDR response as measured using ɣH2AX foci ( Supplementary Fig. 4d, e). Given this result, it is very unlikely that the significant increase in lifetime measurements observed in the helicase mutant experiments reflects an increase in binding of DAOTA-M2 to other DNA structures that might form in these genetic backgrounds. With these factors relating to chromosomal packing and DNA damage accounted for, we can say with confidence that FLIM analysis of DAOTA-M2 reports directly on the prevalence of G4 structures in live mammalian cells.
G4 DNA targeting is emerging as a novel design strategy in the development of therapeutic drugs for diseases such as cancer 65 , thus, a robust method to test their G4 binding in live cells is highly desirable. We have demonstrated that DAOTA-M2 can be used to investigate G4 dynamics and their interaction with G4 binders in cellulo in real time. This approach holds promise for monitoring the sensitivity of newly developed G4 targeting drugs (e.g., testing against various cancer cell lines, including patientderived samples) as well as for further understanding of the in vivo role of G4 DNA.
In summary, we have used DAOTA-M2 in combination with FLIM to monitor the formation of G4 DNA in the nuclei of live and fixed cells. We have developed a cellular assay to study the interaction of small molecules with G4 DNA, which can be applied to a wide range of drugs, which do not have intrinsic fluorescence. Our finding that cell fixation has a significant effect on the DAOTA-M2 lifetime, which must reflect a change in DNA topology, has wider implications for G4 studies that require fixations protocols. Additionally, our method can be used to study G4 DNA dynamics in live cells, for example when the expression of helicases known to unwind G4 in vitro is reduced or abolished. The knock-down of FancJ in U2OS or loss of FancJ and/or RTEL1 in MEF cells increase the nuclear lifetime, consistent with a reduction in amount of these enzymes. These experiments pave the way to directly study G4-related biological phenomena in live cells and to correlate the biological role of small molecules with their ability to target G4 DNA structures in live cells.

Methods
DAOTA-M2 and metal-salphen (Ni, Cu, VO, Zn) complexes were synthesised as previously reported 32,44,45 . PDS was kindly donated by Dr. Marco Di Antonio, synthesised according to literature methods 66,67 . DAPI and cisplatin were purchased from commercial sources. Oligonucleotides (c-Myc and ds26; sequences = 5′-TGAGGGTGGGTAGGGTGGGTAA-3′ and 5′-CAATCGGATCGAATTC-GATCCGATTG-3′, respectively) were purchased from Eurogentec, and dissolved in 10 mM lithium cacodylate buffer at pH 7.3. KCl was added to a final concentration of 100 mM, then the oligonucleotides were annealed at 95°C for 5-10 min. Calf thymus DNA (CT-DNA, Sigma) was dissolved in the same cacodylate buffer, and KCl added to a final concentration of 100 mM. All oligonucleotide concentrations were determined unannealed using the molar extinction coefficients 13200 M −1 cm −1 (base pair for CT-DNA), 228700 M −1 cm −1 (strand for c-Myc), and 253200 M −1 cm −1 (strand for ds26). Concentrations of G4 and dsDNA are expressed as per strand, and per base pair, respectively.
In vitro time-correlated single photon counting (TCSPC). Time-resolved fluorescence decays were obtained using an IBH 5000 F time-correlated singlephoton counting (TCSPC) device (Jobin Ybon, Horiba) equipped with a 467 nm NanoLED as an excitation source (pulse width < 200 ps, HORIBA) with a 100 ns time window and 4096 time bins. Decays were detected at λ em = 575 nm (± 12 nm) after passing through a 530 nm long-pass filter to remove any scattered excitation pulse. For experiments using Ni-, Cu-or VO-salphens, a 404 nm NanoLED excitation source was used. For experiments involving ≤ 0.2 µM DAOTA-M2, 256 time bins were used detected at λ em = 575 nm (±16 nm) to increase signal intensity per bin. Decays were accumulated to 10000 counts at the peak of fluorescence decay. A neutral density filter was used for the instrument response function (IRF) measurements using a Ludox solution, detecting the emission at the excitation wavelength. Data collection was performed using DataStation v2.2 (HORIBA Scientific) and analysed using DAS6 v6.8 (HORIBA Scientific). Traces were fitted by iterative reconvolution to the equation I(t) = I 0 (α 1 e −t/τ1 + α 2 e −t/τ2 ) where α 1 and α 2 are variables normalised to unity. The intensity-weighted average lifetime (τ w ) was calculated using the equation: 68 A prompt shift was included in the fitting to take into account differences in the emission wavelength between the IRF and measured decay. The goodness of fit was judged by consideration of the deviations from the model via a weighted residuals plot.
In vitro fluorescence lifetime displacement assay. DAOTA-M2 (2 µM) was dissolved in 10 mM lithium cacodylate buffer (pH 7.3) supplemented with 100 mM KCl and the time-resolved fluorescence decay was recorded. If needed, dsDNA (CT-DNA, 20 µM) was mixed with DAOTA-M2 and the decay recorded. Prefolded G4 DNA (c-Myc, 4 µM) was then added and the decay recorded. Increasing amounts of the corresponding molecules under study were added and the fluorescence decay recorded until no further changes were observed.
In vitro fluorescence lifetime measurements with nucleic acid-depleted egg extracts. Decay data for nucleic acid-depleted cell extract studies was acquired by mixing 33 µL of nucleic acid-depleted Xenopus Laevis egg extracts 38 , with 12 µL of 10 mM lithium cacodylate buffer (pH 7.3) containing 100 mM KCl, DAOTA-M2 (final concentration 2 µM) and oligonucleotide (final concentration: c-Myc = 4 µM or ds26 = 44 µM). For samples where cell extract was not used, the same volume of 10 mM lithium cacodylate buffer (pH 7.3) containing 100 mM KCl was used in its place. Lifetimes decays were recorded using a home-built TCSPC method described previously 69 with data collected using SPCM v9.83 (Becker & Hickl GmbH). Samples were excited using a pulsed diode laser (Becker & Hickl GmbH, 477 nm, 20 MHz) and emission collected at 575 nm (±15 nm), using a 550 nm long-pass filter to remove scattered excitation photons. Decay traces were fitted using the FLIMfit software tool developed at Imperial College London 70 (v5.1.1, Sean Warren, Imperial College London) to a bi-exponential function, and the mean weighted lifetime (τ w ) calculated using Eq. (1).
General cell culture. Human Bone Osteosarcoma Epithelial Cells (U2OS, from ATCC) were grown in high glucose Dulbecco's modified Eagle medium (DMEM) containing 10% fetal bovine serum at 37°C with 5% CO 2 in humidified air.
In cellulo fluorescence lifetime displacement assay. Cells were seeded on chambered coverglass (1.5 × 10 4 cells, 250 μl, 0.8 cm 2 ) for 24 h, before washing with phosphate-buffered saline (PBS) and adding fresh media containing DAOTA-M2 (20 μM, 250 μl) for a further 24 h. For 24 hr co-incubation experiments, the compound under study was added with DAOTA-M2 and imaged 24 hr later. For < 24 hr co-incubation experiments, the compound udder study was added directly to the incubation media after 24 hr DAOTA-M2 incubation, and images were taken over time. Cells were imaged directly in the final incubation media. For time dependant measurements (Fig. 5), t = 0 cannot be measured directly as cells require equilibrating before imaging. Therefore, t = 0 was calculated from an average of all the control time points. Cell culture for helicase knock-down and knock-out experiments. For FancJ helicase knock-down treatment, a solution of 30 nM siRNA (SigmaAldrich, 5′-GUACAGUACCCCACCUUAU -3′) and DharmaFECT reagent (used as per manufacturer's instructions, Dharmacon) in serum-free DMEM was prepared and incubated at RT for 30 min. 400 µL of this solution was mixed with 2.1 mL DMEM containing 10% FBS and used to seed cells (1.6 × 10 5 cells, 2.50 ml, 9.6 cm 2 ) for 16 hr at 37°C. The growth medium was replenished, and cells further incubated for 24 h before reseeding on chambered coverglass (5.0 × 10 4 cells, 400 μl, 0.8 cm 2 ) with the addition of DAOTA-M2 (10 µM) for a further 24 hr before imaging. Cells treated analogously using an siRNA for Luciferase (Dharmacon, 5′-UCGAA-GUAUUCCGCGUACG -3′) were used as a control. The knockdown efficiency was assessed by Western blot for FancJ compared to actin [ Figure S10a].
Mutant mouse embryonic fibroblast (MEF) cells were used in which the FancJ gene is compromised by a gene trap cassette, eliminating expression of the FancJ ORF, resulting in a null allele 43  MTS cytoxicity assay. U2OS cells were seeded (5 × 10 3 cells, 100 μl, 32.2 mm 2 ) in a 96-well plate. After 24 h, compounds under study were added at the appropriate concentration in triplicate (150 μl). After a further 24 h, the MTS/PMS reagents were added according to the Promega MTS Assay protocol. The average absorbance of the triplicate wells was recorded at 492 nm, and where applicable, readings were corrected for compound absorbance by subtracting the compound-only control run in parallel. Wells were maintained with DMSO (only cells) as 100% and the maximum compound concentration as 0% viability. For experiments involving co-incubation with DAOTA-M2, cells with only DAOTA-M2 incubation were used as 100% and no cells as 0%. Absolute IC 50 was determined from the doseresponse curve of absorbance vs. logarithm of concentration of compound.
Fluorescence lifetime imaging microscopy (FLIM). FLIM was performed through time-correlated single-photon counting (TCSPC), using an inverted confocal laser scanning microscope (Leica, SP5 II) and a SPC-830 single-photon counting card (Becker & Hickl GmbH). A pulsed diode laser (Becker & Hickl GmbH, 477 nm, 20 MHz) was used as the excitation source, with a PMC-100-1 photomultiplier tube (Hamamatsu) detector. Software for collecting FLIM images was SPCM v9.83 (Becker & Hickl GmbH). Fluorescence emission (550-700 nm) was collected through a 200 μm pinhole for an acquisition time sufficient to obtain signal strength suitable for decay fitting, or a maximum of 1000 s. For all live-cell imaging, cells were mounted (on chambered coverglass slides) in the microscope stage, heated by a thermostat (Lauda GmbH, E200) to 37 (±0.5)°C, and kept under an atmosphere of 5% CO 2 in air. A 100x (oil, NA = 1.4) or 63x (water, NA = 1.2) objective was used to collect images at either 256 × 256 pixel resolution or at 512 × 512 pixel resolution, as stated in the text. The IRF used for deconvolution was recorded using reflection of the excitation beam from a glass cover slide.
Lifetime data were fitted using the FLIMfit software tool developed at Imperial College London (v5.1.1, Sean Warren, Imperial College London) 70 to a bi-exponential function, and the mean weighted lifetime (τ w ) calculated using Eq. (1). 7 × 7 and 9 × 9 square binning was used to increase signal strength for images recorded at 256 × 256 and 512 × 512 resolution, respectively. A scatter parameter was added to the decay fitting to account for scattered excitation light. Before fitting, a mask was applied to the images to analyse individual cell nuclei. Fitted lifetime data for each pixel within a single cell nucleus were pooled to find average and median nuclear τ w values. A threshold was applied to the average of each nucleus to require a minimum of 300 at the peak of the decay and a goodness-of-fit measured by χ 2 of less than 2. Average nuclear intensity in FLIM images was calculated from the peak maximum of the fitted decay (excluding scatter), averaged across the nucleus.
Gamma irradiation. Chamber slides containing either U2OS or MEF cells were inserted into the Irradiator (GSR D1 Cell Irradiator) for 50 s to irradiate cells to 2 Gy. Control cells were left outside for the equivalent time. Samples for FLIM were then imaged as above, except MEF cells which were imaged using a 20x (NA = 0.7) objective. To confirm DNA damage, slides were further incubated for 30 min to allow initiation of damage repair by the cells, then imaged for ɣH2AX immunofluorescence staining.
Fluorescence-activated cell sorting (FACS). Cells in culture were treated with BrdU (10 µM) in medium for 30 min before collection with trypsin and resuspended in PBS (300 µL). Ice-cold methanol (700 µL) was added dropwise with mixing and the samples were stored at −20°C for a minimum of 2 h. Fixed cells were harvested by centrifugation and incubated with pepsin (2 mL, 0.2 mg/mL in 2 M HCl, RT) for 20 min. Samples were washed with PBS and resuspended in 0.5 m: block solution (PBS, 0.5% Tween20, 0.5% BSA) with αBrdU-FITC conjugate (20 µL) for 1 h at RT. Samples were washed with the same solution then resuspended in PBS containing RNaseA (0.5 mg/mL, 1 mL, 37°C, 15 min). Propidium iodide (10 µg/mL) was added for 5 min at RT and samples were stored at 4°C for FACS. Data and graphs were analysed and generated using FlowJo software (BD biosciences), using Watson (Pragmatic) module.
Statistics. Results are expressed as mean ± SD (unless stated otherwise). Twotailed non-paired t-tests were performed using OriginPro v9.55. p-values, t-values, and degrees of freedom (DF) are stated in the figure captions. Significance: ns p > 0.05, *p < 0.05, **p < 0.01, ***p < 0.001. n values are the total cell nuclei that meet the threshold requirements. For box plots, the box represents 25-75% range, error bars are the mean ± SD, horizontal line is the median and the grey dot is the mean.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data that support this study are present in the manuscript and supplementary information, and are available from the corresponding author upon request. Source Data are provided with this paper.