A series of compounds that targeted the helicase activity of human BLM were identified in a quantitative high-throughput screen (qHTS) (Rosenthal, 2010), where the results were made publicly available from the PubChem data repository [https://pubchem.ncbi.nlm.nih.gov/bioassay/2528]. Filtering the 627 reported active compounds for preferential physicochemical properties (e.g. Lipinski’s rule of five) and excluding those with potential pan-assay interference activity (PAINS) allowed us to group the compounds into several distinct clusters according to chemical similarity. The inhibitory activity of exemplars from each cluster were tested in a fluorescence-based DNA unwinding assay (Rosenthal, 2010) against recombinant human BLM-HD (HD = helicase domain; amino acids 636–1298). However, only a single compound produced an IC₅₀ lower than 10 µM (compound 1, IC₅₀ = 4.0 µM; Figure 1A).

Figure 1 with 2 supplements see all

Download asset Open asset

We synthesised and purified six close analogues of this compound with the aim of generating preliminary structure-activity relationship data and confirmed their inhibitory activity in the unwinding assay (Materials and methods, Appendix 1). Compounds 2 to 6 inhibited the 3’ → 5’ helicase activity of recombinant human BLM-HD with IC₅₀ values ranging from 2.2 to ~60 µM, whereas 7 did not inhibit BLM-HD over the concentration range tested (Figure 1A and Figure 1—figure supplement 1, Table 1). An IC₅₀ of 4 µM was determined for ML216, a compound reported to be a semi-selective inhibitor of human BLM (Nguyen et al., 2013; Rosenthal, 2010), which was included as a positive control (Figure 1A).

Table 1

		unwinding	turnover
#	Chemical drawing	IC50 [95% CI]; µM	IC50 [95% CI]; µM
2		2.2 [1.7–2.7]	3.2 [2.3–4.0]
3		3.5 [2.4–5.2]	5.3 [4.7–6.1]
4		6.6 [3.4–12.7]	11.2 [8.3–15.3]
5		12.8 [4.5–36.8]	47.86 [18.28–180.7]
6		56.9 [25.4–171.3]	40.94 [15.9–139.8]
7		No inhibition	No inhibition
ML216		4.0 [3.7–4.3]	4.4 [4.0–4.8]

In a malachite green-based assay that measures ATP turnover, we observed robust stimulation of hydrolysis by BLM-HD when the protein was incubated with a short single-stranded 20-base oligonucleotide (Figure 1B). Here, we determined IC₅₀ values ranging from 3.2 to ~50 µM for each of our active analogues and 4.4 µM for ML216 (Figure 1C and Figure 1—figure supplement 1, Table 1). Whilst the values of IC₅₀ obtained in our orthogonal assay did not agree in absolute value with those determined in the first, it ranked each analogue with a similar order of potency.

We could readily observe changes in fluorescence, indicative of binding, upon titration of both ADP and ATP-γS into BLM-HD using microscale thermophoresis (MST, Figure 1—figure supplement 2). We could not, however, observe any interaction for our most potent compound 2. In the absence of biophysical evidence for binding, we sought to confirm that 2 wasn’t just a false positive generated by interference with the fluorescent readout of the unwinding assay. An alternative gel-based assay allowed direct visualisation of the conversion of a forked DNA-duplex into its component single-stranded oligonucleotides via the helicase activity of BLM-HD (Figure 2A). Titration of 2 clearly inhibited production of the single-stranded DNA product in a dose-dependent manner, with a calculated IC₅₀ of 1.8 µM (Figure 2B).

Figure 2 with 1 supplement see all

Download asset Open asset

Another potential false positive could be generated by compounds that bind directly to DNA, thus preventing BLM-HD from productively engaging with its substrate. To test this hypothesis, we used a commercial assay that utilises recombinant Topoisomerase I (Topo I) to relax a supercoiled plasmid. Compounds that intercalate or bind to the major or minor groove of the plasmid DNA prevent relaxation. At the manufacturer’s recommended concentration of 200 µM, the positive control m-Amsacrine (mAMSA) strongly inhibited relaxation of the supercoiled plasmid. In contrast, no effect was observed with 2 at the same concentration. However, partial inhibition of relaxation could be observed for a reaction containing ML216 (Figure 2C). To confirm this observation, we purchased ML216 from an alternative commercial supplier (ML216-A) and also resynthesised and purified the compound in-house (ML216-B; Materials and methods, Appendix 1). In both cases, a similar level of inhibition was observed when the compounds were included in the relaxation assay, indicating that this was both a real and reproducible effect (Figure 2C).

To explore further the possibility that ML216 might interact directly with DNA, we tested its ability to displace SYBR Green II (SG2) from a DNA substrate in a dye displacement assay (Del Villar-Guerra et al., 2018; Tse and Boger, 2004). When SG2 binds to DNA, a concomitant increase in its fluorescence can be measured. If an added compound can compete with the dye for binding to the DNA, a corresponding decrease in the fluorescent signal is observed. We titrated ML216 into a forked-50mer dsDNA substrate, that had been pre-incubated with SG2, observing a clear time- and dose-dependent displacement of the dye, indicating that ML216 can directly interact with a DNA substrate (Figure 2D).

With confidence that 2 was, in fact, a bona fide inhibitor of BLM, we repeated the unwinding assay in the presence a 10-fold higher concentration of ATP to examine if the compound was directly competitive with nucleotide binding to the active site of the enzyme. As the resulting IC₅₀ value was identical to that previously determined, it ruled out this mode of inhibition, and suggested that the compound bound elsewhere (Figure 2—figure supplement 1).

ATP-turnover experiments, under Michealis-Menten conditions, allowed us to generate a Lineweaver-Burk plot with data taken from DNA substrate titrations in the presence of 0, 5, and 10 µM of 2. The resultant plot indicated a non-competitive (allosteric) mode of inhibition for 2 (Figure 2E). With this information, we postulated that 2 might only bind to BLM-HD when it was engaged with a DNA substrate. We therefore revisited MST, first confirming the interaction of BLM-HD with the single-stranded 20mer used in our malachite green assay, plus a shorter 15mer that would be taken into crystallographic trials (Figure 2F). We next titrated 2 into the two pre-formed BLM-HD/ssDNA complexes. This time changes in fluorescent signal could be detected, confirming our hypothesis, with dissociation constants of 1.7 and 2.6 µM determined for the interaction with the 15mer and 20mer, respectively (Figure 2G).

We created the expression construct BLM-HD^ΔWHD to remove the conformationally flexible Winged Helix domain (WH) that requires the presence of either a stabilising nanobody, or interaction with a large DNA substrate to facilitate crystallogenesis (Newman et al., 2015; Swan et al., 2014) replacing it with a short poly-(glycine/serine) linker that serves to connect the Zinc-binding domain (Zinc) directly to the Helicase and RNAse C-terminal domain (HRDC, Figure 3A). In validation of this approach, we were able to crystallise the protein in complex with ADP and magnesium co-factor, and to determine its structure at a resolution of 1.53 Å; a significant increase in resolution over structures previously deposited in the PDB (4CDG, 2.8 Å; 4CGZ, 3.2 Å; 4O3M, 2.3 Å; see Appendix 1—table 1).

Figure 3 with 3 supplements see all

Download asset Open asset

Superposition of the structures of BLM-HD (PDB: 4CDG) and BLM-HD^ΔWHD produces a rmsd of 0.86 Å over 2450 atom positions (D1 + D2 + Zn; PyMOL), indicating the overall conformation and geometry of the two recombinant proteins is highly similar, despite deletion of the WH domain (Figure 3—figure supplement 1). Furthermore, BLM-HD^ΔWHD binds both ssDNA-15mer and 2 with a similar affinity to that of BLM-HD (Figure 3—figure supplement 2).

We crystallised 2 in complex with BLM-HD^ΔWHD, ADP/magnesium co-factor, and ssDNA-15mer (liganded complex); determining its structure at a resolution of 3.0 Å (Appendix 1—table 1). The complex crystallised in space group P1, with six molecules of BLM-HD^ΔWHD and associated ligands forming the asymmetric unit. Interestingly, the co-crystallised ssDNA-15mer helped drive formation of the crystal lattice, due to its partial self-complementary at the 5’ end (5’-CGTAC-3’) that serves to form four consecutive base pairs between two oligonucleotides (Figure 3B); the cytosine at the 5’ end of the oligonucleotide is not readily discernible in electron density maps and is therefore likely to be disordered. An extensive series of interactions are made to the bound nucleic acid by amino acids from all four sub-domains of the BLM-HD^ΔWHD expression construct (Figure 3—figure supplement 3).

Compound 2 sits in a small pocket found on the opposite face of the protein to that which binds nucleotide (Figure 3B, inset), and integrates amino acid side chains from both the D1 and D2 subdomains of the helicase core, as well as several from the Zn-binding domain. The oxygen of the amino group at the centre of the compound is hydrogen-bonded to the side chain of Asn1022, whilst the nitrogen of the same moiety is in hydrogen-bonding distance to both the backbone oxygen and side chain of Ser801. The side chains of His805 and Thr1018 stack up against, and provide Van der Waals contacts to, the central ring system of the 3-amino-4,5-dimethylbenzenesulfonamide pendant group as part of a pocket lined by the side chains of residues Asp806, His1014, Thr1015, His1019 (Figure 3C). The nitrogen of the sulphonamide group is hydrogen-bonded to the side chain of Asp840, which itself is bonded to the side chain of His805. The 2-methyl-thiazole moiety of 2 sits against the surface of the alpha-helix containing Gly972 and is sandwiched by additional packing interactions with the side chains of Gln802 and Glu971. The side chains Thr832 and His798 also contribute to this section of the binding pocket, which is ‘capped’ by Gln975. The central benzene ring of 2 is also contacted by the side chains of Gln802 and Glu971.

The aromatic-rich loop (ARL) is a highly conserved motif in RecQ helicases that serves as a molecular ‘sensor’, detecting binding of single-stranded DNA and coupling it to structural rearrangements that enable ATP hydrolysis (Manthei et al., 2015; Zittel and Keck, 2005). In our high-resolution structure of BLM-HD^ΔWHD, the ARL is disordered and is not visible in electron-density maps (Figure 4A). By contrast, it can be fully modelled in the liganded complex, but its conformation is distinct from that observed in PDB entries 4CDG and 4CGZ where the single-stranded extension of bound nucleic acid substrates does not extend across to the D1 domain (Figure 4B, Figure 4—figure supplement 1).

Figure 4 with 3 supplements see all

Download asset Open asset

Structures with a high degree of structural similarity to the liganded complex were identified with PDBeFold (Krissinel and Henrick, 2004). The search produced PDB entries 6CRM and 4TMU, with Q-scores of 0.47 and 0.46 respectively, which both describe structures of the catalytic core of Cronobacter sakazakii RecQ helicase (CsRecQ) in complex with different DNA substrates (Manthei et al., 2015; Voter et al., 2018). Comparison, in each case, reveals a close to identical conformation of the ARL to that observed in our liganded complex, as well as nucleic acid interactions that include the D1 domain (Figure 4—figure supplements 1, 2 and 3).

For CsRecQ, Manthei et al., 2015 described concerted movements of residues Phe158 and Arg159, within the ARL to interact with the 3’ single-stranded extension of their co-crystallised DNA substrate in our liganded complex the equivalent residues undergo a similar transition (Phe807 and Arg808, respectively). We observe that Phe158 moves to make base-stacking interactions with G8 and A9 of the bound ssDNA-15mer. By comparison to 4CGZ, we also see that Arg808 switches from interacting with Asp806 of the ARL to the backbone oxygen of Pro715 and the side chain of Glu768 (Figure 4A and B). Notably, mutation of residues equivalent to Arg808 or Glu768 in EcRecQ (Arg159 and Glu124) have been shown to perturb enzyme function (Manthei et al., 2015; Zittel and Keck, 2005). Interestingly, in BLM, Asp806 is ‘freed’ to interact with the side chain of His1019, a residue within the Zn-binding domain (Figure 4A and B).

The HRDC (Helicase and RNase D C-terminal domain) was originally identified as a putative nucleic-acid binding motif in both BLM and WRN (Werner syndrome helicase) and named in part for its similarity to a domain found at the C-terminus of E. coli RNase D (Morozov et al., 1997). However, only very weak ssDNA binding (K_d ~100 µM) has been reported for this domain in isolation (Kim and Choi, 2010).

In their paper describing the crystal structure of BLM in complex with DNA, Newman et al. observe that the HRDC domain packs against a shallow cleft formed between the D1 and D2 domains of the helicase core, with the interface between the different modules being highly polar in nature. Their follow-on small angle X-ray scattering experiments also indicate that the HRDC domain of BLM is free to disassociate and re-bind to the helicase core (Newman et al., 2015). In our high-resolution crystal structure of BLM-HD^ΔWHD, we observe the same ‘parked’ interaction for the HRDC, even in the absence of the WHD, but in our liganded structure we see that the HRDC domain swings across the face of the core helicase fold (Figure 5A) to make a series of polar interactions with the ssDNA-15mer, which is presented on the surface of the D1 domain of a second protomer within the asymmetric unit (Figures 5B and 3B). The side chains of HRDC residues Asn1242 and His1236, contained within the ‘hydrophobic 3₁₀ helix’ (Kim and Choi, 2010), are hydrogen-bonded to the O4 group of the T12 base and the N3 of the G11 base, respectively. The side chain of Phe1238 is also involved in an edge to ring-stacking interaction with the G11 base.

Figure 5

Download asset Open asset

Each of these interactions is consistent with chemical shift changes previously observed in HSQC spectra – as a result of titrating ssDNA into ¹⁵N-labelled BLM-HRDC (Kim and Choi, 2010) – suggesting that the observed interactions have biological relevance, and that our structure represents the first to capture HRDC interactions with ssDNA. Furthermore, amino acids residues Lys1227, Tyr1237, Thr1243, and Asn1239 are also in close proximity to the bound DNA (Figure 5B) and could be expected to undergo changes in chemical environment upon interaction; again consistent with the reported perturbations in HSQC spectra (Kim and Choi, 2010). Asn1239 might also be expected to pick up an additional contact with the 5’-phosphate of a subsequent nucleotide in an extended substrate.

With a robust molecular understanding for the binding mode for 2, we next examined if the compound displayed selectivity for members of the RecQ-helicase family. In our ATP-turnover assay, we saw no inhibition of recombinant helicase domains (HD) from human WRN, human RecQ5 or the unrelated E. coli helicase UvrD over the concentration range tested, whilst at higher concentrations inhibitory effects started to appear against human RecQ1. In contrast, ML216 robustly inhibited all four RecQ-family helicases tested and at higher concentrations also affected UvrD; in line with, and in support of, our observation that ML216 is non-specific and elicits at least part of its inhibitory effect by binding directly to DNA (Figure 6A).

Figure 6 with 1 supplement see all

Download asset Open asset

Whilst compound solubility prevented generation of a complete inhibition curve for UvrD and thus a robust estimate of IC₅₀, the estimated Hill coefficient (nH) was close to 1 — in contrast to those calculated for titrations of ML216 against the RecQ helicases, which were generally steeper (ranging from 1.8 to 4.3), again suggesting that instead of a forming a 1:1 protein to inhibitor complex, there is in fact, a more complex (possibly mixed) mode of binding for this compound to this class of enzymes.

As binding of 2 has no direct effect on the ability of BLM-HD^ΔWHD to bind either ssDNA or nucleotide, we hypothesised that it might act to ‘lock’ the helicase into a conformational state where DNA substrates remain bound but cannot be unwound. In support of this idea, we undertook MST assays with a labelled single-stranded oligonucleotide in both the presence and absence of 2. Here, we observed a clear concentration-dependent decrease in K_d when 2 was added, consistent with a decrease in the off-rate for DNA-binding, supporting our hypothesis that the compound acts to ‘trap’ BLM in its interaction with ssDNA (Figure 6B).

During their catalytic cycle, the RecQ-family of helicases undergo a sequential set of conformational transitions, driven by ATP-binding and hydrolysis, then the subsequent release of inorganic phosphate and ADP (Newman et al., 2017). An initial ‘encounter’ complex is generated when the D2 domain of a RecQ-helicase binds to single-stranded DNA (ssDNA). This then leads to a set of motif / domain movements, including the ARL, that serve to couple ATP-binding and hydrolysis to movement of the bound ssDNA, such that it is now functionally engaged with the D1 domain. The process of DNA-unwinding is thought to proceed via an ‘inchworm’ type of mechanism, where the D2 domain sequentially binds and releases the DNA substrate, in order to ‘feed’ it onto the D1 domain, translocating it one base at a time. The WHD and HRDC domains of the RecQ helicase-family appear to contribute to the binding, recognition and unwinding of different DNA substrates via their ability to bind to ds and ssDNA respectively.

As noted earlier, our liganded complex is structurally most similar to PDB entry 4TMU (CsRecQ)—a structure that has been previously characterised as representing a ‘pre-ATP hydrolysis’ conformation in the catalytic cycle of RecQ-family helicases (Newman et al., 2017). Although we have co-crystallised BLM-HD^ΔWHD with ADP, we can see that the active site of the liganded complex is also compatible with ATP-binding, through comparison to the crystal structure of E. coli RecQ in complex with ATP-γ-S (Figure 6—figure supplement 1; Bernstein et al., 2003). This in turn suggests that binding of compound 2—to a small pocket hereinafter referred to as the allosteric binding site (ABS)—acts to ‘trap’ or stall BLM by blocking the set of conformational changes required for progression to the next step of the catalytic cycle.

The ARL must be in a particular conformation to permit compound binding, otherwise the side chain of Trp803 would occupy and occlude the ABS (Figure 4A, Figure 4—figure supplement 1). This condition is satisfied when the D1 domain of BLM is engaged with single-stranded DNA, providing one explanation as to why we only observe compound binding in the presence of oligonucleotide. However, it is equally possible that binding of 2 serves to promote rearrangement of the ARL into the configuration that then permits the D1-ssDNA interaction.

The side chains of amino acids Gln975 and His798 contribute to the ABS (Figure 3C right and Figure 4A). In hsRECQ5, a polar contact between the equivalent residues (Gln345 and His160) is reported to be broken as a consequence of ATP-binding, freeing the ARL to interact productively with the single-stranded 3’-overhang of a bound DNA substrate (Newman et al., 2017). Importantly, mutation of the glutamine to alanine (Q345A) prevents stimulation of ATPase activity by binding to DNA, but does not perturb the basal rate of hydrolysis (Newman et al., 2017). The importance of the conserved glutamine is perhaps more apparent in E. coli RecQ, where mutation of the equivalent residue (Q322A) essentially ablates all ATPase capability (Manthei et al., 2015). Consistent with the ‘pre-ATP hydrolysis’ conformation for our liganded complex, the polar contact between Gln975 and His798 is broken, being ~3.8 Å apart. We note, however, that proximity of the 2-methyl-thiazole moiety from 2, may sterically prevent reformation of this important contact.

It is clear that more detailed kinetic studies would be required to unambiguously distinguish between a ‘passive’ or ‘induced’ mode of binding for 2. However, compounds bound stably to the ABS should sterically prevent the ARL from reverting to its initial conformation/structurally disordered state found at the beginning of the catalytic cycle (Figure 4A,B). The observed hydrogen bond between Asp806 and His1019 may also act to stablise the ternary interaction between BLM-HD^ΔWHD, 2 and ssDNA (Figure 4).

In addition to the ARL, several other regions of BLM interact with 2 when it is bound to the ABS, including amino acids from helicase motifs I and III, plus a short region just upstream of motif IV (Gorbalenya and Koonin, 1993; Hall and Matson, 1999) (pre-motif IV, Figure 4A). Unsurprisingly, the amino acid sequence identity of each region across the RecQ-family is extremely high, and do not therefore provide a facile explanation for the observed selectivity of 2 (Figure 7A). In particular, two of the amino acid side chains involved in hydrogen bonds with 2 are absolutely conserved in identity (motif I, Ser801; motif III, Asp840, Figures 7A and 3C). Likewise, amino acids within these motifs involved in hydrophobic contacts with 2 are also highly conserved in identity/chemical property. However, the third hydrogen bonding interaction (made by Asn1022 to 2) and its position within the ‘Helical Hairpin’ of the Zn-binding domain provides some insight, as both the length and amino acid composition of this loop is highly divergent across the RecQ-family and absent from RecQ4 (Figure 7A). There is no obvious consensus for any of the residues, in equivalent positions to those in BLM, involved in compound interaction. This observation provides a plausible route, although the addition or alteration of chemical groups to the core scaffold of 2, for generating potent and highly selective inhibitors for the individual members of the RecQ-family of helicases.

Figure 7

Download asset Open asset

During the catalytic cycle, it is clear that the HRDC must be ‘released’ in order to interact with ssDNA, indeed amino acid residues within the aforementioned ‘hydrophobic 3₁₀ helix’ contribute to both the ‘parked’ and ‘engaged’ interfaces; as exemplified by the side chain of Phe1238 that moves from a packing interaction with Pro956, a residue in the D1 domain, to an interaction with a base in the bound ssDNA-15mer (Figure 5A). As a consequence of the extended ssDNA interface visualised in our liganded complex (also seen in PDB entry 4TMU), we can see that the size and shape of the pocket that serves to bind the HRDC is substantially altered. Interaction of the D1 domain with ssDNA alters its spatial relationship with the D2 domain, leading to a widening of the pocket and to disruption of the hydrogen bonds that have previously been described to anchor the HRDC in place (Newman et al., 2015; Figure 4—figure supplement 3). Binding and release of the HRDC has previously been linked to nucleotide-status, indicating that the domain only becomes disengaged when BLM is not bound to ADP or ATP (Newman et al., 2015). Our data suggest an additional nuance: if the HRDC is able to engage in an ssDNA interaction, it is prevented from re-binding to the catalytic core, thus isolating it from the catalytic cycle of the enzyme.

During model building and evaluation of our liganded structure, we found that the nucleic-acid interactions made by the different sub-domains of BLM-HD^ΔWHD explain how the HRDC might contribute to the ability of BLM to unwind different types of DNA substrate—incorporating information taken from our own structure, as well as that for the interaction of the WHD with dsDNA from PDB entry 4CGZ (Figure 7B).

Simple superposition of the two structures results in a clash of the HRDC (in our structure) with the WHD; however, this is readily resolved by a small horizontal translation of the HRDC (roughly equivalent to adding an additional nucleotide to the single-stranded portion of the bound DNA substrate). We also note that, in our liganded structure, the polarity of the single-stranded DNA interacting with the HRDC is opposite to that of our model but with the understanding that is actually dictated by the packing arrangement of the molecules that serve to generate the crystal lattice; on examination, each of the observed HRDC interactions is fully compatible with binding to ssDNA in either orientation.

However, using the trajectories for each of the bound DNA substrates as positional markers allows generation of a model for unwinding of a simple DNA duplex. The double-stranded portion of the substrate is held in place through the previously described set of interactions with the WHD domain (Newman et al., 2015). The β-hairpin of the WH serves to separate the DNA duplex, with one strand ‘actively’ engaged with the D1/D2 domains of the helicase core and the Helical Hairpin of the Zn-binding domain. The second ‘inactive’ strand passes along the opposite face of the WH β-hairpin, to subsequently interact with the HRDC domain, potentially acting to prevent reversion and re-annealing of the DNA duplex. The relatively poor binding affinity of the HRDC for ssDNA is compatible with this model, as it would allow iterative release and recapture of the ‘inactive’ strand as the DNA substrate is unwound.

A model for how BLM might unwind a Holliday Junction has previously been reported (Kitano, 2014) but this treats the HRDC domain as a static object, leaving in it the ‘parked’ position and making no interactions with nucleic acid. Data published subsequent to this paper has indicated that the HRDC plays a more fundamental role, for example the charge-reversal mutation N1239D to has been shown to ablate interaction of the HRDC with both ssDNA and a Holliday junction substrate (Kim and Choi, 2010). The HRDC has also been reported to confer DNA-structure specificity to BLM, with Lys1270 playing a role in mediating interactions with DNA and for efficient dissolution of double-Holliday junction substrates in vitro (Wu et al., 2005). Consistent with this, our structural data reveals that both Asn1239 and Lys1270 of the HRDC are poised to interact with a section of single-stranded DNA just one to two nucleotides longer than that captured in our crystal lattice (Figure 5B). Finally, our model also suggests the existence of a transient direct interaction between the WHD and the HRDC.

Compound 2 itself, is unlikely to be suitable as a therapeutic agent, due to its poor ability to penetrate the cell membrane of mammalian cells (data not shown). Indeed, subsequent iterations of 2, or compounds that bind to the same allosteric site, may lead a tool compound that could be used to discover or confirm synthetic lethal relationships with BLM across different tumour backgrounds (Datta et al., 2021).

However, as the first described bona fide selective allosteric inhibitor of human BLM, the understanding of its mode of action will aid ongoing efforts to develop molecules targeting this class of enzymes for the treatment of human disease.

Details for synthesis and purification of compounds is provided in Appendix 1.

Synthetic genes, codon-optimised for expression in E. coli, were purchased from GeneArt (ThermoFisher Scientific, Loughborough, UK). With the exception of RECQ5 (see below) the coding sequence was subcloned into an in-house modified pET-17b vector at the NdeI and EcoRI sites of the multiple cloning site.

ML216 was purchased from Merck KGaA (Darmstadt, Germany), product code: SML0661. ML216-A was purchased from Cayman Chemical (Ann Arbor, Michigan, USA), product code: 15186.

Assay uses the PiColorLock Gold Phosphate Detection System from Novus Biologicals following the manufacturer’s recommended protocol. Briefly, assays were carried out in 96-well clear flat-bottomed plates, with absorbance measurements taken at a wavelength of 630 nm in a CLARIOstar multimode plate reader (BMG Labtech). Assay buffer: 50 mM Tris-HCl pH 7.5, 50 mM NaCl, 2 mM MgCl₂, 0.05% v/v Tween-20, 0.5 mM TCEP.

165 μl of BLM-HD and ssDNA-20mer (at a concentration of 2.4 nM and 121 nM, respectively) was pre-incubated with 10 μl of compound (2 mM stock dissolved in 100 % v/v DMSO, over a range of final concentrations up to 100 µM) for a period of 15 min at room temperature. Next, 25 µl of Mg-ATP substrate (at 16 mM) was added. After 20 min, reactions were stopped by the addition of 50 μl Gold mix (a 100:1 ratio of PiColorLock:Accelerator reagents). After 2 min, 20 μl of stabiliser solution was added to each well. After a further 30 min absorbance measurements were taken.

Assay conditions: 2 nM BLM-HD, 100 nM ssDNA-20mer and 2 mM Mg-ATP in a reaction volume of 200 µl over a 20-min incubation period.

Assay buffer: 50 mM Tris-HCl pH 8.0, 50 mM NaCl, 2 mM MgCl₂, 0.01% v/v Tween-20, 2.5 µg/ml poly(dI-dC), 1 mM DTT. 28 μl of BLM-HD (at a concentration of 2.9 nM) was pre-incubated with 2 μl of compound (2 mM stock dissolved in 100 % v/v DMSO, over a range of concentrations up to 100 µM) for a period of 15 min at room temperature. Next, 10 μl of substrate (300 nM forked-50mer dsDNA and 4.8 mM Mg-ATP) were added. After 10 min, reactions were terminated by the addition of 1 x loading dye (6 x solution: 10 mM Tris-HCl pH 7.5, 0.03% w/v bromophenol blue, 60% v/v glycerol, 60 mM EDTA). The samples were then loaded onto a 15% native gel (29:1 acrylamide:bis-acrylamide, 0.5 x TBE), separated by electrophoresis, and then visualised using a FLA-1500 fluorimager [Fujifilm, Bedford, UK]. The intensity of each species on the gel was quantified using the analysis tools provided in the software package Fiji (Schindelin et al., 2012).

Assay uses the DNA Unwinding Assay Kit from Inspiralis (Norwich, UK) following the manufacturer’s recommended protocol. Resultant samples were applied to a 1% w/v agarose gel (in 1 x TAE buffer), separated by electrophoresis, stained with ethidium bromide, and then visualised with a UV-transilluminator/digital gel documentation system.

Fluorescence intensity was measured in a CLARIOstar multi-mode plate reader (BMG Labtech) with excitation and emission wavelengths of 485 nm and 520 nm respectively, in 384-well black plates. Twenty-eight µl of forked-50mer dsDNA (at a final concentration of 800 nM) was pre-incubated with 10 μl of SYBR Green II (1:200 dilution) for a period of 20 min at room temperature. Two µl of compound (1 mM stock dissolved in 100 % v/v DMSO, over a range of final concentrations up to 50 µM) was then added. Measurements were taken after incubation times of 20, 45, and 60 min. Assay buffer: 50 mM Tris-HCl pH 8.0, 50 mM NaCl, 2 mM MgCl₂, 0.01% v/v Tween-20, 1 mM DTT.

Assay conditions: 800 nM annealed DNA substrate and 1:800 SYBR Green II in a reaction volume of 40 µl over a 20-min incubation period.

Diffraction data were automatically processed at the synchrotron by the xia2 pipeline (Winter et al., 2013), using software packages DIALS (Beilsten-Edmands et al., 2020; Winter et al., 2018) or XDS (Kabsch, 2010) and Aimless (Winn et al., 2011). For BLM-HD^ΔWHD/ADP, coordinates corresponding to the helicase domain were extracted from PDB entry 4O3M and provided as a search model for molecular replacement using Phaser (McCoy et al., 2007). For BLM-HD^ΔWHD/ADP/ssDNA-15mer/compound 2, the rebuilt and refined coordinates for BLM-HD^ΔWHD were used as the search model. Initial models were extended and improved by iterative rounds of building in Coot (Emsley and Cowtan, 2004) and refinement in either PHENIX (Liebschner et al., 2019) or BUSTER (Bricogne, 2020) to produce the final deposited models. Crystallisation and refinement statistics are provided in Appendix-table 1.

All experimental data were plotted and analysed using GraphPad Prism, 2020.

All reactions were conducted under an atmosphere of nitrogen unless otherwise stated. Anhydrous solvents were used as purchased or were purified under nitrogen as follows using activated molecular sieves. Thin layer chromatography was performed on glass plates pre-coated with Merck silica gel 60 F254. Visualisation was achieved with U.V. fluorescence (254 nm) or by staining with a phosphomolybdic acid dip or a potassium permanganate dip. Flash column chromatography was carried out using pre-packed columns filled with Aldrich silica gel (40–63 µm) on an ISCO Combiflash Rf, or a Biotage Isolera Prime. Proton nuclear magnetic resonance spectra were recorded at 500 MHz on a Varian 500 spectrometer (at 30°C), using residual isotopic solvent (CHCl₃, δH = 7.27 ppm, DMSO δH = 2.50 ppm, 3.33 ppm (H₂O)) as an internal reference. Chemical shifts are quoted in parts per million (ppm). Coupling constants (J) are recorded in Hertz (Hz). Carbon nuclear magnetic resonance spectra were recorded at 125 MHz on a Varian 500 spectrometer and are proton decoupled, using residual isotopic solvent (CHCl₃, δC = 77.00 ppm, DMSO δC = 39.52 ppm) as an internal reference. Carbon spectra assignments are supported by HSQC and DEPT editing and chemical shifts (δC) are quoted in ppm. Infrared spectra were recorded on a Perkin Elmer FT-IR One spectrometer as either an evaporated film or liquid film on sodium chloride plates. Absorption maxima are reported in wave numbers (cm⁻¹). Only significant absorptions are presented in the data, with key stretches identified in brackets. LCMS data was recorded on a Waters 2695 HPLC using a Waters 2487 UV detector and a Thermo LCQ ESI-MS. Samples were eluted through a Phenomenex Lunar 3µ C18 50 mm ×4.6 mm column, using acetonitrile and water acidified by 0.01% formic acid in three methods: method 1 (3:7 to 7:3 acetonitrile and water over 7 min), method 2 (3:7 to 7:3 acetonitrile and water over 4 min) and method 3 (19:1 to 1:19 acetonitrile and water over 10 min), High resolution mass spectrometry (HRMS) spectra were recorded on Bruker Daltonics Apex III ESI-MS, with an Apollo ESI probe using a methanol spray. Only molecular ions, fractions from molecular ions and other major peaks are reported as mass/charge (m/z) ratios.

To 4-(2-methylthiazol-4-yl)benzoic acid (120 mg, 0.55 mmol), HBTU (153 mg, 0.66 mmol), N,N-diisopropylethylamine (0.19 mL, 1.09 mmol) in N,N-dimethylformamide (3 mL) was added, 3-amino-4,5-dimethylbenzenesulfonamide (110 mg, 0.55 mmol). The reaction mixture was stirred at ambient temperature for 16 hr. Upon completion, the solvent was removed under reduced pressure. The crude product was taken up in ethyl acetate (5 mL), washed with saturated aq. NaHCO₃ (4 mL), brine (4 mL), dried over MgSO₄, filtered and concentrated under reduced pressure. The crude was purified by column chromatography (silica 12 g, 0% to 60% ethyl acetate in petroleum ether) to yield the desired amide 2 as a white solid (15 mg, 7%). Rf 0.12 (petroleum ether:ethyl acetate 1:1); m.p. 246–248°C; ¹H-NMR (500 MHz, DMSO-d6) δ 10.17 (s, 1H, CONH), 8.13–8.08 (m, 3H, H-3, H-6), 8.05 (d, J 8.4, 2H, H-2), 7.64 (d, J 1.9, 1H, H-4’), 7.54 (d, J 1.9, 1H, H-3’), 7.28 (s, 2H, SO₂NH₂), 2.74 (s, 3H, 8-CH₃), 2.36 (s, 3H, 3’-CH₃), 2.17 (s, 3H, 2’-CH₃); ¹³C-NMR (126 MHz, DMSO-d6) δ 166.4 (CO), 165.7 (C-8), 153.3 (C-5), 141.6 (C-5’), 138.5 (C-1), 137.6 (ArC), 137.5 (ArC), 137.1 (ArC), 133.7 (C-4), 128.7 (C-2), 126.3 (C-3), 124.6 (C-6’), 122.3 (C-4’), 116.1 (C-6), 20.7 (3’-CH3), 19.4 (8-CH3), 15.0 (2’-CH3); IR (neat, ν_max, cm⁻¹) 3258, 2923, 1630, 1572, 1516, 1444, 1403, 1304, 1154; LCMS (LCQ) Rt = 2.0 min (method 1), m/z (ESI+) 402.1 [M+H]⁺; HRMS m/z (ESI): calcd. for C₁₉H₁₉N₃O₃S₂ [M+H]⁺ 401.0868, found 401.0866

To 4-(2-methylthiazol-4-yl)benzoic acid (80 mg, 0.36 mmol) in dichloromethane (3 mL) was added oxalyl chloride (37 µL, 0.44 mmol) in a dropwise manner followed by the addition of a few drops of N,N-dimethylformamide (10 µL). The reaction mixture was stirred at ambient temperature for 16 hr. The crude acyl chloride was then added to a stirred mixture of 3-amino-2-methylphenol (54 mg, 0.44 mmol), N,N-diisopropylethylamine (397 µL, 2.28 mmol) and dichloromethane (1 mL) and stirred at ambient temperature for 2–16 hr. The solvent was removed under reduced pressure. The crude product was taken up in saturated aq. NaHCO₃ (5 mL) and extracted with ethyl acetate (3 × 10 mL). The combined organic components were then washed with brine (10 mL), dried over MgSO₄, filtered and concentrated under reduced pressure. The crude was purified by column chromatography (silica 12 g, 0% to 40% ethyl acetate in petroleum ether) to yield the desired amide three as a white solid (13 mg, 10%).Rf 0.47 (petroleum ether:ethyl acetate 1:1); m.p. 238–240°C; ¹H-NMR (500 MHz, DMSO-d6) δ 9.84 (s, 1H, CONH), 9.35 (s, 1H, OH), 8.10 (s, 1H, H-6), 8.07 (d, J 8.2, 2H, H-3), 8.03 (d, J 8.3, 2H, H-2), 7.00 (t, J 7.9, 1H, H-5’), 6.79 (d, J 7.8, 1H, H-6’), 6.73 (d, J 8.0, 1H, H-4’), 2.74 (s, 3H, 8-CH3), 2.03 (s, 3H, 2’-CH3); ¹³C-NMR (126 MHz, DMSO-d6) δ 166.4 (CONH), 165.3 (C-8), 156.2 (C-5), 153.4 (C-3’), 137.9 (ArC), 137.3 (ArC), 134.1 (ArC), 128.7 (H-2), 126.2 (H-3), 126.0 (ArC), 121.3 (ArC), 118.0 (C-6’), 116.0 (C-6), 112.9 (C-4’), 19.4, (C-8) 11.4 (C-2’); IR (neat, ν_max, cm⁻¹) 3291, 1640, 1607, 1499, 1466, 1307, 1174; LCMS (LCQ) Rt = 2.5 min (method 1), m/z (ESI+) 325.1 [M+H]⁺; HRMS (ESI): calcd. for C₁₈H₁₆NaN₂OS [M+Na]⁺ 347.0825, found 347.0827.

To 4-(2-methylthiazol-4-yl)benzoic acid (800 mg, 3.65 mmol) and methyl 3-amino-4-methylbenzoate (52 µL, 9.12 mmol) in tetrahydrofuran (15 mL) was added phosphorus trichloride (0.32 mL, 3.65 mmol). The reaction mixture was heated in a microwave reactor for 20 min at 150°C. The crude product was taken up in saturated aq. NaHCO₃ (5 mL) and extracted with ethyl acetate (3 × 10 mL). The combined organic components were then washed with brine (10 mL), dried over MgSO₄, filtered and concentrated under reduced pressure. The crude was purified by column chromatography (amino silica 12 g, 0% to 50% ethyl acetate in petroleum ether) to yield the desired amide 4 as a white solid (920 mg, 65%). Rf 0.23 (petroleum ether:ethyl acetate 3:1); m.p. 179–181°C; ¹H-NMR (500 MHz, DMSO-d6) δ 10.03 (s, 1H, CONH), 8.14–8.07 (m, 3H, H-3, H-6), 8.05 (d, J 8.3, 2H, H-2), 8.00 (d, J 1.8, 1H, H-6’), 7.76 (dd, J 7.9, 1.8, 1H, H-4’), 7.44 (d, J 7.9, 1H, H-9’), 3.85 (s, 3H, OCH3), 2.74 (s, 3H, 8-CH3), 2.33 (s, 3H, 2’-CH3); ¹³C-NMR (126 MHz, DMSO-d6) δ 166.4 (CO), 165.6 (C-8), 153.3 (C-5), 140.0 (C-8’), 137.6 (C-1), 137.3 (C-7’), 133.7 (C-4), 131.3 (C-3’), 128.8 (C-2), 128.1 (C-6’), 127.5 (C-5’), 126.9 (C-4’), 126.3 (C-3), 116.1 (C-6), 52.5 (OCH3), 19.4 (8-CH3), 18.6 (2’-CH3). COO not visible; IR (neat, ν_max, cm⁻¹) 3256, 1726, 1643, 1523, 1432, 1296, 1171, 1110; LCMS (LCQ) Rt = 2.6 min (method 1), m/z (ESI+) 367.0 [M+H]⁺; HRMS m/z (ESI): calcd. for C₂₀H₁₈N₂O₃S [M+Na]⁺ 389.0930, found 389.0931.

To 4-(2-methylthiazol-4-yl)benzoic acid (80 mg, 0.36 mmol) in dichloromethane (3 mL) was added oxalyl chloride (37 µL, 0.44 mmol) in a dropwise manner followed by N,N-dimethylformamide (10 µL). The reaction mixture was stirred at ambient temperature for 16 hr. The crude acyl chloride was then added to a stirred mixture of 2-amino-6-methylpyridine (40 mg, 0.36 mmol) followed by the addition of N,N-diisopropylethylamine (397 µL, 2.28 mmol). The reaction mixture was stirred at ambient temperature for 2–16 hr. The solvent was removed under reduced pressure. The crude product was taken up in saturated aq. NaHCO₃ (5 mL) and extracted with ethyl acetate (3 × 10 mL). The combined organic components were then washed with brine (10 mL), dried over MgSO₄, filtered and concentrated under reduced pressure. The crude was purified by column chromatography (silica 12 g, 0% to 45% ethyl acetate in petroleum ether) to yield the desired amide 7 as a colourless solid (54 mg, 45%). Rf 0.48 (petroleum ether:ethyl acetate 1:1); m.p. 195–197°C; ¹H-NMR (500 MHz, DMSO-d6) 10.69 (s, 1H), 8.13 (s, 1H), 8.10 (d, J 8.2, 2H), 8.06 (d, J 8.0, 2H), 8.02 (d, J 8.2, 1H), 7.73 (t, J 7.8, 1H), 7.03 (d, J 7.3, 1H), 2.74 (s, 3H), 2.46 (s, 3H); ¹³C-NMR (126 MHz, DMSO-d6) 166.4, 165.9, 157.0, 153.3, 152.0, 138.8, 137.6, 133.5, 129.1, 126.1, 119.5, 116.2, 112.1, 24.0, 19.4; IR (neat, ν_max, cm⁻¹) 3314, 2923, 1610, 1539, 1522, 1290, 1169; LCMS (LCQ) Rt = 3.1 min (method 1), m/z (ESI+) 310.1 [M+H]⁺; HRMS (ESI): calcd. for C₁₇H₁₆N₃OS [M+H]⁺ 310.1009, found 310.1012.

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

General: We thank Will Pearce (Sussex Drug Discovery Centre) for discussions relating to assay development. We are also grateful to the Diamond Light Source Ltd., Didcot, UK for access to synchrotron radiation. Funding: This work was supported by funding from: Chinese Scholarship Council (XC and FGMP), Wellcome Trust 110578/Z/15/Z (SEW), Wellcome Trust 105630/Z/14/Z (FGMP, AWO and LHP), Cancer Research UK Programme Grant C302/A24386 (LHP and AWO).

© 2021, Chen et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.