Gene editing using oligo targeting in T. brucei simply requires delivery of single-stranded oligodeoxynucleotides (ssODNs) by electroporation, using approx. 50 b ‘reverse-strand’ ssODNs [20]; a single allele is typically edited. We recently developed Multiplexed Oligo Targeting (MOT) library screening in T. brucei to examine a proteasomal drug target [18] and have now applied this approach to CPSF3 (Tb927.4.1340). Our previous analysis [12] revealed twenty-six amino acids located within 5 Å of the acoziborole binding site in CPSF3 (Fig 1A); with the cognate codons for these amino acids distributed over a region of approx. 1,200 bp in the CPSF3 gene. Prior to assembling a mutant library, we determined the CPSF3 allele replacement frequency using oligo targeting. Mismatch-repair supresses editing efficiency in T. brucei [20], and we found that this was also the case at the CPSF3 locus. Using an ssODN to introduce a CPSF3 N²³²H edit, we obtained 6.5-times more acoziborole-resistant cells following transient MSH2 knockdown for 24 h, with an estimated allele replacement frequency of approx. 0.01%. A similar mock assay but with no ssODN failed to yield any drug-resistant cells.

To assemble a CPSF3 mutant library, we designed twenty-six reverse-strand, 53-b, ssODNs, each with a centrally located degenerate ‘NNN’ (N = A, C, T, G) codon (S1 Table). Mismatch-repair was transiently knocked down for 24 h and each degenerate ssODN was delivered individually to avoid combinatorial editing in individual cells (Fig 1B), which could allow for the enrichment of bystander edits, complicating codon-based genotype to phenotype assessments when screening MOT libraries [18]. We estimated an average yield of approx. 325 codon edits per ssODN, which in the pooled library equated to approx. 6,500 edited cells among 50 million cells in a 150 ml culture volume; an average approx. 5-fold redundancy for each alternative codon. To assess editing at each of the targeted codons, we extracted genomic DNA before, and six hours after library assembly, PCR-amplified the edited region in the CPSF3 gene, and deep-sequenced the amplicons (Fig 1B). A scan to quantify variant codons across the edited region revealed highly specific editing at all twenty-six targeted sites (Fig 1C). We concluded that the substantial majority of all 1,664 possible alternative codons encoding 520 CPSF3 variants were likely represented in our pooled MOT library.

The pooled MOT library was split to generate a pair of MOT libraries which were grown with acoziborole at 1 μM; approx. three times the EC₅₀ (Effective Concentration of drug to inhibit growth by 50%). The selected libraries were then split three ways and grown with acoziborole at either 3, 9 or 27 μM. In this case, only 3 μM selection yielded resistant cells. We then extracted genomic DNA from resistant cells, PCR-amplified the edited region in the CPSF3 gene, deep-sequenced the products (Fig 1B), and quantified variant codons. The heatmap in Fig 1D shows relative representation of all 1,664 codon variants prior to drug-selection and following 1 or 3 μM acoziborole selection. Only the known N²³²H resistance-associated mutation [12,20], represented by both possible histidine codons, was enriched following selection; one by a single-nucleotide edit (AAT - CAT) and the other by a double-nucleotide edit (AAT - CAC).

Several mutations in Apicomplexan CPSF3 have been linked to benzoxaborole resistance, including Toxoplasma Y³²⁸H, Y⁴⁸³N, S⁵¹⁹C and E⁵⁴⁵K [14], and Plasmodium Y⁴⁰⁸S and D⁴⁷⁰N [8]. These sites are equivalent to sites targeted for editing above in T. brucei; N²³², Y³⁸³, S⁴²¹, and N⁴⁴⁸ (S1 Fig), with N²³²H already linked to acoziborole resistance in T. brucei (see above). Pairs of co-transfected mutagenic ssODNs that target proximal sites are incorporated at high frequency in T. brucei [18] and a similar phenomenon is observed in yeast [19]. Given the paucity of acoziborole resistance associated mutations identified above, we exploited combinatorial oligo targeting to edit CPSF3 and to seek to identify other sites that impact acoziborole efficacy.

We combined ssODNs designed to introduce an N²³²H^CAT edit, and three degenerate ssODNs designed to target Y³⁸³, S⁴²¹ and N⁴⁴⁸ for site saturation mutagenesis. Mismatch-repair was transiently knocked down for 24 h, five T. brucei populations were transfected with the ssODN mix, pooled, split into two cultures, and grown with acoziborole at 3 μM. Resistant populations emerged, and ten sub-clones were assessed by Sanger sequencing, revealing three distinct combinatorial edits, all incorporating N²³²H^CAT, as expected, and additionally with Y³⁸³F (either TTC or TTT) and N⁴⁴⁸Q or N⁴⁴⁸H (Fig 2A); all observed edits were heterozygous, while S⁴²¹ edits were not observed.

Acoziborole EC₅₀ dose response analysis was carried out for an N²³²H mutant and for a representative clone of each combinatorial mutant. The N²³²H mutant registered the expected moderate 3-fold increase in EC₅₀, but this was not significantly different to wild type cells (P = 0.73; one-way analysis of variance). In contrast, the N²³²H/ Y³⁸³F (HFN), N²³²H/ Y³⁸³F/ N⁴⁴⁸H (HFH) and N²³²H/ Y³⁸³F/ N⁴⁴⁸Q (HFQ) mutants all registered a significantly different EC₅₀ (P < 0.0001) relative to wild type cells (Fig 2B). In particular, the HFQ triple mutant registered >40-fold increased EC₅₀ suggesting synergy between the N²³²H and Y³⁸³F edits and additional synergy with the N⁴⁴⁸Q edit.

Recovery of triple-edited cells using the combination of ssODNs detailed above suggested remarkably efficient incorporation of multiple ssODNs, targeting sites 651 bp apart, likely at a common DNA replication or transcription fork. At this point, it remained unclear as to whether the heterozygous combinatorial edits were present on the same CPSF3 allele, however. To determine whether this was indeed the case, we engineered HFQ triple mutant strains to express a native CPSF3 allele fused to green fluorescent protein [12]. Amplification and Sanger-sequencing of the CPSF3^GFP alleles from a series of clones revealed either wild-type sequence or triple-mutant sequence, indicating tagging of either allele, and demonstrating the presence of all three edits on the same allele (Fig 3A).

We next asked whether acoziborole-resistant triple mutant CPSF3 was resistant to acoziborole-induced turnover [17]. We analysed a pair of acoziborole resistant CPSF3^GFP strains, one with the triple-mutant HFQ allele tagged with GFP and another with the wild-type NYN allele tagged with GFP. We treated these cells with 3, 6 or 10 μM acoziborole for 30 minutes and assessed the samples by protein blotting. The wild-type protein was rapidly turned over by acoziborole, and pre-treatment with the proteasome inhibitor MG132 blocked turnover (Fig 3B), confirming that NYN-CPSF3^GFP turnover was proteasome-dependent [17]. In contrast, mutant HFQ-CPSF3^GFP was resistant to turnover (Fig 3B). We concluded that triple mutant CPSF3 resisted rapid acoziborole and proteasome-dependent turnover.

Identification of multiple resistance-associated mutations around the acoziborole binding site, which is also part of the RNA substrate binding site, provided compelling evidence for CPSF3 binding by this small molecule. To further explore interactions between acoziborole and CPSF3, we modelled and compared wild-type NYN-CPSF3 and multiple CPSF3 mutants. We used crystal structures for related RNAse’s as templates and generated T. brucei CPSF3 homology models docked with acoziborole. To select a binding pose, we prepared single mutants for all possible residues at N²³², Y³⁸³, and N⁴⁴⁸, and used ΔΔG Molecular Mechanics General Born Surface Area (MM/GBSA) free energy calculations to track those mutations [21]. Among the ensemble of predicted poses, we selected pose two derived from the 3IEM-based homology model, which exhibited sensitivity to mutations at all three sites (S2 Fig).

In the selected pose, CPSF3 N²³², Y³⁸³ and N⁴⁴⁸ surround the acoziborole binding site, with distances from the ligand of 2.17, 3.67 and 5.45 Å, respectively. Mutations at these sites may have steric effects or impact acoziborole affinity. To explore these potential impacts, we ran molecular dynamics simulations, which suggested interactions with N²³² and Y³⁸³; via a water network and a hydrophobic interaction, respectively (Fig 4A). Simulations were then clustered by Root Mean Square Deviation (RMSD), and three representative snapshots were taken for mutation analysis. We created single, double, and triple mutant combinations of N²³²H, Y³⁸³F, N⁴⁴⁸Q, and N⁴⁴⁸H, and determined the ΔΔG MM/GBSA associated with each mutant. Upon mutation, measures of RMSD indicate a number of steric effects. Notably, the four mutants we describe above (Fig 2) register the greatest increase in RMSD, which is particularly pronounced in the ‘HF’ double mutant and in the ‘HFQ’ triple mutant (Fig 4B), consistent with our in cellulo experimental data. Finally, free energy (ΔΔG MM/GBSA) was predicted to be increased for multiple double mutants compared with single mutants and further increased for triple mutants compared with double mutants (Fig 4C).

Our predictions also potentially explain why we failed to recover mutations at S⁴²¹ above; because the side chain is not predicted to directly face the ligand (Fig 4D, upper panel). We can also see why N⁴⁴⁸ does not appear in the molecular dynamics interaction network (Fig 4A). Although N⁴⁴⁸ appears in close proximity to the ligand in the initial structural model, this proximity is predicted to be transient due to a highly flexible and dynamic loop (Fig 4D, lower panel). While mutation at N⁴⁴⁸ alone is predicted to have no impact on RMSD (Fig 4B) or free energy (Fig 4C), mutation at N⁴⁴⁸ is predicted to impact free energy in a synergistic manner with N²³²H and Y³⁸³F (Fig 4C), suggesting that these additional mutations impact the flexible loop. This is also consistent with recovery of N⁴⁴⁸ mutations only in the context of triple mutants above (Fig 2A). We conclude that the predictions were in agreement with our experimental data, which suggested synergy between the resistance-associated N²³²H and Y³⁸³F edits and additional synergy with the N⁴⁴⁸Q edit (Fig 2B).

Thus, in silico modelling and molecular dynamics simulations suggested that triple-mutant CPSF3 was resistant to acoziborole due to both steric clash between the ligand and the mutated residues, and disruption of hydrophobic and water-mediated interactions, thereby reducing affinity. The poses shown in Fig 4E indicate how the N²³², Y³⁸³, and N⁴⁴⁸ residues cluster around the fluorinated benzene ring in acoziborole (upper panel) and how acoziborole stability is impacted in the HFQ triple-mutant (lower panel).

Bloodstream form Lister 427 T. brucei wild-type and MSH2 RNAi [20] strains were cultured in HMI-11 (Gibco) supplemented with 10% fetal bovine serum (Sigma) at 37°C and with 5% CO₂ in a humidified incubator. Tetracycline inducible 2T1-RPaⁱMSH2 RNAi cells were generated previously [20]. Genetic manipulation was carried out using electroporation with a Nucleofector (Lonza), and a human T-cell kit (Lonza), with the Nucleofector set to Z-001 (Amaxa). To calculate allele replacement frequency, transient MSH2 knockdown was induced 24 h prior to transfection by the addition of 1 μg/mL tetracycline (Sigma). Acoziborole (MedChem Express) was then added 6 h after transfection of an ssODN designed to introduce a CPSF3 N²³²H edit and cells were serially diluted in 96-well plates. Number of resistant clones was determined 5 days later considering 50% survival following transfection. GFP-tagged CPSF3 was expressed in a wild-type T. brucei background as described previously by Wall et al. [12].

For site saturation mutagenesis, we used a set of twenty-six degenerate single-stranded oligodeoxynucleotides (ssODNs, Thermo Fisher Scientific), as described [20]. Briefly, we used 40 μg of each ssODNs in 10 μl of 10 mM Tris-HCl, pH 8.5, mixed with 5 million T. brucei ‘inducible MSH2 RNAi’ cells in 100 μl transfection buffer. MSH2 knockdown was induced 24 h prior to transfection. Each ssODN was transfected individually and the cells were then pooled in 150 ml of medium to generate the library. After 6 h, cells from 50 ml of the culture were cryo-preserved, cells from another 25 ml were collected for (pre-selection) DNA extraction, and the remainder were split into two cultures and subjected to selection with 1 μM acoziborole in a total volume of 75 ml per culture. Following eight or ten days, 50 ml of each culture was collected for DNA extraction, and the remainder was split into three cultures that were subjected to selection with either 3 μM, 9 μM or 27 μM acoziborole. Only 3 μM selection yielded resistant cells, and these cultures were collected for DNA extraction. Genomic DNA was extracted using a Qiagen DNeasy Kit followed by PCR amplification of the CPSF3 gene using Q5 DNA Polymerase (NEB). PCR products were purified using a PCR Purification Kit (Qiagen).

To identify edits, CPSF3 amplicons were subjected to deep sequencing using DNBSEQ (BGI Genomics), as described [18]. Briefly, filtering of sequencing reads was performed using SOAPnuke and codon variant scoring was performed with the OligoSeeker (0.0.5) Python package [23]. To visualise codon variant scores, we performed a normalization step by dividing each codon variant score by the total reads for that position and converted the fraction of reads to a percentage. This was followed by background correction whereby we subtracted the values for the control sample. Negative values were replaced with 0, and average values for the duplicate libraries were calculated to give codon variant scores. Edits for codons registering >100 read-counts in the control sample, including all single nucleotide variants, were excluded from the analyses shown in Fig 1C.

Oligo targeting was carried out using ssODNs essentially as described above, but in five batches with 10 μg of each ssODNs mixed with 10 million T. brucei ‘inducible MSH2 RNAi’ cells in 100 μl transfection buffer. The batches were pooled and selection with 3 μM acoziborole was applied 6 h later. Genomic DNA extraction and PCR amplification were carried out as above and, to identify edits, amplicons were subjected to Sanger sequencing (Genewiz, Azenta Life Sciences).

To determine the Effective Concentration of drug to inhibit growth by 50% (EC₅₀), cells were plated in 96-well plates at 1 x 10³ cells/ml in a 2-fold serial dilution of selective drug. Plates were incubated at 37°C for 72 h, 20 µl resazurin sodium salt (AlamarBlue, Sigma) at 0.49 mM in PBS was then added to each well, and plates were incubated for a further 6 h. Fluorescence was determined using an Infinite 200 pro plate reader (Tecan) at an excitation wavelength of 540 nm and an emission wavelength of 590 nm. EC₅₀ values were derived using Prism (GraphPad), or Graphit in the case of biphasic dose response curves.

Twenty-five million T. brucei cells were treated with acoziborole for 30 min and collected by centrifugation. The proteasome inhibitor MG132 was added to a culture at 20 μM, 30 mins prior to acoziborole treatment. RIPA buffer and proteinase inhibitors were added, and samples sonicated for 10 cycles (30 sec on/off), centrifuged, and the supernatant collected. Proteins were resolved using SDS-PAGE in 4–12% Bis-Tris gels (Invitrogen) and then transferred to PVDF membrane using the iBlot 2 system (Invitrogen). Membranes were blocked with blocking buffer (50 mM Tris-HCl pH 7.4, 0.15 M NaCl, 0.25% BSA, 0.05% (w/v) Tween-20, and 2% (w/v) fish skin gelatin). The following antibodies were used in blocking buffer: rabbit α-GFP (Invitrogen, A-11122, 1:2000) and mouse α-EF1α (Millipore, 1:10000) overnight at 4°C. Secondary antibodies were α-rabbit IRDye800 and α-mouse IRDye680 (1:15000 and 1:10000, respectively, LI-COR) for 1 h at room temperature. Blots were analysed using the LI-COR Odyssey CLx Imager and Image Studio 6.0.

T. brucei CPSF3 homology models were generated using the crystal structures with PDB ID: 3IEM (Thermus thermophilus HB8) [24], 6Q55 (Cryptosporidium hominis) [9], and 8T1Q (Homo sapiens) [25] as templates, and the homology modelling tool implemented in Maestro (Schrodinger inc.). Acoziborole was docked using Glide XP [26], and the top two poses for each model were selected for further analysis. Two docking poses were also generated using Boltz2, a state-of-the-art generative model for protein–ligand docking [27]. The ligand and protein complex was placed in a cubic box of water and 0.15 M NaCl at 27 °C. We ran 500 ns of molecular dynamics simulations following a multistage equilibration protocol designed to gradually relax positional restraints and achieve thermodynamic stability of the system before production sampling. Initially, the system was constructed and subjected to energy minimisation to remove steric clashes and optimise atomic geometries. This was followed by a short Brownian dynamics simulation in the canonical (NVT) ensemble at 10 K for 100 ps, using small integration time steps. During this simulation, positional restraints were applied to all solute heavy atoms to allow the solvent and ions to equilibrate around the solute. The simulation was then continued under NVT conditions for an additional 12 ps with the same restraints to ensure complete thermal equilibration of the solvent environment. Subsequent equilibration was performed under NPT conditions at 10 K and 1 bar for 12 ps with solute restraints to allow the system density to adjust. A further 12 ps NPT simulation was carried out under the same conditions to ensure pressure stability. The positional restraints were then released, and the system was equilibrated for 24 ps under NPT conditions to allow for the complete relaxation of all atoms. Finally, an unrestrained production simulation was performed under stable NPT conditions for 500 ns to collect trajectory data for analysis. Post-simulation analyses were conducted on the resulting trajectories to evaluate structural stability, conformational changes, and energetic properties throughout the simulation. We determined the ΔΔG MM/GBSA associated with the change. In the case of histidine, we analysed all the possible protonation states. We tracked the movement of the ligand in the binding site using RMSD, employing an MCS protocol on the ligand.