BF cells were grown in HMI-9 (Hirumi and Hirumi, 1989) with 10% FBS at 37°C, 5% CO₂. PF cells were grown in SDM-79 (Brun and Schonenberger, 1979) with 10% FBS at 27°C. Transfections of BF cell lines with the Amaxa Nucleofector (Lonza), and of PF cell lines with the BTX transfection device (Harvard Apparatus, Inc.), were carried out as described (Merritt and Stuart, 2013), with the exception that Tb-BSF buffer (90 mM sodium phosphate buffer (Na₂HPO4/NaH₂PO4), 5 mM KCl, 0.15 mM CaCl₂, 50 mM HEPES pH 7.2) was used for BF nucleofection instead of the Human T Cell Nucleofector Kit (Schumann Burkard et al., 2011). Unless otherwise stated, concentrations of drugs used for selection and tetracycline (tet)-regulated expression of transgenes are as follows. For BF: 2.5 µg/mL G418, 5 µg/mL hygromycin, 2.5 µg/mL phleomycin, 0.5 µg/mL tet, 0.1 µg/mL puromycin, 5 µg/mL blasticidin, 12 µg/mL nourseothricin, 29 µg/mL trimethoprim. For PF: 15 µg/mL G418, 25 µg/mL hygromycin, 2.5 µg/mL phleomycin, 0.5 µg/mL tet, 1 µg/mL puromycin, 10 µg/mL blasticidin.

The structure of B4 was predicted using a ColabFold notebook-based version of AlphaFold2/MMseqs2 that incorporates modifications for analyzing proteins from organisms within the Discoba clade that include Trypanosoma (Steinegger and Soding, 2017; Wheeler, 2021). Unstructured and disordered regions of very low (AlphaFold pLDDT score < 50) and low (AlphaFold pLDDT score < 70) confidence at the N- and C termini of B4 were removed for clarity.

Predicted structures were modeled onto Saccharomyces cerevisiae Rnt1p RNase III crystal structure (5T16) using the Matchmaker function in Chimera (Pettersen et al., 2004). Modeling parameters are described in Supplementary Data . Prediction of contacts between B4 and modeled dsRNA, as well as crosslink visualization and distance measurements on predicted structures and models were also performed using Chimera.

Cleared lysate was prepared by lysis of 6 × 10⁸ BF or 4 × 10⁸ PF cells in 1.0 mL IPP150 (10 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1% Nonidet P-40) with 1% Triton X-100, followed by centrifugation at 10,000 × g and 4°C. For each immunoprecipitation, 0.5 mL cleared lysate (3 × 10⁸ BF or 2 × 10⁸ PF cells) was incubated overnight with 2 μL of rabbit antibody specific for the V5 epitope tag (Rockland Immunochemicals; #600-401-378). Magnetic beads (25 μL; Protein G Mag Sepharose Xtra; GE Healthcare) were washed twice with 1 mL of 1× phosphate-buffered saline–0.1% bovine serum albumin and once with 1 mL IPP150. The beads were then incubated for 4 h with rotation at 4°C with cleared lysate/antibody. After incubation, the supernatant was removed, and the beads were washed four times with 1 mL of IPP150. Complexes bound to beads were eluted by heating with 100 μL of 2× SDS sample buffer for 5 min at 95°C. Samples containing purified protein complexes were resolved on 10% SDS-polyacrylamide gels (Criterion Tris-HCl; Bio-Rad). For Western blotting, resolved proteins were transferred to Immobilon-P polyvinylidene difluoride membranes (Millipore) and probed using mouse monoclonal antibodies against KREPA1, KREPA2, KREL1, and KREPA3 as previously described (Panigrahi et al., 2001). Blots were sequentially stripped and reprobed using mouse monoclonal primary antibody against the V5 epitope tag (Thermo Fisher Scientific; #R960-25) at 1:5,000 with goat anti-mouse immunoglobulin secondary antibody conjugated with horseradish peroxidase at 1:5,000. Blots were developed with an enhanced chemiluminescence kit (Thermo Scientific) per the manufacturer’s instructions and imaged using X-ray film (Kodak).

We sought to establish a high-throughput complementation assay for scoring the function of multiple protein variants in T. brucei. We focused on the RNase III-like domain of B4 and used our BF B4 CN cell lines, in which both endogenous B4 alleles have been deleted, and in which a tet-inducible WT B4 allele has been inserted into the rRNA locus (McDermott and Stuart, 2017) ( Figures 1A–C ). First, we created a library of alleles in which 249 bases encoding 83 amino acids spanning the B4 RNase III-like domain were randomly mutated using error-prone PCR ( Figure 1A ). The library was enriched for full-length alleles in E. coli as previously described (Gray et al., 2007; McDermott et al., 2015a), and subcloned into a vector for constitutive expression from the tubulin array in T. brucei. Approximately 150,000 plasmids were present in the final T. brucei expression construct library. Thirty individual alleles were sequenced which revealed an average of 2.3 amino acid substitutions per encoded variant. Importantly, the plurality (37%) contained just a single amino acid substitution, 50% contained either one or two substitutions, and 30% encoded wild-type (WT) variants that did not contain any substitutions (i.e., had no or only silent mutations) ( Figure 1B ). All sampled alleles also encoded full-length variants.

To increase the efficiency and reproducibility of transfection of our large allele library into our BF CN cells, we developed an inducible I-SceI meganuclease-based system to introduce double-strand breaks at a single β-tubulin locus (Alsford et al., 2005; Glover et al., 2007, 2008; Glover and Horn, 2009; Alsford et al., 2011; Glover et al., 2015). Because our CN cell line already contained a tet-regulated WT B4 construct, we used a trimethoprim (TMP)-stabilized dihydrofolate reductase (DHFR) destabilizing domain (ddDHFR) to regulate I-SceI expression (Iwamoto et al., 2010; Rakhit et al., 2011; Ma et al., 2015; Podesvova et al., 2017). Briefly, we integrated a construct containing ddDHFR-tagged I-SceI, an embedded I-SceI cleavage site, and nourseothricin and herpes simplex virus-thymidine kinase (HSVTK) selectable markers into a β-tubulin locus of BF B4 CN cells ( Figure 2A ). The resulting I-SceI construct containing cell lines respond to tet-withdrawal with the same kinetics as the parental BF B4 CN cells (McDermott and Stuart, 2017) i.e., levels of B4 mRNA are reduced by >97% 2 days following tet withdrawal, resulting in growth inhibition that is apparent 3 days following tet withdrawal ( Supplementary Figures 2A, B ). To test the system, three independent BF B4 CN cell lines containing the I-SceI construct were treated in triplicate with increasing concentrations of TMP at 0, 10, and 100 µM for 6 h prior to transfection of a second β-tubulin-targeted construct containing a WT B4 allele and puromycin selectable marker. Following transfection, cells were positively selected in puromycin and negatively selected in ganciclovir. The approach increased transfection efficiency in a TMP dose-dependent manner, and by approximately two orders of magnitude from ~4.2 x 10^-5 to ~2.1 x 10^-3 upon treatment with 100 µM TMP ( Figure 2B ), generating up to ~70,000 unique cell lines in transfections of 3 x 10⁷ cells ( Table 1 ). Sequence analysis and growth of the resulting cell lines both in the presence and absence of tet ( Supplementary Figure 2C ) showed that site-specific cleavage at the tubulin locus had promoted the desired integration of the WT B4 allele, where it had replaced the I-SceI construct.

The mutated RNase III-like domain allele library was transfected into two independent I-SceI expressing BF B4 CN cell lines. Serial dilution revealed that we obtained libraries containing approximately 100,000 and 120,000 transfected cells respectively. Each of the T. brucei libraries were grown to logarithmic phase in the presence of tet and then diluted into media that lacked tet. Since B4 mRNA levels are significantly reduced by 2 days, and cell growth is inhibited by 3 days, following tet withdrawal ( Supplementary Figures 2A, B ) we collected cells both before (input; plus tet) and after 4 days of growth minus tet. We extracted genomic DNA, PCR amplified the segment that had been mutated, and carried out Illumina sequence analysis for each time-point. We identified 75,879 distinct nucleotide sequences, including the WT nucleotide sequence and sequences containing between 1-10 nucleotide changes, across all time-point and replicate samples. Of the 75,878 sequences that had nucleotide changes, 3,303 contained silent mutations which did not encode any amino acid substitutions, 20,909 encoded sequences with a single amino acid substitution (735 unique amino acid sequences), 34,907 had two amino acid substitutions (27,619 unique amino acid sequences), and the remaining 16,759 had between three and nine amino acid substitutions from WT sequence (15,708 unique amino acid sequences) ( Supplementary Table 2 ). After filtering variants based on having a minimum read count of 50 in both replicate input libraries generated from cells grown in the presence of tet ( Supplementary Figure 3 ), we calculated fitness scores based on changes in frequency from input plus tet to minus tet cell populations (Faure et al., 2020). These fitness scores serve as a proxy for cell growth and thus function of each variant. Variant fitness scores were highly correlated in our replicate experiments ( Supplementary Figure 4 ). The filtered dataset quantifies the effects of 350 single amino acid changes on cell growth with high reproducibility, which we focused on to allow for easier deconvolution of mutant phenotypes ( Figure 3 and Supplementary Table 3 ). Within this group we identified substitutions at every site in the 83 amino acid mutated region, with on average ~4 different substitutions at each position. Of these, 102 single substitutions at 61 of the 83 sites had a detrimental effect i.e., fitness score below -0.4 and p value < 0.05, a further 183 single substitutions in 77 of the 83 amino acids had no significant effect, and 65 single substitutions in 44 amino acids appeared to result in a slight beneficial effect on cell fitness ( Figure 3A ). We calculated a mean site-level fitness score using the fitness scores for all individual substitutions per amino acid position ( Figure 3B and Supplementary Table 3 ). Perhaps unsurprisingly, residues with the lowest i.e., most detrimental, mean fitness scores were also sites of the lowest individual substitution fitness scores ( Figures 3A, B ). We confirmed that selected single substitutions were responsible for the growth defects observed in our high-throughput complementation screens by recreating WT and site-directed V5-tagged mutant constructs and transfecting them into the parental BF CN cell line ( Figure 3C and Supplementary Figure 5 ). Expression of the WT or mutant alleles was confirmed by Western analysis that probed for the V5 tag ( Supplementary Figure 5 ). Growth analyses confirmed the fitness phenotypes of selected detrimental loss-of-function (LOF) substitutions, as well as of control substitutions that did not result in growth defects in our screens ( Figure 3C , Supplementary Figure 5 , and Supplementary Data ).

Conservation analysis is a widely used method for prediction of sites that are important for protein function, as mutational sensitivity and evolutionary conservation are often strongly correlated (Stone and Sidow, 2005). To study the correlation between B4 RNase III-like domain conservation and function as defined by our screen, we quantified the degree of conservation for each B4 amino acid position across a wide range of orthologs from 42 kinetoplastid species including T. cruzi and L. major ( Figure 4A and Supplementary Figure 6 ). We then calculated a conservation score for each site as the mean pairwise identity over all pairs per column in an alignment of all orthologs ( Supplementary Table 3 ). Calculation of the Pearson correlation coefficient between conservation and site-level fitness scores revealed no appreciable linear correlation ( Figure 4B ), demonstrating that B4 sequence conservation cannot be solely used to predict function. Considerable variation in correlation between conservation and variant effects has been observed across previous deep mutational scanning experiments in other model systems and is protein dependent (Hoie et al., 2022), further highlighting the power of this mutational scanning approach.

The release of AlphaFold2 and Discoba-specific improvements (Wheeler, 2021) together provide powerful resources to predict T. brucei B4 structure and to understand the effects of our amino acid substitutions on B4 RNase III function ( Supplementary Figure 8 ). For clarity, we removed low confidence regions at the N and C-termini of the modelled B4 that were unstructured i.e., pLDDT score < 70 ( Figure 5A , Supplementary Table 3 , and Supplementary Figure 7 ). These low confidence regions also correlate with regions of predicted disorder, particularly in the mutated RNase-like domain ( Supplementary Table 3 and Supplementary Figure 8 ) (Horvath et al., 2020; Miskei et al., 2020; Hatos et al., 2022, 2023). Comparison with the crystal structures of archetypal eukaryotic (Saccharomyces cerevisiae Rnt1p; PDB 5T16) ( Figure 5A ) and bacterial (Aquifex aeolicus RNase III; PDB 2NUF) (not shown) RNase III proteins confirmed the presence of the RNase III domain-like fold in B4, as expected from our previous sequence searches and homology modelling (22,46,47). Furthermore, comparisons with the AlphaFold predicted structures of other RECC RNase III paralogs that all contain an RNase III Associated Motif (RAM) flanking their RNase III or RNase III-like domains (Carnes et al., 2022), revealed that B4 also has a RAM. We modelled the Discoba AlphaFold2 predicted structure for B4 onto the S. cerevisiae Rnt1p RNase III crystal structure and built a dsRNA substrate-containing model ( Figures 5A, B ). We further validated our model using our previous BS3 crosslinking mass spectrometry data (22). BS3 has a linker arm of 11.4 Å when fully extended and can crosslink two residues whose Cα atoms are up to 30 Å apart (48). We measured the distances between the B4 residues that could be mapped onto the predicted structure and model. We were able to measure the distances for three intralinks ( Figure 5C and Supplementary Table 4 ). The distances between crosslinked residues were <30 Å, indicating that the model is consistent with available experimental data and provides a reasonable representation of the structure. We also highlighted B4 residues that crosslinked with other RECC proteins on our model ( Figure 5C and Supplementary Table 5 ). These analyses did not identify crosslinks to other proteins in the vicinity of the modelled dsRNA but did identify several crosslinks to other RECC RNase III and OB-fold proteins in other regions.

To assess fitness and protein function in the context of structure, we mapped the mean site-level fitness scores onto our structural model ( Figures 6 , 3B , and Supplementary Table 3 ) (Hilton et al., 2020). Comparison with the previous protein-protein crosslinking data ( Supplementary Figure 9 ) (McDermott et al., 2016) revealed detrimentally substituted residues that are proximal to regions implicated in RECC protein binding, and therefore that potentially interfere with B4 protein-protein interactions within RECCs. We tested this via immunoprecipitation of V5-tagged detrimental K211I and control R174H mutant B4 ( Figure 6B and Supplementary Figure 5 ). K211 is close to sites of B4 interlinks with other RECC proteins (K129 and K110) and intralinks within B4 ( Supplementary Figure 9 ). As observed previously for cells that lack B4 (McDermott and Stuart, 2017), cells that exclusively expressed the K211I B4 variant had much-reduced signals for other RECC components in input and co-immunoprecipitation compared to cells exclusively expressing WT B4 or the control R174H variant ( Figure 6C ). We interpret the reductions in total levels of RECC components as due to turnover of proteins that cannot be incorporated into RECCs without functional B4.

Several RECC protein mutagenesis studies, including of B5-B8 that contain RNase III-like domains, have identified single amino acid substitutions with different consequences on BF vs. PF cell growth and RECC integrities (Carnes et al., 2011; McDermott et al., 2015a, b; McDermott and Stuart, 2017; McDermott et al., 2019; Carnes et al., 2022; Davidge et al., 2023). Therefore, we hypothesized that we might also observe life cycle stage differences in the growth of cells containing substitutions that were identified in our BF B4 deep mutational scan. To test this hypothesis, we transfected selected site-directed V5-tagged LOF mutant constructs into our parental PF B4 CN cell line (McDermott and Stuart, 2017), where expression of the WT or mutant alleles was confirmed by Western analysis that probed for the V5 tag ( Supplementary Figure 9 ). Exclusive expression of several B4 variants with BF LOF substitutions in the RNase III-like domain residues did not affect PF cell growth ( Figure 7A and Supplementary Data ). This included K211I, which in contrast to BF ( Figure 6C ), did not affect total levels or co-immunoprecipitation of RECC components in PF. The exception was the F207V variant that caused strong growth defects in both BF and PF cells ( Figures 3B , 7A ), reduced total levels of RECC components in BF, and reduced co-immunoprecipitation of RECC components in PF ( Figure 7B ) compared to WT B4. In conclusion, as hypothesized, our screen identified single amino acid substitutions in the B4 RNase III-like domain that have different consequences on BF vs. PF cell growth and RECC integrities.