Chemogenomic Profiling of Antileishmanial Efficacy and Resistance in the Related Kinetoplastid Parasite Trypanosoma brucei

The arsenal of drugs used to treat leishmaniasis, caused by Leishmania spp., is limited and beset by toxicity and emergent resistance. Furthermore, our understanding of drug mode of action and potential routes to resistance is limited. Forward genetic approaches have revolutionized our understanding of drug mode of action in the related kinetoplastid parasite Trypanosoma brucei.

cutaneous leishmaniasis can be self-limiting, infections with Leishmania braziliensis (and other members of the Viannia subgenus) can develop into mucocutaneous leishmaniasis, a profoundly disfiguring form of the disease (4). Visceral leishmaniasis (VL), also known as kala-azar, is typically fatal if untreated.
There are four current antileishmanial drugs, sodium stibogluconate (SSG), paromomycin, miltefosine, and amphotericin B, which are unsatisfactory due to toxicity, emerging drug resistance, complex administration protocols, and variable efficacy depending on the disease type or infecting Leishmania species (5). With the exception of miltefosine (in use against leishmaniasis since 2002), the current antileishmanial drugs have been in use for many decades. Until recently, efforts have focused on the development of more-effective drug delivery regimens and combination therapies, with the aim of reducing dosages (and, therefore, side effects) and combating the emergence of resistance. The rise of antimonial-resistant Leishmania donovani on the Indian subcontinent now precludes the use of SSG (6), while miltefosine-resistant L. donovani has been confirmed in the clinic (7). Consequently, the World Health Organization recommends various combination therapies, depending on the Leishmania species and geographical region (8). However, it is relatively easy to generate Leishmania parasites resistant to combination therapies in the laboratory (9,10). More recently, new drugs have entered the clinical development pipeline. However, the most advanced of these, fexinidazole, which recently passed phase 2/3 clinical trials against HAT (11) and has antileishmanial activity in vitro (12), lacks efficacy in vivo (13).
Given the ease with which Leishmania parasites become resistant to the available drugs, it is critically important to understand how this resistance might develop. Identification of the genetic changes underlying drug resistance will enable the development of molecular diagnostics to inform treatment choice (14). Leishmania genome and transcriptome analyses have identified large numbers of candidate genes (15,16), but relatively few have been directly linked to drug action. While some drugs can freely move across membranes, many are taken up via specific surface receptors and transporters. For example, miltefosine uptake is dependent on a Leishmania amino phospholipid-transporting (P4)-type ATPase (or flippase) and its ␤-subunit/CDC50 cofactor, Ros3 (17,18), while the Sb(III) form of SSG is taken up via an aquaglyceroporin, AQP1 (19). There is also evidence that the ABC transporter MRPA influences SSG uptake and sequestration (20), and several other proteins have been implicated in SSG efficacy (reviewed in reference 14). In addition, the generation of drug-resistant Leishmania in the laboratory and various omics analyses have provided insights into antileishmanial drug action and resistance mechanisms. Proteomic analyses of paromomycin-resistant L. donovani revealed a complex picture, with a range of proteins being upregulated, including several involved in translation regulation, vesicular trafficking, and glycolysis (21). A similar analysis of amphotericin B-resistant Leishmania infantum highlighted the differential expression of metabolic enzymes and the upregulation of proteins involved in protection against reactive oxygen species (22). Metabolomic analyses suggested that oxidative defense also contributes to SSG-amphotericin B and SSG-paromomycin resistance in L. donovani (23).
The studies described above highlight the phenotypic consequences of changes in drug sensitivity but not necessarily the genetic changes responsible. Forward genetic approaches can identify genes that contribute to drug action and resistance. For example, genome-scale RNA interference (RNAi) library screening, coupled with RNA interference target sequencing (RIT-seq), has revolutionized our understanding of anti-HAT drug action and resistance (24,25). In addition, cosmid sequencing (Cos-seq) has enabled gain-of-function screening in Leishmania (26), leading to target validation for N-myristoyltransferase (27) and the identification of a panel of putative antimony and miltefosine resistance genes (28). While undoubtedly a powerful technique, Cosseq is unable to identify drug uptake or activation mechanisms, which can be characterized by loss-of-function approaches such as RIT-seq. However, due to the absence of the RNAi machinery in most Leishmania species (with the notable exception of L. braziliensis [29]), this loss-of-function approach is not possible for these parasites.
Although T. brucei and Leishmania have distinct life cycles, they are phylogenetically related kinetoplastid parasites that exhibit a high degree of biochemical and genetic similarity (30). Indeed, the majority of orthologous genes are syntenic, indicating little change in gene order since divergence from a common ancestor. Perhaps not surprisingly, then, several "dual-purpose" drugs display activity against both parasites, including pentamidine (5), fexinidazole (11,12), and the proteasome inhibitor GNF6702 (31). T. brucei is also susceptible to in vitro killing by the four current antileishmanial drugs. Therefore, we hypothesized that T. brucei RNAi library selection with the antileishmanial drugs would enable the identification of candidate drug efficacy determinants with orthologues in Leishmania.
Here, we describe RIT-seq library screening using each of the current antileishmanial drugs. We identified 44 high-confidence putative drug efficacy determinants, including T. brucei orthologues of the Leishmania SSG and miltefosine transporters (MTs). Among many previously unknown drug efficacy determinants, we found that the vesicleassociated membrane protein TbVAMP7B contributes to miltefosine and amphotericin B efficacy and highlight a role for a cohort of amino phospholipid-transporting P4-type ATPases (or "flippases") in driving amphotericin B efficacy. This collection of validated and putative antileishmanial drug efficacy determinants provides new insight into mode of action and potential resistance mechanisms and represents an important resource to guide future study.

RESULTS
Orthology-based chemogenomic profiles for antileishmanial drugs. The four current antileishmanial drugs SSG, paromomycin, miltefosine, and amphotericin B have in vitro 50% effective concentration (EC 50 ) values against T. brucei of 1.8 g · ml Ϫ1 , 17 M, 30 M, and 260 nM, respectively (Fig. 1A). The equivalent values versus intracellular L. donovani amastigotes in mouse peritoneal macrophages are approximately an order of magnitude higher (SSG and paromomycin) or lower (miltefosine and amphotericin B) (32). To identify factors whose loss renders T. brucei less sensitive to each antileishmanial drug, a bloodstream-form (BSF) T. brucei RNAi library was induced for 24 h, and each drug was then added at 1ϫ to 3ϫ EC 50 ; selection and induction were maintained thereafter (Fig. 1B). After selection for approximately 10 days, populations with reduced drug sensitivity emerged and grew consistently under continued selection (Fig. 1C).
Following robust growth for at least 2 days, genomic DNA was isolated from the drug-selected populations and subjected to RNAi construct-specific PCR, generating distinct banding patterns for each (Fig. 1C). We sequenced the amplified RNAi target fragment populations from the selected RNAi libraries on an Illumina HiSeq platform (see Table S1 in the supplemental material). For each selected RNAi library, we mapped more than 3 million individual sequence reads, representing antileishmanial enriched RNAi target fragments, to the TREU927 T. brucei reference genome (33) using our established RIT-seq methodology (34) (Fig. 1B). The presence of the RNAi constructspecific barcode identified "high-confidence" hits, i.e., those represented by more than 99 barcoded reads/kb/predicted transcript (open reading frames plus predicted untranslated regions, as annotated in the TREU927 reference genome available at www .tritrypdb.org) and recovery of at least two independent RNAi target fragments (Fig. 2,  Fig. S1, and Table S1).
Importantly, we identified T. brucei orthologues of two known Leishmania determinants of antileishmanial drug efficacy. RNAi target fragments that mapped to the TbAQP2-3 locus (Tb927. 10.14160-70), which encodes two aquaglyceroporins, dominated the SSG-selected RNAi library; L. donovani AQP1 (LdBPK_310030.1) is a key mediator of SSG uptake (19). Another significant hit identified following miltefosine selection was a putative flippase (Tb927.11.3350); the corresponding coding sequence is syntenic with the L. donovani miltefosine transporter (LdBPK_131590.1) (17). The identification of T. brucei orthologues of these known antileishmanial efficacy determinants highlights the power of this chemogenomic profiling approach in the identifi-cation of mechanisms of action and resistance that are also relevant to Leishmania parasites. In addition to these hits, our RIT-seq analyses yielded a further 42 highconfidence hits (Fig. 2, Fig. S1, and Table S1).
TbAQP3, an orthologue of Leishmania AQP1, is linked to antimonial action. Aquaglyceroporin defects in T. brucei and in Leishmania have been linked to arsenical and antimonial resistance (see above), but specific relationships among drugs and AQPs have not been fully elucidated. For example, TbAQP2 is responsible for pentamidine and melarsoprol uptake (35), possibly via receptor-mediated endocytosis in the former case (36), and mutations that disrupt TbAQP2 are responsible for melarsoprol resistance in patients (37). L. donovani AQP1 has also been linked to antimonial resistance in patients (38). Notably, TbAQP3 and Leishmania AQP1 have the same set of selectivity filter residues (NPA/NPA/WGYR), while TbAQP2 has a divergent set (NSA/ NPS/IVLL) (39). Therefore, we investigated the specificity of the interaction between SSG and TbAQP2/TbAQP3, the major hits in the SSG screen.
Sequence mapping of the RNAi target fragments following SSG selection revealed that approximately 71% and 29% of mapped reads containing the RNAi construct-specific barcode corresponded to TbAQP2 (Tb927. 10.14170) and TbAQP3 (Tb927. 10.14160), respectively (Fig. 3A); only 0.08% of reads mapped elsewhere in the genome. These data are consistent with the idea that both aquaglyceroporins contribute to SSG action. However, the TbAQP2 and TbAQP3 coding sequences are 82.3% identical; thus, while an  50 ), and adjusted as indicated (black arrows); induction with 1 g · ml Ϫ1 tetracycline was maintained throughout. Genomic DNA was prepared at the indicated times (red arrows). Insets show RNAi library-specific PCR.
RNAi fragment may unambiguously map to TbAQP2, it may be sufficiently similar to TbAQP3 to elicit its depletion. Therefore, we tested the relative contribution of the encoded aquaglyceroporins to SSG action against T. brucei using aqp2-3-null and reexpression cell lines (35).
Deletion of the TbAQP2-3 locus led to a 6.7-fold increase in the SSG EC 50 (Fig. 3B), consistent with the output from the screen. Inducible expression of green fluorescent protein-tagged TbAQP2 ( GFP TbAQP2) in the null cell line had little effect on T. brucei SSG sensitivity (Fig. 3C, left); however, GFP TbAQP3 expression reduced the SSG EC 50 5.5-fold (Fig. 3C, right). In contrast, and as shown previously (35), GFP AQP2 expression complemented the pentamidine resistance of aqp2-3-null T. brucei (Fig. 3D, left), while GFP AQP3 expression had no effect on pentamidine sensitivity (Fig. 3C, right). Therefore, SSG sensitivity and resistance are specifically determined by TbAQP3 expression. This indicates that the NPA/NPA/WGYR selectivity filter, present in both TbAQP3 (39) and Leishmania AQP1, may be selective for antimonial uptake.
T. brucei lysosomal MFST influences aminoglycoside action. Selection of the BSF T. brucei RNAi library with the antileishmanial aminoglycoside paromomycin identified 50 hits, 28 of which fulfilled our high-stringency criteria (Table S1). Twenty-one of the high-confidence hits were functionally annotated and included several associated with transport and nucleic acid processing. The top three hits with functional annotations were Tb927. 9 (40), targeted by approximately 84%, 1.7%, and 0.9% of the mapped reads, respectively (Fig. 4A, Fig. S2, and Table S1). However, while parasites able to deplete AAT15 and TSN persisted in the population over the 12 days of selection with paromomycin, we were unable to detect a significant advantage versus wild-type T. brucei during the course of a standard 72-h EC 50 assay (Fig. S1). Therefore, we focused our attention on the MFST genes.
The genes at the Tb927.9.6360-80 locus share at least 92% sequence identity and encode three putative MFSTs, a ubiquitous family of proteins responsible for membrane transit of a wide range of solutes, including drugs (41). Comparison with the sequences annotated "MFS" or "major facilitator superfamily transporter" in the L. major reference genome confirmed that the syntenic coding sequence, LmjF.15.0870, is most closely related to Tb927.9.6360-80 ( Fig. 4B and Fig. S3). The Leishmania and T. brucei proteins share similar trans-membrane (TM) domain organizations and the cytoplasmic loop between TM6 and TM7, which is characteristic of MFST proteins ( Fig. 4C) (42).
We previously identified the Tb927.9.6360-80 locus as a key contributor to suramin efficacy against T. brucei, with RNAi depletion of the three transcripts leading to a 10-fold reduction in parasite sensitivity to suramin; localization studies also indicated that at least one of these transporters is lysosomal (24). Deletion of the whole locus ( Fig.  4D) revealed that the three encoded proteins are collectively dispensable in cultured BSF T. brucei (Fig. 4E) and enabled us to confirm that these proteins influence not only suramin efficacy (Fig. 4F) but also those of paromomycin (Fig. 4G) and the related aminoglycoside neomycin (Fig. 4H). While loss of these MFST proteins dramatically reduces suramin efficacy, the effect on paromomycin and neomycin sensitivity is less pronounced (1.5-and 2.8-fold EC 50 increases, respectively) although significant. Our mutant BSF T. brucei parasites also exhibited better tolerance than wild-type parasites to the aminoglycosides at concentrations equivalent to or greater than the EC 99 during the first 24 h of exposure (Fig. S4).
TbVAMP7B, a cross-efficacy determinant for amphotericin B and miltefosine. To identify antileishmanial cross-efficacy determinants, we next used pairwise comparisons of RNAi library screen outputs (Fig. 5). We first identified a small cohort of hits represented by at least two RNAi target fragments and Ͼ99 reads/kb/transcript in more than one screen. This group included the AQP2-3 locus, represented by at least 100 reads in all four screens. We did not explore this locus further since the read count was at least 3 orders of magnitude lower in each screen than in the SSG screen, and leishmanial AQPs have not been implicated in resistance to the other drugs (see above). Two other loci fulfilled our stringency criteria, and both were enriched following  (Table S1); further analysis of the former hit is considered in this section, while the contribution of Tb927.11.3350 to drug action is addressed subsequently.
RIT-seq analysis revealed that 2.2% and 97% of mapped reads identified Tb927.5.3550-70 in the amphotericin B and miltefosine screens, respectively (Fig. 6A). This locus encodes a thioredoxin-like protein (Tb927.5.3550); a vesicle-associated membrane protein, TbVAMP7B (Tb927.5.3560) (43); and a hypothetical protein (Tb927.5.3570). Analysis of the RNAi target fragments mapping to Tb927.5.3550-70 revealed that few uniquely targeted the TbVAMP7B coding sequence (Fig. 6A). Instead, the RNAi target fragments that mapped to the flanking genes overlapped either the TbVAMP7B coding sequence (Tb927.5.3550 RNAi target fragments) or the 3= untranslated region (Tb927.5.3570 RNAi target fragments). This pattern is consistent with the poor tolerance of TbVAMP7B depletion. Our previous high-throughput phenotypic analysis indicated that TbVAMP7B RNAi knockdown is associated with a significant loss of fitness, while depletion of the flanking transcripts had a less dramatic effect (Table S1) (44). Taken together, these data suggested that TbVAMP7B is an amphotericin B-miltefosine cross-efficacy determinant, while the identification of the flanking genes was due to bystander effects.
To test this hypothesis, we generated stem-loop RNAi BSF T. brucei cell lines targeting TbVAMP7B and Tb927.5.3570. As predicted, depletion of Tb927.5.3570 had no effect on growth or sensitivity to amphotericin B or miltefosine (Fig. S5). In contrast, knockdown of TbVAMP7B following induction with tetracycline at 2 ng or 1 g · ml Ϫ1 resulted in a significant growth defect (Fig. 6B). To assess the contribution of TbVAMP7B to drug efficacy, we induced RNAi with 2 ng · ml Ϫ1 tetracycline for 24 h and assessed drug sensitivity over a further 30 h under inducing conditions. Incubation with lowconcentration tetracycline and a shorter EC 50 analysis (as opposed to the standard 72-h protocol) ensured that the growth defect due to TbVAMP7B RNAi knockdown was minimized while still allowing us to test the protein's contribution to drug action.
Unexpectedly, RNAi knockdown of TbVAMP7B reduced the amphotericin B EC 50 , by 24% (Fig. 6C). However, TbVAMP7B depletion also resulted in a significant decrease in the Hill coefficient. Consequently, while the EC 50 decreased upon TbVAMP7B depletion, the EC 90 and EC 99 increased 1.45-fold and 3-fold, respectively (Fig. 6D); the EC 25 decreased by 44%, consistent with the effect on the EC 50 and the change in the Hill coefficient. Therefore, small changes in TbVAMP7B expression can lead to a significant loss of sensitivity to high-concentration amphotericin B while enhancing sensitivity to the drug at a low concentration. This relative resistance to high-concentration ampho- tericin B explains the enrichment of TbVAMP7B-targeting RNAi fragments following selection of the RNAi library at 1.5ϫ EC 50 . In contrast, miltefosine at relatively low concentrations complemented the TbVAMP7B RNAi growth defect and further increased growth at lower concentrations ( Fig. 6E and F).
Our findings indicate specific interactions between TbVAMP7B and both amphotericin B and miltefosine. VAMP7 proteins are involved in endosome and lysosome membrane fusion (45), and it is notable in this respect that amphotericin B disrupts membranes and that miltefosine is a phospholipid drug. TbVAMP7B depletion does not significantly increase the EC 50 for either drug, but nevertheless, these interactions may be important in a clinical setting where exposure will be variable in different tissues and at different times following dosing.
Multiple hits link amphotericin B action to phospholipid transport and metabolism. Our amphotericin B screen yielded 13 high-confidence hits, for which Gene Ontology (GO) term profiling revealed links to membranes and lipids (Table S2 and The inset shows confirmation of RNAi knockdown by Northern blotting following 24-h induction with 1 g · ml Ϫ1 tetracycline; an ethidium bromide-stained gel is shown as a loading control. (C) Representative data from a 30-h amphotericin B EC 50 assay following TbVAMP7B RNAi knockdown induced with 2 ng · ml Ϫ1 tetracycline. The inset chart summarizes Hill coefficient data for five biological replicates. (D) Effect of TbVAMP7B RNAi knockdown on EC X for five biological replicates. Data for each replicate were derived from EC 50 values and Hill coefficients presented in panel C. (E) Representative data from a 30-h miltefosine EC 50 assay following TbVAMP7B RNAi knockdown induced with 2 ng · ml Ϫ1 tetracycline. Data are plotted to show population growth relative to untreated T. brucei (uninduced or induced). The dashed ellipse highlights miltefosine-mediated complementation of the Tb927.5.3560 RNAi growth defect. (F) Chart summarizing T. brucei population growth in the presence or absence of TbVAMP7B RNAi at a subset of miltefosine concentrations from five independent biological replicates. Individual growth (B) and EC 50 (C and E) assays were carried out in triplicate and quadruplicate, respectively. Error bars represent standard deviations. P values were derived from paired Student's t test (**, P Ͻ 0.01; ***, P Ͻ 0.001). Fig. S6). This is consistent with disruption of membranes by amphotericin B. Miltefosine uptake in Leishmania is dependent on a flippase (17,18), which also contributes to the antileishmanial action of amphotericin B (46). RNAi fragments targeting the syntenic locus in T. brucei, Tb927.11.3350, were enriched following selection with amphotericin B and miltefosine ( Fig. 5 and Fig. 7A). Depletion of Tb927.11.3350, while having no effect on parasite growth in culture (Fig. 7B), led to a reproducible increase in amphotericin B and miltefosine EC 50 s ( Fig. 7C and D). RNAi knockdown also significantly enhanced short-term survival in high-concentration amphotericin B and miltefosine (Fig. S6). Therefore, as in Leishmania, the T. brucei miltefosine transporter orthologue contributes to the action of miltefosine and amphotericin B.
In addition to the Leishmania miltefosine transporter orthologue, Tb927.11.3350, RNAi fragments targeting the flippases Tb927.11.13000 and Tb927.6.3550 and the ␤-subunit Tb927.11.13200 were enriched following selection with amphotericin B, with Tb927.11.13000 represented by 78% of mapped reads (Fig. 8B and Table S1). Targeted RNAi depletion of Tb927.11.13000 led to a mild growth defect (Fig. 8C) and a Ͼ2-fold EC 50 increase, validating this protein as an amphotericin B efficacy determinant in T. brucei (Fig. 8D). The impact of Tb927.11.13000 depletion was most  11.3350. The inset shows confirmation of RNAi knockdown by Northern blotting; an ethidium bromide-stained gel is shown as a loading control. (C and D) Representative data from amphotericin B and miltefosine EC 50 assays following RNAi knockdown of Tb927.11.3350. Inset charts summarize data from three independent biological replicates. Individual growth (B) and EC 50 (C and D) assays were carried out in triplicate and quadruplicate, respectively. Error bars represent standard deviations. P values were derived from Student's t test (*, P Ͻ 0.05; **, P Ͻ 0.01). RNAi inductions were carried out with 1 g · ml Ϫ1 tetracycline. pronounced during the initial 24 h of drug exposure, enabling the parasite population to increase approximately 1.3-fold and 4-fold over 8 and 24 h, respectively, in the presence of 0.7 M (ϾEC 99 ) amphotericin B (Fig. S6). The uninduced population declined by more than 40% and 60% over the same periods. In addition, while exposure to 1.8 M (ϾEC 99.9 ) amphotericin B led to an 80% decline in the induced population over 24 h, cultures of uninduced cells were cleared within 4 h of exposure to this drug concentration (Fig. S6). Depletion of this putative phospholipid-transporting ATPase had no effect on miltefosine efficacy (Fig. 8E), confirming its specific contribution to amphotericin B action.
Our results reveal that multiple T. brucei flippases drive the efficacy of amphotericin B, all of which have syntenic orthologues in Leishmania (Fig. 8A). Therefore, in addition to the well-characterized miltefosine-transporting flippase, other Leishmania flippases may play significant, and potentially specific, roles in the antileishmanial action of amphotericin B and miltefosine.  11.13000. The inset shows confirmation of RNAi knockdown by Northern blotting; an ethidium bromide-stained gel is shown as a loading control. (D and E) Representative data from amphotericin B and miltefosine EC 50 assays following RNAi knockdown of Tb927.11.13000. Inset charts summarize data from three independent biological replicates. Individual growth (C) and EC 50 (D and E) assays were carried out in triplicate and quadruplicate, respectively. Error bars represent standard deviations. P values were derived from Student's t test (*, P Ͻ 0.05; **, P Ͻ 0.01). RNAi inductions were carried out with 1 g · ml Ϫ1 tetracycline, unless otherwise stated.

DISCUSSION
In the current absence of an effective genome-scale loss-of-function screen in Leishmania, we speculated that selection of a T. brucei RNAi library would provide insights into antileishmanial drug action while also revealing novel T. brucei biology. By selecting our genome-scale BSF T. brucei RNAi library with the current antileishmanial drugs followed by RIT-seq analysis, we identified a panel of putative antileishmanial drug efficacy determinants (see Table S1 and Fig. S1 in the supplemental material). SSG and miltefosine selection identified TbAQP3, an orthologue of the known Sb(III) transporter, and Tb927.11.3350, the T. brucei orthologue of the Leishmania miltefosine transporter (MT), respectively, confirming the power of this approach. In addition to these known drug transporters, we validated several novel drug efficacy determinants identified by our selective screens: Tb927.9.6360-80 (paromomycin), Tb927.5.3560 (miltefosine and amphotericin B), and Tb927.11.13000 (amphotericin B). Our results highlight the role of a lysosomal transporter in paromomycin efficacy, emphasize the importance of membrane composition for the action of amphotericin B and miltefosine, provide insight into the substrate selectivity of the trypanosomatid aquaglyceroporins, and present several new candidate antileishmanial drug efficacy determinants (Fig. 9).
SSG contains Sb(V), which is not reduced in Leishmania medium, limiting its efficacy against the extracellular promastigote stage (48). However, once it enters the host The key T. brucei proteins identified in our antileishmanial loss-of-function screen (left) and their Leishmania orthologues (right) represent candidate antileishmanial drug efficacy determinants. Red denotes known Leishmania drivers of antileishmanial efficacy whose loss of function reduces drug efficacy (see the text for details). The strain prefix for the truncated gene identifications is at the top of each panel, with the exception of the sterol biosynthetic enzymes recently shown to contribute to amphotericin B efficacy against L. mexicana (73). Gray-filled circles (endosomes) and ellipses (lysosome) represent the endocytic system. The purple blocks represent membranes modified by changes in sterol biosynthesis and the putative action of the flippases and their ␤-subunit; changes in membrane composition anywhere in the endocytic system may influence the intracellular transit of amphotericin B or its ability to form ionpermeable channels. Tb927.11.3350 has an intracellular localization in procyclic-stage T. brucei; however, in the current absence of data from BSF T. brucei, we speculate that this protein localizes to the plasma membrane in BSF T. brucei, as per its Leishmania orthologue. macrophage, it is thought to be reduced to the toxic Sb(III) form, which can be taken up by intracellular Leishmania amastigotes via AQP1 (19,49). In contrast, we speculate that Sb(V) is reduced to Sb(III) in T. brucei HMI9 medium due to the presence of supplementary L-cysteine (50). Thus, in common with intracellular Leishmania amastigotes, cultured extracellular T. brucei parasites treated with SSG are exposed to toxic Sb(III). T. brucei RNAi library selection with SSG and our subsequent validation experiments identified a single efficacy determinant, TbAQP3. Aquaglyceroporins are ubiquitous transporters of water, glycerol, and other small solutes, whose specificity is defined by their selectivity filter residues. Leishmania AQP1 and the T. brucei proteins TbAQP1 and TbAQP3 have the same selectivity filter, NPA/NPA/WGYR, while TbAQP2 possesses a divergent filter, NSA/NPS/IVLL (39). TbAQP2 is a key drug transporter in T. brucei, mediating the uptake of pentamidine and melarsoprol, and its loss contributes to clinical drug resistance (35)(36)(37). In addition, TbAQP2 plays an important role in glycerol transport, as its loss increases parasite sensitivity to alternative oxidase inhibition, which leads to elevated intracellular glycerol levels (51). The in vivo roles of the other T. brucei aquaglyceroporins remain unknown, although all three are capable of arsenite and antimonite transport in yeast and Xenopus laevis heterologous expression systems (52). In contrast, our data demonstrate that in T. brucei, these transporters are selective for arsenic-containing melarsoprol (TbAQP2 [35]) and Sb(III) (TbAQP3). Intriguingly, RNAi library selection with SSG failed to identify TbAQP1, even though it contains the same selectivity filter as TbAQP3. This suggests important functional and regulatory differences between TbAQP1 and TbAQP3, which may influence their ability to contribute to Sb(III) uptake in bloodstream-form T. brucei. For example, TbAQP3 is localized to the plasma membrane in bloodstream-form T. brucei, and TbAQP1 localizes to the flagellar membrane (35,53). This differential localization may influence their ability to mediate antimonial uptake.
The aminoglycoside paromomycin is thought to inhibit protein synthesis in Leishmania and enters the cell via endocytosis (21,54,55). However, RNAi library selection did not identify a surface receptor, suggesting that, at least in T. brucei, paromomycin entry is not dependent on a specific ligand-receptor interaction. Rather, the high endocytic flux associated with variant surface glycoprotein (VSG) internalization (56) may drive drug uptake. In addition, our screen did not identify proteins with identifiable roles in translational regulation. This perhaps surprising observation may be due to the relative essentiality of such proteins, whose loss, while capable of affecting paromomycin action, may also cause a substantial loss of fitness. Instead, RNAi fragments targeting Tb927.9.6360-80 dominated the paromomycin-selected RNAi library, with the remaining 28 high-confidence hits constituting only 9% of mapped reads. This locus encodes a set of closely related MFST proteins, at least one of which localizes to the lysosome, and has previously been associated with suramin efficacy (24). In contrast to paromomycin, several other endocytic pathway proteins, including three lysosomal proteins (p67, cathepsin L, and the MFST proteins), influence suramin efficacy (24). This led to the proposal that proteolytic processing in the lysosome releases suramin from bound proteins, enabling neutralization in the acidic environment or association with an alternative endogenous carrier and escape to the cytoplasm via one or more of the lysosomal MFSTs (57). In contrast, the absence of hits targeting other endocytic components following paromomycin RNAi library selection suggests little reliance on the endocytic network per se. Therefore, the lysosomal MFST proteins may influence paromomycin efficacy indirectly. MFST proteins mediate the transit of a diverse range of molecules, including polyamines and amino acids (41), and changes in the intracellular flux of these molecules may affect translation efficiency, which in turn may influence paromomycin efficacy. Deletion of the Tb927.9.6360-80 locus from T. brucei yields only a 2-fold increase in the paromomycin EC 50 . However, the MFST protein encoded by the syntenic single-copy gene in Leishmania (e.g., LmjF. 15.0870) remains to be characterized and may make a more substantial contribution to paromomycin action against this parasite.
Combination therapies are increasingly being used to treat leishmaniasis, enabling reduced dosing and treatment duration, resulting in fewer side effects (8). For example, a single dose of liposomal amphotericin B in combination with a short course of oral miltefosine or intramuscular paromomycin is an effective treatment for visceral leishmaniasis (VL) in the Indian subcontinent (58). In East Africa, SSG-paromomycin combination therapy is effective against VL (59). However, L. donovani parasites resistant to these and other antileishmanial drug combinations can be selected for in vitro (9,10), and oxidative defense upregulation and changes in membrane fluidity have been associated with cross-resistance in laboratory-derived lines (23). Therefore, we carried out pairwise comparisons of our RNAi library screen data to identify potential crossefficacy determinants. Only two hits fulfilled our stringency criteria, both of which influence amphotericin B and miltefosine action: TbVAMP7B, an endosomal SNARE protein responsible for endosome-lysosome fusion in other eukaryotes (45,60), and Tb927.11.3350, the T. brucei orthologue of the Leishmania miltefosine transporter (17). However, while both of these proteins may influence membrane fluidity (see below), it seems unlikely that either one contributes significantly to oxidative defense. Recent Cos-seq gain-of-function analyses in L. infantum identified several candidate proteins whose overexpression reduces sensitivity to multidrug exposure (26); these proteins also lack an obvious connection to oxidative defense. Therefore, rather than being dependent on the increase or decrease in the expression of a single protein, changes in oxidative defense that lead to antileishmanial resistance are likely to be multifactorial. Our findings also suggest that amphotericin B-miltefosine combination therapy is most vulnerable to loss-of-function mutation, while others may be less susceptible to the downregulation of a single protein. This finding is particularly significant given that recent trials have confirmed the efficacy of amphotericin B-miltefosine combination therapy in treating VL (61,62). In contrast to the other antileishmanial drug efficacy determinants described here, TbVAMP7B depletion does not simply increase the drugs' EC 50 s. Instead, TbVAMP7B RNAi knockdown reduces the amphotericin B EC 50 and has little effect on the miltefosine EC 50 . The drop in the amphotericin B EC 50 is due to a substantial decrease in the amphotericin B Hill coefficient, which has the opposite effect on EC 90 and EC 99 , increasing both and enabling TbVAMP7B-depleted parasites to persist at these drug concentrations. Our data show that T. brucei has limited tolerance for TbVAMP7B depletion, presumably due to impairment of endosome-lysosome fusion (45). Intriguingly, exposure to low-concentration miltefosine complements the growth defect seen following TbVAMP7B depletion, suggesting that miltefosine treatment is able to promote vesicle membrane fusion in the endocytic system, a possible consequence of the enhanced membrane fluidity seen upon miltefosine exposure (63). TbVAMP7B has also recently been identified as a putative T. brucei apolipoprotein L1 (apoL1) sensitivity determinant (64), and other workers have highlighted the importance of the intracellular transit of the apoL1-carrying membrane for trypanolysis (65,66). Our findings suggest that such transit also contributes to amphotericin B and miltefosine action. The VAMP7 proteins are highly conserved between T. brucei and Leishmania (43), suggesting that Leishmania parasites will also be sensitive to VAMP7B loss (LmjF.08.0030). However, subtle changes in VAMP7B expression that can be tolerated may enable parasites to take advantage of variations in amphotericin B and miltefosine tissue penetrance.
Miltefosine uptake in Leishmania is dependent on a phospholipid-transporting flippase (the MT) and its ␤-subunit, Ros3 (17,18); both in vitro-selected lines and miltefosine-resistant L. donovani clinical isolates harbor mutations in the MT (7,67,68). Consistent with this, T. brucei RNAi library selection with miltefosine led to enrichment for RNAi fragments mapping to the syntenic sequence in T. brucei (Tb927.11.3350). RNAi library selection with amphotericin B also enriched for RNAi fragments mapping to this gene, consistent with recent findings in Leishmania (46), as well as two other flippases and a putative ␤-subunit (Tb927.11.13200). Interestingly, the ␤-subunit targeted was not the syntenic orthologue of Ros3, previously shown to interact with the MT (18). Therefore, different flippase/␤-subunit dependencies may have evolved following the divergence of the Leishmania and T. brucei lineages. A further difference in the behaviors of these proteins between Leishmania and T. brucei lies in their localization. The MT and Ros3 localize to the plasma membrane in Leishmania (18), whereas in procyclic-form T. brucei, the MT orthologue (Tb927.11.3350) and a second flippase (Tb927.11.13000) localize to an intracellular structure reminiscent of the endosomal system (www.TrypTag.org) (69); their localization in BSF T. brucei is unknown. Therefore, while flippases influence drug action against Leishmania and T. brucei, they may mediate drug and/or phospholipid transit across different membranes in each parasite.
Phospholipid transport by flippases maintains the membrane asymmetry necessary for membrane fusion, vesicle trafficking, and sterol homeostasis (47). The identification of a single flippase following miltefosine selection is consistent with its role as a drug transporter (17). In contrast, amphotericin B selection identified three flippases, suggesting an indirect role in drug action, possibly through changes in membrane composition and transit through the endosomal system (Fig. 9). Amphotericin B acts by binding membrane ergosterol (70), leading to the formation of ion-permeable channels and downstream oxidative damage (71). Consistent with the importance of ergosterol to amphotericin B action, resistant clinical isolates exhibit elevated membrane fluidity and reduced ergosterol content (72). Recent findings have highlighted the mutation of key sterol biosynthetic enzymes, such as CYP51, and reduced ergosterol production as drivers of resistance in laboratory-derived amphotericin B-resistant Leishmania mexicana (73,74). In contrast, amphotericin B selection of the BSF T. brucei RNAi library did not identify TbCYP51 (Tb927.11.6210) or any other recognizable component of the ergosterol biosynthetic pathway, suggesting that limiting ergosterol production in BSF T. brucei is unable to reduce amphotericin B efficacy. However, our data indicate that changes in flippase expression may provide an alternative route to amphotericin B resistance. We speculate that reduced flippase activity may lead to changes in membrane ergosterol content or accessibility, thereby reducing the efficiency of amphotericin B binding and uptake. Therefore, functional characterization of the syntenic Leishmania orthologues of the T. brucei flippases may provide additional insights into the processes that drive the antileishmanial action of amphotericin B.
In summary, using our genome-scale BSF T. brucei RNAi library, we have identified a panel of putative antileishmanial drug efficacy determinants, highlighting two candidate cross-efficacy determinants, as well as roles for multiple flippases in the action of amphotericin B. The findings from this orthology-based chemogenomic profiling approach substantially advance our understanding of antileishmanial drug mode of action and potential resistance mechanisms and should facilitate the development of improved therapies as well as surveillance strategies to identify drug-resistant parasites.
Drug sensitivity assays. InvivoGen) were determined over 78 or 30 h. BSF T. brucei parasites were seeded at 2 ϫ 10 3 cells · ml Ϫ1 (or 2 ϫ 10 5 cells · ml Ϫ1 ) in 96-well plates with a 2-fold dilution series of each drug; assays were carried out in the absence of other antibiotics. After 72 or 24 h, resazurin (Sigma) in phosphate-buffered saline (PBS) was added to a final concentration of 12.5 g · ml Ϫ1 per well, and the plates were incubated for a further 6 h at 37°C. Fluorescence was determined using a fluorescence plate reader (Molecular Devices) at an excitation wavelength of 530 nm, an emission wavelength of 585 nm, and a filter cutoff of 570 nm (78). Data were processed in Microsoft Excel, and nonlinear regression analysis was carried out with GraphPad Prism. The short-term kinetics of killing at high concentrations of the drug (ϾEC 99 ) were determined in triplicate over 24 h from a starting cell density of 1 ϫ 10 5 cells · ml Ϫ1 .
T. brucei RNAi library screening and RIT-seq. RNA library screening was carried out as previously described (34). Briefly, library expression was induced with 1 g · ml Ϫ1 tetracycline (Sigma) for 24 h prior to selection with each antileishmanial drug at 1ϫ to 3ϫ EC 50 . Cell density was assessed daily using a hemocytometer and diluted to no less than 20 million cells in 100 ml medium; induction and antileishmanial drug selection were maintained throughout. Once robust growth had been achieved for at least 2 days, genomic DNA was prepared for RNAi target identification. The RNAi cassettes remaining in the antileishmanial-selected RNAi libraries were amplified from genomic DNA using the primer pair LIB2F/ LIB2R and sequenced on an Illumina HiSeq platform at the Beijing Genome Institute.
The sequenced RNAi target fragments were mapped against the T. brucei strain TREU927 reference genome (release 6.0), as described previously (34). Briefly, mapping was carried out using Bowtie2 (79) set to "very sensitive local" alignment, and output SAM files were processed using SAMtools (80). The resultant BAM files were viewed against the reference genome in the Artemis genome browser (81). Reads containing the RNAi construct-specific 14-base barcode were identified using a custom script (34) and corresponded to at least 22% of reads from each selected RNAi library. This subset of reads was mapped against the TREU927 reference genome, as described above. Plots were generated using the Artemis graph tool and processed in Adobe Photoshop Elements 8.0. Stacks of reads that included the 14-base barcode on the positive strand were used to define RNAi target fragment junctions and to assign high-confidence hits as those identified by at least two RNAi target fragments. RNAi target fragment read numbers were converted to RPKM (reads per kilobase per million reads mapped) to account for interlibrary read depth variations when comparing RNAi library sequencing outputs.

SUPPLEMENTAL MATERIAL
Supplemental material for this article may be found at https://doi.