All the analyses were performed using genomes available on TriTrypDB (release 46). Trypanosoma cruzi: CL Brener Esmeraldo-like, T. brucei: DAL972, and L. major: Friedlin.

Experimental structures for TriTryps were obtained from the Protein Data Bank (PDB). For all the remaining proteins, we attempted to predict their structure by homology modeling as described in (Sosa et al., 2018). Structural models, template information and validation of the models are available on Target Pathogen ⁴ database.

The structural druggability of T. cruzi, T. brucei and L. major proteins was assessed using the fpocket program (Schmidtke et al., 2010) and Druggability Score (DS) index (Schmidtke and Barril, 2010; Schmidtke et al., 2010) as defined in (Sosa et al., 2018). This assessment was performed both on crystals and on homology-based models.

All the proteins of TriTryps were blasted against the human proteome (NCBI assembly access GCF_000001405.36) to identify close human homologous. Hits with an identity greater than 40% and E-value less than 10^–5 were ruled out, as they may share a high degree of structural preservation that could produce side effects if the parasite protein is used as target of a putative drug.

The PathoLogic module within Pathway Tools v. 20.0 (Karp et al., 2002) environment was used to build each metabolic network using the respective annotated genome as input. The metabolic reconstruction included the determination of reaction-protein-gene associations, which are primarily based on the corresponding enzyme commission (EC) number. EC number annotations were previously added to the genomic annotation. After manual curation of the metabolic networks, choke-point (CP) analysis was conducted within Pathway Tools. Choke-point reactions (CPs) are those that either uniquely produce or consume a given product or substrate, respectively (Yeh et al., 2004). In this sense, it is assumed that CP blockade may lead to the lack of an essential compound or the accumulation of a toxic metabolite in the cell; thus, these types of reactions have great significance in drug targeting.

All previously calculated data was integrated in Target Pathogen. Target Pathogen is our own web server developed for drug target prioritization. This webserver was previously used to obtain attractive targets for drug development projects in other relevant pathogens (Defelipe et al., 2016; Ramos et al., 2018; Farfán-López et al., 2020; Palumbo et al., 2022; Serral et al., 2022). The integrated data was used to obtain a list of drug target candidates. In this sense, proteins of the studied TriTryps were filtered and ranked using the Target-Pathogen database. At first, proteins with DS < 0.5 (non-druggable or poorly druggable proteins) and proteins with cross-reaction potential with the human host were filtered out. We then defined a scoring function as follows to assign a score to each protein. This equation assigns a score that defines a protein’s potential as a drug target based on its druggability, human off-target, and metabolic context (CPs).

Scores are assigned using the following function: S = 1 ∗ D S + 1 ∗ H O T + 5 ∗ C P where DS is the Druggability Score, HOT is the Human Off-Target Score and CP defines whether the protein is associated with a CP reaction. HOT reflects the results of a blastp search of the pathogen protein in the human proteome database with the scale: 1-identity of the best hit of the queried protein and the human proteome. DS and HOT can take values between 0 and 1, while CP is equal to 1 if the protein is involved in a CP reaction, otherwise is set to zero. We have given the CP parameter a weight of 5, so that the proteins involved in CPs (usually essential proteins for the organism metabolism) lead the ranking.

For clustering proteins among these 3 parasites, we used the MMSeqs2 software (Steinegger and Söding, 2017). A minimum coverage of 70%, a sequence identity greater than 50% and an E-value less than 0.001 were used as cut-off points to determine a significant hit. Alignment method number 3 was used, as it is described as the most accurate approach, and clustering mode 0 which uses the Greedy Set Cover algorithm was selected.

Once the clusters were obtained, it was observed that many of them contained more than one paralog gene per parasite. Common clusters were reduced to a single gene from each of the parasites, looking for candidates with similar DS and HOT. To carry out this cluster reduction, genes with DS < 0.5 and HOT < 0.7 were first removed. Then, DS standard deviation (stdDS) and HOT standard deviation (stdHOT) were calculated for all combinations involving a single gene from each of the parasites. Finally, the combinations with stdDS > 0.1 and stdHOT > 0.1 were filtered out, obtaining triplets with similar DS and HOT in each of the clusters. As result, each cluster is composed of similar proteins from each of the parasites with DS ≥ 0.5 and HOT ≥ 0.7. Finally, if a cluster has two or more combinations that pass these filters, the one with the highest average DS was kept.

The defined clusters were ranked based on the S-score obtained for each gene in each of the parasites, that was calculated as described in the “Target prioritization pipeline” section. Using the S-score of each of the three genes belonging to a cluster, the mean and the standard deviation were calculated. With these values, a ranking of clusters was determined using the mean in descending order as first factor and the standard deviation in ascending order as a second factor.

Drugs affect the parasites during stages in the mammalian host, therefore it is important to analyze candidates’ expression at these life cycle stages. For T. cruzi, they are the trypomastigote and amastigote stages, for T. brucei the blood trypomastigote stage, and for L. major the metacyclic promastigote and amastigote stages. The gene expression analysis was performed using data from Li et al. (Li et al., 2016) for T. cruzi, data from Naguleswaran et al. (Naguleswaran et al., 2018) for T. brucei and data from Inbar et al. (Inbar et al., 2017) for L. major. Candidate mRNA levels were evaluated at the trypomastigote and amastigote (48 h post-infection) stages for T. cruzi, at the slender and stumpy forms of blood trypomastigote stage for T. brucei and at the metacyclic promastigote and amastigote stages for L. major. We compared the expression of our candidates with the total mRNAs for each stage of the life cycle.

For L. major raw reads were downloaded, trimmed and Kallisto version 0.46.1 (Bray et al., 2016) was used to obtain expression estimates for the reference transcriptome (TriTrypDB v.6.1 L. major Friedlin) (Aslett et al., 2010). Read pseudo counts were normalized by sequencing depth for each replicate. For T. cruzi and T. brucei analysis counts tables were available and used in this work. Log2 transformation and quantile normalization were applied to all counts tables. We considered a gene to be expressed if this value is greater than 0.

Once the candidate list was obtained, functional analyses were performed to better understand the roles of these proteins in the parasite’s biology. We first evaluated the Gene Ontology (GO) terms overrepresentation, using the trypanosomatid database TritrypDB (Aslett et al., 2010). The analysis was performed for Molecular Function, Biological Process and Cellular Component ontology terms with default parameters, reporting both curated and computed terms, and using an FDR cut-off of 0.05 to determine enrichment.

To better understand the biological potential of the candidates as putative drug targets, a literature search was performed, aided by IdMiner ⁵ software. It allows linking candidate identifiers to published articles and search for overrepresented terms. The search term “drug” was used to find an association between our candidates and drug research and development articles.

DS of each protein pocket were calculated for all protein structures we were able to obtain for T. cruzi, T. brucei, and L. major. DS was then assigned to 4592, 4197, and 6656 proteins from T. cruzi, T. brucei and L. major respectively ( Table 1). The difference in the number of proteins assigned a DS for L. major is striking and possibly reflects that both trypanosomes are more phylogenetically related, but this observation will require further analysis. Figure 1A shows the distribution of DS for the three parasites as well as for the RCSB Protein Data Bank (PDB), with a clear enrichment in highly druggable proteins for all the TriTryps.

Given that trypanosomatids are ancient eukaryotes, the observed similarity of their proteins to their human counterparts is low (Figure 1B) (Burki et al., 2020). However, this analysis allowed us to exclude highly conserved proteins that are not suitable candidates for drug targeting candidates as they are more likely to exhibit off-target effects.

Metabolic network reconstruction was performed with Pathway Tools followed by manual curation. TriTryps networks were analyzed allowing the determination of CPs. For T. cruzi 575 proteins were assigned to reactions distributed across 98 metabolic pathways. A total of 466 proteins participated in CPs, of which 70 were associated with producing CPs, 374 with strictly consuming ones and 22 were mapped with both types. Trypanosoma brucei metabolic network is composed of 234 predicted pathways. From a total of 386 proteins involved in reactions, 156 were annotated as CPs. 57 were classified as production CPs, 58 as consuming ones and 41 on both producing and consuming sides.

Finally, L. major metabolic model resulted in 142 pathways, composed of 307 proteins assigned to reactions, with 135 classified as CPs. A total of 45 and 62 were annotated as producing and consuming CPs, respectively. Additionally, 28 proteins were classified as CPs on both types.

The shared biological characteristics of TriTryps give us the opportunity to search for common druggable proteins, enabling the design of drugs that might be effective against more than one species. Similarity clustering analysis resulted in 3,333 protein groups common to T. cruzi, T. brucei and L. major (Figure 2A).

By using the tools included in Target Pathogen database, we combined all the previous results and filtered the candidates with DS ≥ 0.5 and HOT ≥ 0.7, obtaining a list of 319 protein clusters with interesting features such as druggable pockets and low sequence similarity with human proteins (Figure 2B). Of these, 82 are annotated as hypothetical proteins in all three TriTryps and 139 have this annotation in at least one of them. The rest of the clusters are functionally annotated for the three parasites.

Since drugs should be effective against the parasite forms present in the mammalian host, we analyzed the gene expression patterns of the candidate proteins at these stages. We observed that the mRNAs coding for these proteins are expressed in all analyzed forms for the three parasites (Figure 3), excepting LmjF.01.0140 in both analyzed L. major stages, LmjF.34.0190 in L. major amastigote stage and Tbg.972.2.3880, Tbg972.8.3790 for both T. brucei stages. These expression values in the mammalian stages make them interesting drug target candidates.

The GO analysis does not report any enrichment either for Molecular Function or for Biological Process terms. For the Cellular Components ontology, only mitochondrion (GO:0005739, FDR <0.05) and kinetoplast (GO:0020023, FDR <0.05) were reported as enriched in our list. This is an interesting observation as proteins involved in the mitochondrial oxidative stress response are considered putative drug targets (see discussion below). Proteins related to the kinetoplast are also reasonable candidates given the uniqueness and relevance of this structure in the biology of the TriTryps. Manual inspection of the list revealed two proteins with relevant features that are discussed in detail below (TcCLB.510,295.30, Tbg972.10.6190, LmjF.36.0700); (TcCLB.506,287.200, Tbg972.7.500, LmjF.26.1340).

No functional relationships are expected to be a priori in the list, so, as expected, no other evident functional relationship is observed among the candidates. This prompted us to manually inspect the functions of interesting candidates on our list and discuss their relevance as potential drug targets (Supplementary Table S1 ). As can be observed, their functions are varied but they are involved in key biological aspects of parasite biology.