Background

The single-celled flagellated eukaryotes of the group kinetoplastids include notorious human pathogens such as Trypanosoma and Leishmania. Mitochondrial (mt) genomes of numerous trypanosomatids have been sequenced, with complete and nearly complete mtDNA sequences available for five species: Leishmania tarentolae (GenBank Accession No: NC_000894), Trypanosoma brucei (M94286), T. cruzi (DQ343645), Crithidia oncopelti (X56015), Leptomonas seymouri (DQ239758), and major portions of mtDNA for two other members of the group: Leishmania major (AH015294), Leptomonas collosoma (AH015822). For a review, see [1].

The unassigned open reading frame (ORF) murf1 in T. brucei mtDNA has been known for 25 years, but until today, there is no protein of known function that shares significant sequence similarity with this ORF [2]. In a recent study, we conducted a comprehensive function prediction of all hypothetical mitochondrion-encoded proteins using a machine-learning-based classifier MOPS [3]. This classifier does not rely on sequence similarity but rather on a host of other features including physico-chemical properties of proteins, and hence should be able to detect remote homologs. MOPS predicted, but only with moderate support, MURF1 of the kinetoplastid Phytomonas serpens as subunit 2 (NAD2) of the NADH-Ubiquinone Oxidoreductase (NADHdh) or Complex I of the electron transport chain – a multi-complex pathway embedded in the inner mitochondrial membrane. NADHdh is the largest complex of this pathway with ~45 distinct subunits, seven of which are usually encoded in the mitochondria. We chose to scrutinize this function assignment in detail, motivated by several reasons: the long-standing controversy surrounding MURF1, the large available body of related biological knowledge, and the significance of this organismal group for human health [2, 4–6].

Results

As mentioned in the Background, the hypothetical protein MURF1 was predicted by the automated similarity-free classifier MOPS to be a divergent NADHdh subunit 2 (NAD2). To test this prediction, we conducted the following analyses.

Sequence – Sequence Comparison

BLAST searches of Phytomonas MURF1 sequence against NRDB or UniProt did not result in any informative hits, but identified all the MURF1 homologs from other kinetoplastids such as T. brucei, L. tarentolae, etc. In contrast, FASTA searches against UniProt returned, after MURF1 homologs, NADHdh subunit 5 from the kinetoplastid Crithidia as top informative hit with an e-value of 6.5e-09, followed by NAD2 from the red alga Chondrus crispus with an e-value of 8.8e-07. A list of all hits and their corresponding e-values is compiled in Table 1.

Table 1 List of FASTA hits for P. serpens MURF1 searched against UniProt
Full size table

Profile – Sequence Comparison

For the identification of distantly related sequences, methods that exploit profiles (i.e., position-specific descriptions of the consensus of a multiple sequence alignment) are more sensitive than those based on pairwise alignment such as BLAST and FASTA. Here, we used PSI-BLAST to generate a MURF1 profile and searched it against NRDB, but no other proteins beyond kinetoplastid MURF1 sequences were found.

Profile HMM – Profile HMM Comparison

Our hypothesis is that MURF1 is a highly derived distant homolog of NAD2. We used Profile HMM – Profile HMM comparison because it is the most sensitive method in identifying distant homologs. In contrast to simple sequence profiles, Profile Hidden Markov Models (HMMs) contain extra information about insertions/deletions and gap scores. HHsearch (the first implementation of this approach), was shown to outperform profile – sequence comparison methods such as PSI-BLAST and HMMER, profile – profile comparison tools such as PROF_SIM and COMPASS and the other HMM – HMM comparison tool PRC [7].

We built a profile HMM for MURF1 from the multiple alignment of several kinetoplastid MURF1 sequences. Using HHsearch, we searched this profile HMM against the profile HMMs available in Pfam, PANTHER, COG and TIGR. In most cases, the top hit was to the "NADH-Ubiquinone/plastoquinone (Complex I)" profile HMM, which was built from 12 distinct subunits of different function. Though these subunits are non-homologous proteins, Pfam puts them all together in to a single family because they share high hydrophobicity (transmembrane domains). Only the search against the COG database returned a specific subunit as top hit, i.e., NAD2. HHsearch results are summarized in Table 2.

Table 2 Best informative hits for the MURF1 profile HMM when searched against profile HMMs from various databases
Full size table

To narrow down the exact function of MURF1, we generated profile HMMs for all 12 subunits of NADHdh. For that, we clustered the protein sequences of all NADHdh subunits at different identity thresholds ranging from 40% to 75%, constructed a multiple sequence alignment for each of the subunits at each threshold, and generated a total of 84 profile HMMs. We then searched the MURF1 profile HMM against all the profiles of NADHdh subunits. As expected for remote homologs, the scores are relatively low. The six top hits are NAD2 with an e-values ranging from 2.70e-15 to 1e-11. The e-value of the other subunit best hits is 4 orders of magnitude worse (Table 3).

Table 3 Best hits for the MURF1 profile HMM when searched against the profile HMMs of all NADH dehydrogenase subunits using HHsearch. The hits are ranked based on E-valuesa
Full size table

Discussion

While sequence – sequence comparison and profile HMM – profile HMM comparison point to MURF1 being a subunit of NADHdh, profile – profile comparison against the profile HMMs of individual subunits of NADHdh is able to clearly assign MURF1 to NAD2. In the following, we will confront this in silico prediction with the available biological knowledge. If the MURF1 protein of trypanosomes is indeed NAD2, then the following criteria must apply.

1. There should be no previously annotated nad 2 gene in either mitochondrial or nuclear genomes of kinetoplastids. A nad 2 gene has not been reported in any mitochondrial genome of kinetoplastids. Recently, the sequence of the nuclear genome became available for the P. serpens [4]. Neither genome nor EST data (2,190 ESTs) indicate the presence of this gene.

2. There should be numerous precedents for nad 2 being encoded by mtDNA. The nad 2 gene is mtDNA-encoded by the large majority of eukaryotes (see GOBASE, 'Gene Distribution' http://gobase.bcm.umontreal.ca/searches/compilations.php). The rare species that lack this mitochondrial gene also lack other NADH subunits (Apicomplexa, yeast).

3. The murf1 gene should be transcribed. Evidence for murf1 being expressed rather than being a spurious ORF is provided by several observations. First, the deduced amino acid sequence is conserved across trypanosomes, despite considerable divergence at the nucleotide level. Second, transcription of this gene has been demonstrated in P. serpens [5].

4. Rotenone-sensitive NADH dehydrogenase Complex I should be present in kinetoplastids. The presence of Complex I has been biochemically confirmed in Trypanosoma and Phytomonas [6, 8].

Conclusion

On all accounts enumerated above, the biological knowledge reinforces the in silico prediction. Together, this provides convincing evidence that MURF1 is a highly derived homolog of NAD2. For illustration purpose, Fig. 1 depicts the multiple protein sequence alignment of the most conserved block of known NAD2 proteins and kinetoplastid MURF1 sequences.

Figure 1
figure 1

Multiple sequence alignment of kinetoplastid MURF1 sequences with NAD2 sequences from other eukaryotes. The top five sequences are kinetoplastid proteins. Only the most conserved region of the protein is depicted. The range of amino acid positions included in the alignment is indicated by the numbers following the species name. Dashes specify alignment gaps.

Full size image

Outlook

Notably, a functional NADHdh is crucial to the survival of trypanosomes. Under aerobic conditions (procyclic, insect stage), NADHdh is required as a component of the respiratory chain, to catalyze electron transport toward complex IV. The thus generated proton gradient is utilized for ATP synthesis. Under anaerobic conditions (bloodstream form), a functional NADHdh is equally essential. In the blood stream of mammals, NADHdh provides electrons for the alternative oxidase, a pathway required for maintaining the balance of NADH/NAD+ in the cell. This confirms that trypanosomes depend on a functional NADHdh. In fact, Atovaquone, an anti-leishmanial drug, inhibits the NADHdh activity in P. serpens and this inhibition was suggested to underlie the anti-leishmanial activity of that drug [6]. In this context, the identification of MURF1 as a divergent NAD2 could offer new avenues to the prevention or treatment of trypanosomatid-caused diseases.

Methods

Dataset

All function-known protein sequences used in this study were retrieved from the organelle genome database GOBASE release 12.0 [9]. The homologs for MURF1 were retrieved from Entrez, and their accession numbers are given in Table 4[10].

Table 4 List of kinetoplastid MURF1 sequences with GenBank Accession Numbers
Full size table

Assignment of MURF1

For the function assignment of MURF1, we chose to use sequence-sequence, sequence-profile and profile-profile methods described below, which are most sensitive methods to detect remote homologs.

Sequence – Sequence Comparison

A BLAST (blastp) search was conducted for the MURF1 protein sequence against NCBI's NRDB (non-redundant protein database) (October, 2006; 4,565,699 sequences), with default parameters [11]. In addition, a FASTA search was conducted for the MURF 1 protein sequence against UniProt (release 10.4) with default parameters, at the EBI website http://www.ebi.ac.uk/fasta33[12].

Profile – Sequence Comparison

This comparison was conducted in two different ways. First, PSI-BLAST was employed to search MURF1 remotely against NCBI's NRDB, with four iterations [13]. Second, we performed profile HMM – sequence comparison using profiles from Pfam version 21.0, executed at the Pfam website http://www.sanger.ac.uk/Software/Pfam[14].

Profile HMM – Profile HMM Comparison

For Profile HMM – profile HMM comparison, we used HHsearch of the HHpred package, which takes the MURF1 sequence as input and searches against NRDB using PSI-BLAST [15]. The MURF1 homologs obtained from the PSI-BLAST search are then used to generate a profile HMM. As a next step, this MURF1 profile HMM is searched against all profile HMMs of function-known proteins available from the public databases Pfam, PANTHER, SMART, COG, PDB and SCOP.

In addition, we generated our own profile HMMs for each of the 12 NADHdh subunits (1–11 and 4L) from all known sequences of these protein classes. These sequences were clustered at eight different identity thresholds (40, 45, 50, 55, 60, 65, 70 and 75%) using CD-HIT, followed by multiple sequence alignment performed with MUSCLE [16, 17]. (Note: The number of instances for subunit NAD8 and NAD10 are less than 3 at identity thresholds 65 and 75% respectively and hence profile HMMs were not generated below these thresholds for these two subunits) [see Additional file 1]. The multiple alignment served as input for generating profiles using hmmbuild of HMMER version 2.3.2, 2003 package [18]. In order to verify whether profile HMM-profile HMM comparison is efficient in distinguishing the subunits, we tested this approach on the function-known sequences. Herefore, we used NAD2 and NAD5 subunits – the most difficult subunits to distinguish. For evaluating NAD2-profile HMMs, all NAD2 sequences were divided randomly into ten non-overlapping subsets of equal size. A test-profile HMM was generated using one of the subsets, while the remaining nine subsets were used for generating a "master" profile HMM. The NAD2 test-profile HMM was then searched against the NAD2 "master" profile HMM and the NAD5 profile HMM (generated using all NAD5 sequences) using HHsearch. This procedure is repeated ten times. The same test was done for NAD5. All test-profile HMMs were correctly identified at 100%. Finally, the MURF1 profile HMM was searched against all the 84 profiles using HHsearch with default parameters.