The PPIs of P. horikoshii OT3 were explored using the mammalian two-hybrid system that we had already established 11. A flow chart of the assay process is shown in Figure 1. P. horikoshii OT3 has 2,061 ORFs and we cloned 1,390 of these (data not shown). These clones were the starting material for our analysis. The protein interactions of membrane proteins and secreted proteins are generally hard to analyze using the two-hybrid system because the process occurs in the nucleus. Using the SOSUI program 17, we examined the ORFs to deduce which proteins included membrane-spanning sequences or signal peptide sequences; 410 clones were removed because they were predicted to code for membrane or secreted proteins.

The basis of the mammalian two-hybrid assay is the bait and prey protein interaction, where interaction initiates transcription of the luciferase reporter gene; an increase in expression of the luciferase reporter gene corresponds to the interaction between the bait and prey proteins. Assay samples expressing bait and prey proteins (Gal4- and VP16-fusion proteins, respectively) were constructed by PCR (see Materials and methods, and Additional data file 1). We pooled the samples for the first assay. As two bait samples were excluded from the assay due to self-activation, a total of 479 two-mixture (two-mix) bait samples and 480 two-mix prey samples were systematically tested in the first assay. The positive combinations in the first assay were then examined using combinations of single bait and prey samples to identify the interacting pairs. We finally obtained 170 interactions from the assay.

The frequency of self-activation proteins in P. horikoshii OT3 was 0.2% in our mammalian two-hybrid system, which is relatively lower than the frequencies of self-activation proteins observed in other organisms 910111216. One of the reasons for this might be that many of the Pyrococcus proteins were not sufficiently expressed in our system because of different preferences for codon usage between P. horikoshii and mammalian cells. Thus, we randomly selected 34 bait samples and examined expression of the fusion proteins in CHO-K1 cells using western blot analysis. All of the samples exhibited signals with expected molecular size (Figure 2; the data shows the representative results of 11 samples). However, the amount of detected proteins seems to be dependent on molecular size; proteins with a size less than about 50 kDa showed strong signals whereas there was a tendency for larger proteins to show relatively weaker signals.

We detected characteristic hetero-interactions consisting of α and β subunit proteins, such as indolepyruvate ferredoxin oxidoreductase (PH1138-PH0229 or PH1138-PH0764), and 20S proteasome (PH1553-PH0245), showing that we successfully identified at least some of the protein interactions in P. horikoshii OT3. Several reports indicate, however, that recombinant Pyrococcus proteins expressed in Escherichia coli cells assume their mature conformations or activities only after heat activation 1819, suggesting that the other interactions might be artificial and are only detected at 37°C, the temperature used in our analysis method. Thus, we evaluated the interactions derived from our mammalian two-hybrid system using the in vitro pull-down assay, with or without heat activation. In three of the protein pairs, all of the ³⁵S-labeled proteins were successfully precipitated (Figure 3), showing that these protein pairs can interact with each other regardless of heat activation. The results indicate that at least some Pyrococcus proteins form their native conformations when expressed in cultured mammalian cells at 37°C.

Generally, PPIs obtained from the two-hybrid method have many false positives, which may complicate elucidation of the biological importance of the interactions. For further analysis it is best to select reliable interactions using positive and negative training interaction sets that can be made from known interaction information, which was done in the analysis of Drosophila interactions 9. This approach seems impossible to apply to the Pyrococcus interactions, however, because few of the PPIs in P. horikoshii OT3 or related species are known. To overcome this problem, we considered the activity of the luciferase reporter gene. In our mammalian two-hybrid system, the activity of the luciferase reporter is used to judge protein interactions. We classified the interactions into levels 1 to 3 (weak to strong) depending on the strength of luciferase activity (see Materials and methods). We decided to take the level 2 and 3 interactions as the selected interaction set, resulting in 107 interactions consisting of 51 self-interactions and 56 hetero-interactions, including 7 bi-directional interactions (Figure 1).

We evaluated this selected interaction set using the interaction generality (IG) measure, a method for computationally assessing the reliability of PPIs 20. Interactions with lower IG values are more likely to be reliable than interactions with higher IG values. The IG values for all interactions ranged from 1 to 18, whereas the IG values for the selected interaction set ranged from 1 to 4 (Figure 4). The average IG value of 5.52 for all interactions decreased significantly to 1.71 in the selected set. The latter value is also significantly lower than the average IG value of 3.80 (P < 0.0001) that was calculated from 10,000 mathematical trials that randomly removed the same number of interactions (by a jack-knife calculation). Furthermore, of the seven P. horikoshii OT3 hetero-interactions for which we found corresponding interactions in other species (interlogs), all were in the selected interaction set (asterisks in Table 1 and Figure 5). The result suggests that we successfully concentrated the true-positive interactions.

In P. horikoshii OT3, the IDs of the predicted ORFs are systematically numbered according to their genomic location. We mapped the ORFs of the interacting protein pairs onto the P. horikoshii OT3 genome and found that in 11 out of 49 hetero-interactions (22%), the corresponding ORF pairs were mapped to adjacent loci on the genome, and that several other ORFs located close were orientated in the same direction (Table 2; Figure 6). We hypothesize that these ORF pairs belong to operons, where many of the functionally related genes in eubacteria and archaea are transcribed as a polycistronic mRNA.

To confirm our hypothesis, we explored the interlogs of the Pyrococcus interactions (Figure 6). The ORFs corresponding to the interacting pair PH1978 and PH1983 are separated on the OT3 genome by the ORFs for PH1981 and PH1980; although close to one another, and close to five downstream ORFs (PH1972, PH1974, PH1975, PH1976, and PH1977), all the ORFs are encoded in the same direction. The proteins PH1983 to PH1972 all show high similarity to archaeal-type H⁺-ATP synthase protein subunits in Methanosarcina mazei and Methanococcus jannaschii, and vacuolar-type H⁺-ATP synthase protein subunits in Thermus thermophilus (Figure 6a). In M. mazei, the gene cluster for these proteins is reported to be organized into a single operon 21. In addition, the structure of the protein complex of this ATP synthase has been reported in M. jannaschii 22, where subunits E and H, corresponding to PH1978 and PH1983, interact directly on the intracellular side (Figure 6a inset). It is highly plausible, therefore, that this ORF cluster in OT3 is in an operon. We also found that the interacting pair of PH0487 and PH0490 (not annotated) have high similarity to the Bacillus subtilis chemotaxis proteins CheC and CheD, the genes for which are located adjacently on the Bacillus subtilis genome and compose an operon 23, suggesting that the ORFs for PH0487 and PH0490 are also expressed in P. horikoshii OT3 as an operon and that their functions are similar to those of CheC and CheD.

Another example is PH1354 and PH1355, which are similar to SNO1 (32% identity) and SNZ1 (56% identity), respectively, from Saccharomyces cerevisiae (Figure 6b). In the Pfam database (The Sanger institute, UK) 24, SNO1 and SNZ1 are shown to be widely conserved proteins categorized as SNO (PF01174) and SNZ family (PF01680) proteins. Members of these families are related to the pyridoxine biosynthetic pathway and are ethylene-responsible proteins, or glutamine aminotransferases. All known orthologous SNO/SNZ genes in eubacteria and archaea are adjacently located in the same direction on the genome, suggesting that they are likely to be organized in an operon (Figure 6b). Further, it has been shown using the yeast two-hybrid method and DNA microarray analyses, that SNZ1 and SNO1 in S. cerevisiae interact with each other and that their genes are co-regulated 25.

As very few proteins in P. horikoshii OT3 have had functions assigned to them, we predicted the functions of the interacting proteins by using the annotations from orthologous proteins. We classified all of the 960 explored P. horikoshii OT3 proteins into 4 classes depending on the existence of orthologous proteins: class A, proteins with eubacteria, eukaryote and archaea orthologs (177); class B, proteins with eubacteria and archaea orthologs (228); class C, proteins with eukaryote and archaea orthologs (69); and class D, archaea-specific proteins (486). The set of selected self-interacting proteins (Table 1) and the set of selected hetero-interacting proteins (Figure 5) were also arranged according to these classes and by the annotation information of orthologs. Gene IDs for the hetero-interacting proteins are available in the DOGAN database of P. horikoshii OT3 26.

Of the selected 51 self-interacting proteins, 20 of 29 proteins in classes A to C were well annotated, whereas for the archaea-specific proteins in class D, only 1 out of 22 proteins was annotated (Table 1). The proteins in classes A and B were often annotated as enzymes involved in cellular metabolism, which is consistent with previous reports that many of these enzymes form homo-oligomer structures 27282930313233. PH0119 and PHS042 in class C are not enzymes; the first is similar to the DNA repair RadA/Rad51 protein, and the second to the RNA-binding small nucleoprotein (Sm protein), which are both known to form homo-heptamer structures 3435. PHS053 was the only annotated protein in class D, showing similarity to archaea-specific DNA-binding protein AlbA of Archaeoglobus fulgidus, which has been shown by X-ray crystallography to form a homo-dimer structure and to possess DNA-binding properties 36.

As with the self-interacting proteins, many of the hetero-interacting proteins in classes A to C were well annotated, but few of the class D archaea-specific proteins were described (Figure 5). More than half of the hetero-interactions between proteins of the same class consisted of the archaea-specific protein pairs (17 out of 30 interactions), and in the interactions between proteins of different classes, most of the pairs included archaea-specific proteins as one of the interaction partners (18/19 interactions, 95%). In the interactions for which both proteins in a pair are annotated, the annotations are related. For instance, PH1022-PH1645 are orthologs of proteins related to sugar phosphate metabolism, mannose-1-phosphate guanylyltransferase and ADP-specific phosphofructokinase, respectively. These results are reasonable as many proteins play a role in the network of cellular biological processes by associating with other related proteins (guilt-by-association) 3. Based on this concept, several research groups have successfully predicted the functions of uncharacterized proteins using data on their interaction with other proteins 1011121337. We obtained 13 hetero-interactions between annotated proteins and hypothetical proteins (Figure 5), in which the functions of such hypothetical proteins are likely to be related to the functions of their interaction partners.

Dividing the hetero-interactions into two groups according to the classes described above - hetero-interactions consisting of interactions between proteins of the same class (30 pairs, 61.2%) and interactions between proteins of different classes (19 pairs, 38.8%) (Figure 5) - we found that the percentage of hetero-interactions consisting of interactions between proteins of the same class was significantly (P < 0.01) higher than the expected value of 35.1%, which was calculated by assuming that the interactions are not biased by class.

Forward and reverse primers specific to the P. horikoshii OT3 genes were used to construct the assay samples that expressed the P. horikoshii proteins fused with the Gal4 DNA-binding domain (BIND) or the VP16 transcriptional activation domain (ACT). Mammalian two-hybrid assays, including the transfection method, were carried out as previously described 11, with slight modifications. The positive combinations in the assay were categorized by the fold value of luciferase reporter activity as follows: level 1, ≥ 3 to <5 times as high as the background activity; level 2, ≥ 5 to <10 times as high; and level 3, ≥ 10 times as high. For a more detailed description, see Additional data file 1.

Bait sample (10 μl) was transfected to 10⁵ CHO-K1 cells in six-well culture plates using Lipofectamine2000 (Invitrogen, Carlsbad, CA, USA). After 24 h of incubation, cells were washed once with ice-cold TBS (50 mM Tris-HCl, pH 8.0, 137 mM NaCl, 2.68 mM KCl) and harvested using 200 μl of Lamuli sample buffer. The sample was boiled for five minutes and suspended with vortex mixer for 30 s. Protein in Lamuli sample buffer (10 μl) was subjected to 12% SDS-PAGE and transferred electrically onto a polyvinylidene fluoride (PVDF) membrane. The membrane was blocked by TBS/0.05% w/v Tween 20 (TBS-T) containing 6% w/v skim milk for 1 h and incubated with a polyclonal antibody against the Gal4 DNA binding domain (dilution 1:200; Santa Cruz Biotechnology, Santa Cruz, CA, USA) for 1 h. After washing with TBS-T, the membrane was incubated with horse radish peroxidase (HRP)-conjugated anti-rabbit goat IgG (dilution 1:2,000; GE (Amersham Biosciences, Piscataway, NJ, USA) for 1 h and washed with TBS-T. Detection of the signal was performed using the ECL plus system (GE Amersham Biosciences).

In vitro pull-down assays were carried out as previously described 45 with slight modifications. The template DNA was constructed using overlapping PCR, which has the T7 promoter sequence upstream of the P. horikoshii genes. Biotinylated or ³⁵S-labeled proteins were synthesized in vitro according to the manufacturer's protocols, using the Transcend Biotinylated lysine-tRNA (Promega, Madison, WI, USA), redivue L-[³⁵S] methionine (Amersham Biosciences), and TNT^® T7 Quick Coupled Reticulocyte Lysate system (Promega). The samples were applied to the Centrisep spin column (Applied Biosystems, Foster, CA, USA) and diluted with an equal volume of 1 × phosphate-buffered saline without 1 mM CaCl₂ and 0.5 mM MgCl₂ (PBS (-)). Each sample was divided into two microcentrifuge tubes and incubated for 15 minutes at 37°C or 75°C (non-heat and heat, respectively). Samples were then centrifuged at 15,000 × g for 20 minutes at 4°C and the supernatant collected. Recovery of ³⁵S-labeled proteins in the non-heat and heat samples was estimated using SDS-PAGE followed by autoradiography. Equal amounts of biotinylated and ³⁵S-labeled proteins were mixed and incubated for 1 h at 25°C. Dynabeads^® Streptavidin (Dynal Biotech LLC, Milwaukee, WI, USA) were added to the reaction mix and incubated on a rotary shaker for 30 minutes at 25°C. The beads were isolated with the magnet and washed three times with ice-cold TBS-T. The precipitated proteins were subjected to SDS-PAGE followed by autoradiography.

The IG was calculated with reference to its definition 20. The IG value of the PPI between proteins a and b is defined as:

where N is a set of proteins such that x, y, a, b ∈ N. E is the set of PPIs between proteins a and b. In the calculation, we considered the protein a (BIND)-protein b (ACT) interaction and the protein b (BIND)-protein a (ACT) interaction to be a single interaction (a bi-directional interaction in a two-hybrid experiment). Furthermore, we removed the self-interactions from the calculation of the IGs. Therefore, E is the set with conditions such that ∀a,b{(a,b) ∈ E ∧ (b,a) ∈ E} and ¬∃x{(x,x) ∈ E}.

Orthologs for each P. horikoshii OT3 protein were identified using the Sequence Similarity Database (SSDB) 46 and the Gene Function Identification Tool (GFIT) program 47 in the Kyoto Encyclopedia of Genes and Genomes (KEGG) 4849. We screened the orthologous protein sequences from other organisms by using P. horikoshii OT3 proteins as query sequences and selected those that exhibited a Smith-Waterman score of 200 or more.

All PPI data from P. horikoshii OT3 have been stored in a database supported by DDBJ 50. These interaction data have also been deposited with each BIND ID (331136 to 331235) in the Biomolecular Interaction Network Database (BIND) 51, as well as 100 single interaction records (not including reciprocal interaction data), and 3 records determined from in vitro pull-down assays (Figure 3; BIND IDs 331124 (PH0245-PH1533), 331125 (PH0795-PH0017) and 331126 (PH1354-PH1355).