In this study we report the cloning of the complete ORFeome of SARS-CoV and the results of a matrix-based yeast two-hybrid screen of pairwise viral protein-protein interactions. From a number of recent structural studies it is clear that during the viral life cycle large replication complexes are formed, which involve a large number of viral proteins 
. SARS-CoV is a representative of the Coronaviridae, the largest RNA viruses known (27 to 32 kb, plus-stranded). SARS-CoV expresses at least 16 non-structural replicase proteins which are cleaved co- and post-translationally from two precursor polyproteins by two viral proteinases, four structural proteins and a set of eight accessory proteins specific for the individual virus groups 
. Since the polypeptide processing sites of non-structural proteins are well defined, we chose the strategy to subclone all individual ORFs predicted, and not to use the precursor polyproteins. Using this approach we expected to avoid problems in the expression, folding or targeting of the polypeptides due to incorrect processing. Such problems had been reported for yeast two-hybrid assays performed with the plus-stranded RNA viruses Hepatitis C virus 
and wheat streak mosaic virus (WSMV) 
, where interactions had been observed only when random fragments, not mature proteins were used. But there are also reports on potato virus A (PVA) and pea seed-borne mosaic virus (PsbMV), which belong to the potyvirus (+) strand RNA virus family similar to WSMV 
, as well as on a subset of poliovirus proteins 
, where interactions have been detected among cloned mature proteins.
In our screen approximately 14% of the 450 possible non-redundant protein interactions tested were positive and approximately 38% of which were confirmed by CoIP. This result is in the same order of magnitude as the outcome of similar Y2H matrix screens in KSHV and VZV, indicating that this approach can also be applied for plus-strand RNA viruses. The low numbers of Y2H interactions detected in two directions are a common phenomenon in Y2H assays and are probably due to steric constraints of either bait or prey fusion proteins.
Coronavirus replication complexes consist of intricate macromolecular structures in which many of the non-structural replicase proteins are involved. One of the most interesting interactor proteins found in our study is nsp8, which interacts with replicase proteins nsp2, nsp5, nsp6, nsp7, nsp8, nsp9, nsp12, nsp13, nsp14. The importance of this protein is supported by recently reported crystallization studies, which described the multimeric association of various of the non-strucutural proteins. Nsp 8 seems to be one of the proteins involved in these complexes. Nsp8 deletion or irreversible fusion to nsp7 or nsp 9 by mutagenesis of the corresponding cleavage site results in a lethal phenotype supporting the idea that nsp8 is absolutely essential for virus replication (Deming et al., submitted). Evidence has been presented for interaction with nsp9, a ssRNA-binding protein, by analytical ultracentrifugation experiments and by a decrease of the disorder of the nsp8 N-terminal region after the addition of nsp9 
. Furthermore, a hexadecameric nsp7-nsp8 supercomplex was described which was suggested to encircle RNA where it may serve as a general processivity factor for the RNA-dependent RNA Polymerase (RdRp) nsp12 (19). A very recent report described nsp8 as a second RdRp of SARS-CoV. It was shown to initiate the synthesis of complementary oligonucleotides of <6 residues in a low fidelity reaction which eventually might serve as primers for the primer-dependent nsp12 RdRp 
. For MHV it was shown that RdRp co-immunoprecipitates with nsp8, nsp9, nsp5 and the helicase nsp13 
, and that it also colocalizes with nsp7, nsp9 and nsp10 
. Thus, the nsp8 interactions found by us are confirmed by a number of different studies and it seems to play an important role in the viral replication complex.
In this manuscript, we demonstrate interactions between RdRp (nsp12) and nsp8, and with the helicase nsp13 in both directions of the Y2H screen. It is likely that the RdRp interacts with more nsps than were found, but these interactions may require mediator proteins like nsp8.
Nsp2 interacted with seven other nsps including nsp8 and with one of the newly described structural proteins ORF3a. As shown by CoIP, it also self interacts to a dimeric or multimeric complex. The relatively large number of interactions might imply a crucial role of nsp2 in the viral life cycle. However, it was shown by deletion mutants of SARS-CoV and MHV that neither the encoding genomic RNA sequences nor the nsp2 proteins are necessary for the generation of infectious viruses in cell culture 
. Since these viruses displayed slightly reduced phenotypes in growth, RNA synthesis but not protein processing, it was speculated that nsp2 might play a role in global RNA synthesis, and possibly in virus-cell interactions or viral pathogenesis. The reported subcellular localization of individually expressed nsp2 in delayed brain tumor (DBT) cells 
is similar to the diffuse cytoplasmic and nuclear immunofluorescence staining pattern found with our N- or C- terminally tagged nsp2 proteins. Thus, the exogenously expressed nsp2 does not target specific membranes in the absence of infection. However, after coinfection with a MHV mutant virus lacking nsp2, the protein expressed in trans
was reported to be recruited into distinct viral replication complexes. This relocalization of nsp2 to small vesicular foci in the cytoplasm was also confirmed in SARS-CoV-infected Vero cells by immunofluorescence staining with anti-nsp2 antibodies 
In our study, only few interactions were found for the structural proteins, which might be biased by transmembrane sequences preventing the transfer of expressed prey (containing the GAL4 activating domain) and/or bait (containing the GAL4 DNA-binding domain) fusion proteins to the nucleus of the yeast cell where protein-protein interaction leads to transcription. Only the E and ORF3a proteins showed a number of associations whose relevance is unclear. Interactions of ORF3a –M and ORF7a-S fit to the recent finding that the two accessory proteins display structural functions as has been described 
For the group-specific accessory proteins it has recently been shown that deletion of five of the eight ORFs (ORFs 3a, ORF3b, ORF6, ORF7a and ORF7b) alone or in combination did not influence dramatically the level of RNA or the replication efficiency in vitro
or in an in vivo
mouse model 
. The most interesting accessory protein with respect to interactions in our study turned out to be ORF9b. Y2H interactions with nsp8 and nsp14 were found bi-directionally and the self-interaction could also be confirmed by CoIP. Latter result is confirmed by recent structure data 
. The ORF9b protein, which is encoded within the nucleocapsid gene, is an intertwined dimer with an amphipathic outer surface and a long hydrophobic lipid binding tunnel. This suggests that ORF9b is targeted to ER-Golgi compartments via an unusual anchoring mechanism and acts as an accessory protein during virion assembly. Although most of the accessory proteins do not seem to play pivotal roles in viral replication, they might still be important for the virus-host interplay and for pathogenicity. Currently, there is no reasonable explanation for the large number of interactions found for ORF9b by the Y2H screen. As deletion of ORF9b does not seriously reduce virus replication in vitro
consistent with a luxury function, the 9b protein may function to enhance the global stability of the SARS proteome network and play some unrealized role in regulating virus-host protein-protein interactions. It thus might be more important for enhancing in vivo
Immunofluorescence localization of Flag-tagged viral proteins corresponded in most cases to published data on SARS-CoV and other coronaviruses. We found nsp2 proteins in the cytoplasm and to some extent in the nucleus which is in accordance with anti-nsp2 antibody stainings of stably DBT-nsp2 (MHV) expressing cells 
. Many of the non-structural proteins are involved in the replication of the virus and locate to virus-induced cytoplasmic double-membrane vesicular complexes as the sites of viral replication. It is therefore important to take into account that the localization patterns of nsps might be quite different when expressed individually in cells as compared to the situation of viral infection where various viral proteins might help to recruit each other to the sites of active replication.
Accessory protein 3a, for which a number of effects on cellular functions were described 
, we located in our Flag-tagged versions to the Golgi complex as Yuan et al. 
observed using EGFP-tagged constructs. As a structural protein ORF3a interacts with the M protein 
which was also clearly found in the Golgi as Flag fusion proteins. The nuclear localization of ORF3b is also reasonable because it induces cell cycle arrest at the G0/G1 phase and apoptosis 
. Proteins ORF6 and ORF7b, interacting in Y2H and CoIP, were both found in the ER. To our knowledge this localization has not been described for ORF7b before. Not much is known about ORF9b other than it is expressed in infected cells 
and that antibodies to it are found in infected patients 
. As opposed to Meier et al. 
, who located ORF9b to intracellular vesicular structures (293T cells), we found it to be diffusely distributed within cytoplasm and nucleus (HeLa cells).
Analysis of network statistics showed that despite high clustering coefficients the SARS interaction network is not higher clustered than expected at random. It, thus, appears as a single module such as the KSHV network and is not subdivided into separate functional modules as cellular networks. Based on currently known and predicted host-virus interactions, a joint virus-human network was derived in which the viral part of the network appears to be separated from the main host network. In this respect, the SARS network differs from the KSHV viral network which is incorporated into the host interactome. However, this may be due to the small number of virus-host interactions identified so far for SARS. Indeed for KSHV, the predicted virus-host network was based on about twice as many interactions to the host. To better understand the role of the intraviral protein interactions it is necessary to gain more knowledge on the SARS-CoV with it's host during infection.
We certainly missed a considerable number of intraviral protein interactions in our Y2H screen as can be seen for M-N and nsp2-nsp2, nsp5-nsp5 self-interactions, which we could only detect by CoIP. Although, it is generally acknowledged and certainly has to be taken into account that Y2H assays are error-prone by producing false positives and false negative results, we identified a large number of interactions which have not been reported previously and which could be confirmed biochemically. These interactions will be of great help for further studies which are aiming at the elucidation of SARS-CoV replication and pathogenesis. Future experiments with the mutant viruses lacking nsp2, nsp8 or ORF9b will show the relevance of the interactions detected for virus replication, growth and pathogenicity in vitro and in vivo model systems.