Functional genomics and proteomics of the cellular osmotic stress response in `non-model' organisms

Major advances in high-throughput technologies for the detection and quantification of nucleic acids, proteins and metabolites have led to a paradigm shift in biological research. The new field of systems biology attempts to integrate the complex data sets generated by high-throughput approaches to develop a holistic understanding of complex biological structures, their dynamics and their responsiveness to external stimuli such as salinity stress. The level of complexity of structures modeled by systems biology ranges from molecular networks via cells and organs to whole organisms (Kitano, 2002). Systems biology organizes data obtained by genomic, transcriptomic, proteomic and metabolomic approaches to attempt to build descriptive and mechanistic models of integrative biological phenomena such as development or interactions of organisms with their environment.

The ultimate goal of this approach is to generate a mathematical model that describes the biological system and has predictive power(Aggarwal and Lee, 2003). Achieving this tremendously ambitious goal depends on in-depth knowledge about each element constituting the system of interest. For instance, experimental data about the expression, regulation, function, compartmentation,interaction, modification and stability of individual RNAs and proteins have to be collected and integrated for each state of the system that is described by the model. Systems biology approaches have largely been applied to organisms whose genomes have been sequenced (so-called `model' organisms)because many of the available bioinformatics tools are based on prior genome sequencing and annotation of the encoded transcriptome and proteome. However,most high-throughput technologies for analyzing the transcriptome, proteome and metabolome do not strictly depend on prior knowledge of genomic sequence. Therefore, these approaches can also be applied to `non-model' organisms for which few if any genome sequence data are available. Such organisms account for the majority of species used in traditional comparative biology.

We utilized high-throughput transcriptomics and proteomics approaches for identifying key molecular constituents associated with osmoregulatory functions in euryhaline tilapia and other animals. Our work shows that these data can be used for bioinformatics analysis to generate models about biological processes, cellular pathways and molecular functions associated with osmotic stress responses. Nevertheless, many obstacles are encountered when attempting a systems biology approach with non-model species. In the following, we briefly review our efforts to apply high-throughput transcriptomics and proteomics methods to euryhaline tilapia, dogfish shark and an intertidal sponge to pave the way towards a systems biology approach for studying osmoregulation.

Organisms have osmoregulatory mechanisms that span all levels of biological organization from molecules to behavior. Therefore, understanding responses of fishes and other animals to salinity stress requires knowledge of the biological system as a whole, starting with its molecular components. A very popular approach for gaining insight into genes that are regulated by environmental stimuli is based on the use of DNA microarray chips. However, to harness the full power of this approach, sequence knowledge of the entire transcriptome of the species of interest is required. Although DNA microarrays can also be used very effectively for non-model species(Buckley, 2007; Gracey, 2007), prior sequence knowledge greatly facilitates their production and use and maximizes coverage of the transcriptome during the analysis. Serial analysis of gene expression(SAGE) and recent modifications of SAGE represent alternative approaches to DNA microarrays that identify very short sequence tags that can then be matched against existing sequences(Maillard et al., 2005). In addition to SAGE and DNA microarrays, the identification of a subset of RNAs whose levels are regulated by salinity can be achieved by methods that are based on differential analysis of RNA levels and less dependent on prior sequence knowledge of the transcriptome. Such methods include differential display (Stein and Liang,2002), representational difference analysis(Hubank and Schatz, 1999) and suppression subtractive hybridization (SSH)(Diatchenko et al., 1999).

Because SSH works optimally even in the absence of prior sequence knowledge, we have used this technique for enriching a subset of genes that are rapidly upregulated by salinity stress in gill epithelial cells of tilapia. Tilapia are strongly euryhaline teleosts that are capable of adaptation to a wide range of environmental salinity. Therefore, mechanisms and molecules involved in salinity adaptation are very prominent in this fish and can be effectively studied.

SSH was performed with mRNA from control fish transferred for 4 h from freshwater (FW) to FW (as the Driver sample) and the corresponding mRNA from fish transferred for 4 h from FW to seawater (SW) (as the Tester sample). The final resulting PCR products were cloned into pGEM-Teasy vector (Promega,Madison, WI, USA) to generate a subtracted cDNA library. This library was screened by colony PCR to yield 650 individual clones. Only 72 clones were selected for sequencing because many of the 650 clones had similar lengths and likely represented the same sequence. Thirty-two unique cDNAs were represented by the 72 clones that were sequenced. Increased mRNA abundance after 4 h SW transfer was confirmed for 22 of these 32 cDNAs by quantitative real-time PCR(qPCR). This result demonstrates that the rate of false-positive identification by SSH is low (less than one-third of the analyzed clones were false positives). Full-length coding sequences for most of the 22 cDNAs were obtained by RACE-PCR and degenerate primer PCR, starting with original SSH clones, many of which represented 3′ terminal fragments of cDNAs(Fiol and Kültz, 2005; Fiol et al., 2006a). Based on amino acid homology in the coding sequences, we were able to identify more than 90% of the SSH genes cloned from tilapia(Table 1).

Hyperosmotic induction of tilapia SSH genes was confirmed and the kinetics of their induction analyzed by real-time qPCR. Most SSH genes show a rapid and transient increase in mRNA abundance, with peak levels observed between 2 and 8 h after SW transfer. An example of a time course depicting the salinity-induced induction of a tilapia SSH gene (protein phosphatase 2A,catalytic subunit) is shown in Fig. 1. Most of the SSH genes differ only slightly in the kinetics of their induction, suggesting that they participate in a common cellular network as part of an overall essential and coordinated adaptive response that involves changes in metabolism, transport, damage repair, etc.(Fiol and Kültz, 2005; Fiol et al., 2006a). A notable single exception is a cDNA clone (SSH#7) that displays a robust sustained induction in response to SW transfer. This cDNA increases as rapidly as the other SSH genes but it remains high even after a prolonged stay of tilapia in SW, and the degree of its increase is much greater than for any other SSH gene, exceeding five orders of magnitude(Fiol et al., 2006a). Unfortunately, we were not able to match the sequence of the SSH#7 cDNA to any known sequence. The lack of an open reading frame (ORF), even after extensive attempts to extend the sequence by RACE-PCR, may suggest that SSH#7 represents a non-protein-coding RNA (ncRNA). Because ncRNAs are involved in transcriptional and translational regulation, modulation of protein function,and regulation of RNA and protein localization, SSH#7 may have an important role for salinity adaptation even if it is an ncRNA(Goodrich and Kugel, 2006). For one of the SSH genes that encodes a protein [osmotic stress transcription factor 1 (OSTF1)], we have already performed a follow-up study to analyze the mechanisms of its regulation in more depth. The results of this study show that hypertonicity per se, rather than indirect systemic factors, is responsible for the salinity-induced OSTF1 increase, that the mechanism of increase is based on transient mRNA stabilization and that a return to normal OSTF1 levels at later time points is promoted by systemic factors(Fiol et al., 2006b).

The rapid and transient induction kinetics of most SSH cDNAs during hyperosmotic stress is characteristic of immediate-early genes (IEGs). IEGs were originally identified in neuronal tissue and encode many inducible transcription factors but also other proteins such as scaffolding proteins,signaling proteins and proteases (Lanahan and Worley, 1998). IEGs are key regulators during environmental adaptation because they control the transcriptional regulation of effector genes (delayed response genes) necessary for adaptation. IEGs represent the first set of genes that are induced in response to environmental signals and encode mRNAs that do not require de novo protein synthesis for their induction (Morgan and Curran,1989). Induction of many IEGs is often mediated by mitogen-activated protein kinases (MAPKs)(Thomson et al., 1999), which we have shown to be activated by salinity stress in euryhaline teleosts(Kültz and Avila, 2001). Moreover, IEGs function within a network of other, constitutively expressed,proteins to control cellular regulation in response to environmental cues(Lanahan and Worley, 1998). Thus, we utilized the gene set identified by SSH for a systems biology approach that is based on bioinformatics tools to gain insight into the regulatory network that underlies salinity adaptation in tilapia.

To gain insight into the nature of regulatory networks that are activated early in response to salinity stress in tilapia gill cells, we used bioinformatics tools that extract and synthesize most available information about any given gene. The SSH gene set was compared to templates for known biochemical pathways, molecular functions and biological processes. Using Pathway Studio software (Ariadne Genomics, Inc., Rockville, MD, USA), we distilled all published information about biological relationships of mammalian orthologs of the salinity-induced tilapia genes and modeled the interaction of the SSH genes within a common stress response signal transduction network (Fiol et al.,2006a). This network contains 13 of the 22 identified tilapia SSH genes (OSTF1, TFIIB, BMPR2, PPP2CB, MIG-6, MAP3K7IP2, ENO1, IDH2, FABP6,APOD, GSN, LGALS4, ANXA11) (see Table 1). Furthermore, a different type of bioinformatics approach,PANTHER (Mi et al., 2005),revealed that mammalian homologs of 11 of the 22 identified tilapia SSH genes participate in apoptotic and cell cycle regulatory pathways (OSTF1, TFIIB,BMPR2, PPP2CB, MIG-6, MAP3K7IP2, RFP128, PAD2, GSN, LGALS4, ANXA11) (see Table 1). Apoptosis and cell cycle regulation are key processes of the cellular stress response, indicating that these biological processes are targeted during osmotic stress in tilapia gills (Kültz, 2005). In addition to apoptosis and cell cycle signaling, cellular functions associated with the 22 tilapia SSH genes are organic osmolyte accumulation, energy metabolism, lipid transport and membrane protection, modulation of actin-based cytoskeleton dynamics, and control of mRNA and protein stability(Fiol et al., 2006a). Regulation of these processes reflects the need for extensive remodeling and rapid cellular turnover in tilapia gill epithelium during salinity stress.

Although the pathway analysis described above yielded remarkably consistent and meaningful results, some potential pitfalls need to be considered. First,comparisons are made based on data that are heavily biased towards a few model organisms such as human, mouse and rat. However, the involvement of a particular protein in cellular functions, molecular pathways and biological processes in non-model species may differ from that in model species. This first aspect represents a significant disadvantage for work with non-model species. Second, when working with gene sets that are based on altered levels of mRNA abundance, corresponding changes in protein abundance are often assumed but frequently not observed (Hack,2004). Thus, further analyses of proteins encoded by genes that have been identified or independent proteome analysis are logical next steps.

We have started confirming the upregulation of the immediate-early osmotic stress response genes identified by the SSH approach. The most practical approach for achieving this task is by using antibodies that specifically recognize the proteins of interest. Antibodies are much more readily available for model species, and currently projects are underway to generate specific antibodies for every human and every mouse protein. Therefore, this approach is less work-intensive for model species. We have used two strategies for antibody-based quantification of salinity-induced tilapia proteins by western immunodetection. First, we have tested available antibodies that were made against mammalian TFIIB for cross-reactivity and specificity with tilapia TFIIB. After successful identification of a suitable antibody that is cross-species reactive and specific (detection of a single band of expected molecular mass on western blots), we used it to confirm that tilapia TFIIB is also salinity-induced at the protein level(Fiol and Kültz, 2005). Second, we have generated a new antibody against tilapia OSTF1 based on immunization of rabbits with an immunogenic peptide. This antibody was used to confirm the induction of OSTF1 protein in gill cells of tilapia after transfer from FW to SW (Fiol and Kültz,2005). Currently, we are working on generating additional antibodies for more tilapia SSH genes. However, this approach is work-intensive, expensive and time-consuming and, therefore, not well-suited for the high-throughput data generation needed for systems biology approaches that utilize unbiased discovery-driven screens. Nevertheless, such a more traditional approach will greatly advance our knowledge of osmotic stress signaling in tilapia because it is firmly focused on prior knowledge of a limited set of elements involved in a specific biological process.

An alternative approach for protein analysis that is much faster, less time-consuming and more economical is based on comparing protein patterns of two different samples. This can be achieved by separating proteins via two-dimensional gel electrophoresis (2DGE) or two-dimensional liquid chromatography (2DLC) followed by subsequent mass spectrometry (MS)analysis. Both techniques (2DGE and 2DLC) are capable of resolving in the order of 1500 proteins and they are complementary, with different advantages and drawbacks (Wagner et al.,2000; Ishihama,2005; Hu et al.,2005; Bunai and Yamane,2005; Okano et al.,2006). A common limitation of both techniques is their bias towards high-abundance proteins, which is a result of the huge dynamic range of protein abundance in biological samples. This limitation can be addressed by extensive sample pre-fractionation, but such an approach requires a lot of starting material and greatly increases the cost, time and work load for the analysis. High-throughput proteomics using protein chips represents an alternative approach that is capable of detecting low-abundance proteins. However, protein arrays are expensive and depend on very high-quality antibodies that are often not available for a large number of proteins from non-model organisms.

Keeping these limitations in mind, we have compared the proteomes of four osmoregulatory tissues (rectal gland, kidney, gill epithelium, intestinal epithelium) with heart and brain from dogfish shark (Squalus acanthias) by 2DGE and MS (Lee et al., 2006). The rationale of this proteomics experiment was to determine whether highly abundant proteins that are common to osmoregulatory tissues can be identified and whether they are indicative of specific signaling pathways and biological functions in osmoregulatory tissues. The number of consistently resolved protein spots ranged from 984 in brain to 1230 in intestine, and overall 1465 unique protein spots were detected in the six tissues. Thirty-six percent of these protein spots (535 spots) were present at comparable abundance in all tissues. Two hundred and seventy protein spots that were significantly over-represented in one specific tissue or common to osmoregulatory tissues were identified using the Delta2D software package(Decodon GmBH, Greifswald, Germany). To identify such proteins we applied a robust and accurate method of protein quantitation that is based on 10 internal standard spots that are used for normalization of gel-to-gel loading and staining differences (Valkova and Kültz, 2006; Lee et al.,2006). Gel-based protein quantitation procedures are of course not without pitfalls but in our experience this method is more accurate than isotope-labeling or other approaches that rely on MS quantitation. The 270 identified protein spots were cut out from the gel, in-gel digested with trypsin, and analyzed by MS to determine their identity(Lee et al., 2006).

Of the 270 selected shark proteins, we were able to identify 62 proteins(=23%) with high confidence (Lee et al.,2006). Fifty-four of these 62 proteins were correctly identified using the MS-BLASTP2 algorithm (=87%), compared to 43 that were identified with the Mascot search engine (=69%) at the same mass accuracy limit of 50 p.p.m. High-quality spectra and de novo sequence information were obtained for many of the 270 analyzed protein spots, but a large portion was not identified by Mascot or MSBLASTP2, probably because of significant amino acid sequence variation in the shark proteins compared to their orthologs in model species (Lee et al.,2006). The genome of dogfish sharks has not been sequenced and few cDNA/protein sequences of this species are available. Therefore,identification of shark proteins by mass spectrometry poses unique challenges that are not encountered when working on model species. In particular,approaches based on peptide mass fingerprinting (Mascot, etc.), which generally work very well for model organisms (exceptions are proteins that are extensively post-translationally modified), are less successful in non-model species. Even if homologous peptide sequences differ by only a single amino acid, identification of the corresponding protein is prevented because peptide masses will not match. Nevertheless, fish proteins with highly homologous sequence stretches can still be identified using peptide mass fingerprinting(Smith et al., 2005; Monti, G. et al., 2005; Zupanc et al., 2006; Lee et al., 2006). Success rates of protein identification in non-model species can be increased using an MSBLAST approach that is exclusively based on sequence similarity and does not require matching peptide masses (Lee et al., 2006; Kjaersgard et al.,2006; Russell et al.,2006). Success rates of protein identification can be further increased by combining different methods of peptide fragmentation for MS/MS[e.g. collision-induced dissociation (CID) and post-source decay (PSD)] to generate a greater number of informative daughter ions(Lee et al., 2006).

Even invertebrate proteins for which close homologs from a related and fully sequenced reference organism are currently unavailable can be identified by MS. For instance, using MSBLASTP2 and a combination of CID and PSD for MS/MS, we identified many proteins from the evolutionarily primitive sponge Tetilla mutabilis (Fig. 2). In this experiment, we were interested in proteins that are upregulated during emersion and post-emersion stress(Fig. 2; Table 2). Such stress is experienced by this intertidal sponge during ebb-tide in its natural habitat at Estero La Cruz in the Gulf of California. The nature of stressors represents a mixture of hyperosmotic, heat, UV and oxidative stress. Sponges were collected in their natural habitat before emersion, after being emersed for 3 h and after being re-submersed for 2 h following 3 h emersion. Twenty-seven upregulated proteins and nine other proteins were picked for MS analysis, in which 25 of these 36 proteins (69%) were identified. This high success rate seems surprising compared with the 25% success rate with shark proteins (see above). However, for shark proteins we only considered proteins successfully identified if at least three out of the four search methods(PSD–Mascot, CID–Mascot, PSD–MSBLASTP2, CID–MSBLASTP2)yielded consistent results (Lee et al.,2006). If the same criteria are applied to the sponge samples,then the success rate drops to 44%. Again, the MSBLASTP2 approach identified five additional proteins (14% more) compared with the Mascot approach(Table 2), illustrating that database searches based on de novo peptide sequence alignments are more successful in non-model species than PMF approaches.

To overcome current limitations for proteomics in non-model species and to increase the success rate of protein identification, a number of technical improvements can be made. In particular, efforts directed at increasing protein sequence coverage in mass spectra are promising. Increased sequence coverage can be achieved by combining multiple proteases with different sequence-specific cutting patterns(Biringer et al., 2006). Likewise, top-down mass spectrometry approaches increase sequence coverage but are currently limited by the low efficiency of elution of intact proteins from polyacrylamide (PAG) gels (Bogdanov and Smith, 2005). Thus, elution of whole proteins from PAG gels is another area of research that should be addressed to improve the versatility of proteomics approaches. The development of new methods for N- and C-terminal protein sequencing by MS will also increase success rates for protein identification based on cross-species comparison(McDonald et al., 2005). If terminal protein sequences are known, then BLAST searches can be greatly narrowed and degenerate primers for full-length cloning and sequencing of the corresponding cDNA can be designed. Finally, improvements in bioinformatics tools such as MSBLASTP2 will increase success rates of protein identification approaches that are based on sequence similarity searches(Shevchenko et al., 2001).

Once protein sets have been identified, they can be used for pathway analysis as described for the SSH gene set above. We have performed such analysis for the set of shark proteins that are common to osmoregulatory tissues and unique to a single tissue. Even though the number of shark proteins used for the analysis (62) was relatively small, it was sufficient to provide some insight into molecular functions associated with osmoregulatory tissues in shark (Lee et al.,2006). Our analysis of molecular functions, cellular pathways and biological processes using PANTHER revealed that stress response proteins such as molecular chaperones and oxidoreductases are enriched and highly abundant in shark osmoregulatory tissues. In addition, we identified the Rho-GTPase pathway of cytoskeletal regulation (PANTHER pathway 00016) as being significantly over-represented in shark osmoregulatory tissues(Lee et al., 2006). Thus, the bioinformatics analysis of protein sets identified by proteomics technology generates specific targets for hypothesis-driven, focused and in-depth follow-up research in non-model organisms.

An important limitation of using bioinformatics approaches for pathway analysis based on identified protein sets in non-model organisms is the use of pathway models and other information from model species as templates for non-model species (see discussion above for SSH pathway analysis). In addition, virtually all proteins have multiple functions that depend on their subcellular compartmentation, posttranslational state and interaction with other cellular constituents. However, currently, functions of proteins stored in databases often do not take into account the aforementioned parameters and,even if they do, information about these parameters is often not available for protein sets of interest. This problem is exacerbated by mixing information about protein functions across species boundaries, which is inevitable when using high-throughput approaches with non-model species. These pitfalls illustrate that, despite the rapid growth of sequencing data, there is still an enormous need for the experimental analysis of proteomes and metabolomes. Even for a single-celled organism, experimentally determining all possible states of posttranslational modification, subcellular compartmentaion and interaction of proteins and correlating those with environmentally induced and developmental cell states may not be feasible. Thus, a narrow focus on a few model species seems warranted for collecting as many data as possible to achieve the ultimate goal of systems biology. Currently, this need for focusing on a handful of model species appears to outweigh the classical advantage of traditional comparative biology, as formulated elegantly by the August Krogh principle: `For many problems there is an animal on which it can be most conveniently studied'(Krebs, 1975). For instance,using stenohaline zebrafish for studying mechanisms of osmotic stress adaptation and euryhalinity in fishes may be counterintuitive. Medaka is better suited for such studies, but extremely euryhaline fishes such as tilapia, killifish or desert pupfish are the organisms of choice from a physiological perspective even if they are considered non-model species from a genomic perspective. Future breakthroughs in high-throughput technologies and bioinformatics, in combination with a reductionist focus on a single problem(or biological process) that lends itself exceptionally well for study in a non-model species, may provide a solution for optimally combining the strengths of traditional comparative biology and systems biology.

A significant advantage of identifying sets of genes and proteins associated with a particular biological process (e.g. salinity adaptation or osmoregulation) in non-model species is that such sets represent well-focused and manageable arrays of elements that can be studied in-depth in the context of a specific biological process. Such follow-up studies may include reverse genetics assays (using siRNA, overexpression, etc.), biochemical assays and imaging methodologies that provide information about the regulation and function of each gene/protein involved in osmotic stress adaptation or any other biological process of interest. In the following, we briefly outline selected examples of biochemical and histological approaches for characterizing proteins associated with salinity adaptation.

Posttranslational protein modification in response to osmotic stress can be studied by using phospho-specific or other state-specific antibodies(Kültz et al., 1997; Kültz and Avila, 2001) or MS following protein separation by 2DGE or 2DLC(Dihazi et al., 2005; Salih, 2005; Valkova and Kültz, 2006). For the latter approach, protein abundance is a limiting factor. If protein sets have previously been identified by a proteomics approach, this is not a problem because all identified proteins exceed the detection threshold for 2DGE or 2DLC. However, when further studying gene sets identified by transcriptomic approaches, this limitation may represent a serious obstacle.

Another property of proteins that critically controls protein regulation and function is their association with other proteins, nucleic acids and micromolecules (inorganic ions, organic osmolytes, metabolic intermediates). The interactome (sum of bound macro- and micromolecules) can be determined for each of the proteins previously identified to be involved in salinity adaptation. Protein–protein interactions can be revealed using genetic or biochemical approaches (Cusick et al.,2005; Monti, M. et al.,2005; Suter et al.,2006). Genetic approaches require a cell line and knowledge about strong promoters that may not be available for many non-model organisms. Therefore, biochemical approaches may be easier to adapt to non-model organisms. However, most of them rely on high-quality antibodies. If not already available, such antibodies can be generated for a limited set of proteins of interest. Once an antibody is available, co-immunoprecipitation of proteins followed by separation of protein complexes by 2DGE or 2DLC and MS can be used to identify interacting proteins. Chromatin immunoprecipitation and variations of this method can be used to identify nucleic acid sequences that bind a protein of interest. Finally, using these approaches,salinity-dependent changes in the interactomes of osmotic stress response proteins can be identified.

The tissue-specific, cellular and subcellular compartmentation of osmotic stress response proteins can also be studied using genetic and biochemical approaches (Giltnane and Rimm,2004; Suter et al.,2006). Again, biochemical/histological approaches for studying protein localization seem more amendable to non-model organisms than genetic approaches. However, like biochemical methods used for studying the interactome, they also rely to a great extent on high-quality antibodies. In particular, high-throughput immunohistochemical techniques such as tissue microarrays are promising tools for large-scale studies of protein compartmentation. We have recently used this technique to localize Na⁺/K⁺-ATPase to chloride cells in gills of FW- and SW-adapted tilapia (Lima and Kültz,2004). This approach can be extended to other proteins involved in salinity adaptation once appropriate antibodies are available.

High-throughput transcriptomics and proteomics technologies are powerful tools for identifying sets of genes and proteins associated with salinity adaptation and the osmotic stress response. They can be effectively used in`non-model' organisms, although some technical improvements are required to further increase success rates of protein identification. These more holistic and global approaches have the advantage of being less biased and more informative with regard to the biological process of interest when compared to a reductionist single-gene/protein approach that requires prior knowledge about which genes/proteins are involved.

Sets of genes or proteins associated with salinity adaptation can be effectively identified in non-model organisms as illustrated by the SSH and proteomics examples outlined above and by recent cDNA microarray experiments that are based on normalized cDNA libraries from non-model species or on cross-species hybridization (Gracey et al., 2001; Podrabsky and Somero, 2004; Buckley et al.,2006). A more severe barrier for work with non-model species is the reliance on models of biochemical pathways, gene ontology databases and other information that is largely based on data obtained with model species. Although many biochemical pathways and biological processes are evolutionarily highly conserved, a great amount of plasticity is inherent in cellular metabolic and signal transduction networks. We would expect such plasticity to be particularly pronounced when a non-model organism differs in its properties, with regard to the biological process under study, from model organisms. For instance, how valid are models of biochemical pathways/networks based on sets of osmotic stress response proteins for a euryhaline fish when modeled against templates based on information available for stenohaline zebrafish? Even though the individual elements (proteins) may be evolutionarily highly conserved, their posttranslational modification,interaction, compartmentation and function during salinity stress may differ substantially. Currently, there are no alternatives to this approach, but this area undoubtedly requires further study.

In our opinion, the most promising and feasible avenue for merging strengths of traditional comparative biology with strengths of systems biology approaches is to select a single biological process and focus on it (rather than attempting to describe the system as a whole) using an organism selected according to the August Krogh principle. Gene and protein sets associated with the biological process of interest (e.g. salinity adaptation) can then be identified. Next, tools for studying the regulation (including posttranslational modification, interaction and compartmentation) and function of these proteins (e.g. antibodies, cell lines, siRNA) should be generated for a manageable set of proteins. Then, the regulation and function of these proteins can be analyzed in detail using such tools. Finally, models describing the biological process of interest can be generated based on the resulting data. This avenue of research depends on extensive collaboration and, therefore, represents an exciting path for the future.

This project was supported by grant number IOB-0542755 from the National Science Foundation (NSF), UCMexus-CONACYT, an NIA from the MDIBL, and grant number 5 P42 ES004699 from the National Institute of Environmental Health Sciences (NIEHS), NIH. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the NSF or NIEHS, NIH.