The idea that some genes have been trapped in organellar genomes stems largely from the idea that the proteins encoded by these genes are difficult to transport back to the organelle for assembly when synthesized in the cytosol. Intrinsic to this idea is that there has been a hierarchical loss of organellar genes, whereby those that were first to be successfully relocated were those encoding proteins that are easiest to transport back, while those that were last to be transferred encode proteins that are difficult to transport back. Many bacterial proteins have or are predicted to have mitochondrial targeting properties, or can be targeted to mitochondria without the acquisition of a targeting presequence, and these are predicted to be the first organellar genes to be successfully relocated to the nucleus 1011. As there seem to be no limitations on the transfer of genetic material, organellar gene loss should, according to this theory, have continued until the cell solved the targeting and assembly problems for all proteins, and organellar genomes would then no longer exist 12. This may yet occur in plants, where transfer of organellar genes to the nucleus continues to erode the organelle genomes 4.

Why, then, has the transfer of genes to the nucleus not gone to completion? The difference in genetic code between mitochondria and the nucleus in eukaryotic organisms other than plants is one plausible explanation for the apparent 'freeze' on gene relocation 13. Regardless of the reason, organellar genomes linger, and they are enriched in genes that encode hydrophobic, membrane-embedded proteins that are difficult to transfer from cytosol to organelle. Notably, even the reduced plastid genome of dinoflagellates is enriched in genes encoding hydrophobic proteins 14. These initial observations instigated the 'hydrophobicity hypothesis', which was first proposed by von Heijne in 1986 15 and was later expanded by others 1617 (see Box 1). Since then there has been a steady accumulation of both bioinformatic observation and experimental evidence suggesting that hydrophobic regions in proteins are a major obstacle to both targeting and import, and that proteins must overcome this obstacle if their genes are to relocate.

The preferential 'maintenance' of a core set of organellar genes is encompassed by the CORR theory (co-location for redox regulation; see Box 1), which was first proposed in 1992 3031. This hypothesis proposes that there is a direct link between coding location and regulation, either transcriptional or post-transcriptional, which gives a fitness advantage compared with nuclear-encoded genes. In this way, expression of a gene within the organelle gives an advantage, thus preventing transfer of the gene to the nucleus. An example often quoted to support the hypothesis is that if more subunits of a protein complex are needed in a particular chloroplast, for example the DI subunit of photosystem II, which might be required because of photo-oxidative damage, it is more efficient to have the gene encoded within the organelle, as nuclear encoding would mean that the protein would be sent even to those chloroplasts that did not require this subunit 32. Although rich in predictions, there is no direct experimental evidence for CORR (for mitochondria) 5, in contrast to several experimental investigations that support the validity of the hydrophobicity hypothesis 516171819202122232425262829.

On the surface, the hydrophobicity hypothesis does not appear to adequately explain the retention of all organelle-encoded genes. Notable flaws are, firstly, that not all protein-coding genes encoded in organelles encode hydrophobic proteins, the most obvious example being the large subunit of ribulose 1,5-bisphosphate carboxylase/oxygenase (Rubisco-LSU) in chloroplasts; and secondly, both mitochondria and chloroplast already import hydrophobic proteins. The mitochondrial carrier family and the light-harvesting protein of the light-harvesting chlorophyll-protein complex are cited examples for mitochondria and chloroplasts, respectively.

Similarly, the CORR hypothesis also has some deficiencies. Firstly, redox control has as yet been demonstrated for only a handful of plastid-encoded genes. Secondly, the expression of many nuclear-encoded mitochondrial and chloroplast proteins is under redox control and yet these proteins are not organelle-encoded 3334; thus, why only some redox-controlled genes must be organelle-encoded is not explained by CORR. And thirdly, even for the redox-regulated components encoded by chloroplasts, the products are functional only when combined with additional nuclear-encoded subunits; thus, being organelle-encoded does not offer any immediate advantage in terms of protein function.

Thus it might be possible that there are a variety of reasons why genes are organellar and the reason for each gene might differ, or even be a combination of a number of different factors. It is worth examining the exceptions to each hypothesis to see whether there is evidence to validate or invalidate the objection. Also, there is a need to question how being chloroplast-encoded and under redox control is an advantage in evolutionary terms.

The hydrophobicity hypothesis centers on the problem of targeting and importing a protein following its synthesis in the cytosol. Although hydrophobic proteins clearly present targeting problems, many organellar genes do not encode hydrophobic proteins. If we look at Rubisco-LSU, as it is the obvious and cited counterexample to the hydrophobicity hypothesis, can it be synthesized in the cytosol and imported to produce a functional protein in chloroplasts? The answer is yes, as this has been successfully achieved in some dinoflagellates 14. In other plants where Rubisco-LSU is normally plastid-encoded the reported attempts to express the gene allotopically have been successful qualitatively but not quantitatively. Although it is possible to express and import the protein, only a small proportion of the wild-type activity can be achieved 35. One reason for this failure might relate to the assembly of the complex. As the holoenzyme of Rubisco is the most abundant protein in a cell, efficient assembly is critical, and specific chaperone systems are involved in this process (for review, see 36). Thus, although it is possible to import the protein, the limitations of assembly - that is, reduced assembly efficiency - may reduce fitness and thus successful gene transfer. An unfolded protein response, a stress-induced pathway, has been described for mitochondria and thus, in addition to the fact that inefficient assembly may reduce fitness, unfolded or unassembled proteins may be degraded and/or may induce stress pathways, as has been demonstrated in mitochondria 3637. It should be noted that as the small subunit of Rubisco is nuclear-encoded, the reason for the failure to express Rubisco-LSU adequately from a nuclear location is unlikely to be due to a gene dosage effect of plastid genome versus nuclear genome.

This assembly concept could be extended further to explain the organellar coding location of other proteins that are not encompassed by the hydrophobicity hypothesis. One common feature of almost all organelle-encoded genes is that the products are assembled into multisubunit complexes that contain at least one other protein - for example the Rubisco holoenzyme - but usually many others, as is evident for the electron-carrying components of the photosynthetic and respiratory chains. Studies on how such complexes assemble indicate that there are ordered sequential assembly pathways. The order of assembly is critical to producing a functional complex, and importantly organelles have specific protease and chaperone systems for degrading proteins that have not assembled correctly 3839.

The extensively studied photosystem II from chloroplasts indicates "a hierarchy in the protein components that allows a stepwise building of the complex" 40. An excellent example is the DI protein, which is encoded by the chloroplast gene psbA. This protein is inserted into the thylakoid membrane in a co-translational manner with the aid of the chloroplast signal-recognition particle, and requires the presence of several other subunits of photosystem II 4041. Studies of the unicellular green alga Chlamydomonas reinhardtii indicate that specific sequences in the 5'-untranslated region of the mRNA bind specific proteins that might define thylakoid membrane targeting 42. Allotopic expression of genetically altered psbA resistant to herbicide demonstrated that the protein product could be imported into chloroplasts but plants were still sensitive to herbicide (albeit less than wild-type plants) 43, possibly as a result of ineffective or inefficient assembly of the cytosolically synthesized protein. This example indicates that assembly may define an organelle-encoded location. An excellent review containing more details of this process is available 44.

The second apparent 'flaw' in the hydrophobicity hypothesis is that both mitochondria and chloroplast import many hydrophobic proteins. Why then should some hydrophobic proteins be resistant to this process? The answer may lie in the protein itself; many nuclear-encoded mitochondrial proteins are imported across the organellar membranes to the matrix, and then rerouted via conserved sorting pathways. This import pathway requires that all but the last transmembrane stretch must pass through the import machinery. If, however, a transmembrane stretch is recognized as a 'stop-transfer' sequence, the import process stops 22, the offending stretch of amino acids is moved laterally into the membrane, and the protein is unable to fold to its active conformation 45. Clearly, all nuclear-encoded organellar proteins have evolved so that their transmembrane stretches do not resemble stop-transfer sequences, enabling polytopic proteins to be easily imported and assembled. Thus it is the subtle signals contained in a transmembrane stretch that can prevent import, something we are not yet able to predict from gene sequence alone.

The often-cited examples of mitochondria and chloroplasts importing hydrophobic proteins do not contradict the principle that assembly can define organelle-coding location. Members of the mitochondrial carrier family, present in the inner membrane, may be hydrophobic but function in homodimeric complexes; thus there is no sequential assembly required 46. The hydrophobic light-harvesting chlorophyll proteins were derived from simpler forms in cyanobacteria that are single-membrane-spanning 4748. These two sets of proteins represent 'eukaryotic' proteins and thus import and assembly pathways were invented de novo by eukaryotic cells. But, when multisubunit protein complexes were derived from the endosymbiotic ancestor, the assembly pathways were dictated.

As outlined above, there is compelling experimental evidence that the targeting and import of some proteins might be the major determinant for their organelle-coding location. But not all organelle-encoded proteins pose targeting and import problems, and it is becoming increasing clear that assembly should be added to the list of difficulties. We therefore feel that the term 'importability' better encompasses the difficulties experienced by some proteins when expressed in the cytosol, and therefore the retention of organellar genomes. The importability concept does not ignore the observations of the CORR hypothesis: rather, the elegant redox regulation of some chloroplast-encoded genes 34 may be a mechanism for ensuring that these organelle-encoded subunits are synthesized in the correct sequential manner, so as to ensure correct assembly. Redox regulation may have specialized to a stage at which it facilitates the ordered assembly of multisubunit complexes and may now represent a barrier to gene relocation.

The importability hypothesis also encompasses the proposal that some products may be toxic if synthesized in a cytosolic location, as has been demonstrated for the apocytochrome b and ND4 proteins 29. An important point to note is that even if, under experimental conditions, allotopic expression of some organelle genes can be achieved and can rescue mutant phenotypes, in evolutionary terms it is the efficiency of import and assembly that can be a selective factor. Thus, reducing the growth rate by achieving allotopic expression may reduce fitness and result in an organellar location for a gene even though a nuclear location can be achieved in the laboratory.

Importability may not be sufficient to explain an organellar coding location for all genes in all organisms. There might be alternative reason(s) why some genes are retained in organelles. The evolution of organellar and nuclear genomes must be a complementary process in cells. Effective cross-talk takes place between nuclear and organellar genomes to coordinate function, as is evident with retrograde regulation for nuclear-encoded genes for mitochondrial and chloroplast proteins 495051. There is also evidence for an additional form of regulation, termed intergenomic communication 5253; this is based on the physical presence and expression of a gene within an organelle genome independent of the function of the encoded protein 54. Furthermore, it appears that mutations in the mitochondrial genome in yeast increase the rate of nuclear mutation 55. Mutations in mitochondrial genomes cause defects or alterations in development in mammalian and plant systems 5657. Thus, perhaps an additional reason that genes are encoded in organelles is that some genes must be encoded there in order for expression of organelle and nuclear genomes to be coordinated. Genes encoding protein products that present additional barriers for successful gene transfer will also most often be observed in organelle genomes. With all the genome information that is now available, care needs to be taken to look at genomes rather than focusing solely on individual genes. This approach may yield insights that would not be possible with single-gene analysis and may provide more inclusive hypotheses for explaining organelle genome maintenance.