Proteins that look like enzymes but aren't

This essay is taken from the introduction to my thesis, which is pretty much the only part of my thesis I was pleased with. It described way it's not always easy to assign a function to a protein and was used as the basis for my first paper Moonlighting enzymes in parasitic protozoa. At some point I would like to write a version that is more accessible to the lay reader (with diagrams and shorter) as it demonstrates evolution leave vivid traces in genomes.

Proteins With The Hallmarks Of Enzymes

Proteomic and genomic data provide us with a wealth of information, for example, facilitating the identification of proteins within an organelle. For instance, one can identify proteins within flagella that exhibit significant sequence similarity to enzymes. However, one must be cautious in how we interpret this information. Initially, it may seem logical to think that if a protein sequence has significant sequence similarity to that of an enzyme then the protein will function as that enzyme, but there are an increasing number of proteins where such a conclusion would be incorrect. Many proteins have recently been identified which display the hallmarks of an enzyme, but have another, distinct function. In some cases, these other functions exclude the predicted enzymatic function, in other cases, both enzymatic and non-enzymatic functions are carried out by the same protein. This phenomenon is not limited to proteins produced by alternative splicing, DNA rearrangements or gene-fusions, or to multi-domain proteins.

Moonlighting enzymes

One explanation for a protein having strong sequence similarity to an enzyme while having a non-enzymatic function is that the protein is in fact a bona fide enzyme but one with a so-called “moonlighting” role [1]. These moonlighting proteins function as enzymes under some conditions, but have one or more additional roles under different conditions. Multi-domain proteins and fused proteins are different from moonlighting proteins because though they can have multiple roles, the regions they use for the different activities are distinct. Additionally, moonlighting proteins tend not to exhibit their functions simultaneously [2]. There is an ever-increasing number of examples of moonlighting proteins and it is becoming apparent that moonlighting may represent the norm for enzymes. As an example, seven of the ten enzymes in the glycolytic pathway have been shown to possess additional functions [3]. Of these, phosphoglucose isomerase (PGI), glyceraldehyde-3-phosphate dehydrogenase (GAPDH) and enolase are particularly promiscuous.

Looking further at the example of GAPDH, this enzyme has traditionally been considered a housekeeping enzyme, catalysing one of the reactions of glycolysis in the cytoplasm. However, in the last 25 years GAPDH has been implicated in numerous cellular processes in diverse subcellular locations [4]. On membranes, GAPDH catalyses membrane fusion, [5, 6] and binds the inositol 1,4,5-triphosphate receptor, modulating the release of calcium from the endoplasmic reticulum through the generation of NADH [7]; a single amino acid substitution in GAPDH disrupts endocytosis [8] In the microsomes of skeletal muscle, GAPDH is capable of autophosphorylation and phosphorylating other proteins [9, 10]. Interestingly, when an enzyme named uracil DNA glycosylase (UDG), which repairs DNA damage caused by cytosine deamination was first isolated from nuclei in human placenta, it was discovered that the protein sequence was identical to that of GAPDH [11]. It was subsequently found that GAPDH has the same level of UDG activity as UDG extracted from human placenta. GAPDH has also been found in the nucleus promoting gene transcription [12, 13], exporting tRNA [14] and maintaining telomeres [15].

GAPDH has also been implicated in inducing apoptosis on its translocation to the nucleus in response to oxidative stress [16, 17]. Translocation of GAPDH to the nucleus is thought to be a result of nitric oxide-induced S-nitrosylation of GAPDH, which allows it to bind a nuclear localisation sequence-containing protein Siah1 [18]. It has been speculated that this role for GADPH is important in neurodegenerative diseases [19]. Exposing cortical neurons to the amyloid b peptide found in plaques from brains of Alzheimer's disease patients, causes GAPDH to become disulphide-linked and to accumulate in the nucleus [20]. GAPDH has also been found localised to Lewy bodies, aggregates of proteins most commonly associated with Parkinson's disease [21] and bound to the disease products of Huntington's disease [22]. This hints at a role in these diseases, though it may simply be that the ability of GAPDH to bind so many proteins results in its frequent occurrence in protein aggregates.

GAPDH has also been demonstrated to bind diadenosine tetraphosphate [23], RNA polymerase [24], microtubules [25], and mRNA [26]. Binding mRNA, thus altering its stability and translation efficiency appears to be a common feature of enzymes [27]; the best characterised example is cytosolic aconitase, which turned out to be identical to a protein previously named iron-responsive element-binding protein (IRE-BP). This protein binds a conserved hairpin structure in mRNAs, altering translational efficiency in an iron-dependent manner [28]. Among the specific interactions with RNAs, GAPDH binds an AU-rich region of the 3' UTR of the glucose transporter GLUT1 mRNA [29], an iron-response element in the 5'UTR of hepatitis A virus RNA [30] and a U-rich element in the 3'UTR of human parainfluenza virus [31].

Enzymes with a structural role

Another role for GAPDH and for several other enzymes has been discovered in the lens [32]. Lens crystallins are a well documented example of moonlighting enzymes, where the phenomenon has been referred to as “gene-sharing” [33, 34]. Crystallins make up about 80% - 90% of the water-soluble protein in a lens [35] and by differential accumulation of these proteins in the layers of lens fibre cells light entering the lens is refracted [36]. Historically, crystallins presented a puzzle because while they are highly conserved within some taxa, when studied more generally, evolutionarily diverse animal taxa are found to possess very different crystallin proteins [37, 38]. It appears therefore, that the specialised role of refraction strongly constrains crystallin evolution, and yet many different proteins are able to fulfil the role.

The puzzle was compounded when sequence comparisons revealed that crystallins are similar, and in some cases identical, to small heat shock proteins and enzymes [39, 33]. For example ε-crystallin, specific to bird and reptile lenses is similar to lactate dehydrogenase [40], whereas τ-crystallin from turtle lens is similar to α-enolase [41]. This suggests that the catalytic activity of the proteins is not important for their function in the lens. These proteins turned out, in many cases, to be examples of moonlighting enzymes or “gene-sharing”: ε-crystallin is actually encoded by the gene for lactate dehydrogenase B4 and τ-crystallin is the encoded by the gene for α-enolase. There is no difference in how the genes are spliced or in how the proteins are post-translationally modified. The proteins are expressed at low levels in non-lens tissues for an enzymatic role and at high levels in the lens in a structural role; evolution has altered the regulatory sequences that determine gene expression rather than the coding sequence of the gene itself.

However, not all lens crystallins with homology to enzymes are moonlighting enzymes. In some cases organisms have a gene for an enzyme, which is expressed ubiquitously, at a low level, and a second gene for a homologous lens crystallin, which is expressed only in lens-tissue and at a high concentration. These crystallin genes may encode active or inactive enzymes. For example, ducks and chickens have two d-crystallin genes, arranged in tandem and both homologous to arginosuccinate lyase [42]. d1-Crystallin is inactive whereas d2-crystallin is the active enzyme. Interestingly, in chickens, d1-crystallin makes up 99% of the d-crystallin expressed in the lens, but in duck lens, d1-crystallin and d2-crystallin are expressed in about equal amounts in the lens [43]. The squid Loligo opalescens provides a more extreme example: it has at least 24 S-crystallins [44], each homologous to glutathione S-transferase. All but one of these crystallins has an insertion in the coding sequence and numerous amino acid substitutions, and are thus inactive [45].

By way of contrast, there are the ubiquitous a-crystallins found in all vertebrate lenses. These proteins are encoded by two genes: aA and aB, the result of a gene duplication that occurred at least 500 million years ago [46]. aA-Crystallin is highly expressed in lens tissue and only expressed at a low concentration in a few other cells whereas aB-crystallin is constitutively expressed in many tissues and is stress-inducible [47]. But both aA and aB encode functional molecular chaperones, despite aA-crystallin being highly specialised for lens tissue [48, 49, 50].

There is some debate over whether the lens crystallins that have retained a catalytic activity use that catalytic function in the lens. For example, it has been suggested that the generation or interaction with substrates, such as NADPH can stabilise proteins [51] or even filter UV light [52, 53]. Many crystallins are similar to enzymes involved in stress responses or detoxification pathways [54, 55] and there is evidence that α-crystallin functions as a small heat shock protein to prevent protein aggregation in the lens [48, 56]. However, it seems unlikely that different catalytic activities of the various enzymes employed as crystallins are required in the lenses of different species. In any case, the concentrations of these enzymes in the lens is almost certainly much more than would be required for any catalytic role. Instead, it is thought, that the enzymes in the detoxification pathways were highly expressed in the ancestral lens tissue, refracting light to a small degree which evolution could improve on [55].

Enzymes with a regulatory role

Lens crystallins provide examples of active enzymes or inactive enzyme homologues having a purely structural role. Surveys of genomic data increasingly identify proteins with significant sequence similarity to enzymes, but which lack critical catalytic residues, and are later found to have adopted regulatory roles [57]. One common way of regulating enzymes is by having an inactive regulatory subunit bind modulate enzyme activity allosterically. For example, PKA forms an inactive heterotetramer of two catalytic and two inactive subunits in the absence of cAMP. When cAMP is present, it is bound by the regulatory subunits causing them to dissociate from the catalytic subunits, which are then free to bind and phosphorylate other protein substrates [58]. In some cases, the regulatory subunit of an enzyme has turned out to be highly homologous to the enzyme itself, just lacking a few key residues in the catalytic core.

For example, T. brucei has an inactive paralogue of S-adenosylmethionine decarboxylase (AdoMetDC), which lacks a catalytic cysteine residue. However, this protein stimulates the activity of bona fide AdoMetDC by 1,200 fold [59]. AdoMetDC catalyses a step in the polyamine biosynthesis pathway; polyamines are vital for cell growth and thus inhibiting AdoMetDC can clear T. brucei infection [60]. This trypanosome-specific mechanism of regulating the pathway could therefore provide an effective drug target. Similarly, ADP-glucose pyrophosphorylase, an enzyme that regulates synthesis of polysaccharides, is regulated in plants by a catalytically inactive homologue. In this case however, the inactive subunit functions by modulating the effect of allosteric effectors [61]. The active and inactive subunits form a heterotetramer, so by altering the relative expression of the two types of subunit, plants can create tetramers with different allosteric properties [62].

The inactive subunit of ADP-glucose pyrophosphorylase in the potato Solanum tuberosum is so similar to the bona fide enzyme that it can be converted into an active enzyme by making two amino acid substitutions [63]. A similar experiment has been carried out with Gal3p in Saccharomyces cerevisae. Gal3p is highly homologous to galactokinase (Gal1p), but is in fact the transcriptional activator of the GAL genes (which includes Gal1p) [64]. Gal3p does not function as an active enzyme, but inserting just two amino acids from Gal1p into Gal3p converted Gal3p into a galactokinase [65]. This modified protein retains its ability to induce the GAL genes. The result of this experiment is perhaps unsurprising given that in Kluyveromyces lactis, Gal1 is both a galactokinase and an inducer of the genes for galactose metabolism [66]; it is a moonlighting enzyme. The difference between S. cerevisae and K. lactis seems to be that about 100 million years ago, S. cerevisae underwent a whole-genome duplication event [67], providing the cell with two copies of what had been a moonlighting enzyme. The two proteins were then able to diverge, each specialising in a single role. The evolution of moonlighting proteins is described in more detail in section 1.4.4

Another group of proteins for which inactive versions have been identified are the protein kinases. Protein kinases are a diverse family of enzymes that are involved in regulating almost every cellular process. Analyses of proteins encoded in genomes from humans [68], mice [69], Caenorhabditis elegans [70], Dictyostelium [71] and yeast [72] has revealed that 2-3% of eukaryotic genes encode a protein with a protein kinase domain. However, about 10% of the proteins found in these analyses lack at least one conserved catalytic residue in the kinase domain, so have been labelled pseudokinases [73]. Pseudokinases are thought to be inactive and are found throughout the various protein kinase subfamilies. Many have homologues from humans to yeast, suggesting that pseudokinases evolved from different active proteins kinases and have been conserved [68].

Since pseudokinases have been conserved from yeast to man, it seems likely that they have a significant function. A function has now been demonstrated for some of these pseudokinases: they generally bind true protein kinases and regulate their activity [73]. For example STE20-related adaptor proteins (STRAD) a and b share sequence similarity with STE20 family protein kinases but lack a DFG motif and catalytic arginine residue found in most active protein kinases. It has been demonstrated experimentally that STRAD a and b are not catalytically active [74]. Instead, they function by binding the kinase domain of a tumour suppressor protein kinase, LKB1, increasing the activity of LKB1 by more than 100 fold and anchoring it the cytoplasm. It has been speculated that LKB1 activation is a result of STRAD proteins (and a scaffolding protein MO25) stabilising LKB1 in an active conformation [75]. Interestingly, STRAD a has retained the ability to bind ATP and ADP, though this has no apparent effect on its ability to activate or anchor LBK1 [76].

Janus kinases (Jaks) are an example of proteins that contain both an active kinase domain (the JH1 domain) and pseudokinase (JH2) domain. Jak2 is a receptor tyrosine kinase that associates with cytokine receptors and is involved in transducing growth signals [77]. Mutations in the JH2 domain are associated with leukaemia and haematopoietic disorders [78, 79]. The JH2 domain interacts with the JH1 domain [80]. Deletion of, or mutations in, the JH2 domain deregulate signalling through Jak2 [81, 80]. This, and structural evidence suggest that the role of the pseudokinase domain is to bind and inhibit kinase activity in the absence of cytokines [82]. It has been suggested that the three regulatory nucleotide-binding domains of the DHC, are not catalytically active, but involved in the ATP-dependent inhibition of the protein ATPase activity described in section 1.2.2 [83].

Mammalian Type I and Type III hexokinase (HK) isoforms use a similar regulatory mechanism. HK catalyses the first reaction of glycolysis, and in most eukaryotes it is inhibited by the products of its reaction, namely ADP and glucose-6-phosphate. Mammalian Type I and III HKs have two monomer-like HK domains fused together, of which, only the C-terminal domain is catalytically active; the N-terminal domain is inactive, but regulates the C-terminal domain by binding glucose-6-phosphate [84].

Early experiments on HK in T. brucei found no evidence that the enzyme is regulated [85]. Instead, it was thought that glycolysis in T. brucei is regulated by the compartmentalisation of the first enzymes of glycolysis into a specialised organelle called the glycosome [86, 87]. However, recent experiments suggest that HK activity in T. brucei is regulated, and it is regulated through the interaction between different HK domains. It had been known that HK in T. brucei forms oligomers of up to six units, while in most eukaryotes HK is either monomeric or dimeric [88]. The functional significance of this was not understood until after sequencing of the T. brucei genome, when two HK genes were found in tandem. The proteins produced from these genes, TbHK1 and TbHK2, are expressed in both PCF and BSF parasites and are 98% identical [89]. Recombinant TbHK1 has been shown to have HK activity in vitro and RNAi of TbHK1 results in a reduction of HK activity in vivo [89, 90]. Recombinant TbHK2, conversely, was not shown to exhibit HK activity, and while RNAi of TbHK2 causes loss of HK activity in BSF, it causes an increase of HK activity in PCF [90, 87]. Further experiments with recombinant TbHK1 and TbHK2 showed heterohexomers have approximately three times the HK activity per polypeptide than oligomers of TbHK1 alone [91]. However, it is not simply that the apparently inactive TbHK2 regulates TbHK1 because an inactive variant of TbHK1 is actually able to stimulate activity in TbHK2. TbHK1 and TbHK2 regulate one another, and myristate, the most potent known inhibitor of HK activity in T. brucei, functions by dissociating the oligomers [91].

TbHK2 demonstrates how apparently inactive enzymes, may be active in vivo. This is important to bear in mind, even when bioinformatics suggests that key catalytic residues are absent. It has also been shown that some pseudokinases function as true kinases with the lack of otherwise conserved catalytic residues compensated by changes elsewhere in the protein [92]. For example, Ca2+/CaM-activate serine-threonine kinase (CASK) has been labelled a pseudokinase as a result of lacking key catalytic residues required to coordinate magnesium ions. However, it is catalytically active under physiological conditions [93]. Similarly, WNK (with no lysine (K)) protein kinases are so-called because they lack a key lysine residue normally found on the third b-sheet, binding ATP. However, a crystal structure of WNK1 revealed that a lysine residue from the second b-sheet entered the active site instead [94]; WNK1 is in fact able to phosphorylate a protein called synaptotagmin2 and influence its activity [95]. Choline and aminoglycoside kinases also compensate for the lack the same lysine residue [96].

A protein with demonstrable serine/theronine protein kinase activity is the anti-tumour protein known as CC3 or TIP30, which can phosphorylate itself and the heptapeptide repeats of the largest subunit of RNA polymerase II [97]. Bioinformatic studies [98] and then the crystal structure [99] demonstrated the CC3 is structurally homologous to the short-chain dehydrogenase/reducatase family, particularly UDP-galactose epimerase. The active site residues for reductase activity appear to be present in the CC3 structure, but in the incorrect orientation. It is not clear whether the protein has a reductase activity, but it is able to bind NADPH [99].

NmrA is a negative regulator of the transcription factor, AreA, which is involved in nitrogen metabolite repression in fungi. NmrA is also structurally similar to the short-chain dehydrogenase/reducatase family, but lacks a conserved catalytic tyrosine residue, so is unlikely to be a functional dehydrogenase [100]. It is however, able to bind NAD, though there is no known biological requirement for this activity. Similarly, carboxyl-terminal binding protein (CtBP) is a transcriptional co-repressor with homology to NAD-dependent 2-hydroxy dehydrogenase. It is enzymatically active, although this activity does not appear to be required for its co-repressor function [101]. It has been proposed that the ability of CtBP to differentially bind NAD and NADH would allow it to modulate its control of transcription in response to the cell's redox state [102]. Interestingly, these hydrogenase/reductase-like proteins bind nucleotides such as NAD with a Rossman fold, which is the fold in GAPDH that binds AU-rich RNA [103]. These proteins may therefore have conserved an enzymatic fold in order to bind RNA for a regulatory purpose, thus giving them a superficial similarity to metabolic enzymes.

Why do enzymes accumulate new functions and when do genes duplicate?

Traditionally, proteins are thought of as having a single, well defined role for which they are specifically adapted to. However, evolution works by adapting what is already present (see Shubin et al. 2009 for a recent review). It has been suggested, that the most common mechanism by which a protein with a novel function is generated begins with gene duplication. This provides the cell with a gene that can evolve a divergent function, free from the pressures of natural selection [104, 105]. It has been demonstrated that the yeast cells that evolve to cope with the loss of an essential motor protein are those that first duplicate entire chromosomes [106].

However, extra copies of genes are likely to be lost from a population unless there is a selective pressure to maintain them, and yet a lack of selection pressure is required for the gene to diverge [107]. Instead, it has been proposed that a protein first gains a new function after a small change on its surface, which creates an additional binding site, or by a change in where the protein is expressed that allows it to interact with a pre-existing binding site. Evolution can then select for ever-increasing activity in this new function. However, advantageous changes will be limited because many mutations will be detrimental to the protein's original function. It is only then that there will be sufficient pressure for a gene duplication event. In many cases this is likely to lead to the two genes diverging and specialising for the separate function. One of the proteins may lose any enzymatic function but its enzymatic hallmark will be preserved as a molecular fossil. Another model of protein evolution is the duplication-degeneration-complementation model [108]. In this model, duplicated genes are preserved if they acquire loss-of-function mutations or lose regulatory element regions, which are complemented for by the other gene, which has itself lost different functions.

The transcriptional activator of GAL genes in yeast shines some light on protein evolution. As described above, K. lactis has a single protein that functions as both a galactokinase and a transcriptional activator [66], while S. cerevisiae has two proteins: Gal3p which induces transcription, and Gal1p which is a galactokinase (but is also able to induce transcription) [109], thought to be the result of a whole genome duplication event that occurred in S. cerevisiae. An assessment of yeast fitness with various combinations of promoter elements demonstrated that duplication of the ancestral bifunctional gene in S. cerevisiae relieved an adaptive conflict; the genes and promoters could subsequently evolve in different directions, leading to loss of galactokinase activity in Gal3p and allowing the two genes to be expressed at different levels [110].

It seems therefore that evolution is only likely to maintain duplicated genes, if that gene already has multiple functions. This then presents the question: why is it that enzymes in particular have accumulated so many functions? One answer is simply that enzymes are abundant and therefore likely to come into contact with many of the other proteins of the cell. For example, GADPH is known to be an abundant and 'sticky' molecule, so binds many other proteins. Hence, the relevance of GAPDH being found bound to protein aggregates in various neurodegenerative diseases is unclear [19]. The fact that enzymes come into contact with many proteins makes them prime candidates for evolving new functions, by improving binding interactions or providing a scaffold on to which new regulatory functions can be built. Enzymes tend to have a relatively small active site, but often require a large structure to correctly orientate catalytic residues within the active site. This results in proteins with large surfaces that are mainly unused [2].

In the case of protein kinases, the ability to interact with many other proteins is required as part of their function. They are therefore ideal candidates for taking on new regulatory roles, independent of their ability to phosphorylate proteins. This may explain the existence of numerous pseudokinases, which often regulate other protein kinases by binding the activation loop, but not phosphorylating it [73]. Another common interaction partner of enzymes is the enzyme itself (i.e. they form oligomers). The cell therefore has an opportunity to evolve new regulatory mechanisms should the gene for such an enzyme duplicate because hetero-oligomers with different proportions of subunits can then be assembled. This appears to have occurred with the enzyme potato ADP-glucose pyrophosphorylase [61].

Another reason for enzymes acquiring additional roles is that they frequently bind co-factors, such as ATP or NADH. This property could potentially allow them to alter local substrate concentrations. For example GAPDH modulates Ca2+ release from endoplasmic reticulum by binding and releasing NADH, which in turn activates inositol 1,4,5-triphosphate receptors [7]. Another reason for a protein maintaining an ancestral substrate-binding site is that the enzymatic fold required for binding a substrate can be used to bind other nucleotides, just as the NAD-binding Rossman fold in GAPDH binds AU-rich mRNA [103]. There is also evidence that binding nucleotides stabilises proteins [51, 111]. There are in fact, a relatively small number of stable proteins folds, so evolution is forced to recycle the same few motifs repeatedly [112].

Finally, there are the crystallins, where it seems that the abundance and stability of an enzyme is the property that has been exploited for a new role. The fibre cells of the lens lose their nuclei and most of their other organelles, so damaged or denatured proteins cannot be replaced [32], and the crystallin protein must exist at high concentrations without aggregating. This problem appears to have solved through the accumulation of mutations in cis- and trans-regulatory elements, rather than in the gene itself [113]. Proteins with the hallmark of enzymes may therefore have a multitude of catalytic, regulatory and structural roles independent of their ability to catalyse the reaction associated with their homologues.

  1. Ramasarma 1994
  2. Jeffery 1999
  3. Sriram et al. 2005
  4. Sirover 1999
  5. Morero et al. 1985
  6. Glaser & Gross 1995
  7. Patterson et al. 2005
  8. Robbins et al. 1995
  9. Kawamoto & Caswell 1986
  10. Engel et al. 1998
  11. Meyer-Siegler et al. 1991
  12. Morgenegg et al. 1986
  13. Zheng et al. 2003
  14. Singh & Green 1993
  15. Sundararaj et al. 2004
  16. Ishitani et al. 1996
  17. Sawa et al. 1997
  18. Hara et al. 2005
  19. Hara et al. 2006
  20. Cumming & Schubert 2005
  21. Tatton 2000
  22. Burke et al. 1996
  23. Baxi & Vishwanatha 1995
  24. Mitsuzawa et al. 2005
  25. Kumagai & Sakai 1983
  26. Ryazanov 1985
  27. Cieśla 2006
  28. Kennedy et al. 1992
  29. McGowan & Pekala 1996
  30. Chang et al. 1993
  31. De et al. 1996
  32. Piatigorsky 1998
  33. Piatigorsky & Wistow 1989
  34. Jeffery 2003
  35. Piatigorsky 1981
  36. Benedek 1983
  37. Piatigorsky 1984
  38. Lubsen et al. 1988
  39. Ingolia & Craig 1982
  40. Hendriks et al. 1988
  41. Wistow et al. 1987
  42. Piatigorsky et al. 1988
  43. Wistow & Piatigorsky 1990
  44. Tomarev et al. 1995
  45. Tomarev et al. 1992
  46. de Jong et al. 1993
  47. Sax & Piatigorsky 1994
  48. Horwitz 1992
  49. Ganea 2001
  50. Horwitz 2003
  51. Rao et al. 1992
  52. Bateman et al. 2003
  53. Bando et al. 2007
  54. Piatigorsky et al. 1994
  55. Jong et al. 1989
  56. Borkman et al. 1996
  57. Pils & Schultz 2004
  58. Krebs & Beavo 1979
  59. Willert et al. 2007
  60. Bacchi et al. 1992
  61. Ballicora et al. 2004
  62. Crevillén et al. 2003
  63. Ballicora et al. 2005
  64. Bhat et al. 1990
  65. Platt et al. 2000
  66. Meyer et al. 1991
  67. Wolfe & Shields 1997
  68. Manning et al. 2002
  69. Caenepeel et al. 2004
  70. Plowman et al. 1999
  71. Goldberg et al. 2006
  72. Hunter & Plowman 1997
  73. Boudeau et al. 2006
  74. Baas et al. 2003
  75. Boudeau et al. 2003
  76. Boudeau et al. 2004
  77. Ihle 1995
  78. Cools et al. 1999
  79. Ihle & Gilliland 2007
  80. Luo et al. 1997
  81. Shashidharan et al. 1999
  82. Lindauer et al. 2001
  83. Yoshimura et al. 2007
  84. Cárdenas et al. 1998
  85. Nwagwu & Opperdoes 1982
  86. Bakker et al. 2000
  87. Albert et al. 2005
  88. Misset et al. 1986
  89. Morris et al. 2002
  90. Morris et al. 2006
  91. Chambers et al. 2008
  92. Kannan & Taylor 2008
  93. Mukherjee et al. 2008
  94. Min et al. 2004
  95. Lee et al. 2004
  96. Kannan et al. 2007
  97. Xiao et al. 2000
  98. Baker 1999
  99. El Omari et al. 2005
  100. Stammers et al. 2001
  101. Kumar et al. 2002
  102. Fjeld et al. 2003
  103. Nagy & Rigby 1995
  104. Ohno 1970
  105. Kimura & Ohta 1974
  106. Rancati et al. 2008
  107. Bergthorsson et al. 2007
  108. Force et al. 1999
  109. Lohr et al. 1995
  110. Hittinger & Carroll 2007
  111. Sellick et al. 2009
  112. Caetano-Anollés & Caetano-Anollés 2003
  113. Piatigorsky 2003

Post new comment

The content of this field is kept private and will not be shown publicly.