We applied the NMT predictor for glycine myristoylation to taxonomic subsets of publicly available databases (SWISS-PROT, GenBank). The NMT prediction methodology is described in detail elsewhere 16. Then, we first removed redundancy within our predictions using the program mcd-hit 1920 with a 40% amino-acid identity threshold. This procedure already results in a dramatic reduction of the prediction lists to a much smaller set of representative sequences (for example, from approximately 4,400 sequences to approximately 2,000 for the eukaryotic subset). So as not to lose the information in the removed sequences, they were assigned to groups according to their clustering by mcd-hit. The conservative threshold of 40% amino-acid identity allows the interpretation that sequences within the clusters are homologous to their representative sequence but also results in the appearance of more remotely related sequences in separate clusters. The size of these clusters was used to rank the representative sequences in the tables in the database and is linked to view the full set of sub-tables listing all cluster members.

We have also estimated the taxonomic distribution of sequences within a cluster, as occurrence in a larger set of organisms might reflect the functional need for conservation of the motif during the aeons of evolution. Special emphasis is given to the known experimental status of myristoylation to distinguish new predictions. Therefore, we indicate the annotation of a cluster member for myristoylation in the SWISS-PROT 21 database below the cluster size and add further details to the sequence description. Besides this short description of the protein, the predictions are linked to their corresponding GenBank entries, an RPS-Blast with CD-search 22 to view the domain architecture and PSI-Blast 23 to collect the set of homologous sequences.

Then, the position in the sequence of the glycine that is predicted to be myristoylated (position 2 often implies prior cleavage of a leading methionine) and the full-length motif follow. We highlight cysteines that could be subject to palmitoylation in yellow and positive charges by a blue background. Both factors can not only influence membrane affinity but also induce changes in subcompartment-specific localizations 5. Finally, the score assigned by the NMT predictor (yielding details of the single score components on mouse-over), the estimated probability of false-positive prediction and a simple prediction attribute are listed.

Entries in MYRbase can be accessed by simple browsing or entering the corresponding MYRbase identifier on the index page (MYRbase identifiers correspond to GenInfo numbers of entries from the latest database built). Alternatively, we recommended using our MYRbase-specific BLAST-engine (also available from the index page) to be linked directly to the cluster of the best hit of the query protein with MYRbase entries.

When analyzing single eukaryotic genomes of human, mouse, fly, worm, yeast and plant, we found a conserved relative occurrence (around 0.5-0.8% of all proteins in the genome) of predicted myristoylated proteins, with an elevated percentage for Arabidopsis 16 that is also supported by combined theoretical/experimental data for this organism 24. In our genome-spanning analysis, we find that the most prominent and taxonomically widespread groups of myristoylated proteins comprise GTP-binding proteins (several alpha subunits of G proteins and ADP ribosylation factors), serine/threonine kinases, calcium-binding EF-hand proteins with diverse functions (for example, recoverin, calcineurin B, guanylate cyclase activators) and tyrosine kinases (see Table 2 for complete listing).

It is interesting that for proteins with conserved myristoylation there are a few families with a large number of members, and a large number of families with only a few members. An interpretation of this situation, which is reminiscent of a power-law distribution (see Figure 1 and legend), in evolutionary terms might be that the few larger families are the oldest cases of eukaryotic protein glycine myristoylation. The many smaller families, on the other hand, appear to be examples of functional specialization, as they are often confined to taxonomic subgroups. Most of the larger families with a conserved motif include proteins with experimentally verified myristoylation (Table 2). Hence, future research will focus on the role of the predicted lipid anchors for the specialized smaller groups.

We do not only observe myristoylation of proteins specific to taxonomic subgroups but also examples of myristoylation of proteins from some species that have multiple orthologs in other species that lack the motif for this lipid modification. In several of these cases, the myristoyl anchor has been substituted by other factors that can facilitate membrane association, such as other lipid anchors (palmitoyl, farnesyl, geranylgeranyl), transmembrane regions or specific protein-lipid or protein-membrane protein interaction domains. Such substitutions can have occurred not only in orthologs and paralogs but even in isoforms of the same protein.

For example, several G_α subunits have either a myristoyl-, both myristoyl- and palmitoyl-, or only palmitoyl-anchors 25. While the Arabidopsis Rab5 ortholog Ara7 is geranylgeranylated on carboxy-terminal cysteines just like Rab5 in other species, its closely related paralog Ara6 lacks the carboxy-terminal cysteines but has an experimentally verified amino-terminal myristoylation motif instead 26. Furthermore, it has been shown that an artificially introduced transmembrane helix can substitute for acyl-mediated membrane targeting of eNOS 27. While most phospholipase C-gamma's are expected to be targeted to membranes by amino-terminal pleckstrin homology domains (PH, which bind phospholipids), a phospholipase C in Trypanosoma cruzi utilizes a myristoyl-anchor instead 28. Indeed, myristoylation was also able to substitute for PH domain-mediated membrane targeting of the scaffolding protein Gab1 29.

Diverse shuffling, similar to that of globular domains, of several membrane-attachment factors (MAFs) within homologous proteins apparently occurred during evolution (by substitution, addition or deletion). It should be noted, however, that most of these changes also imply drift in targeting specificity or strength, and that the contributions and importance of single MAFs for membrane attachment might differ.

The notion of evolutionary modularity should not be misunderstood as a common ancestry for all sequence segments comprising myristoylation motifs. On the contrary, sequence-based comparisons appear to favor a model of multiple independent evolution (for example, for the emergence of those single motifs; data not shown). The modularity we describe here is that targeting specificity is achieved by combining different modules for membrane attachment in arrangements that are not fixed but can change dynamically during evolution. The evolutionary steps required for these changes from one targeting motif to the other are most likely to involve intermediates that contain several of the interchangeable factors.

Such evolution has apparently led to lipid-anchor-mediated targeting to a wide range of subcellular localizations. The role of glycine myristoylation is not limited to the classical attachment of proteins to the plasma membrane. There, several myristoylated proteins are found to target cholesterol and sphingolipid-enriched detergent-resistant compartments, designated as lipid rafts 30. In addition, the myristoyl anchor is involved in targeting proteins to membranes throughout the cell, ranging from endoplasmic reticulum (ER), Golgi and mitochondrial membranes to the nuclear envelope and even to extracellular localization (see Table 2). Extracellular localization is the result of an alternative myristoylation- and palmitoylation-dependent export mechanism identified in Leishmania and possibly conserved to higher eukaryotes 31.

Specificity and strength of targeting can be achieved by optimizing lipid interactions by the type and number of lipid anchors (for example, acyl (myristoyl, palmitoyl) 6, prenyl (farnesyl, geranylgeranyl) 3233 and GPI 3435), possibly aided by additional features (such as clusters of positive charges 36), but also by specific protein-lipid (for example, PIP₂ 37) and protein-protein (for example, PDZ domain 38) interactions.

As several factors in addition to myristoylation contribute to subcellular localization and membrane attachment, we have analyzed their distribution within our sets of known and predicted myristoylated proteins. We distinguish five major classes that are defined in Table 3 and depicted in Figure 2. From our analysis, subsequent cysteine palmitoylation (class I) and clusters of positive charges (class II) seem to occur in comparable amounts in both known and predicted subsets, while there is a clear tendency for the remaining classes (class III-V) to be more common among the new predictions. The prediction of the myristoylation of eukaryotic transmembrane proteins (class IV), in particular, will have to be verified in the future, although it seems generally reasonable, as viral examples exist 3940 and the very similar acyl modification with palmitoyl anchors is commonly found in combination with transmembrane regions 41.

We also expect that final targeting specificity will be determined by not one but a combination of several MAFs, and the subcellular localization will remain difficult to predict, especially in regard to the involvement of protein-protein interactions (class V). Reversibility of membrane attachment can be regulated by depalmitoylation, disruption of the cluster of positive charges through introduction of negative charges (phosphorylation, deamidation of asparagine to aspartate), a pH-dependent switch when there are histidines within the positive charge clusters, or conformational changes due to proteolytic cleavage, binding of small molecules or altered protein-protein interactions.

From Table 2 we can see that the functional spectrum of experimentally verified myristoylated proteins is much more diverse than their well-known involvement in signaling. However, complete analysis of eukaryotic sequences in GenBank suggests that, after subtraction of known examples and their homologs, a significant number of proteins remain whose function has not been implicated with myristoylation so far (Table 1). To investigate the spectrum of these predictions with functions 'new' to myristoylated proteins, we have systematically analyzed the domain composition of a nonredundant (90% identity) subset of the respective proteins (HMMer 42 against PFAM 43; E-value 0.01) and summarize the results in Table 4 (ranked by the number of occurrences of domains and corrected for repeated occurrence of the same domain in one entry). As a rule, the proteins associated with domain hits in Table 4 do not belong to a unique protein family but to several families that differ in their overall domain architecture. Consequently, we also analyzed the set of new predictions in respect of clearly defined protein families, and summarize them in Table 5. This list only includes families with at least 10 different proteins that, furthermore, begin with a methionine in their sequence (for exclusion of possible non-amino-terminal fragments). Not surprisingly, as myristoylation of remotely related family members has already been established, we find additional examples of kinases, phosphatases and phosphodiesterases. Several 'new' predictions will be discussed in the following sections. If not listed in Tables 5 or 6, a MYRbase identifier (MI) that equals the GenInfo number of the predicted entry in the analyzed database will be given to access the corresponding entries in MYRbase.

We also picked a few interesting predictions from MYRbase and investigated the capacity of the corresponding amino-terminal peptides (Table 6 and Additional data file 1) to be myristoylated. Our in vitro experiments with synthetic peptides and bacterially expressed NMT (see Materials and methods) confirm the computational results for specific immunoglobulin μ-heavy and λ-light chains, 47 kDa GTPase IIGP, ubiquitin hydrolase Ubq-M, lung cancer candidate FUS1, potassium channel Kir2.1 and potassium channel interacting protein KChIP1. Hence, we show that our predictor recognizes the capability of substrates to productively interact with NMT. However, it is clear that in vitro myristoylation of synthetic peptides only gives limited information about the situation of the full-length protein in vivo. Some positively predicted substrates might be myristoylatable in vitro, but whether they come into contact with NMT in vivo depends on the cellular context and needs to be proven or at least shown to be plausible.

For example, MYRbase also includes a series of specific immunoglobulin chains. By our in vitro myristoylation experiments, we show for a μ-heavy and a λ-light chain (Table 6, and Additional data file 1) that peptides derived from the amino termini of the corresponding database entries are in principal agreement with the physicochemical requirements for productive substrate-NMT interaction as modeled by our predictor.

It is unclear whether the predicted entries only represent non-amino-terminal fragments in the database. Differing sequence constructs for that region could be attributable to V(D)J recombination 44. However, we could not find convincing expressed sequence tag (EST) evidence for variants with a starting methionine just before the predicted glycines. Proteolytic cleavage is an alternative possibility for glycines to become amino-terminal (for example, BID 45 and several viral proteins). Immunoglobulin chains are normally targeted to the ER via signal peptides and the existence of NMT activity in the ER has been proposed 2. None of the predicted cleavage sites 46 of immunoglobulin precursors after the removal of the signal peptide appeared to result in an amino-terminal glycine. However, a different subcellular targeting mechanism, as well as alternative cleavage sites 47 and cleavage by other proteases, cannot be excluded. Myristoylation at the predicted sites may occur in vivo only if the sequences, starting with glycine as they appear in the database, can interact with NMT within the cell.

Early work exists describing in vivo incorporation of radiolabeled myristate into the same specific immunoglobulin chains as those predicted 48. In this light, the possible alternative of myristoylation of an internal lysine (a theoretical suggestion of Pillai and Baltimore 48) has to be addressed experimentally. Reports of internal lysine myristoylation are rare in the literature 4950. These potentially myristoylated immunoglobulin chains should have a yet-unknown cellular function (possibly, in a cell-type and cell-state-dependent manner) 4851 in which the lipid anchor influences subcellular targeting.

Plant-specific disease-resistance proteins, with the nucleotide binding motif NB-ARC and leucine-rich repeat domains (for protein-protein interactions), are prominent in Table 5. While this manuscript was in preparation, octapeptides derived from the amino terminus of respective proteins were shown to be myristoylatable by plant NMT 24. Myristoylation of those proteins that are involved in the innate immune response 52 fits well into the scheme that they act through interaction with avirulence proteins that are injected into the plant host cell through the type III secretion system of the parasitic bacteria 53. The lipid anchor may facilitate co-localization since a series of type-III secreted bacterial proteins are also myristoylated by host plant NMT 4.

In the immune response of mammals to intracellular pathogens, expression of families of 65 kDa and 47 kDa GTPases is induced by interferon-gamma 54. In this work, we show in vitro myristoylation (Table 6 and Additional data file 1) of the amino terminus of a human homolog of 47 kDa GTPase IIGP 55. This protein seems to be the only family member, besides very close homologs, that has a myristoyl anchor. Lipid modifications are quite common within the large superfamily of GTP-binding proteins, for example, Ras and Ras-related as well as Rho and Rab small GTPases are most often either farnesylated or geranylgeranylated 32. Also, members of the 65 kDa family of the interferon-inducible GTPases 54 share the CaaX box motif that most likely directs their geranylgeranylation. In the case of the 47 kDa GTPase IIGP, a myristoyl-anchor could be involved in localizing the protein to membranes of the ER and Golgi 5556. In the context of myristoylation and mammalian innate immune responses, it is noteworthy that interferon-inducible, double-stranded (ds) RNA-activated 2'-5'-oligoadenylate synthetase isoform 2 is also known to be myristoylated 57.

The amino terminus of the human ubiquitin-specific protease Ubp-M is myristoylated by NMT in vitro (Table 6, and Additional data file 1). This enzyme is suspected to be involved in the deubiquitination of histone 2A, cell-cycle regulation and caspase-dependent apoptosis 5859. However, the observations were made with a protein amino-terminally tagged with green fluorescent protein (GFP), which questions the functional importance of a putative amino-terminal myristoyl anchor. Interestingly, only a GFP-tagged mutated inactive version was localized to the nucleus, while the GFP-tagged protein lacking inactivating mutations was found in the cytoplasm 58. This indicates that several factors influence its localization, and it cannot be ruled out that a myristoylated amino terminus would result in a different localization pattern, possibly also targeting the 'active' wild-type protein to its actual place of action.

In addition to Ubq-M, we predict myristoylation for another group of ubiquitin-specific proteases conserved in rat, fly, worm and plants (for example, MI 27683149). These proteins are substantially shorter, lack additional domains (such as an amino-terminal zinc finger in Ubq-M), and most probably belong to the ubiquitin carboxy-terminal hydrolase subfamily 60. Furthermore, we predict that the Caenorhabditis elegans cytokinesis defect protein 3 61 has a myristoyl anchor followed by a calcium-binding EF-hand and an ubiquitin hydrolase homology domain (MI 16950531).

It might be interesting to check whether the predicted lipid modification for some of these ubiquitin-specific proteases could result in subcellular targeting similar to that of a 26S proteasome variant whose regulatory subunit 4 was predicted 16, and subsequently shown to be, myristoylated in vivo 62.

In this context, it is also interesting that we predict myristoylation for a group of proteins containing putative RING zinc finger domains (Tables 4 and 5) that are commonly found in ubiquitin ligases. Although RING finger domains do exist in some proteins known to be myristoylated (for example, rapsyn, see Table 2), most of the hits with this top-ranking domain in Table 4 seem to have 'new' functions in the context of myristoylated proteins. For example, a subgroup of the RING finger proteins are isoforms of the Notch pathway 63 protein neuralized 64, with a conserved myristoylation motif in human, mouse, frog, fly and worm (for example, MI 15128197). Another example of a RING domain-containing protein with conserved myristoylation motif in human, mouse, fly and worm is mahogunin (MI 27229238), which is involved in spongiform neurodegeneration 65. Also, the myristoylation motif of the RING-containing peripheral nerve injury gene nin283 (for example, MI 14150005) is conserved in human, mouse, fly and worm.

Furthermore, several F-box-containing proteins, parts of which could act as receptors for ubiquitination targets, are found within our set of proteins predicted to be myristoylated (for example, MI 6912466 and MI 7299840).

As well as proteins involved in apoptosis or the mitochondrial respiratory chain that are known to be myristoylated (Table 2), we also predict amino-terminal myristoylation of a homolog of TOM40, a protein thought to be a central component of the mitochondrial import machinery 66, presumably acting as a pore-forming 67 sorting station 68. Unexpectedly for a proposed all-beta integral membrane protein of the outer mitochondrial membrane 69, sequences from human, mouse, rat, frog and fly reveal a conserved predicted myristoylation site at the amino terminus (Table 5).

Interestingly, the amino termini of cytochrome c-type heme lyase (CCHL) homologs in worm, fly, zebrafish, frog, mouse and human (for example, MI 7512486) also share a conserved predicted myristoylation motif. It should be noted that, in the human sequence, a proline at motif position 4 is rather unfavorable in terms of flexibility of the substrate to adopt the proper conformation in the NMT binding pocket. Fungal homologs also have amino-terminal glycines. Although possible myristoylation cannot be excluded, three large aromatic residues that are difficult to accommodate in the narrow NMT substrate-binding pocket follow these conserved glycines, which might also suggest a new, as yet unknown, role for the amino-terminal glycine conservation. CCHL catalyzes the conversion of apocytochrome c to holocytochrome c by attaching a heme group 70, which is an important step required for cytochrome c biogenesis and transport through the membranes 71. Deletion of the mammalian gene encoding CCHL is associated with X-linked microphthalmia with linear skin defects syndrome 72.

We also show in vitro myristoylation of the amino terminus of human FUS1 (Table 6 and Additional data file 1), an apparent tumor suppressor. This protein is missing or carboxy-terminally truncated in lung-cancer cells 7374. The exact mechanism of inactivating its function as a tumor suppressor is unknown, but is most likely to be independent of the myristoylation status as the truncated protein form would still contain the predicted myristoylation motif. It is possible that CpG methylation downstream of the unmethylated 5' promoter region 73 could account for inactivation, as the CpG island appears to extend over the whole coding region (as tested with methods described in 75). For example, in the case of the HIC1 tumor suppressor, CpG methylation was found to occur predominantly in introns and exons instead of the 5' promoter region. This methylation appears to be characteristic for cancer types and stages 76.

Myristoylation has been predicted for specific modulatory proteins interacting with voltage-gated potassium channels 77. In this work, we demonstrate in vitro myristoylation of the human representative of the KChIP1 subgroup of Kv4 channel-interacting proteins (Table 6 and Additional data file 1). The myristoylation motif is conserved in mouse, rat and human and, in support, there exists homology to hippocalcin, frequenin and neuronal calcium sensor proteins 77, whose myristoylation has already been shown experimentally to be of functional importance 78. However, amino-terminal truncation experiments with KChIP1 suggest that, in the investigated system, the modulatory activity is sufficiently represented by the more conserved central domain containing EF-hand Ca²⁺-binding motifs 79. On the other hand, the role of a myristoyl lipid anchor for KChIP1 could also lie in subcellular targeting specificity. Voltage-gated potassium channel isoforms (for example, Kv2.1 and Kv1.5) can localize to distinct lipid-raft populations 80. In the context of the similar targeting specificity of acyl chains 81, it is interesting that isoforms KChIP2 and KChIP3 that lack the predicted myristoylation motif are palmitoylated, and their lipid modification controls the plasma-membrane localization of the associated channels 82.

Furthermore, several mammalian and avian homologs of the strong inward rectifying potassium channel Kir2.1 are predicted to be myristoylated. The fact that the motif is evolutionarily retained over 11 organisms would rather support the functional importance of such conservation. The in vitro results show that the amino terminus of human Kir2.1 can productively interact with NMT (Table 6 and Additional data file 1). None of the other Kirs is predicted to be myristoylated, and a lipid anchor unique to Kir2.1 could be responsible for differing subcellular and submembrane localizations of isoforms, similar to those observed among other potassium channels 8083.

Consequently, it is not surprising that artificial introduction of a myristoylation motif into the isolated cytosolic carboxy terminus of another Kir (Kir3.1) is sufficient to co-localize the fragment with intact channels 84. After synthesis in the ER, carboxy-terminal signals direct the export of Kir2.1 to the Golgi complex 8586. Stockklausner et al. 87 showed that a cluster of positive charges in the amino terminus that are conserved with Kir1.2/4.1 can be held responsible for post-Golgi trafficking of Kir2.1 to the plasma membrane. There, Kir2.1 is localized to lipid rafts 87. It was hypothesized that interaction with phosphoinositides (PIP₂) could be necessary for lipid-raft association but Stockklausner et al. 87 found that mutation of the corresponding residues had no effect on lipid-raft targeting. Other work suggests the involvement of PIP₂ interaction in regulating gating processes 88.

The functional role of the predicted myristoyl anchor for these channel proteins with respect to submembrane localization (lipid rafts), conformational flexibility of the amino terminus or protein-protein interactions remains to be investigated.

Many pieces of evidence suggest that not only the many different membranes in the cell, but also the microcompartments therein, can be distinguished by their lipid and protein compositions 8990. Proteomic analyses of detergent-resistant lipid-raft fractions from different cells 91929394 unveil partly overlapping but also cell-type specific populations of proteins. This includes several known myristoylated proteins that are involved in various signaling pathways (for example, G_α 95, Src-family tyrosine kinases 96, NAP-22 97, endothelial nitric oxide synthase 98, rapsyn 99, annexin XIII 100, MARCKS 101, CAP23 102) and Nef 103 or participate in viral budding (for example, HIV Gag 104).

In addition, we predict myristoylation for two more proteins of unknown function that co-localized with detergent-resistant membrane fractions. Each appears in a separate proteomic analysis of lipid rafts in Jurkat T-cells 9194. The first (161 amino acids, MI 8923579) protein has a conserved myristoylation site plus potentially palmitoylated cysteines in human, rat and mouse homologs. Interestingly, the second protein (227 amino acid, MI 12834233), which has a predicted central coiled coil region, was not only detected in detergent-resistant fractions of T cells in a dynamic, receptor-activation-dependent manner 94, but also in a chaotrope-resistant fraction derived from nuclei of neuroblastoma N2a cells that contains clusters of proteins localizing to the outer nuclear membrane 105. It will be interesting to establish possible dynamic trafficking between distinct subcellular compartments of this protein (with highly conserved orthologs in human, mouse and frog, and more diverged in zebrafish and fly).

Existing proteomic analyses focus on the most abundant protein representatives within membrane fractions and can also only identify a temporally limited subset of dynamically targeted proteins. Therefore, it would not be surprising to find several more myristoylated proteins to participate in highly specific signaling events by reversible trafficking to and from membrane microdomains.

Targeting experiments suggest differences in submembrane localization that depend on the anchor type. For example, membrane partition clustering of non-dimerizing green fluorescent protein variants fused to peptides containing double acylation (for example, myristoyl+palmitoyl) or prenylation (for example, farnesyl, geranylgeranyl) consensus signals was shown to differentially depend on lipid raft disruption by cholesterol depletion 81. However, targeting experiments with a few artificial constructs varying only in lipid anchor type cannot take into account the complexity of the interplay between different MAFs that coexist in real proteins. Hence, it is difficult to generally attribute targeting specificity purely to occurrences of particular lipid anchors. Nevertheless, a myristoyl anchor followed by a palmitoyl anchor seems to be a strong indication for targeting to the plasma membrane and also eventually to lipid rafts.

The amino-terminal eight amino acids of selected predictions (Table 6) were synthesized with two minor modifications. To avoid racemization of the carboxy-terminal cysteine, one serine residue was added to the terminus of MI 4761381. Threonine and valine were added to the octapeptide of MI 5454156, as the dominance of positive charges was expected to result in difficulties in purification and handling. Peptides were purchased from Sigma Genosys (Japan). NMT was prepared as described previously 106 and the cDNA of yeast NMT was a gift from J. Gordon. Peptides were myristoylated in vitro essentially as described in 107 and analyzed by mass spectrometry. MALDI-TOF mass spectrometry was carried out on a Voyager DE Pro (PE Biosystems) and the matrix was 10 mg/ml alpha-cyano-4-hydroxycinnamic acid (Sigma) in 0.1% TFA-50% acetonitrile solution. The spectra were displayed and analyzed using the GRAMS-MS software and the spectra were calibrated using calibration mixture 1 or 2 (PE Biosystems). The spectra are available in portable document format (PDF) from the MYRbase homepage 18. As positive controls, peptides of proteins known to be myristoylated in vivo have been analyzed, as well as corresponding peptides without modification by NMT as negative controls. Differences between theoretical and experimental masses for peptides ending in serine are most likely due to carboxy-terminal sodium salt formation. MYRbase identifiers can be found in the upper left corner of each page.