Sequence Dependence and Differential Expression of Gγ5 Subunit Isoforms of the Heterotrimeric G Proteins Variably Processed after Prenylation in Mammalian Cells*

Between 1 and 2% of proteins coded for in the human genome, including all G protein γ subunits, are predicted to be prenylated. Subsequently, prenylated proteins are proteolytically cleaved at the C terminus and carboxymethylated. These reactions are generally obligatory events required for functional expression of prenylated proteins. The biological role of prenyl substrates has made these reactions significant targets for anticancer drug development. Understanding the enzymology of this pathway will be key to success for this strategy. When Gγ1, -2, -4, -10, -11, -12, and -13 were expressed in HEK293 cells they were completely processed according to the current understanding of the prenylation reaction. In contrast, Gγ5 was processed to two forms; a minor one, fully processed as predicted, and a major one that was prenylated without further processing. When the Ca1a2X motif of Gγ5, CSFL, was exchanged for that of Gγ2, CAIL, Gγ5 was completely processed. Conversely, Gγ2-SFL was incompletely processed. Differential processing of Gγ5 was found due to the presence of an aromatic amino acid in its Ca1a2X motif. Retrieving endogenous Gγ subunits from HEK293 or Neuro-2a cells with FLAG-Gβ constructs identified multiple Gγ subunits by mass spectrometry in either cell, but in both cases the most prominent one was Gγ5 expressed without C-terminal processing after prenylation. This work indicates that post-prenylation reactions can generate multiple products determined by the C-terminal Ca1a2X motif. Within the human genome 10% of predicted prenylated proteins have aromatic amino acids in their Ca1a2X sequence and would likely generate the prenylation pattern described here.

One to two percent of proteins coded for in the human genome are thought to be substrates for prenylation (1)(2)(3)(4)(5)(6). These include a number of proteins affecting cell growth, including the Ras and Rho proteins, numerous other small GTPases, many phosphatases and kinases, the nuclear lamins A and B, and all of the G␥ subunits of the heterotrimeric G proteins involved in cell signaling. The role of prenylation in pro-tein function appears to be multiple and complex; it affects membrane localization, intracellular trafficking, and proteinprotein interactions. In general, prenylated proteins require correct processing to mediate their biological functions (7)(8)(9)(10).
Protein prenylation is catalyzed by one of three prenyl transferases that adds either a farnesyl or a geranylgeranyl group to a Cys residue four amino acids from the C terminus of target proteins (1,5). The prenyl group transferred is primarily, but not exclusively (11), determined by the C-terminal amino acid of the protein where a geranylgeranyl group is transferred in the case of Leu and sometimes Phe, and a farnesyl is generally transferred in the case of Ser, Met, Asn, Ala, and sometimes Cys (5,12). Following prenylation, the last three amino acids are cleaved by a specific protease (13). In yeast, there are two prenyl proteases, Rce1 and Afc1/Ste24 (13), each with fairly broad specificity (14). In vertebrates, Rce1 appears responsible for most proteolysis of prenyl proteins (15,16), whereas the specificity of the homolog of Afc1/Ste24, Zmpste24, is more limited, targeting in particular prelamin A (17). Finally, the new C terminus of the protein is carboxymethylated (5,6) and mediated by a specific carboxymethylase (18).
The involvement of prenylated proteins in the regulation of cell growth, their relatively low number, and the specialized enzymology of this modification have resulted in all three steps in this enzymatic cascade being targeted for production of anticancer drugs (5,19,20), as well as for treatment of other diseases where prenylation or prenylated proteins play a role (21). Prenyl transferase inhibitors are currently in Phase III clinical trials, with both advantages and disadvantages. They appear to be promising for treatment in a number of tumors, but they have lower efficacy, particularly for solid tumors, than first envisioned (19), and alternative strategies have been suggested for their use, such as adjunct treatment with other cancer drugs (22). One issue with the further development of these drugs is that their relevant substrates have turned out to be ambiguous and difficult to determine. For example, it is uncertain that the Ras proteins for which the drugs were originally designed are related to the pharmacological effects of these drugs (23). Another issue is that the complex enzymology involved seems to include effects of inhibitors causing a switch in the biological targets of different enzymes (24).
An alternative strategy for future therapies using prenyl transferase inhibitors is the development as adjunct therapy drugs targeting the carboxymethylation and proteolysis steps (5). The mammalian Rce1 enzyme that catalyzes proteolysis of the C terminus of prenyl proteins is a unique metalloprotease primarily targeting prenylated proteins (25). In general, it is thought to target most or all naturally occurring prenylated proteins, and a recent review identified no significant or important exceptions to this (5). Deletion of this enzyme is lethal during embryonic development or shortly after birth (15), indicating the importance of this reaction. Undoubtedly, the successful development of Rce1 as a therapeutic target, as has proved true for the farnesyl/geranylgeranyl transferase inhibitors (5,19,23,24), will depend upon correctly understanding the substrate specificity and diversity of Rce1. Previously, we found that G␥5 associated with purified bovine brain G proteins is largely unprocessed by either proteolysis or carboxymethylation after prenylation (26). The significance of this observation has remained unclear, however, and this processing pattern has been perceived as likely explained by isolation of a precursor or an abnormally processed protein. Here, we have investigated the origin and significance of this unprocessed protein after expression of G protein subunits in mammalian cells.

EXPERIMENTAL PROCEDURES
Materials-FLAG-G␥ and FLAG-G␤ constructs in pcDNA3 were recently described (27) and corresponded to human (␥ 1 , ␥ 4 , ␥ 12 , and ␥ 13 ), bovine (␥ 2 ) or rat (␥ 5 ) sequences. FLAGtagged ␥ 5 and ␥ 2 were further modified using the QuikChange kit (Stratagene) to generate chimeras by swapping the three C-terminal residues of each isoform: G␥ 5 -AIL and G␥ 2 -SFL. Finally, a series of FLAG-tagged ␥ 5 constructs were generated with QuikChange substituting codons for all combinations of aromatic amino acids into the aliphatic positions within the Ca 1 a 2 X motif ( Table 1). The sequence of all constructs was verified by complete DNA sequencing of the final insert.
Immunoprecipitation-Frozen cell pellets were thawed on ice before addition of 600 l of lysis buffer (50 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA, 2.5 mM MgCl 2 , 10 M GDP, 5 mg/ml DNase I, 10 mg/ml RNase A, 1ϫ protease mixture mix 1 (Calbiochem), 1% cholate) and incubation for 1 h at 4°C in a rotating mixer. Lysate was then centrifuged for 15 min (16,000 ϫ g, 4°C), and the supernatant transferred to a new tube. FLAG monoclonal antibody-conjugated agarose beads (Sigma, 10 l of 50% slurry) were added, and the mixture was incubated for 3 h while rotating at 4°C. Beads were precipitated by 10-s centrifugation at 300 rpm in a microcentrifuge, the supernatant removed, and the beads were washed twice with cold Tris-buffered saline (50 mM Tris-HCl, 150 mM NaCl) and twice with cold H 2 O. Bound proteins were eluted with 70% acetonitrile/0.1% trifluoroacetic acid (50 l).
MALDI-MS-Eluted FLAG-tagged proteins were concentrated by vacuum spin drying to Ͻ1 l volume and resuspended with 5 l of 50% acetonitrile/0.1% trifluoroacetic acid. The matrix (sinapinic acid, 10 mg/ml) was solubilized with acetonitrile-trifluoroacetic acid. Protein mass standards (Calibration mixture 3, Applied Biosystems) were mixed with the matrix, and both matrix alone and matrix with calibrant were pre-spotted for each sample and allowed to fully air dry. Each protein sample (1 l) was mixed with matrix (1 l), and 0.5 l of protein-matrix mixture was applied to both the matrix and calibration mixture dried spots to obtain spectra with internal and external calibration. MALDI 2 was performed on an Applied Biosystems Voyager DE-STR Biospectrometry workstation in linear mode. Calibration was performed using the two closest neighbors bracketing the mass of the sample. Observed values are reported with mass to charge ratio (m/z) of singly protonated proteins (MϩH) ϩ .
Immunoblots-One microliter of concentrated eluted protein sample was dried down and resuspended in SDS sample buffer. Samples were separated by SDS-PAGE on 8 -16% gradient Tris-HCl gels (Bio-Rad Criterion gel) and transferred to 0.1 m of nitrocellulose. Membranes were incubated with FLAG polyclonal antibody (1:5,000, Sigma) for 1 h, washed four times with Tris-buffered saline/Tween 20, and incubated with

Primers for FLAG-G␥ C-terminal mutations and chimera
Indicated are the forward primers used for site-directed mutagenesis using QuikChange to prepare G␥2 and G␥5 C-terminal chimera or variable Ca 1 a 2 X sequences in G␥5. Target codons for site-directed mutagenesis are underlined. Reverse primers were the reverse complements of the forward primers. Also indicated are the parent G␥2 and G␥5 sequences for the corresponding regions.

RESULTS
Mass spectrometry (MS) has proven to provide an effective approach for characterizing the post-translational modifications of the G␥ subunits of the heterotrimeric G proteins puri-fied from brain (11, 28 -30). Here we have used MS to characterize the processing of expressed and endogenous G␥ subunits in cultured cells. G␥ subunits were expressed as FLAG-tagged proteins in HEK293 cells and isolated by immunoprecipitation on anti-FLAG beads after detergent solubilization of cell pellets. Although expression and recovery after FLAG immunoprecipitation were variable, as evaluated on FLAG immunoblots ( Fig. 1, IP), all the expressed proteins generated robust MALDI-TOF MS signals (Fig. 2). Samples from cells expressing FLAG-G␥1 ( Fig. 2A), FLAG-G␥2 (Fig. 2E), FLAG-G␥4 (Fig.  2B), FLAG-G␥12 (Fig. 2C), and FLAG-G␥13 (Fig. 2D) all generated MS spectra containing a single very prominent mass in the 8-to 10-kDa mass range characteristic of singly charged FLAG-G␥ proteins. In contrast, precipitates from control pcDNA-transfected cells generated no appreciable signal in this mass range (data not shown). The average masses observed in repeated independently isolated samples were characteristic of the G␥ isoform expressed, had errors (S.D.) of 1.6 Da or less, and were within 1.1 Da or less (113 parts per million; range 45-190 ppm) of the predicted mass of the correct subunit isoform modified as expected from past studies of prenylated proteins ( Fig. 2). Thus, FLAG-G␥1 had a mass compatible with that of a farnesylated protein, whereas all of the others had masses compatible with geranylgeranylated proteins, as predicted by their C-terminal Ca 1 a 2 X sequence. Additionally, all of the masses were compatible with those of a FLAG-tagged protein with an acetylated N terminus.
In contrast to the other G␥ isoforms, the most prominent mass after expression of FLAG-G␥5 was substantially different from that predicted according to accepted understanding of prenyl processing. The predicted FLAG-G␥5 should have a mass of 8294.6, but the most prominent signal had an m/z ratio of 8628.0 Ϯ 1.1 (Fig. 2G, peak a). This corresponds to the pre-FIGURE 1. FLAG immunoblot for expression and recovery of FLAG-G␥ proteins expressed in HEK293 cells. HEK293 cells were transfected with FLAG-G␥ cDNAs as described under "Experimental Procedures," which included co-expression with untagged ␣ i1 and ␤ 1 . Detergent lysates were prepared and used to recover FLAG-tagged proteins. Equivalent aliquots (1 l) of tagged proteins eluted from (mouse) Anti-FLAG Immunobeads were processed for SDS-PAGE and immunoblotting as described under "Experimental Procedures." A polyclonal rabbit antisera was used for immunoblotting. FIGURE 2. MALDI-TOF spectra of FLAG-G␥ isoforms expressed in HEK293 cells. HEK293 cells were transfected as described in Fig. 1 with FLAG-G␥ cDNAs. Detergent lysates were prepared and used to recover FLAG-tagged proteins for MALDI-TOF MS, as also described in Fig. 1. Masses reported are based on measurements with internal calibrants and are reported as mean Ϯ S.D. of data from three independent experiments. Also indicated in each panel is the structure of the predicted or most probably modified protein corresponding to the major peak in the spectrum. The cDNAs expressed corresponded to FLAG-G␥1 (A), FLAG-G␥4 (B), FLAG-G␥12 (C), FLAG-G␥13 (D), FLAG-G␥2 (E), and FLAG-G␥2-SFL (F). The most prominent mass (peak a) corresponded, within Ͻ1 Da, to Ac-FLAG-G␥2-(gg)-SFL without proteolytic loss of the three terminal amino acids. The other masses detected corresponded to: peak b, the unacetylated version of peak a with predicted mass of 9176.6 and observed mass of 9176.0; peak c, the acetylated protein expressed without further processing with expected mass of 8937.3 and observed mass of 8938.1; peak d, the fully processed Ac-FLAG-G␥2-(gg)-OCH3 with predicted mass of 8876.3 and observed mass of 8876.8; and peak e, an expressed Ac-FLAG-G␥2 that has been cleaved prior to the prenylated Cys with predicted mass of 8486.6 and observed mass 8488.5. G, FLAG-G␥5. Mass in peak a corresponds to the previously observed FLAG version of G␥5 without proteolytic processing of the prenyl signal and with the structure Ac-FLAG-G␥5-Cys(gg)-SFL. Peak b corresponds to the G␥5 processed as predicted and with observed mass of 8294.4 and predicted mass of 8294.6. H, FLAG-G␥5-AIL. In the structural notations: Ac, N-acetylation; far, farnesyl; and gg, geranylgeranyl. dicted mass of 8628.0 for the FLAG-tagged version of a G␥5 protein prenylated with a geranylgeranyl moiety but lacking the expected C-terminal proteolysis and carboxymethylation. Such a G␥5 protein was previously characterized in bovine brain G proteins (26). Just as in that case, there was a minor peak ( Fig.  2G, peak b) at 8294.4 m/z that was compatible with the protein processed as predicted. The similarity of the spectrum in Fig.  2G to that previously found for bovine brain G protein G␥5 suggests that differential proteolysis of the C terminus of G␥5 is a characteristic of the processing of this protein in mammalian cells. An initial hypothesis was that this processing pattern depended directly on the C-terminal Ca 1 a 2 X prenylation sequence of the protein. To test this possibility, we expressed a chimeric FLAG-G␥5 protein containing the Ca 1 a 2 X sequence from G␥2 (CAIL) instead of its parent sequence (CSFL). This chimeric FLAG-G␥5-AIL protein generated a single prominent peak (Fig. 1H) characteristic of the predicted processed G␥5 protein, which was geranylgeranylated with proteolysis of AIL and carboxymethylation of the new C terminus. In contrast, expression of a chimeric FLAG-G␥2-SFL construct generated multiple peaks (Fig. 2F), the most prominent of which (peak a) was within 1 Da (33 ppm) of the geranylgeranylated protein lacking C-terminal proteolysis and carboxymethylation. Another mass (peak d), as with G␥5, was compatible with correctly processed G␥2, suggesting that the -SFL sequence of G␥5 is weakly processed by Rce1.
Whereas expressed FLAG-G␥5 has a primary mass indicative of prenylation without further processing (Figs. 2G and 3A), the FLAG-G␥5-AIL chimera is processed normally (Figs. 2H and 3B). This indicates that the G␥5 Ca 1 a 2 X sequence determines the susceptibility of the protein to proteolysis by Rce1. G␥5 has the hydrophilic amino acid Ser at position a 1 of its Ca 1 a 2 X sequence instead of an aliphatic amino acid. To determine if this Ser determines susceptibility to proteolysis we tested a mutant FLAG-G␥5-AFL construct (Fig. 3C), which was also found to be insensitive to proteolysis by Rce1. Thus, the presence of a hydrophilic amino acid at position a 1 does not seem to account for differential processing by Rce1; a conclusion supported also by the observation that two other G␥ subunits, ␥ 4 and ␥ 13 , which are processed normally (Fig. 2, B and  D), contain a threonine in that position. More significantly, the Ca 1 a 2 X sequence of G␥5 differs from that of all other G␥ subunit isoforms in that it contains an aromatic amino acid at position a 2 (CSFL) instead of a canonical aliphatic amino acid. In contrast to unmodified G␥5 (Fig. 3A) ending in -SFL, the -SIL mutant (Fig. 3D) was correctly processed, as was the protein ending in the native G␥2 -AIL sequence (Fig. 3B). This indicates that it is the Phe in the G␥5 Ca 1 a 2 X motif that is responsible for poor proteolytic processing by Rce1. Replacing the Phe at a 2 Post-prenylation Processing of G␥5 MAY 11, 2007 • VOLUME 282 • NUMBER 19 with either Tyr (Fig. 3E) or Trp (Fig. 3F) also generated a G␥5 protein resistant to Rce1 proteolysis. Finally, the effect of an aromatic amino acid in the Ca 1 a 2 X motif was not unique to the a 2 position but also resulted from any aromatic amino acid in the a 1 position (Fig. 3, G-I). Thus, any aromatic amino acid (Phe, Tyr, or Trp) at either the a 1 or a 2 position in the Ca 1 a 2 X sequence resulted in expression of a protein that was prenylated without further C-terminal processing. Interestingly, most (but not all) of these point mutants showed more complete suppression of the C-terminal processing than was even found in the case of the unmodified -SFL sequence. Most, in fact, lacked any evidence of production of a processed protein.
The data in Figs. 2 and 3 were generated with G␥ constructs containing a FLAG epitope tag. To substantiate that the results obtained were not dependent upon the epitope tag, an alternative expression strategy was used whereby untagged G␥ constructs were co-expressed with FLAG-G␤1 (Fig. 4). In the case of G␥2, once again, FLAG immunoprecipitation generated a single prominent peak with a singly charged m/z of 7750.1 consistent with the predicted N-and C-terminal modifications (Fig. 4C) previously observed for native bovine brain G␥2 (31). Co-expression of G␥5 with FLAG-G␤1 resulted in recovery of a primary peak of 7502.1 corresponding to prenylated G␥5 without further C-terminal processing, but with correct N-terminal processing (Fig. 4C), as reported previously (26). A second, less prominent peak, was also observed at 7168.7 that corresponded to the fully processed G␥5 with a predicted mass of 7168.4, as also had been previously observed (26). In fact, all of the remaining G␥ constructs characterized in Fig. 2, when expressed as untagged proteins in the presence of G␤1, generated a primary mass very consistent with that predicted for both correct N-and C-terminal processing (Fig. 4C). These results substantiate correct processing of multiple forms of expressed G␥ isoforms and their functional formation of a G␤␥ dimer based upon co-immunoprecipitations with FLAG-G␤1.
Although the data in Fig. 4 indicated preferential association of expressed FLAG-G␤1 with co-expressed G␥ constructs, the spectra obtained often had minor masses that seemed to be compatible with predicted masses of processed G␥ isoforms. When FLAG-G␤ isoforms were expressed without an associated G␥ construct in HEK293 cells, mass spectra of FLAG immunoprecipitation samples showed numerous peaks in the G␥ subunit range, at least for G␤1-G␤4 (Fig. 5, A-D). In contrast, as reported by others (32) and as found in vitro (27), FLAG-G␤5 failed to form stable dimers with any proteins in this mass range (Fig. 5E). The distinct signals found associated with G␤1-G␤4 were generally consistent with one another and had average masses compatible with expression of multiple G␥ proteins processed as predicted, including, in particular, G␥5, G␥10, G␥11, and G␥12 (Fig. 5G). Of these, the one that appeared to be differentially associated with G␤ isoforms was G␥11, which appeared to be associated with G␤1 and G␤4 (Fig.  5, A and D) but not with G␤2 and G␤3 (Fig. 5, B and C). In addition, there were also weaker but consistently observed signals compatible with G␥7, G␥8, and three masses that did not correspond to any known human G␥ variant. Most notable about these experiments, however, in all of the samples (Fig. 5,  A-D), the most prominent signal was that corresponding to the unprocessed G␥5-SFL protein characterized as the primary variant of expressed G␥5. In all cases too, however, there was a mass characteristic of the fully processed G␥5 protein, generally at lower intensity.
To determine the generality of the approach used in HEK293 cells to retrieve endogenous G␥ proteins with FLAG-G␤ constructs, similar experiments were carried out with Neuro-2a cells, a mouse neuroblastoma cell line (33). In this case, a signal corresponding to G␥2 was found that was not present in HEK293 cells and is indicative of the neural origin of these cells (33). Even in this case, however, the other signals identified corresponded to those for G␥5 and the most prominent signal FIGURE 4. MALDI-TOF spectra of expressed untagged G␥ constructs recovered from cells co-transfected with FLAG-G␤1. HEK293 cells were transfected as described under "Experimental Procedures" with untagged G␥ constructs and FLAG-G␤1, and detergent lysates were prepared and used to recover FLAG-G␤1 and associated G␥ proteins for MALDI-TOF as described under "Experimental Procedures." Masses reported are based on measurements with internal calibrants and are reported as mean Ϯ S.D. of data from three independent experiments. A, MALDI spectrum for cells transfected with G␥2 and FLAG-G␤1; B, MALDI spectrum for cells transfected with G␥5 and FLAG-G␤1; and C, summary of data from cells transfected with FLAG-G␤1 and different G␥ isoforms. Structural notation is as described in the legend of Fig. 1.  MAY 11, 2007 • VOLUME 282 • NUMBER 19 was for the G␥5-SFL variant. These results indicate that variable processing of G␥5 is a general phenomenon in mammalian cells, that the proteolytically unprocessed form is the predominant variant expressed in cells and that, at least as assayed by MALDI-MS, G␥5-SFL is the most prevalent G␥ isoform and variant observed.

Post-prenylation Processing of G␥5
To estimate the potential biological significance of such a variant processing pattern, the size of the pool of potentially differentially processed protein was determined. Prenylated proteins were identified in the NCBI annotated human protein data base from Build 34.3 using a Perl script matching a C-terminal Cys-a 1 -a 2 -X motif, where a 1 or a 2 is an aromatic amino acid instead of an aliphatic one and X is a Lys or Phe for geranylgeranylation or Ser, Cys, Met, Asn, or Ala for farnesylation. The human reference protein data base contained 27,970 protein sequences, 852 of which had a Cys at position Ϫ4 from the C terminus. Of these, 167 would be predicted to be geranylgeranylated and 250 farnesylated, based upon their C-terminal amino acid. Of these predicted prenylated proteins, 17 containing geranylgeranyl and 27 farnesyl, or ϳ10% of the total, contained an aromatic amino acid at a 1 or a 2 ( Table 2).

DISCUSSION
The studies reported here were designed primarily to elucidate the requirements of the prenylation pathway for variable processing of the G␥5 subunit isoform. In addition, however, these studies have established a number of other findings related to G␥5 or to general questions about the expression, processing, and function of G␤ and G␥ subunits in intact cells. These observations include the following: 1) These studies extend the use of MS to studying processing of G␥ subunits in cultured cells either by capturing them directly or by capturing them indirectly by their association with G␤ subunits. Part of the power of this approach is to be able to evaluate completely the processing of these proteins. 2) These studies verify that many G␥ isoforms, including G␥1, -2, -4, -12, and -13, are expressed and processed in cells as predicted by past studies of the processing of prenylated proteins (Fig. 2). These proteins are also processed as predicted at the N terminus as to, for example, whether or not they are acetylated (Fig. 4). 3) For all of the G␥ isoforms studied, essentially all of the expressed G␥ made in the cell either forms a functional dimer with G␤ or is

or a 2 positions in their Ca 1 a 2 X signal sequence
The NCBI annotated human protein database from Build 34.3 was downloaded and scanned using a Perl script matching a C-terminal Cysa 1 a 2 X motif where a 1 or a 2 is an aromatic amino acid instead of an aliphatic one, and X is a Lys or Phe for geranylgeranylation or Ser, Cys, Met, Asn, or Ala for farnesylation. Given in the table are the accession numbers, the C-terminal Ca 1 a 2 X motif sequence, the predicted prenyl group modification, and the entry description.
degraded. This conclusion is based upon past studies indicating that dimer formation is required for prenylation (34). Because little or no unprocessed protein or intermediates or aberrantly formed complexes are detected for the range of G␥ isoforms expressed, either when retrieved directly as N-terminally tagged proteins (Fig. 2) or retrieved indirectly in association with tagged G␤ subunits (Fig. 4), all of the expressed protein that is not degraded is functional and is fully processed. 4) Tagged-G␤ constructs can be used to recover either expressed (Fig. 4) or endogenous G␥ from cells (Fig. 5). 5) Co-expression of G␥ and G␤ subunits preferentially form dimers of the expressed proteins rather than with endogenous proteins. This is most clearly shown for G␥ where FLAG-G␤ is predominantly associated with expressed and not endogenous proteins (Fig. 4). Signals for endogenous G␥ are significantly associated with FLAG-G␤ only when G␥ cDNA is omitted from the transfection (Fig. 5). This is compatible with and more completely defines the significance of the observation that G␤ and G␥ stabilize each other's expression when co-transfected into cells (35). 6) G␥5 is processed differently from all of the other G␥ subunits characterized here, and most of the expressed G␥5 protein, whether expressed (Figs. [2][3][4] or endogenous (Fig. 5), is not proteolytically processed at the C terminus. This is also compatible with isolation of G␥5 from an in vivo tissue source such as bovine brain (26). 7) G␤1-G␤4 all bind both variants of G␥5 (Fig. 5). 8) G␥5 is a prominent G␥ expressed in HEK293, Neuro2a, and NIH3T3 (data not shown) cells (Fig. 5). In all of these cells, G␥5-SFL is the strongest MS signal observed. In all of them too, a fully processed form is also generated. 9) Besides G␥5, prominent G␥s found in HEK293 cells are G␥12 and G␥11; with G␥7, -8, and -10 detected at lower signal intensity. In Neuro2a cells, G␥2 is also seen, but at lower signal intensity than G␥5. Although the expectation might be that G␥2 would be the primary G␥ in a neuro-related source, previous studies suggest that G␥5 is preferentially expressed during neural development and G␥2 in differentiated neurons (36). This would be consistent with the properties of the Neuro-2a mouse neuroblastoma cell line (33). 10) Of the endogenous G␥ detected, only G␥11 appeared to differentially interact with G␤. This may be related to recent studies suggesting that G␤ isoforms form dimers of differential stability with G␥11 (37).
It has been suggested that most biologically relevant Ca 1 a 2 X motif sequences are substrates for Rce1 (5,38,39). One notable exception, however, is the ␣ and ␤ isoforms of liver phosphorylase kinase with Ca 1 a 2 X sequences of CAMQ and CLVS that are not cleaved (40). Their identification was instrumental in differentiating the two prenyl protease activities in yeast, where these sequences are not cleaved by yeast Rce1p (13). Recently we reported a small fraction of G␥7 with Ca 1 a 2 X sequence CIIL that is unprocessed (11). In these cases the sequence-dependent reasons for the lack of processing of these substrates, partial or complete, is not clear from past studies of prenyl protein proteolysis and may relate either to the combined effects of the residues comprising the Ca 1 a 2 X sequence or to determinants outside the Ca 1 a 2 X prenylation motif for the individual proteins.
Characterized here is the lack of processing of G␥5, which ends in CSFL (26) and is remarkable compared with most other prominent Ca 1 a 2 X substrates in its inclusion of an aromatic amino acid. This residue appears to define the insensitivity of this protein to proteolytic processing after prenylation (Fig. 3). The G␥5 Ca 1 a 2 X sequence is both unique to G␥ sequences in containing an aromatic amino acid and is highly evolutionarily conserved throughout the vertebrate phylum (Fig. 6). This is a strong argument for the importance of this variation in prenyl processing for this protein. In addition, there are at least 44 proteins or predicted proteins in the human genome ( Table 2) that are potentially modified in a similar way. These proteins belong to many different classes of proteins found in varying cellular compartments and organelles, suggesting that the functional role of this processing would relate to specific proteinprotein interactions rather than general targeting to a common cellular site (7)(8)(9)(10).
Prior to our identification of the processing pattern of G␥5 (26) and its sequence dependence described here, past studies were inconclusive that this variation in prenyl processing would be relevant to major prenyl substrates. Based upon analysis of 3and 4-amino acid peptides, many of those with aromatic amino acids a 1 and a 2 were thought likely to be substrates of a purified rat liver Rce1-like activity (41). This is not the case for G␥5 where any aromatic amino acid at either the a 1 or a 2 position prevents processing (Fig. 3). Although other endogenous prenyl substrates containing an aromatic amino acid have not yet been characterized, a mutant Ras protein containing Phe at a 2 is poorly processed by Rce1, although it is also poorly prenylated (7), and small Ca 1 a 2 X peptide sequences containing aromatic amino acids are often prenylation inhibitors (42).
In yeast, aromatic amino acids at a 1 or a 2 also often decrease prenylation, which complicate in this case too rigorous analysis of the effects of these residues on the proteolytic step. Nevertheless, several Ca 1 a 2 X sequence that contain aromatic amino acids do not appear to be substrates of either Afc1p/Ste24p or Rce1p (14). Other Ca 1 a 2 X sequences with aromatic amino acids, however, often do appear to be cleaved, but to the degree that they are, they appear to be preferentially substrates of Afc1p/Ste24p rather than Rce1p (14). The apparently greater role of Rce1 in targeting most prenyl substrates in vertebrates (15,16) would then be compatible with general lack of processing of G␥5 and other prenyl proteins containing an aromatic amino acid in their Ca 1 a 2 X sequence. From this, it is interesting to speculate that the small but consistent amount of fully pro-  cessed G␥5 we see in cells could result from the activity of Zmpste24, which seems to otherwise target somewhat selectively prelamin A (17).
In all cases where we have observed the variant processing of G␥5 there are two components: one unprocessed as the major one and a second processed as predicted, including C-terminal proteolysis, and always nearly in the same ratio. This is true of purified bovine brain G proteins (11,26), heterologously expressed G␥5 without (Fig. 4B) or with (Fig. 1G) an epitope tag, and endogenous G␥5 expressed in HEK293 cells or Neuro-2a cells (Fig. 5). This expression ratio appears to be evolutionarily conserved also, because even a conservative substitution with, for example, a Tyr (Fig. 3E) would eliminate production of the minor form, and this does not appear to be evolutionarily observed (cited from above). Thus, not only does G␥5 represent an example of a prenylated substrate that is not efficiently processed by Rce1, but it also seems to be one designed to generate two forms of the protein with differential C-terminal processing, pointing to endogenous heterogeneity of prenyl substrates by design.
Another observation from these studies is that, at least as assayed by MALDI-MS, G␥5 (and the G␥5-SFL), in particular, is one of the more abundant G␥ subunits expressed in cells, at least in two diverse cell lines such as embryonic kidney cells (HEK293) and brain neuroblastoma (Neuro-2a). This is also true of NIH3T3 cells. 3 The one caveat to this conclusion is that MALDI-TOF MS signal intensity is highly dependent upon how readily a protein ionizes, as well as on its abundance. G␥5 may ionize more readily than other G␥ isoforms, which would then cause overestimation of its relative abundance in FLAG immunoprecipitation samples based on MALDI-TOF MS signals. However, analysis of expression levels also suggests that G␥5 is one of the most abundant and widely expressed mRNAs in cells, 4 and in MALDI analysis of G␥ subunits in purified G proteins, signal intensities are roughly related to the amount of protein present, including for G␥5 (11, 28 -30). Nevertheless, it is likely that part of the predominance of the signal intensity in cells for G␥5 (Fig. 5) is its abundance and part how readily it can be ionized compared with some other G␥ isoforms.
The studies reported here complement our recently reported proteomic analysis of the variation of G␥ proteins found in purified brain G proteins and, in particular, variation in prenyl processing of these proteins (11). Those studies demonstrated that the normal variation in expression of prenylated proteins is far greater than generally appreciated and nearly all of the observed G␥ isoforms are expressed as variably processed proteins to the level that they could be characterized in these samples. The current work indicates that, for those prenylated proteins containing an aromatic amino acid within their Ca 1 a 2 X motif, there is also likely variation in the proteolytically processed variants of the proteins observed. This variation would likely affect the efficacy or consequences of treatment with drugs targeting this reaction. In addition, the realization of a previously unrecognized interaction of aromatic amino acids with the Ca 1 a 2 X motif may provide information important for the development of more effective therapeutics targeting these reactions.