Comprehensive analysis of CXXX sequence space reveals that Saccharomyces cerevisiae GGTase-I mainly relies on a₂X substrate determinants

Abstract

Many proteins undergo a post-translational lipid attachment, which increases their hydrophobicity, thus strengthening their membrane association properties or aiding in protein interactions. Geranylgeranyltransferase-I (GGTase-I) is an enzyme involved in a 3-step post-translational modification (PTM) pathway that attaches a 20-carbon lipid group called geranylgeranyl at the carboxy-terminal cysteine of proteins ending in a canonical CaaL motif (C—cysteine, a—aliphatic, L—often leucine, but can be phenylalanine, isoleucine, methionine, or valine). Genetic approaches involving 2 distinct reporters were employed in this study to assess Saccharomyces cerevisiae GGTase-I specificity, for which limited data exist, toward all 8,000 CXXX combinations. Orthogonal biochemical analyses and structure-based alignments were also performed to better understand the features required for optimal target interaction. These approaches indicate that yeast GGTase-I best modifies the Cxa[L/F/I/M/V] sequence that resembles but is not an exact match for the canonical CaaL motif. We also observed that minor modification of noncanonical sequences is possible. A consistent feature associated with well-modified sequences was the presence of a nonpolar a₂ residue and a hydrophobic terminal residue, which are features recognized by mammalian GGTase-I. These results thus support that mammalian and yeast GGTase-I exhibit considerable shared specificity.

Introduction

Protein lipidation involves the post-translational attachment of a lipid group to specific sites of a protein. These lipids can be fatty acids (palmitoyl, palmitoleyl myristoyl, octanoyl), isoprenoids (farnesyl, geranylgeranyl), sterols (cholesterol), and phospholipids (glycosylphosphatidylinositol) (Nadolski and Linder 2007; Resh 2013; Jiang et al. 2018).

Protein prenylation is mediated by prenyltransferases—farnesyltransferase (FTase) that utilizes farnesyl pyrophosphate (FPP) as a substrate and 3 geranylgeranyltransferases (GGTase-I, -II, and -III) that utilize geranylgeranyl pyrophosphate (GGPP) (Benetka et al. 2006; Wang and Casey 2016; Kuchay et al. 2019; Shirakawa et al. 2020). FTase and GGTase-I are considered CaaX-type prenyltransferases because their protein targets are defined by a C-terminal Ca₁a₂X motif: C—cysteine; a₁, a₂—typically aliphatic amino acids (e.g. leucine, isoleucine, and valine); X—many amino acids (Fig. 1a). Where investigated, the X residue provides specificity for modification by FTase or GGTase-I (Moores et al. 1991; Caplin et al. 1994; Hartman et al. 2005). FTase modifies a wide range of sequences where X is approximately half of the 20 amino acids, while GGTase-I modifies sequences where X is leucine, and sometimes phenylalanine, isoleucine, methionine, or valine (i.e. Caa[L/F/I/M/V]) (Finegold et al. 1991; Moores et al. 1991; Yokoyama et al. 1995; Hartman et al. 2005; Maurer-Stroh and Eisenhaber 2005; Krzysiak et al. 2010; Berger et al. 2018, 2022; Kim et al. 2023). After CaaX protein prenylation, there is often but not always proteolytic removal of the -aaX residues by CaaX protease (Rce1) and carboxymethylation of the prenylated cysteine by isoprenylcysteine methyltransferase (ICMT). This multistep modification is referred to in this study as canonical CaaX protein modification (Fig. 1a). It is generally accepted that the PTMs of CaaX proteins augment their hydrophobic nature and are needed for their protein-membrane and protein–protein interactions (Maurer-Stroh et al. 2003; Wang and Casey 2016). CaaX proteins serve a critical purpose in many cellular activities, such as signaling, growth, differentiation, and migration and relate to human diseases such as cancer, cardiovascular diseases, microbial infections, and progeria. Hence, CaaX-type prenyltransferases are often attractive targets for human disease therapies (Benetka et al. 2006; Berndt et al. 2011; Palsuledesai and Distefano 2015).

Since the recognition of this 3-step canonical CaaX protein PTM pathway, well-known prenyltransferase targets like the yeast a-factor (FTase), Ras GTPases (FTase), and Rho GTPases (GGTase-I) have led to the general view that the 3 steps of the pathway are coordinated. Emerging evidence reveals, however, that some CaaX proteins, like farnesylated yeast Ydj1 and geranylgeranylated mammalian Gγ5, can undergo prenylation but retain their last 3 amino acids (i.e. aaX) (Kilpatrick and Hildebrandt 2007; Hildebrandt et al. 2016). This single-step prenylation-only modification is referred to in this study as “shunted” CaaX protein modification (Fig. 1a). Shunted modification is required for optimal function of Ydj1, but the importance of shunting for other prenylproteins remains unexplored (Hildebrandt et al. 2016).

Using the yeast Hsp40 Ydj1 protein as a genetic reporter, our studies have revealed that yeast FTase has much broader specificity than that defined by the canonical CaaX motif (Berger et al. 2018; Blanden et al. 2018; Ashok et al. 2020; Kim et al. 2023). Many of the identified noncanonical sequences lack a₁ and a₂ branched-chain aliphatic amino acids and are not cleaved by Rce1, consistent with the observation that an aliphatic a₂ residue is an important recognition determinant for Rce1 (Berger et al. 2018, 2022; Kim et al. 2023). While the ability of mammalian FTase to modify noncanonical CaaX sequences has not been explored, farnesylated mammalian proteins with such sequences do exist, suggesting that shunted prenylproteins are widespread across species (e.g. DNAJA2, Lkb1/Stk11, and Nap1) (Sapkota et al. 2001; Storck et al. 2019).

Studies utilizing in vitro, in vivo, and in silico approaches have evaluated the specificity of FTase, but by comparison, the specificity of GGTase-I is underexplored. Several GGTase-I specificity studies have explored specificity using in vitro prenylation assays with purified GGTase-I and individually purified candidate proteins (Moores et al. 1991; Caplin et al. 1994). Medium throughput approaches have explored specificity using peptide libraries or metabolic probes to identify prenylated proteins within cells (Hartman et al. 2005; Chan et al. 2009; Storck et al. 2019). None of these methods have allowed for evaluating GGTase-I against all possible 8,000 CXXX sequence combinations. A prediction system for identifying GGTase-I targets has been established, but it too has limitations (Maurer-Stroh and Eisenhaber 2005). For example, the geranylgeranylated noncanonical sequence CSFL (Gγ5) is not predicted to be modified.

Overall, studies generally support that mammalian GGTase-I prefers CaaX sequences with aliphatic a₂ and hydrophobic X residues. This specificity is entirely reliant on the β subunit of the dimeric GGTase-I complex, and conserved specificity among distinct GGTase-I enzymes is often observed (Caplin et al. 1994; Mazur et al. 1999; Reid et al. 2004; Benetka et al. 2006). The specificity of yeast GGTase-I, however, has not been fully investigated. Rat and yeast GGTase-I β subunits have limited sequence conservation (27.2% identity; 39.9% similarity; 23.2% gaps per EMBOSS Needle), and active site residues that confer peptide substrate and lipid specificity are only partly conserved (Supplementary Table 1 in File 2) (Taylor et al. 2003; Reid et al. 2004).

Because yeast FTase has broader specificity than previously anticipated, and considering the low sequence conservation between mammalian and yeast GGTase-I, we have investigated the specificity of yeast GGTase-I. This was accomplished by genetic approaches that adapted the normally farnesylated yeast Hsp40 Ydj1 as a yeast GGTase-I reporter and used the established geranylgeranylated yeast GTPase Rho1 as a complementary reporter. Our results indicate that yeast GGTase-I targets the Cxa[L/F/I/M/V] sequence, where x is a wide range of residues, a is primarily one of the 3 branched-chain residues, and the terminal position is restricted to a limited set of nonpolar residues.

Materials and methods

Yeast strains and plasmids

Yeast strains and plasmids used in this study are listed in Table 1 and Supplementary Table 2 in File 2, respectively. Plasmid cloning strategies are described in Supplementary Table 3 in File 2.

yWS3761 was constructed using standard yeast genetic techniques involving a cross between rho1 [RHO1] and ram1Δ haploid strains and subsequent phenotypic and PCR analysis of haploid candidates arising from random sporulation of the diploid. The diploid precursor to yWS3761 was transformed with pWS1807 (2µ LEU2RAM1) to facilitate sporulation, which was not otherwise evident. The rho1Δ parent haploid strain yWS3275 was constructed from a commercially available heterozygous diploid strain ypr165w that was transformed with pWS1835 (CEN URA3RHO1) and subjected to sporulation and random spore analysis to identify haploid rho1Δ [RHO1] candidates with desired genetic markers. The chromosomal deletion of RHO1 was confirmed by PCR and sensitivity to FOA. The ram1 parent haploid strain yWS3204, which is a MATα derivative of yWS1632 (Giaever et al. 2002), was constructed by several backcrosses to isogenic wildtype parent strain BY4742 to eliminate a petite phenotype. The chromosomal deletion of RAM1 was tracked by G418 resistance.

Unless described otherwise, plasmids encoding Ydj1-CXXX and HA-tagged Rho1-CXXX variants were constructed by recombination-mediated PCR-directed plasmid construction in yeast (Oldenburg et al. 1997; Yoshida et al. 2009). Oligonucleotides (IDT, Newark, NJ) encoding for different CXXX variants were PCR amplified and co-transformed into yeast (BY4741) along with NheI/AflII linearized pWS1132 for the Ydj1-CXXX plasmids and BamHI/BstAPI linearized pWS2125 for the HA-tagged Rho1-CXXX plasmids. Colonies with recombinant plasmids were selected on SC-Uracil or SC-Leucine, followed by the isolation of plasmids from individual colonies and their verification by diagnostic restriction digests and sequencing of plasmid DNAs (Azenta, Burlington, MA). Construction of the Ydj1-CXXX encoded plasmids obtained from the Ydj1-CXXX (pWS1775) library has been described previously (Kim et al. 2023).

pWS1807 was constructed in 2 steps. First, a PCR product encoding the RAM1 genomic locus was amplified from BY4741 that was co-transformed into yeast BY4741 with HindIII linearized pRS316 to allow for recombination-mediated PCR-directed plasmid construction in vivo (Oldenburg et al. 1997). The SacI-XhoI fragment encoding RAM1 from the resultant plasmid pWS1767 was then subcloned into pRS425 (Sikorski and Hieter 1989) at the same sites to create pWS1807. Diagnostic restriction digests and DNA sequencing (Azenta, Burlington, MA) were used to identify candidates at each step.

Thermotolerance assay

This assay was performed as described previously (Hildebrandt et al. 2016; Berger et al. 2018; Blanden et al. 2018; Ashok et al. 2020; Kim et al. 2023). Briefly, yeast cells were cultured until saturation in SC-Uracil liquid media at 25°C, and a portion of the saturated culture (100 µl) was added to the first well of a 96-well plate. A series of 10-fold dilutions were prepared and the dilution series spotted onto YPD plates. These plates were incubated at 25°C and 40°C for 96 hours and imaged with a Cannon flat-bed scanner (300 dpi; TIFF format). Photoshop was used for minor adjustments to images (i.e. contrast, rotation, and cropping). This assay was performed twice on separate days with at least 2 technical replicates included in each trial.

Ydj1-based thermotolerance screen

The Escherichia coli derived Ydj1-CXXX plasmid library (pWS1775) has been previously described (Kim et al. 2023). This library contains all 8,000 Ydj1-CXXX variants. The library was transformed into yWS2542 (ram1Δ ydj1Δ) using the Frozen-EZ Yeast Transformation II Kit (Zymo Research, Irvine, CA) per the manufacturer's guidelines. Approximately 367,000 colonies were collected by scraping cells off multiple SC-Uracil plates and washing them in SC-Uracil liquid media. The resuspended solution was centrifuged to obtain a cell pellet that was resuspended in 15% glycerol solution and stored at −80°C as aliquots.

For each replicate of the thermotolerance screen, ∼150,000 CFUs were used. This number of CFUs ensured >99.9% coverage of the Ydj1-CXXX library (http://guinevere.otago.ac.nz/cgi-bin/aef/glue.pl). Briefly, ∼55 × 10⁶ cells were inoculated into 200 mL SC-Uracil liquid media in duplicate, each mixture split into 10 × 20 mL replicates, and each 10-member set subsequently incubated at permissive temperature (25°C) or restrictive temperatures (37°C and 42°C) for 24–48 hours until saturation (A₆₀₀ 1.9–3.3). This growth period was estimated to allow the cells to go through at least 8 rounds of population doubling. Cells from the saturated cultures were harvested by centrifugation, washed, and collected. Plasmids were extracted from the cells using E.Z.N.A. Yeast Miniprep kit following manufacturer's guidelines (OMEGA Bio-Tek, Norcross, GA). Of note, the high temperatures used for liquid-based growth were 37°C and 42°C, while 40°C was chosen for plate-based growth.

Next-generation sequencing

The library preparation for next-generation sequencing (NGS) (i.e. Ydj1-CXXX plasmid libraries derived from the thermotolerance screens performed at 25°C, 37°C, and 42°C) and the NGS method itself have been described previously (Kim et al. 2023). In addition, 10 replicates of the E. coli derived and the naïve yeast-derived Ydj1-CXXX plasmid libraries were sequenced by NGS. The NGS run resulted in over 6 million reads with 97% of reads having a Q30 quality score or better (i.e. 99.9+% base call accuracy). The frequency of each CXXX sequence within each library (i.e. E. coli derived, naïve yeast derived, 25°C, 37°C, and 42°C) was calculated by summing the occurrence of a specific CXXX sequence in all 10 replicates and dividing that value by the sum of all CXXX sequence occurrences in the data set (Supplementary File 1). Ultimately, each CXXX sequence had 2 NGS Enrichment Scores (NGS E-Score), which was the frequency of a specific CXXX sequence at the restrictive temperature (37°C or 42°C) divided by its frequency in the naïve yeast library (Supplementary File 1).

YDJ1-CXXX-based mini-screen to sample CXXX sequences

Ydj1-CXXX library (pWS1775) plasmid library was transformed into yWS2542 (ram1Δ ydj1Δ) and incubated on SC-Uracil media plates at 25°C. Individual transformants (i.e. colonies) were cultured in preparation for the thermotolerance assay at 25°C and 40°C. Briefly, cultures of the individual transformants were subjected to a fixed dilution, and the single dilution spotted on YPD media plates that were incubated at 25°C and 40°C. A collection of 20 candidates displaying thermotolerant and thermosensitive phenotypes at 40°C was identified, and plasmids extracted from each candidate were recovered and sequenced by Sanger sequencing. The collection was reduced to 8 CXXX sequences that were judged to best represent an even distribution over the complete range of the NGS E-Score plots.

Ydj1-based gel-shift assay

This assay was done as described previously (Berger et al. 2018; Hildebrandt et al. 2024). Briefly, yeast strains were cultured in SC-Uracil liquid media to log or late log phase (A₆₀₀ 0.95–1.45) and total cell lysates were prepared by alkaline hydrolysis and trichloroacetic acid precipitation (Kim et al. 2005). Total cell lysates were subjected to SDS-PAGE (9.5% separating gel) followed by immunoblot. The primary antibody used was rabbit anti-Ydj1 (courtesy of Dr. Avrom Caplan) and the secondary antibodies were HRP-conjugated donkey or goat anti-rabbit antibodies (Kindle Biosciences, LLC). The blots were treated with ECL spray (ProSignal Pico Spray) and digital images captured using KwikQuant Imager system (Kindle Biosciences). Photoshop was used for minor adjustments to images (i.e. contrast, rotation, and cropping). This assay was performed with 2 biological replicates for the yWS2542, yWS3169, and yWS4277 transformants.

Rho1-based plasmid-loss assay

yWS3761 (ram1Δ rho1Δ [CEN URA3RHO1]) was transformed with LEU2-marked HA-tagged Rho1-CXXX variants and grown on SC-Uracil and Leucine solid media (SC-UL) at 25°C. The purified transformants were inoculated in SC-Leucine liquid media and cultured to saturation at 25°C. The saturated cultures were diluted to 1 A₆₀₀, and 100 μL of each diluted culture was transferred to an independent well of a 96-well plate, which was used to prepare a 10-fold serial dilution for each sample. The dilution series were collectively pinned onto YPD as a control for preparation of the dilution series and SC complete media with 5-fluoroorotic acid (5-FOA) for the functional assay. The plates were incubated at 25°C for 72 hours and scanned with a Cannon flat-bed scanner (300 dpi; TIFF format). Photoshop was used for minor adjustments to images (i.e. contrast, rotation, and cropping). This assay was performed twice on separate days with at least 2 technical replicates included in each trial.

Rho1-based cell viability selection

yWS3761 (ram1Δ rho1Δ [CEN URA3RHO1]) was co-transformed with BamHI linearized pWS2125 and PCR amplified CXXX mutants. The transformation reactions were plated onto SC-Leucine solid media and incubated at 25°C. Colonies surviving selection were replica plated onto SC complete media with 5-FOA, and 200 FOA-resistant single colonies were individually isolated. Each candidate was patched onto 5-FOA media to confirm the 5-FOA-resistant phenotype, then replica plated onto SC-Leucine to confirm leucine prototrophy, indicative of a CEN LEU2 HA-RHO1-CXXX plasmid being present. Candidates with all requisite phenotypes were cultured in SC-Leucine liquid media, and their associated plasmids recovered and sequenced (Azenta, Burlington, MA). Each plasmid was then individually re-transformed into yWS3761, and the Rho1-based plasmid-loss assay was repeated with the transformant to confirm the FOA-resistant growth phenotype imparted by the plasmid.

WebLogo analysis

This was done as previously described with certain modifications (Berger et al. 2018). The desired groups of sequences were analyzed using the WebLogo website server (https://weblogo.berkeley.edu/logo.cgi) by amino-acid frequency-based analysis (Crooks et al. 2004). A customized amino acid coloring scheme was used: Cys was denoted blue, branched-chain aliphatic amino acids (Val, Ile, and Leu) were denoted red, charged amino acids (Lys, Arg, His, Asp, Glu) were denoted green, polar and uncharged residues (Ser, Gln, Thr, Asn) were denoted black, and all other hydrophobic amino acids (Ala, Gly, Pro, Phe, Trp, Tyr, and Met) were denoted purple.

Results

Ydj1 can be used as an efficient genetic reporter for yeast GGTase-I activity

Growth of yeast at elevated temperature requires prenylation of the Ydj1 Hsp40 chaperone, but not the downstream protease and carboxylmethylation steps of the CaaX modification pathway (Hildebrandt et al. 2016). Due to this property, Ydj1 is a more direct reporter for prenylation. This Ydj1-dependent phenotype has been previously used for probing yeast FTase specificity (Berger et al. 2018; Kim et al. 2023).

Using Ydj1 as a reporter has led to the discovery of many farnesylatable CXXX sequences that do not adhere to the canonical CaaX consensus (Berger et al. 2018; Kim et al. 2023). A lot of these sequences support a robust Ydj1-dependent thermotolerance phenotype equivalent to that supported by wildtype farnesylated and shunted Ydj1 (CASQ). Within that study, a few CXX[L/F] sequences were determined to be prenylated in the absence of FTase activity, ostensibly by GGTase-I. To extend this observation and explore the potential utility of Ydj1 as an efficient genetic reporter for GGTase-I activity, we performed additional thermotolerance assays using the FTase-deficient strain (i.e. ram1Δ) and certain selected Ydj1-CXXL sequences. This study confirmed that neither wildtype Ydj1 (CASQ) nor an unmodifiable Ydj1 variant (SASQ) can support growth at high temperature (40°C) (Fig. 1b). Yet, Ydj1 was able to support thermotolerant growth in the context of CAPL, CRPL, CFAL, CPLL, and CSFL sequences. These sequences all contain Leu at the X position but lack a branched-chain aliphatic amino acid at the a₁ position of the CaaX motif, and in all but one case, also lack such an amino acid at the a₂ position. Moreover, Ydj1-CSFL supported a robust thermotolerant phenotype by comparison to unmodified Ydj1-SSFL, indicating that thermotolerance in the FTase-deficient background was cysteine-dependent. CSFL is associated with mammalian Gγ5, which is the only reported example of a shunted geranylgeranylated sequence; it lacks a₁ and a₂ amino acids needed for recognition and cleavage by the Rce1 CaaX protease (Trueblood et al. 2000; Kilpatrick and Hildebrandt 2007; Berger et al. 2018, 2022). Together, these observations indicate that yeast Ydj1 can be utilized as a reporter for GGTase-I activity. Additionally, the data suggest that GGTase-I specificity is broader than the canonical Caa[L/F/I/M/V] consensus, especially with respect to requiring branched-chain aliphatic amino acids at the a₁ and a₂ positions.

Canonical and noncanonical CaaX sequences impart different effects in the Ydj1-based thermotolerance assay

To investigate the possibility that yeast GGTase-I has broader specificity than anticipated, we designed a study to examine the ability of GGTase-I to modify all 8,000 possible CXXX variants. We adapted a high throughput Ydj1-based thermotolerance screen that had been used previously by Kim et al. (2023) to evaluate yeast FTase specificity (Fig. 2a). In our case, we evaluated thermotolerance using a strain lacking FTase so that any identified thermotolerant phenotypes could be specifically attributed to GGTase-I activity.

Several large data sets were created as part of this study. The E. coli derived Ydj1-CXXX plasmid library has been previously characterized and is known to contain all 8,000 Ydj1-CXXX variants (Kim et al. 2023). This E. coli derived library was transformed into ram1Δ ydj1Δ yeast, and colonies were recovered directly from the transformation plates to create a naïve yeast plasmid library. This yeast-derived library was confirmed by NGS to also contain all 8,000 Ydj1-CXXX variants. Comparison of the frequencies for each CXXX sequence within the E. coli and naïve yeast libraries revealed no obvious enrichment bias for any Ydj1-CXXX variants due to the transfer into yeast (Supplementary Fig. 1 in File 2). The naïve yeast library was inoculated into liquid media and propagated under both permissive (25°C) and restrictive (37°C and 42°C) temperatures, where the restrictive temperatures were expected to enrich for cells expressing geranylgeranylated Ydj1-CXXX variants. NGS was utilized to evaluate the frequency of Ydj1-CXXX variants in all the expanded yeast populations (Fig. 2b–d). Although we expected to observe no significant difference in CXXX frequencies between the naïve yeast and 25°C libraries, we observed both enriched and de-enriched outliers in the 25°C sample (Fig. 2b). Comparison of CXXX frequencies in the naïve yeast and elevated temperature libraries (37°C and 42°C) revealed both enriched and de-enriched CXXX motifs in the 37°C sample and strong enrichment of many CXXX motifs in the 42°C sample (Fig. 2c and d). To determine the relative enrichment of each Ydj1-CXXX variant at the restrictive temperatures, an enrichment score (NGS E-Score) was calculated for each CXXX sequence by dividing the frequency of a specific sequence in the restrictive temperature (37°C or 42°C) library relative to its frequency in the naïve yeast library. Those with higher NGS E-Scores were deemed as having better growth than others at the restrictive temperature. This method of analysis was previously used to interrogate FTase specificity using Ras and Ydj1 reporters (Stein et al. 2015; Kim et al. 2023). While those previous studies used frequency data from 25°C samples as denominators for deriving NGS E-Scores, we rejected this approach due to the observation of outliers in the 25°C sample, which could result in false negative and positive outcomes.

2D plots revealed NGS E-Scores ranging from 0–3.2 (37°C vs naïve yeast) and 0–21.2 (42°C vs naïve yeast), indicative that enrichment (>1) and de-enrichment (<1) of Ydj1-CXXX variants had indeed occurred (Fig. 3a and b). The sequences evaluated in preliminary studies were also observed to be broadly distributed in these 2D plots. Ydj1-CASQ was de-enriched and in the bottom 25% of hits in the 42°C library, which was consistent with its expected lack of geranylgeranylation. At 37°C, which was a less stringent condition, CASQ was neither enriched nor de-enriched. Ydj1 harboring noncanonical sequences CRPL, CPLL, CFAL, CAPL, and CSFL expected to be enriched (see Fig. 1) were present in the top 15% hits in both the 37°C and 42°C libraries, which was consistent with their expected geranylgeranylation. Notably, CXXX sequences associated with yeast (Sc) Rho GTPases (CVLL, CIIL, CTIM, CIIM, CVIL, CAIL), Sc Ras-related GTPase Rsr1 (CTIL), and human (Hs) K-Ras4b (CVIM) were among the de-enriched population in both the libraries. These sequences are frequently touted as examples of canonical geranylgeranylated or cross-prenylated sequences (i.e. modified by both FTase and GGTase-I) (Caplin et al. 1994). While not initially predicted, the de-enrichment of these sequences concurs with observations that farnesylated Ydj1 that is canonically modified has impaired activity (i.e. Ydj1-CVIA), although such modified sequences still outperform unmodified Ydj1 (i.e. Ydj1-SASQ) (Hildebrandt et al. 2016). In this study, by contrast, Ydj1-CXXX variants predicted to be geranylgeranylated and canonically modified vastly underperform unmodified Ydj1 (i.e. Ydj1-CASQ) under the competitive growth conditions of the thermotolerance screen. We were able to recapitulate these differences for individual Ydj1-CaaX variants using a plate-based assay (Supplementary Fig. 2 in File 2). Taken together, these observations indicate that the Ydj1-based thermotolerance screen enriches for shunted CXXX motifs relative to unmodified and canonically modified sequences. Additionally, these observations suggest that an increased hydrophobicity imparted by more hydrophobic geranylgeranyl (vs farnesyl) and carboxylmethylation (vs free carboxyl) can synergistically impair Ydj1 activity.

A comparison of the 2D plots indicated a striking difference in the curve profiles. Significant de-enrichment was observed for some sequences in the 37°C data set that was not observed in the 42°C data set (i.e. compare left most regions of 2D plots). WebLogos were created to better visualize the amino acids associated with this region in both plots, as well as the corresponding enriched regions. Highly de-enriched sequences associated with the 37°C data (NGS E-Scores 37°C/naïve Sc library < 0.2) captured all the Rho/Ras-related sequences clustered in this region of the 2D plot. A WebLogo of this subset (n = 137) had a near canonical-looking profile Cx[V/I/L][L/F/I/M/V]. The dominance of branched-chain aliphatic residues at the a₂ position for these sequences, which are optimal for Rce1 CaaX protease specificity, indicates that this region of the 2D plot appears to be populated mainly by canonical geranylgeranylated sequences (Fig. 3a, bottom panel). By comparison, the equivalent number of top performing sequences in this data set (n = 137) contained a wider range of a₂ amino acids and more variability at the X position (Fig. 3a, top panel).

A similar WebLogo analysis was performed using the 42°C data set. A larger subset of de-enriched sequences (n = 500) was analyzed to capture most, albeit not all, of the Rho/Ras-related sequences clustered in this region (Fig. 3b, bottom panel). The sequence profile revealed a wide range of amino acids, including residues at a₂ (i.e. D/E/K/R) or X (i.e. K/R/P) that interfere with farnesylation of Ydj1, which we propose are also likely to interfere with geranylgeranylation (Berger et al. 2018; Kim et al. 2023). In fact, over half of the de-enriched sequences contained these restrictive amino acids (Supplementary Fig. 3a in File 2). A smaller subset of sequences matching the canonical GGTase-I consensus were also among the de-enriched population (Supplementary Fig. 3b in File 2). The remaining sequences in the population displayed no obvious pattern (Supplementary Fig. 3c in File 2). A WebLogo analysis of the most enriched sequences in this data set was also performed (Fig. 3b, top panel). This analysis revealed no obvious enrichment of aliphatic amino acids at a₁ and a₂ positions, but hydrophobic amino acids were generally enriched with aromatic amino acids F and W being most prevalent. A moderate enrichment of expected amino acids was observed at the X position (i.e. L/F/I/M/V), as were a few other amino acids (e.g. W and Y). Overall, this analysis indicates that the enriched sequences recovered by our screening approach do not fully adhere to the expected Caa[L/F/I/M/V] consensus sequence of GGTase-I.

Enriched noncanonical CXXX sequences confer Ydj1-dependent thermotolerance

To validate the observations from the Ydj1-based thermotolerance screen through an orthogonal assay, we evaluated 15 CXXX sequences by Ydj1-based thermotolerance and gel-shift assays. Eight noncanonical CXXX sequences were derived from a Ydj1-based thermotolerance mini-screen (see Materials and methods), and the others were CASQ (shunted farnesylated sequence of yeast Ydj1), CVLL and CVIL (canonical geranylgeranylated sequences of yeast Rho1 and Rho5, respectively), CSFL (shunted geranylgeranylated sequence of mammalian Gγ5), and CAFL, CPIQ, and CHLF (sequences with high NGS E-Scores). These 15 candidates were chosen to represent broad distribution of scores across the NGS E-Score plots (37°C vs naïve yeast library and 42°C vs naïve Sc library) (Fig. 4a and b).

The thermotolerance assay results indicated that 8 of 15 CXXX sequences could support robust growth at high temperatures. We generally observed consistency between NGS E-Score (42°C vs naïve Sc library) and high-temperature growth (Fig. 4c). Higher E-Scores had better thermotolerance. The thermotolerant phenotype appeared between NGS E-Scores of 0.9 and 5.8; a lack of data points between these NGS E-Scores did not allow for further refinement of a minimum threshold for thermotolerance. The canonical sequences CVIL and CVLL were among the 7 thermosensitive sequences.

Enriched noncanonical CXXX sequences are subject to partial modification by yeast GGTase-I

In past studies of FTase specificity, a strong correlation has been observed between thermotolerance and farnesylation of Ydj1, where the latter was determined by gel-shift analysis (Berger et al. 2018). The basis for the gel-shift assay is that farnesylated Ydj1 migrates faster by SDS-PAGE than unfarnesylated Ydj1. We thus investigated whether this was also the case for Ydj1-CXXX variants that were predicted to be modified in the absence of FTase. Ydj1 harboring canonical sequences (i.e. CVLL and CVIL) exhibited a mobility shift relative to unmodified Ydj1p (i.e. CASQ) (Fig. 5a). An obviously gel-shifted population was not observed for the other sequences evaluated, but we consistently observed a light smear beneath the main Ydj1 band for sequences having NGS E-Scores greater than or equal to 0.9 (i.e. CLIN, CYVM, CWIT, CSFL, CYIY, CHLF, CPIQ, and CAFL). To further investigate this observation, we performed the gel-shift assay with select CXXX sequences and their cysteine to serine mutants (SXXX) (Fig. 5b). CVLL again exhibited faster migration and a gel-shift pattern that was clearly distinguishable from their unmodifiable serine counterpart. CASQ and CNTH, which lacked the very light smear, exhibited no obvious mobility difference relative to their serine counterparts. CLIN, CSFL, CPIQ, and CAFL exhibited a very light smear beneath the main protein band, which was absent in the serine mutants. In some cases, cultures were incubated at 37°C to try and improve gel-shift properties to no avail. Together, these observations suggest that many of these noncanonical geranylgeranylated sequences are modified by GGTase-I, albeit only partially, yet this partial modification is sufficient to impart a Ydj1-dependent thermotolerance phenotype. To confirm that sequences with higher NGS E-Scores were reactive with yeast GGTase-I, we overexpressed yeast GGTase-I in an effort to exaggerate geranylgeranylation. Indeed, we observed improved gel-shift patterns for sequences with higher NGS E-Scores and no noticeable change for those with lower scores (Fig. 5c). Of the modified sequences, none appeared to be fully modified.

To extend our studies to human GGTase-I, we performed the gel-shift assay using a yeast strain that overexpresses human GGTase-I in the absence of yeast GGTase-I (Hildebrandt et al. 2024). For sequences with higher NGS E-Scores, we observed a fully modified pattern (e.g. CSFL, CHLF, and CAFL) or a doublet pattern (e.g. CLIN, CYVM, CWIT, CYIY, and CPIQ), while sequences with lower scores (CETT, CDGE, CASQ, CNTH, CVCG) did not display a mobility shift (Fig. 5d). This “humanized” GGTase-I (HsGGTase-I) expressing strain was better at modifying noncanonical CXXX motifs relative to endogenous yeast GGTase-I. This could be due to a higher amount of the HsGGTase-I enzyme produced in cells under the constitutive PGK1 promoter. But, this could also indicate specificity differences between the 2 species of GGTase-I. Interestingly, mammalian and yeast GGTase-I structures differ in active site architecture and the presence of a nonessential proline-rich region in the mammalian enzyme that is proposed to have regulatory properties but whose functional importance has not yet been resolved (Hagemann et al. 2022). We speculate that these features may contribute to mammalian GGTase-I having distinct specificity relative to yeast GGTase-I in our system.

Rho1 is an effective reporter for yeast GGTase-I activity

Ydj1 is a naturally farnesylated protein that was adapted to use as a GGTase-I reporter in this study. To determine whether the GGTase-I specificity observed in the context of Ydj1 was also apparent in the context of a naturally geranylgeranylated protein, we extended our studies of GGTase-I specificity using Rho1, a well-characterized canonically modified geranylgeranylated yeast protein. To our knowledge, there are no shunted geranylgeranylated yeast proteins, which would have been the preferred starting point for these studies.

RHO1 is essential, and canonical modification of Rho1 (CVLL) is required for its function (Ohya et al. 1993; Yamochi et al. 1994). Hence, we took advantage of an established Rho1 functional assay to examine the impact of different CXXX sequences on Rho1 activity (i.e. plasmid-loss assay) (Fig. 6a). In this assay, rho1Δ yeast complemented by a URA3-marked plasmid encoding wildtype Rho1 (i.e. rho1Δ [CEN URA3RHO1]) are transformed with a LEU2-marked plasmid encoding a Rho1-CXXX variant. Negative selection is applied (i.e. 5-FOA) to recover yeast having lost the RHO1URA3-marked plasmid. In this experimental set-up, yeast can only survive negative selection if they retain a functional Rho1-CXXX variant on the remaining LEU2-marked plasmid. To eliminate the possibility of farnesylation occurring to Rho1 and interfering with our results, we used as a starting point a strain that also lacked FTase activity (i.e. ram1Δ rho1Δ [CEN URA3RHO1]). To confirm the utility of the plasmid-loss assay for studies of Rho1-CXXX variants, wildtype Rho1 (CVLL) was evaluated along with an empty vector. As expected, yeast harboring the CEN LEU2 plasmid encoding Rho1 (CVLL) supported robust growth after 5-FOA selection whereas yeast harboring an empty vector did not grow (Fig. 6b). The same results were obtained using the HA-tagged version of wildtype Rho1 (CVLL), while unmodified HA-tagged Rho1-SVLL did not grow.

To assess the impact of other CXXX sequences on Rho1 function, we applied the plasmid-loss assay to Rho1-CXXX variants designed on the 15 CXXX sequences previously evaluated by Ydj1-dependent thermotolerance and gel-shift assays (Fig. 6c). This analysis revealed a range of growth patterns for the Rho1-CXXX variants. Those with higher NGS E-Scores (i.e. CLIN, CYVM, CWIT, CSFL, CYIY, CHLF, CPIQ, and CAFL) generally grew better than those with lower NGS E-Scores (i.e. CETT, CDGE, CASQ, CNTH, and CVCG). Many of the sequences that supported Rho1-dependent growth were noncanonical. Notable exceptions were CVIL and CVLL, which had the 2 lowest scores within the Test Set and are expected to be canonically modified geranylgeranylated sequences. While these 2 sequences underperformed in the context of the Ydj1-dependent thermotolerance assay, they supported Rho1-dependent growth. Overall, the results obtained with the plasmid-loss assay are consistent with the conclusions that geranylgeranylation coupled with canonical modifications hinders Ydj1 but not Rho1 function, and that noncanonical sequences can support Rho1 function.

Considering the results from the Ydj1 and Rho1-based assays together, we predicted that CXXX sequences that poorly supported growth in the context of both reporters (i.e. CETT, CDGE, CASQ, CNTH, and CVCG) were not modified by GGTase-I. By contrast, CXXX sequences that supported growth in both contexts (i.e. CYVM, CWIT, CSFL, CYIY, CHLF, CPIQ, and CAFL) were predicted to be modified by GGTase-I. These observations can be extended to conclude that the Rho1 function in the context of the plasmid-loss assay is indifferent to canonical vs likely shunted CXXX modification. An exception to the above binning is the CLIN sequence that did not support Ydj1-dependent thermotolerance but did support Rho1-dependent growth. In this instance, we predict that CLIN is yielding a mixed population of canonically modified and unmodified products, leading to moderate toxicity and a partial yeast growth defect in the context of Ydj1 while the modified population of Rho1 is sufficient to support growth.

Varied CXXX sequences can support Rho1-dependent growth

Since the function of Rho1 in the plasmid-loss assay did not seem to be impacted by sequences likely to be shunted, we hypothesized that the assay could be adapted into a Rho1-based genetic selection to identify CXXX sequences that can support Rho1-dependent growth, ostensibly because of geranylgeranylation. Importantly, such a Rho1-based screen would be predicted to recover canonically modified CXXX sequences that were not enriched by the Ydj1-based screen.

To test the utility of the plasmid-loss assay for recovering novel functional Rho1-CXXX variants, a random library of plasmid-encoded Rho1-CXXX variants (LEU2 marked) was expressed in ram1Δ rho1Δ [CEN URA3RHO1] yeast, and the transformed yeast subject to various selections (Fig. 7a). The library was generated by recombining a PCR product encoding random CXXX sequences into a LEU2-marked plasmid encoding a nonfunctional Rho1 sequence that was designed to lack the entirety of the natural CaaX sequence CVLL (i.e. Rho1-BamHI). While this strategy theoretically allows for the creation of all 8,000 possible Rho1-CXXX combinations, only a limited number of plasmid candidates were evaluated due to the labor and time costs of evaluating all of CXXX space. Positive selection (i.e. SC-Leucine) was used to recover colonies harboring recombinant plasmid products, and negative selection of the same colonies (i.e. 5-FOA) was used to identify colonies carrying a functional Rho1-CXXX variant.

The Rho1-based genetic screen yielded many 5-FOA-resistant colonies. Plasmids were extracted from 200 FOA-resistant colonies and sequenced. Within this set of plasmids, 83 LEU2-marked plasmids had the exact DNA sequence of wildtype Rho1 (CVLL) that was encoded in the lost URA3 plasmid, indicative that gene conversion between the LEU2 and URA3 plasmids had likely occurred. Of the remaining 117 sequences, 5 were exact duplicate DNA sequences of other sequences within the set, indicative that their parent colonies were likely double picked during the screening process. Among the remaining 112 unique DNA sequences, some encoded for the same CXXX sequence through different codon usage. In all, 94 distinct CXXX sequences were encoded by the 112 unique DNA sequences, including CSFL and CVIL that are naturally associated with geranylgeranylated proteins (HsGγ5 and ScRho5). The 117 hits not matching the parent plasmid DNA sequence were retested using the cell viability assay, and all supported growth on 5-FOA, with most supporting growth that was indistinguishable from that supported by the CVLL sequence (Supplementary Fig. 4 in File 2).

Overall, the Rho1-based screen retrieved CXXX sequences best categorized as a mix of canonical and noncanonical CaaX sequences. WebLogo analysis of the 112 unique hits revealed a prevalence of branched-chain aliphatic amino acids (i.e. L/I/V) at both a₁ and a₂ positions (more so at a₂) and L/F/I/M/V at the X position (Fig. 7b). These results are consistent with the Caa[L/F/I/M/V] consensus motif reported for geranylgeranylated sequences. Yet, there were clearly sequences recovered from the Rho1-CXXX screen that do not fully match the consensus. About 80% of the hits with a consensus residue at the X position (i.e. L/F/I/M/V) did not have branched-chain aliphatic amino acids at a₁ or a₂, indicating some flexibility at these positions (Fig. 7c). For the approximately 20% of the hits not having a consensus amino acid at the X position (i.e. not L/F/I/M/V), aliphatic amino acids were more strongly prevalent at a₂ (Fig. 7d).

A predicted outcome of the Rho1-based screen was that it would recover both canonical and noncanonical CXXX sequences. Moreover, we expected that noncanonical sequences recovered in the Rho1-based screen would be among those enriched in the Ydj1-based screen, while canonical sequences would be de-enriched. Indeed, this is what was generally observed when the 94 unique CXXX hits from the Rho1-based screen were evaluated in the context of their NGS E-Score plots (Fig. 8a and b). Among the 94 unique sequences, about 1/2 had an NGS E-Score (42°C vs naïve Sc library) above 2 (n = 44; 47%), indicative of enrichment in the Ydj1-based screen. Most of these sequences (n = 38) contained a canonical X residue (i.e. L/F/I/M/V) but had mostly noncanonical a₁ and a₂ amino acids (Fig. 8c). About 1/3 of the unique sequences had an NGS E-Score less than 0.5 (n = 34, 36%), indicative of de-enrichment in the Ydj1-based screen. Most of these sequences (n = 27) had a canonical Caa[L/F/I/M/V] profile (Fig. 8d).

From the set of 94 unique sequences recovered by the Rho1-based screen, 4 were investigated by Ydj1-based thermotolerance and gel-shift assays (Supplementary Fig. 5 in File 2). These sequences were chosen because they either had high NGS E-Scores (42°C vs naïve Sc library) in the Ydj1-based assay (i.e. CATL and CNPL) or low NGS E-Scores (i.e. CAIV and CTVL). As expected, Ydj1-CXXX variants encoding CATL and CNPL were able to support thermotolerant growth, while those encoding CAIV and CTVL were not. Also as expected, Ydj1-CXXX variants encoding CAIV and CTVL exhibited a strong gel-shift, while those encoding CATL and CNPL exhibited a very light smear beneath the main Ydj1 band that was absent in the corresponding serine mutants. These results are fully consistent with the observation that all 4 sequences contain a consensus X residue that is compatible for geranylgeranylation (i.e. L or V), and that the de-enriched sequences have a₂ residues compatible for cleavage by Rce1 (i.e. I and V), while the enriched sequences do not (i.e. T and P) (Trueblood et al. 2000; Berger et al. 2018, 2022).

Discussion

Since the recognition of prenylation on yeast mating factors and subsequently on mammalian Ras and Ras-related GTPases, the protein prenyltransferases FTase and GGTase-I have been reported to target the “CaaX” consensus motif (Kamiya et al. 1978). Recent studies have revealed, however, that noncanonical CXXX proteins can be farnesylated, but the specificity of geranylgeranylation in this regard has not been fully investigated (Hougland et al. 2010; London et al. 2011; Hildebrandt et al. 2016; Berger et al. 2018; Storck et al. 2019; Kim et al. 2023). This work addressed this gap in knowledge through 2 genetic screens whose results demonstrate that yeast GGTase-I has relatively more focused specificity than yeast FTase.

After probing all of CXXX sequence space with the Ydj1 reporter, we determined that consistent features among the best yeast GGTase-I targets were hydrophobic residues at both a₂ and X, with a dominance of L/F/I/M/V at the X position (Fig. 3a, bottom WebLogo; Fig. 5b). Additionally, there was a strong preference for branched-chain aliphatic amino acids (BCAs) at a₂, unlike yeast FTase that accommodates a wider range of residues at this position. These preferences for a₂ and X amino acids are also preserved among the sequences recovered using the Rho1 reporter (Fig. 7b). We speculate that the a₂ requirement for BCAs in GGTase-I targeted sequences is a means to ensure coupling to the Rce1 protease step of the canonical CaaX modification pathway, leading to proteins with a highly hydrophobic C-terminus (Fig. 1). Consistent with this premise, geranylgeranylated proteins are typically associated or predicted to be associated with membranes, whereas farnesylated proteins, especially those that are shunted (e.g. Hsp40s, Nap1, and Lkb1/Stk11), are not necessarily membrane associated.

Our results initially suggested that noncanonical sequences might also be GGTase-I targets (Fig. 3a and b), but orthogonal biochemical validation revealed that these sequences are, at best, weakly modified by yeast GGTase-I (Fig. 5a and b). Nonetheless, it appears that limited modification can impart functional properties to both Ydj1 and Rho1. It remains unclear how many geranylgeranylated proteins require partial modification for their function, but having a noncanonical CaaX sequence may provide a means to this end. It also remains unclear whether upregulation of GGTase-I activity could lead to better modification of otherwise poorly modified CaaX sequences, which was observed when overproducing yeast and mammalian GGTase-I in our system (Fig. 5c and d). In natural biological settings, we speculate that increased GGTase-I production could occur in response to external signaling events (e.g. increased transcription) or regulatory mechanisms (e.g. phosphorylation); the latter has been postulated to modulate mammalian FTase activity (Goalstone et al. 1997). Lastly, it remains unclear whether any noncanonical CaaX sequences displaying partial modification could be better modified in their natural protein context.

Overall, we observe similar target specificities for yeast and mammalian GGTase-I despite substantial differences in active site residues (Supplementary Table 1 in File 2). Crystallographic studies have revealed that the a₂ and X binding pockets of mammalian GGTase-I are hydrophobic in nature (Taylor et al. 2003; Reid et al. 2004; Gangopadhyay et al. 2014). These same studies indicate that mammalian GGTase-I has considerable flexibility at the a₁ position. Yeast GGTase-I appears to have the same specificity features based on the results of our study. This suggests that the yeast and mammalian enzymes have similar active site architecture despite differences in active site residues. Indeed, the overall architecture of the yeast GGTase-I β subunit (Cdc43), as predicted by structural modeling, has a high degree of structural alignment with the established structure of the mammalian GGTase-I β subunit (Fig. 9; RMSD = 1.3 Å). A key structural difference appears to be the disposition of W108β in yeast GGTase-I, which would clash sterically with the geranylgeranyl group if it were positioned in the active site as for mammalian GGTase-I. We expect that future structural studies of yeast GGTase-I will resolve how the enzyme accommodates the geranylgeranyl group in a different position.

Our previous studies have revealed that yeast FTase, and likely mammalian FTase, has broader tolerance than previously predicted, especially for a₁ and a₂ amino acids. By comparison, our evidence reveals that yeast GGTase-I has a much more focused specificity, consistent with reports for mammalian GGTase-I. Both the yeast prenyltransferases can allow various amino acids at a₁. While a broader tolerance is exhibited by yeast FTase at a₂ (i.e. D, E, K, and R are disfavored), GGTase-I appears much more restrictive at this position (i.e. BCAs are favored for well-modified sequences) (Kim et al. 2023). Similarly, yeast FTase has broader tolerance at X (i.e. K, P, and R are disfavored) relative to GGTase-I that strongly prefers a limited set of hydrophobic residues at X (i.e. L/F/I/M/V) (Kim et al. 2023). Altogether, our comprehensive analysis of yeast GGTase-I specificity reveals that it has a distinct and more restricted target sequence specificity profile than yeast FTase.

Data availability

The yeast strains and plasmids used in this study will be available upon request. Supplementary File 1 contains the CXXX sequence frequencies of the E. coli, naïve yeast, 25°C, 37°C, and 42°C libraries and the NGS E-Scores.

Supplemental material available at G3 online.

Acknowledgments

We thank Dr. Avrom Caplan (City College of New York) for the anti-Ydj1 antibody, Drs. David Pellman (Dana Farber Cancer Institute, Harvard Medical School) and Satoshi Yoshida (Waseda University) for the SP319 Rho1 plasmid, Dr. James L. Hougland (Syracuse University) for providing valuable feedback on the manuscript, Dr. Zachary A. Wood (University of Georgia) for assistance with PyMol, and Schmidt lab members for constructive feedback and technical assistance.

Funding

This research was supported by U.S. Public Health Service grant GM132606 from the National Institute of General Medical Sciences (WKS).

Literature cited

Author notes