Functional dissection of the zDHHC palmitoyltransferase 5–golgin A7 palmitoylation complex

Abstract

Small molecules serve as valuable tools for probing nonapoptotic cell death mechanisms. The small molecule caspase independent lethal 56 (CIL56) induces a unique form of nonapoptotic cancer cell death that is promoted by a complex formed between zDHHC palmitoyltransferase 5 (ZDHHC5) and an accessory protein, golgin A7 (GOLGA7, also known as GCP16). The structure and function of this complex in nonapoptotic cell death regulation remain poorly understood. Here, we use coimmunoprecipitation, functional assays, and cryogenic electron microscopy (cryo-EM) to elucidate the structure and function of the Zdhhc5-GOLGA7 complex. We identify key residues in both Zdhhc5 and GOLGA7 that are necessary for complex formation and to promote nonapoptotic cancer cell death in response to caspase independent lethal 56. These results provide new insights into the structure and function of a death-promoting protein complex.

Drugs that can induce cell death are useful for the treatment of cancer. Synthetic small molecules that can induce cancer cell death in novel ways are therefore of great interest. Caspase independent lethal 56 (CIL56) was previously identified in a small molecule phenotypic screen as a new inducer of nonapoptotic cancer cell death (1). CIL56 treatment alters the morphology of multiple organelles, including the Golgi apparatus, the endoplasmic reticulum, and the mitochondria, and is associated with perturbation of anterograde and retrograde protein trafficking (2). CIL56-induced cell death is prevented by 5-tetradecyloxy-2-furoic acid (TOFA), a small molecule inhibitor of lipid synthesis (3, 4). This implies a link between lipid metabolism and CIL56-induced cell death. Chemical genetic screening and subsequent biochemical investigation pinpointed zDHHC palmitoyltransferase 5 (ZDHHC5) as an enzyme necessary for CIL56-induced cell death (2). ZDHHC5 is one of 23 mammalian enzymes that catalyze the S-acylation (“palmitoylation”) of hundreds of proteins in the cell, regulating protein localization, stability, and function (5, 6). Protein palmitoylation can regulate the execution of apoptosis and various forms of nonapoptotic cell death (7, 8, 9). The different ways that ZDHHC proteins regulate cell death is incompletely understood.

The family of 23 mammalian ZDHHC enzymes is defined by an Asp-His-His-Cys (DHHC) active site motif. The terminal cysteine residue of this motif is necessary for lipid transfer to substrate proteins (10). Many ZDHHC-family enzymes localize to the endoplasmic reticulum or Golgi apparatus. However, ZDHHC5, ZDHHC8, ZDHHC20, and ZDHHC21 show full or partial localization to the plasma membrane (11, 12, 13). Some ZDHHC enzymes require accessory proteins to function. This was first established in Saccharomyces cerevisiae, where the accessory protein Erf4 is required for the proper function of the ZDHHC9 ortholog Erf2 (14, 15). In mammalian cells, several different accessory proteins have been identified. ZDHHC6 complexes with SELENOK, ZDHHC9 and ZDHHC5 complex with golgin A7 (GOLGA7, also known as GCP16), and ZDHHC5 also complexes with a GOLGA7 paralog, GOLGA7B (2, 15, 16, 17, 18, 19). These accessory proteins may assist in substrate selection, stabilize enzyme intermediates, and/or enhance the stability of their respective ZDHHC partners (14, 16, 19, 20, 21). The full spectrum of ZDHHC complexes that can form in mammalian cells and how these complexes regulate cell phenotypes is unclear.

We previously identified the ZDHHC5-GOLGA7 complex biochemically and showed that it was essential for CIL56-induced cell death (2, 3). In ZDHHC5 gene-disrupted HT-1080 fibrosarcoma cells (“ZDHHC5^KO”), reexpression of WT mouse Zdhhc5 (which shares 98% sequence identity with the human protein, Fig. S1A) fully restored sensitivity to CIL56. A Zdhhc5 active site Cys to Ser mutant was able to interact with endogenous GOLGA7 but unable to restore cell death, implying that Zdhhc5 enzymatic activity was essential for CIL56 to induce cell death. A Zdhhc5 mutant where three putatively S-acylated C-terminal cysteine residues were mutated was mislocalized inside the cell, unable to interact with endogenous GOLGA7, and unable to restore CIL56 sensitivity when expressed in ZDHHC5^KO cells (2). This implies that the proper localization of enzyme-active ZDHHC5 to the plasma membrane, where it can complex with GOLGA7, is necessary for cell death. However, the detailed, atomic-level structure of the ZDHHC5-GOLGA7 complex is not known.

Here, we use co-immunoprecipitation (co-IP), functional assays, and cryogenic electron microscopy (cryo-EM) to map key residues necessary for the Zdhhc5-GOLGA7 complex to form and promote cell death in response to CIL56. We identify an “RNYR” sequence motif adjacent to the DHHC active site in Zdhhc5 and other ZDHHC-family enzymes as essential for interaction with GOLGA7. We also identify an “RDYS” motif found on GOLGA7 and GOLGA7B that is necessary for complex formation with ZDHHC5. Structural modeling by cryo-EM and AlphaFold 3 of the ZDHHC5-GOLGA7 complex reveals that the ZDHHC5^RNYR and GOLGA7^RDYS motifs contribute to the key interaction surfaces between these proteins. These results underscore structural elements necessary for ZDHHC5-GOLGA7 complex formation and the induction of nonapoptotic cell death.

Results

Multiple Zdhhc family proteins interact with GOLGA7

GOLGA7 was initially identified in complex with ZDHHC9 (17). As part of our previous studies, we showed using co-IP in 293T cells that GOLGA7 can, in addition, complex with both Zdhhc5 and the closely related family member Zdhhc8 (2). The observation that GOLGA7 binds at least three mammalian ZDHHC enzymes initially prompted us to ask whether GOLGA7 could interact with other ZDHHC family proteins as well. If so, we reasoned that the information encoded in these interactions could be used to elucidate key protein features necessary for complex formation.

To explore the scope of GOLGA7 interactors, we overexpressed hemagglutinin (HA)-tagged versions of all 23 mouse Zdhhc-family proteins (22) in 293T cells and assessed their ability to co-IP endogenous human GOLGA7 (Fig. 1A). The decision to use mouse Zdhhc proteins was driven by the availability of this library of constructs spanning the entire family, the fact that mouse Zdhhc proteins typically share high sequence identity with their human orthologs (median = 93%, range = 60–99%) (Fig. S1A), and the empirical observation that these HA-tagged Zdhhc proteins appear to localize and behave like the endogenous proteins in our biochemical and cell biological assays (2, 22). Endogenous GOLGA7 was detected with an antibody that we have previously validated, using GOLGA7 CRISPR knockout cell lines, to specifically detect this protein (2). These co-IP experiments were conducted in previously established 293T ZDHHC5 gene-disrupted (knockout, “KO”) cell lines as well as in unmodified control cells (2), with the hypothesis that in cells lacking ZDHHC5 additional molecules of GOLGA7 might be “freed up” to interact with other Zdhhc proteins. Multiple independent pull-down experiments were performed for each HA-Zdhhc protein, using HA-Zdhhc5 as a positive control. Immunoblot analysis demonstrated substantial expression of all Zdhhc proteins, except for Zdhhc6 and Zdhhc12, which expressed poorly (Fig. S1B). The relative expression of each PAT was calculated, and densitometry quantification of GOLGA7 binding to each overexpressed Zdhhc protein, normalized to the Zdhhc5-GOLGA7 interaction (positive control), was summarized as a heatmap (Figs. 1, B, C and S1B).

Figure 1.

GOLGA7 interacts with six mammalian ZDHHC enzymes.A, schematic illustrating coimmunoprecipitation screen of all mammalian ZDHHC enzymes for endogenous GOLGA7 binding. B, coimmunoprecipitation analysis of endogenous GOLGA7 binding in 293T ZDHHC5 gene-disrupted (”KO”) cell line transfected as indicated. Blot is representative of two independent experiments. C, heatmap representing percent GOLGA7 binding to each tested mammalian Zdhhc-family enzyme, based on western blot quantification, normalized to Zdhhc5 binding. Red color indicates binding to GOLGA7 in our screen. D, Phylo.io-generated neighbor-joining tree of all mouse Zdhhc enzymes based on sequence conservation. E, coimmunoprecipitation analysis of endogenous GOLGA7 binding in control (Ct) and ZDHHC5^KO cell lines transfected as indicated. Blot is representative of three independent experiments. Mm, Mus musculus; ZDHHC5, zDHHC palmitoyltransferase 5.

In ZDHHC5^KO cells, endogenous GOLGA7 was recovered in co-IPs from cells expressing Zdhhc1, Zdhhc5, Zdhhc8, Zdhhc9, Zdhhc14, and Zdhhc18, with the highest recovery observed with Zdhhc5. We compared this interaction pattern to the overall sequence similarity between all 23 Zdhhc proteins using a neighbor-joining tree (Fig. 1D). Zdhhc5, Zdhhc8, Zdhhc9, Zdhhc14, and Zdhhc18 were all found within a common branch, correlating with their ability to complex with GOLGA7. Zdhhc19 also clustered within this branch and was the only PAT in the cluster that failed to co-IP with endogenous GOLGA7 (Figs. 1C and S1B). By contrast, Zdhhc1 did co-IP with endogenous GOLGA7 despite sharing little sequence conservation with Zdhhc5 outside the cysteine-rich domain, DHHC active site, and two short motifs immediately downstream of the active site (discussed further below). Zdhhc1 therefore occupied a distant site on the tree from the Zdhhc5 cluster (Fig. 1D). Similar patterns of complex formation were observed in 293T control cells, albeit with Zdhhc14 showing higher recovery of endogenous GOLGA7 compared to ZDHHC5^KO cells, and a loss of detectable GOLGA7 recovery by Zdhhc1 (Figs. 1E and S1C, D). In sum, these data suggested that GOLGA7 was capable of forming complexes with at least six distinct ZDHHC proteins.

An RNYR motif is required for ZDHHC interactions with GOLGA7

The 6 mouse Zdhhc proteins that complexed with GOLGA7 were generally as similar to their human orthologs at the amino acid primary sequence level (94 ± 4%) as the Zdhhc proteins that failed to complex with GOLGA7 (89 ± 11%) (Mann-Whitney U, p > 0.05, Fig. S1A). Thus, differences in overall amino acid sequence did not appear to account for differences in complex formation. To investigate whether more specific sequence features better explained Zdhhc-GOLGA7 complex formation, we pursued a two-pronged approach that combined deletion mapping and computational inference.

First, we generated a series of Zdhhc5 deletion mutants lacking various regions of the protein, including the N terminus and transmembrane (TM) domains (HA-Zdhhc5^ΔN-term), the cytosolic loop containing the enzyme DHHC active site and cysteine rich domain (Zdhhc5^ΔDHHC), or the C terminus (HA-Zdhhc5^ΔC-term) (Fig. 2A). Note that the long C-terminal tail of Zdhhc5 may be prone to proteolytic degradation, resulting in the multiple bands observed on immunoblot. In addition, Zdhhc5 and other GOLGA7 interactors exhibit decreased stability and a propensity for oligomerization or aggregation when not occupied by GOLGA7 (21), sometimes resulting in an SDS-resistant cross-linked ladder on immunoblot (Fig. S2B). co-IP experiments demonstrated that endogenous GOLGA7 was recovered from 293T cells overexpressing full-length Zdhhc5 and, to a lesser degree, HA-Zdhhc5^ΔC-term (Figs. 2B and S2A). Consistent with these biochemical results, expression of full-length HA-Zdhhc5, and to a lesser degree HA-Zdhhc5^ΔC-term, restored CIL56-induced cell death in 293T ZDHHC5^KO cells, as detected by counting of SYTOX Green (SG)-positive dead cells (Fig. 2C). By contrast, overexpression of HA-Zdhhc5^ΔN-term and HA-Zdhhc5^ΔDHHC failed to restore CIL56 lethality, or to recover GOLGA7 in immunoprecipitates (Fig. 2, B and C). Although acknowledging the lower expression of HA-Zdhhc5^ΔDHHC protein (Fig. 2B), these results suggested that the cytosolic loop of Zdhhc5, which was absent from both mutants that did not complex with GOLGA7, might be essential for the interaction of Zdhhc5 with endogenous GOLGA7 in 293T cells.

Figure 2.

RNYR motif is a key interaction domain on GOLGA7 binders.A, schematic of Zdhhc5 deletion mutants. Blue bars represent transmembrane domains, black bars indicate S-acylated cysteines, and red represents the DHHC active site and conserved cysteine rich domain. The RNYR motif is indicated in orange. B, coimmunoprecipitation analysis of endogenous GOLGA7 binding in control (Ct) and ZDHHC5^KO cell lines transfected as indicated. Blots are representative of three independent experiments. C, cell death analysis of ZDHHC5^KO cell lines transfected as indicated. D, schematic of motif mapping strategy for interaction domain identification. Green indicates the protein of interest (POI) binder in this schematic. E, local amino acid alignment of all mouse Zdhhc-family proteins, highlighting the RNYR motif uniquely shared by GOLGA7 binders. F, coimmunoprecipitation analysis of GOLGA7 binding in Ct and ZDHHC5^KO cell lines transfected as indicated. Blots are representative of three to five independent experiments. G, densitometry quantification of GOLGA7 recovery in 293T ZDHHC5 Ct and KO cells transfected with the indicated HA-Zdhhc5 constructs. Recovery by each mutant was normalized to total HA-tagged bait protein for each lane and displayed relative to HA-Zdhhc5, which was set to 1.0. Each datapoint is an individual experiment. H, cell viability analysis of Ct and ZDHHC5^KO cell lines transfected with constructs corresponding to (F). I, coimmunoprecipitation analysis of endogenous GOLGA7 and ZDHHC9 binding in Ct and ZDHHC5^KO cell lines transfected as indicated. Data in (C) and (H) are from three independent experiments, while data in (I) is representative of two independent experiments. ZDHHC5, zDHHC palmitoyltransferase 5.

Protein-protein interactions often rely on short linear motifs within one or both binding partners (23, 24, 25). We therefore hypothesized that specific amino acid motifs may be essential for the binding of Zdhhc proteins to GOLGA7. Accordingly, we searched for conserved amino acid motifs greater than one residue in length that were found in Zdhhc-GOLGA7 interactors (Zdhhc1, Zdhhc5, Zdhhc8, Zdhhc9, Zdhhc14, and Zdhhc18) but absent from the noninteracting set of Zdhhc proteins (Fig. 2D). This analysis identified a single motif that was unique to GOLGA7 binders: RNYR (¹⁴⁴RNYR¹⁴⁷ in Zdhhc5). This motif lies within the cytosolic loop, ten residues C terminal to the DHHC motif, and the region defined by deletion mapping as essential for interaction with endogenous GOLGA7 (Figs. 2E and S2C). Thus, we hypothesized that the RNYR motif was essential for Zdhhc-GOLGA7 complex formation.

To test whether the RNYR motif was required for Zdhhc complex formation with GOLGA7, we generated a series of alanine substitution mutants in Zdhhc5, the strongest interactor. Single alanine substitutions of Arg144 or Tyr146 (“ANYR” or “RNAR”, respectively) did not disrupt interaction, as both mutants retained the ability to co-IP endogenous GOLGA7 and expressed at similar levels to WT protein (Fig. 2, F and G). However, “ANYA” and “RNAA” double mutants failed to complex with GOLGA7. Notably, the fourth residue in the RNYR motif, Arg147, appeared to be essential for GOLGA7 binding, as a single Arg147Ala mutation (“RNYA”) abolished interaction with endogenous GOLGA7 despite effective protein expression (Fig. 2, F and G). Consistent with these results, a triple alanine mutant (Zdhhc5^ANAA), which retained only the universally conserved asparagine residue, failed to co-IP endogenous GOLGA7 (Fig. 2, F and G). This sensitivity was striking given the otherwise high tolerance of the Zdhhc5-GOLGA7 complex to structural perturbation of Zdhhc5, including large deletions (Fig. 2B) and mutations in other conserved residues of this protein (Fig. S2D). To rule out the possibility that Zdhhc5–GOLGA7 co-IP reflected an indirect interaction bridged by another PAT, we confirmed that endogenous ZDHHC9 was not present in the HA-Zdhhc5 pull-downs from any condition (Fig. 2F). Mirroring the co-IP results, sensitivity to CIL56-induced cell death was restored in 293T^N ZDHHC5^KO cells by overexpression of WT HA-Zdhhc5 but not the triple alanine mutant HA-Zdhhc5^ANAA or the single-point mutant HA-Zdhhc5^RNYA (Fig. 2H). Note that in these experiments, we used cells that stably express the live cell marker, nuclear-localized mKate2 (denoted by the “N” superscript), which together with SYTOX Green and imaging, can be used to count live and dead cells and compute a lethal fraction (0 = all cells alive, 1 = all cell dead) (26).

Zdhhc5^RNYR motif mutants could fail to complex with GOLGA7 and restore CIL56-induced cell death because they no longer localize properly to the plasma membrane. For example, we previously found that the Zdhhc5 “3CS” mutant, which cannot be S-acylated on three cysteine residues, cannot traffic to the plasma membrane, complex with GOLGA7 or promote CIL56-induced cell death (2). Using 293T ZDHHC5^KO cells, we confirmed the aberrant accumulation of the Zdhhc5^3CS mutant throughout the cytoplasm surrounding the nucleus; by contrast, WT Zdhhc5 and the most disruptive RNYR motif mutant (Zdhhc5^ANAA) both localized predominantly to the plasma membrane, as determined by colocalization with the plasma membrane marker pan-Cadherin (Fig. S2, F–J). Thus, we reasoned that the RNYR motif might not govern Zdhhc5 intracellular localization but instead regulate cell death through its effects on complex formation with GOLGA7.

Our results thus far indicated that the Zdhhc RNYR motif was necessary for complex formation with GOLGA7. To determine whether this motif was sufficient for GOLGA7 complex formation, we installed the RNYR motif at the corresponding site in a GOLGA7 nonbinding protein, namely Zdhhc4 (¹⁸⁹WNTR¹⁹² → ¹⁸⁹RNYR¹⁹²). We also mutated the closely related motif found in the GOLGA7 nonbinder Zdhhc19 (¹⁵²RNFR¹⁵⁵) to the more optimal motif found in all GOLGA7 binders (¹⁵²RNYR¹⁵⁵). In both cases, installation of the RNYR motif alone did not enable co-IP of endogenous GOLGA7 when these proteins were expressed in 293T Zdhhc5^KO cells (Fig. 2I). Of further note, mutation of RNYR to RNFR in Zdhhc5 (to mimic the sequence of Zdhhc19 at this motif) did not eliminate endogenous GOLGA7 interaction, as assessed by co-IP (Fig. S2E). Thus, the Zdhhc RNYR motif appeared necessary but not sufficient for directing Zdhhc complex formation with GOLGA7.

A GOLGA7 motif required for ZDHHC5 complex formation

We find that ZDHHC5 can complex with GOLGA7 in 293T and HT-1080 cells (2). In other cell lines, ZDHHC5 can complex with GOLGA7B. GOLGA7B is a GOLGA7 paralog that shares 79% primary sequence identity but has a more tissue-restricted pattern of expression (18) (Fig. 3A). We were curious whether GOLGA7 and GOLGA7B could in principle function interchangeably with ZDHHC5 to regulate CIL56-induced cell death. To investigate, we performed co-IP assays in our established 293T GOLGA7^KO cells (2) transfected with FLAG-tagged GOLGA7B or GOLGA7. Overexpressed 3xFLAG-GOLGA7B pulled down endogenous ZDHHC5, as did the positive control constructs 1xFLAG-GOLGA7 and 3xFLAG-GOLGA7, the latter yielding a stronger anti-FLAG signal by immunoblot, consistent with improved antibody affinity for the 3xFLAG tag. Given its stronger expression and signal, 3xFLAG-GOLGA7 and was used in subsequent experiments (Fig. 3B). Of note, the recovery of endogenous ZDHHC5 by different GOLGA7 constructs appeared similar, suggesting that we were capturing the entire pool of “immunoprecipitable” ZDHHC5 in all cases.

Figure 3.

RDYS is a key ZDHHC interaction domain shared by GOLGA7, GOLGA7b, and Erf4.A, schematic of GOLGA7, GOLGA7B, and Erf4 structures indicating the conserved RDYS motif (orange) and acylated cysteines (black). B, coimmunoprecipitation analysis of endogenous ZDHHC5 binding from 293T Control (Ct) and GOLGA7 gene-disrupted (“KO”) cell lines transfected as indicated. Note: the 3xFLAG tag improves GOLGA7 detection. Blot is representative of two independent experiments. C, cell death analysis of cells transfected as in (B) and treated ± CIL56 (2.5 μM) and ± TOFA (1 μM) as indicated. D, schematic of GOLGA7 deletion mutants with S-acylated cysteines in black. E, coimmunoprecipitation analysis of endogenous ZDHHC5 binding in Ct and GOLGA7^KO cells transfected as indicated. Blot is representative of two independent experiments. F, cell death analysis of cells transfected as in (E). G, coimmunoprecipitation analysis of endogenous ZDHHC5, ZDHHC8, and ZDHHC9 in 293T Ct and GOLGA7^KO cells transfected as indicated. Blot is representative of four independent experiments. H, schematic summarizing data from (G), with GOLGA7 in green and GOLGA7B in orange. I, cell viability analysis of 293T^N Ct and GOLGA7^KO cells transfected as in (G). J, schematic of the Zdhhc5 and GOLGA7 complex at the plasma membrane, showing respective binding motifs and the DHHC active site. (C), (F), and (I) data are each from three independent experiments, except for the WT positive control in (F), which is from one replicate. CIL56, caspase independent lethal 56; TOFA, 5-tetradecyloxy-2-furoic acid; ZDHHC5, zDHHC palmitoyltransferase 5.

The ability of GOLGA7 and GOLGA7B to complex with ZDHHC5 corelated with the ability to promote CIL56-induced cell death: overexpression of either GOLGA7 or GOLGA7B in 293T GOLGA7^KO cells restored cell death in response to CIL56 (Fig. 3C). An inhibitor of CIL56-induced cell death, TOFA (2, 4), suppressed cell death across all conditions, indicating that GOLGA7 or GOLGA7B overexpression did not change the mode of cell death in response to CIL56. Thus, GOLGA7 and GOLGA7B can in principle function interchangeably to promote CIL56-induced cell death. Note that most cells appear to express GOLGA7 and not GOLGA7B (18), likely explaining why disruption of GOLGA7 alone in 293T and other cell lines is normally sufficient to cause CIL56 resistance. Nevertheless, we reasoned that we could use the shared ability of GOLGA7 and GOLGA7B to reconstitute cell death in 293T GOLGA7^KO cells in subsequent analyses to pinpoint key residues that were necessary for complex formation with Zdhhc5.

To identify residues in GOLGA7 that were necessary for complex formation with Zdhhc5 and thereby to promote cell death, we designed and tested a series of GOLGA7 deletion mutants encompassing distinct regions of the protein: Gly10-Glu35 (“Δ11–35”), Glu35-Tyr65 (“Δ35–65”), Pro3-Leu60 (“ΔN-term”), and Lys91-Arg137 (“ΔC-term”) (Fig. 3D). We overexpressed these proteins in 293T control and GOLGA7^KO cells and performed anti-FLAG immunoprecipitation, using 3xFLAG-GOLGA7 as a positive control. Although 3xFLAG-GOLGA7 co-immunoprecipitated endogenous ZDHHC5 as expected, none of the four GOLGA7 mutants recovered detectable levels of endogenous ZDHHC5, despite reasonably equal expression of all proteins (Fig. 3E). These biochemical data were mirrored by functional data, which showed that none of the GOLGA7 mutants could restore CIL56-induced cell death when expressed in 293T^N GOLGA7^KO cells (Fig. 3F).

Given that both GOLGA7 and GOLGA7B complex with ZDHHC5 and promote CIL56-induced cell death, we hypothesized that they shared one or more conserved motifs necessary for interaction with ZDHHC proteins. Multiple sequence alignment between GOLGA7 and GOLGA7B, as well as with the distantly related Schizosaccharomyces pombe accessory protein Erf4, identified a conserved four amino acid motif, ¹⁶RDYS¹⁹ in GOLGA7 and ²⁵RDYS²⁸ in GOLGA7 (Fig. 3A). Unlike the corresponding WT proteins, GOLGA7 and GOLGA7B quadruple-alanine mutants (RDYS → AAAA, “GOLGA7^AAAA” and “GOLGA7B^AAAA”) failed to co-IP endogenous ZDHHC5 in 293T GOLGA7^KO cells (Figs. 3G and S3B). This resembled the phenotype of an S-acylation-defective Cys-to-Ser mutant (GOLGA7^2CS) we identified previously (2), which was used here as a control (Fig. 3, G and H). Importantly, the GOLGA7^AAAA mutant did not always express at the same level as the WT protein or the GOLGA7^2CS mutant. However, when high GOLGA7^AAAA expression could be achieved this still failed to recover endogenous ZDHHC5 above mock/background levels (Fig. S3, A and B). Unlike WT GOLGA7 and GOLGA7B, the GOLGA7^AAAA and GOLGA7B^AAAA mutants also failed to restore sensitivity to CIL56-induced cell death when overexpressed in 293T^N GOLGA7^KO cells, much like the acylation-deficient GOLGA7^2CS control (2) (Fig. 3I). Thus, the RDYS motif, conserved between GOLGA7 and GOLGA7B, may be essential for functional interaction with ZDHHC5. The RYDS motif was not obviously related to protein S-acylation, suggesting that it could be important in some other way for complex formation between GOLGA7/B and ZDHHC proteins (Fig. 3J).

Unique functional determinants of GOLGA7B interaction with ZDHHC5

Despite the shared RDYS-dependent binding of GOLGA7 and GOLGA7B to ZDHHC5, their interactions were not identical. We previously suggested (2) that GOLGA7 palmitoylation is likely required for full interaction with ZDHHC5: the S-acylation-deficient GOLGA7^2CS mutant recovers far less endogenous GOLGA7 than the WT ZDHHC5 protein in co-IP assays (Fig. 3G). To investigate whether GOLGA7B S-acylation also promoted ZDHHC5 interaction, we generated an equivalent S-acylation-deficient 3xFLAG-GOLGA7B mutant (GOLGA7B^2CS). GOLGA7B^2CS expressed in GOLGA7^KO cells effectively recovered more endogenous ZDHHC5 than GOLGA7^2CS, albeit still lower amounts than WT GOLGA7 or GOLGA7B (Figs. 3, G, H and S3B). Interestingly, GOLGA7B^2CS overexpression was sufficient to restore CIL56 sensitivity in 293T^N GOLGA7^KO cells to WT levels, whereas GOLGA7^2CS overexpression was not (Fig. 3I). These results suggest that these two cysteine residues may be more important for GOLGA7 interaction with ZDHHC5 and for the induction of nonapoptotic cell death than the equivalent residues of GOLGA7B.

Given that GOLGA7 could complex with approximately one-quarter of all mammalian ZDHHC enzymes (Fig. 1C), and that both GOLGA7 and GOLGA7B interact with ZDHHC5, we hypothesized that GOLGA7B might also interact with other ZDHHC-family proteins in an RYDS-dependent manner. Indeed, co-IP experiments demonstrated that 3xFLAG-GOLGA7B pulled down endogenous ZDHHC8 in addition to ZDHHC5 (Figs. 3, G, H and S3B). However, we also noted important differences in the behavior of GOLGA7 and GOLGA7B in relation to complex formation with ZDHHC-family proteins. First, WT GOLGA7 but not WT GOLGA7B could co-IP ZDHHC9 (Figs. 3G and S3B). Second, the S-acylation deficient (2CS) mutant of GOLGA7B could complex to some degree with ZDHHC5 but not at all with ZDHHC8 (Figs. 3, G, H and S3B). Thus, GOLGA7 and GOLGA7B are not identical in their ability to form complexes with specific ZDHHC proteins (Fig. 3H). Nevertheless, our mapping approaches identified conserved motifs on both ZDHHC enzymes and on GOLGA7/B accessory proteins that appeared necessary for complex formation and the induction of nonapoptotic cell death (Fig. 3J).

Structural analysis of the Zdhhc5-GOLGA7 interaction

We hypothesized that residues identified above through biochemical and phenotypic analysis as necessary for Zdhhc5-GOLGA7 complex formation were located at interfaces necessary for physical interaction between these two proteins. To test this hypothesis, we used cryo-EM to experimentally determine the structure of the Zdhhc5-GOLGA7 complex, obtaining a 3D reconstruction at an estimated resolution of 3.9 Å (Fig. 4A). We built an atomic model of the complex using this map (Figs. 4, A, B, C and S4, A–D, and Table S1), guided by a high-confidence AlphaFold 3 (AF3) prediction of the Zdhhc5-GOLGA7 complex as a reference model. Side chains were manually modeled where resolvable in the cryo-EM map; for regions with limited side chain density, primarily within flexible cytosolic loops, we retained AF3-predicted rotamers during initial modeling. The final model was refined in Phenix using real-space refinement, incorporating secondary structure restraints and optimizing the fit to experimental density.

Figure 4.

Overview of Zdhhc5/GOLGA7 and comparison with ZDHHC9/GOLGA7.A, cryo-EM density map of the Zdhhc5-GOLGA7 complex overlaid with the hybrid model (Zdhhc5 in light green, GOLGA7 in gray). B, ribbon diagram of Zdhhc5–GOLGA7 model with Zdhhc5 colored from blue (N-terminus) to red (C terminus) for each chain, and GOLGA7 in gray. Gray horizontal lines represent estimated plasma membrane boundaries. C, ribbon diagram of Zdhhc5–GOLGA7 model with GOLGA7 colored from blue (N terminus) to red (C terminus), and Zdhhc5 in gray. D, schematic summary of the Zdhhc5/GOLGA7 complex, indicating relative positions of the DHHC active site, RNYR and RDYS motifs, and membrane orientation. E and F, ribbon diagram of ZDHHC9/GOLGA7 complex (PDB 8HF3, from (19)) shown with the same orientation and coloring as in (B) and (C) respectively, for comparison. PDB, Protein Data Bank.

Most of the Zdhhc5-GOLGA7 backbone could be confidently traced, including all four transmembrane helices, the cytosolic zinc finger domains, and the amphipathic helices of GOLGA7. As expected, certain cytosolic loop regions were less well resolved, particularly the TM2-TM3 loop of Zdhhc5, which exhibited moderate to low local resolution. These modeling limitations were consistent with the AlphaFold 3 per-residue confidence scores, which were high for structured regions and lower for flexible elements.

To summarize the overall architecture of the Zdhhc5–GOLGA7 complex, we generated a simplified schematic highlighting the membrane topology, key helices, and the spatial positioning of the RNYR and RDYS motifs within the assembled structure (Fig. 4D). For comparison, we also generated an AlphaFold 3 (27) model of human ZDHHC5 in complex with GOLGA7 and found that this model closely matched the reference structure (Fig. S5), supporting the close conservation between the human and mouse proteins.

In our cryo-EM model, Zdhhc5 comprises four TM helices, a large cytosolic loop between TM2 and TM3, and a cytosolic domain following TM4 that precedes the long C-terminal tail (Fig. 4B), which is largely unstructured and thus unresolved beyond Arg265. The TM2-TM3 loop features one α-helix and four anti-parallel β-strands that together form two zinc finger motifs stabilizing the DHHC active site. Two additional α-helices are present in the cytosolic region following TM4. GOLGA7 contributes two short amphipathic helices (α2′ and α3′) that insert into the membrane along Zdhhc5 (Fig. 4C). These helices occupy a similar position as the α5 helix found in the ZDHHC family member ZDHHC20 (19) (Fig. 4, C and F). This helix in ZDHHC20 plays a significant role in its autoacylation activity, reducing the rate by about half when mutated (10). Notably, these accessory-derived helices are also present in the ZDHHC9-GOLGA7 complex (21). The absence of this α-5 equivalent helix in Zdhhc5 and ZDHHC9 suggests that both enzymes may rely on GOLGA7 to fulfill a structurally analogous role, one that has been functionally validated in the case of ZDHHC9, where mutation of GOLGA7 Tyr76 (α3′) reduces catalytic activity (21).

Comparisons with the ZDHHC9-GOLGA7 complex structure

To assess architectural conservation across the PAT family, we compared our Zdhhc5-GOLGA7 model with a recently reported cryo-EM structure for the ZDHHC9-GOLGA7 complex (Protein Data Bank, PDB: 8HF3) (21), which revealed high structural similarity (Fig. 4, E and F). A TM score was used to assess the topological similarities between Zdhhc5-GOLGA7 and ZDHHC9-GOLGA7. Structural alignment using Foldseek online (28) predicted a tTM score (TM score normalized to the target length) of 0.82, where scores of 0.5 to 1.0 indicate similar fold or family (29).

Our model and the previously determined ZDHHC9 model (PDB: 8HF3) (21) indicated that both DHHC enzymes engage GOLGA7 through four primary interfaces. In our Zdhhc5-GOLGA7 model, the Zdhhc5 RNYR and GOLGA7 RDYS motifs were positioned within two distinct interfaces. Mutations in the Zdhhc5 RNYR motif (RNYR to ANAA or RNYA) disrupted binding, with the second arginine (Arg147) being essential for co-IP interaction (Fig. 2G). In our structural model (Fig. 5A), Zdhhc5 Arg147, located on TM3, interacted with the α2′ and α3′ helices of GOLGA7 (Fig. 5B). Side chains were consistent with a cation-π interaction between Zdhhc5 Arg147 and the aromatic ring of Tyr76 in GOLGA7 (Fig. 5B). These structural insights were fully concordant with the importance of Arg147 for complex formation discovered through biochemical and functional assays.

Figure 5.

Key residues at the ZDHHC-GOLGA7 interaction surface.A, the Zdhhc5-GOLGA7 complex (Zdhhc5 in light green, GOLGA7 in dark green). B, Arg147 from the RNYR motif (TM3) interacts with the α2′ helix of GOLGA7 at Tyr76. Hydrogen bonds are represented by dashed, yellow lines. C, GOLGA7 RDYS motif residues Arg16 and Tyr18, located in the loop following the β1′ strand, interact with Ile96, Pro115, and Glu128 of the zinc finger region. D, the ZDHHC9-GOLGA7 complex (ZDHHC9 in gray, GOLGA7 in dark green, PDB 8HF3, from (19)). E, Arg179 (RNYR motif) helps coordinate a phospholipid (yellow) and the PPII helix of ZDHHC9, which in turn docks into GOLGA7. Tyr183 (TM3) stacks with Tyr76 on the GOLGA7 α3′ helix. F, GOLGA7 RDYS residues Arg16 and Tyr18, in the loop following the β1′ strand, interact ZDHHC9 Phe129, and Pro150 and Glu165 within zinc finger domains. G, in the Saccharomyces cerevisiae Erf2-Erf4 complex (PDB: 8HFC (19)), R216 (RNYR motif, TM3) forms a hydrogen bond with Erf4 Asn164 and interacts with Leu168 on the α5′ helix. Blue indicates nitrogen and red indicates oxygen across all panels. PDB, Protein Data Bank; PPII, type II polyproline.

The GOLGA7 ¹⁶RDYS¹⁹ motif was likewise critical for complex formation with ZDHHC-family proteins. In our structural model, the GOLGA7 RDYS motif was located on a short loop following the β1′ strand and interfaces with the Zdhhc5 zinc finger motifs that stabilized the DHHC active site. Although the residues on this loop were not sufficiently resolved in our cryo-EM map, structural comparison to our reference model suggested that the first residue of the GOLGA7 RDYS motif (Arg16) likely interacted with Zdhhc5 Glu128 through a charge-charge interaction, potentially forming a salt bridge (Fig. 5C). In addition, GOLGA7 Tyr18 was positioned to interact with Zdhhc5 Pro115 through CH-π stacking interactions and may also form weaker interactions, such as transient van der Waals contacts, with Zdhhc5 Ile96. Together, these interactions suggested that GOLGA7 Arg16 and Tyr18 within the RDYS motif helped stabilize the active site conformation through binding to the zinc finger domain, consistent with their functional importance.

The structural importance of the Zdhhc5 RNYR and the GOLGA7 RDYS motifs was conserved in the ZDHHC9-GOLGA7 complex (PDB: 8HF3) (21) (Fig. 5D). ZDHHC9 Arg179 (RNYR) helped create a positively charged patch that coordinated a lipid-like density, proposed to be phosphatidic acid (19) (Fig. 5E). Given the structural conservation between ZDHHC9-GOLGA7 and Zdhhc5-GOLGA7, Zdhhc5 may similarly bind this lipid, though lipid density was unresolvable in our map. In ZDHHC9-GOLGA7, this interaction stabilized the TM2 and TM3 and type II polyproline (PPII) helices of ZDHHC9 and its interaction with GOLGA7. Tyr183 (RNYRY) on ZDHHC9 TM3 directly interacted with GOLGA7 Tyr76 on α3′ helix through π-π stacking (Fig. 5E), an interaction shown to be functionally important as mutation of Tyr76 decreased ZDHHC9 catalytic activity (19), underscoring the relevance of this conserved interface. As in our Zdhhc5-GOLGA7 structure, the GOLGA7 RDYS site also appeared to lie at a key interaction surface in the ZDHHC9-GOLGA7 structure at the ZDHHC9 zinc finger domains. GOLGA7 Arg16 interacted with ZDHHC9 Glu163 through charge-charge interaction, and Tyr18 forms π-π and CH-π stacking interactions with ZDHHC9 Phe129 and Pro150, respectively (Fig. 5F).

Despite the evolutionary divergence between yeast and human DHHC enzymes (only 31% identity between Erf2 and ZDHHC9) (Fig. S6), the key residue of the RNYR motif, Arg216 on Erf2 TM3, also formed an interactive domain with Erf4, the yeast GOLGA7 ortholog (Fig. 5G). The guanidinium group of Arg216 donated a hydrogen bond to the side-chain carbonyl oxygen of Erf4 Asn164 on the α5′ helix and may further stabilize the interaction via hydrophobic contacts with the aliphatic side chain of Erf4 Leu168. Thus, the structural importance of the RNYR motif appears conserved across species, from yeast to humans. Together, these structural insights support a conserved set of interaction interfaces that are recurrently used by ZDHHC-family PATs to engage accessory proteins such as GOLGA7.

Discussion

CIL56 is a lethal small molecule that induces an unusual form of nonapoptotic cell death dependent on the ZDHHC5-GOLGA7 protein complex (2, 3, 4). How this complex promotes cell death in response to CIL56 remains to be clarified. One possibility is that CIL56 directly activates protein S-acylation by the ZDHHC5-GOLGA7 complex, such that deletion of either ZDHHC5 or GOLGA7, or disruptions of protein complex formation, are sufficient to inhibit cell death. A different model is that the normal activity of the ZDHHC5-GOLGA7 complex becomes lethal to cells treated with CIL56, for example due to a toxic imbalance in protein S-acylation versus deacylation. Regardless, here, we could use CIL56 as a probe, leveraging resistance to cell death as a functional readout, to help identify key residues necessary for the assembly and function of the ZDHHC5-GOLGA7 complex itself. Amino acid residues identified through this approach, including the Zdhhc5 RNYR motif, were subsequently validated as essential for complex formation with GOLGA7 in our cryo-EM structure. This strategy may prove generally useful for mapping important amino acids in proteins involved in this and other forms of nonapoptotic cell death.

Our results suggest that assembly of the ZDHHC5-GOLGA7 complex is mediated by specific short motifs and individual amino acid residues present on both proteins. Within the ZDHHC5 RNYR motif, mutation of R147 to Ala (“RNYA”) alone was sufficient to completely disrupt co-IP with endogenous GOLGA7, despite robust expression of this mutant. The finding is rationalized within our Zdhhc5-GOLGA7 cryo-EM model, where Zdhhc5 R147 appears to contribute directly to the interaction interface. In addition to Zdhhc5, GOLGA7 also interacted to some degree with Zdhhc1, Zdhhc8, Zdhhc9, Zdhhc14, and Zdhhc18. We expect that these interactions require the RNYR motifs of each protein, but this remains to be experimentally verified. Zdhhc1, Zdhhc8, Zdhhc9, Zdhhc14, and Zdhhc18 localize to distinct subcellular compartments and function in growth factor signaling, immune signaling, and stem cell function (11, 12, 13, 30, 31). Thus, GOLGA7 may play a broad role in regulating protein S-acylation within the cell, and disruption of GOLGA7 may impair the function of multiple ZDHHC enzymes. At the same time, the ability of GOLGA7 and its paralog GOLGA7B to substitute for one another in some ZDHHC complexes raises the possibility of functional compensation in cells where both proteins are expressed.

Although GOLGA7 exhibits broad interactions, some ZDHHC enzymes, such as Zdhhc9, display striking selectivity. Although AlphaFold 3 predicts that GOLGA7B adopts a binding mode nearly identical to GOLGA7 in complex with ZDHHC9, our experimental data suggest that Zdhhc9 does not complex with GOLGA7B, at least when expressed in 293T cells. The structural basis for this selectivity is not apparent from current models, suggesting that specificity may be determined by subtle differences in surface features or conformational dynamics not captured by current analyses. Unlike ZDHHC5, ZDHHC9 contains unique loop conformations and surface charge features near the interface region, which may contribute to differential recognition of GOLGA7 versus GOLGA7B. Key differences between Zdhhc5 and ZDHHC9 are observed at two variable domains: the N terminus and the TM2-TM3 cytosolic loop. First, the N terminus (residues 9–15) of Zdhhc5 appears flexible and is not modeled, whereas the expanded N terminus in ZDHHC9 (residues 9–32) adopts a more complex loop-like structure. Second, the α-helix (α1) on the Zdhhc5 cytosolic loop between TM2 and TM3 was shorter than that on ZDHHC9, which formed a more distinctive loop of a different orientation. These variations between Zdhhc5 and ZDHHC9 were not structurally essential for interaction with GOLGA7 but may account for the inability of Zdhhc9 to interact with GOLGA7B. These differences were predicted in the AlphaFold 3 reference model and also found in our cryo-EM model; however, the resolution in these regions was limited, and it is unclear whether they reflect true structural divergence or model variability. Higher-resolution data or additional validation will be necessary to determine their functional significance.

Our analysis revealed that deletion of any major portion of GOLGA7 disrupted complex formation with Zdhhc5. However, sequence alignment allowed us to identify the conserved RDYS motif as important for complex formation and stabilization of the zinc-finger domains, an insight supported by our Zdhhc5-GOLGA7 model. Notably, mutation of Pro150 in ZDHHC9, which directly contacts the GOLGA7 RDYS motif, is a cause of X-linked intellectual disorders (19, 32). This raises the possibility that other mutations disrupting RDYS-mediated binding will likewise result in similar neurodevelopmental disorders. More generally, our findings suggest that both the identity and S-acylation state of accessory proteins fine-tune ZDHHC binding specificity. Notably, we find that Zdhhc5, Zdhhc8, and Zdhhc9 each exhibit distinct interaction profiles with GOLGA7 and GOLGA7B, demonstrating that even among closely related PATs, interaction specificity is not conserved. Future studies of other ZDHHC enzymes interacting with GOLGA7 and GOLGA7B could uncover further regulatory complexity and specificity within this critical protein modification pathway.

Our findings also offer insight into how GOLGA7 regulates ZDHHC enzyme activity. In our cryo-EM structure, the GOLGA7 RDYS motif makes direct contact with the zinc finger motifs that stabilize the DHHC active site of Zdhhc5, suggesting a structural mechanism by which GOLGA7 may enhance catalytic function. Prior studies in yeast have shown that Erf4, a GOLGA7 ortholog, enhances the catalytic activity of its DHHC partner Erf2 by stabilizing the active site intermediate (14). Similarly, GOLGA7 has been reported to be essential for ZDHHC9 palmitoylation activity (17). The fact that ZDHHC20, which lacks a requirement for an accessory protein, contains a membrane-inserted α5 helix in the region where GOLGA7 interacts with ZDHHC5 and ZDHHC9, further supports the notion that accessory proteins may functionally compensate for structural features absent in certain DHHC enzymes (10). Accessory proteins like GOLGA7 contribute to DHHC protein stability and membrane trafficking (14, 33), raising the possibility that GOLGA7 both promotes activity and protects ZDHHC5 from destabilization. Together, these observations reinforce the emerging paradigm in which accessory proteins dynamically tune the function, stability, and substrate engagement of membrane-modifying enzymes such as ZDHHC5.

Experimental procedures

Cell lines and culture conditions

The 293T (sex: female) (American Type Culture Collection, ATCC CRL-3216), HEK 293S GnTI^- cells (sex: female) (ATCC CRL-3022), and Sf9 cells (ATCC CRL-1711) were obtained from ATCC expanded for one passage, aliquoted, frozen at −80 °C and stored in liquid nitrogen for subsequent experiments. HT-1080^N (sex: male) cells were described previously (26). HT-1080^N CRISPR/Cas9 Control, ZDHHC5^KO1/2 and GOLGA7^KO1/2, and HEK 293T CRISPR/Cas9 Control, ZDHHC5^KO1/2 and GOLGA7^KO1/2 cell lines were described previously (2). Cell lines are validated periodically by short tandem repeat profiling and tested for contamination with Mycoplasma using the MycoStrip Mycoplasma detection kit (rep-mys-10, InvivoGen). All HT-1080 cell lines were cultured in Dulbecco’s modified Eagle high-glucose plus pyruvate medium (DMEM, Cat# MT-10-013-CV, Gibco) with 10% (v/v) fetal bovine serum (FBS, Cat# 26140-079, Gibco), 0.5 U/ml Pen/Strep (P/S, Cat#5070-063, Gibco), and 1× nonessential amino acids (Cat# 11140-050, Gibco). All 293T cell lines were cultured in DMEM supplemented with 10% FBS (v/v) and 0.5 U/ml P/S. All HT-1080 and 293T media were filtered through a 0.22 μM PES filter (Genesee Scientific) before use. Trypsin (Cat# 25200114, Gibco) was used for passaging of adherent cells. HT-1080 and 293T cells were counted using a Cellometer Auto T4 cell counter (Nexcelom). All HT-1080 and 293T cell lines were grown at 37 °C with 5% CO₂ in humidified tissue culture incubators (Thermo Fisher Scientific). HEK293S GnTI^- cells used for large-scale protein preps and cryo-EM were grown in suspension at 37 °C with 8% CO₂ in Freestyle 293 medium (Cat# 12338018, Gibco) supplemented with 2% (v/v) FBS and 0.5 U/ml P/S. Sf9 cells used for baculovirus production and Sf9 Easy Titer (Sf9-ET) (ATCC CRL-3357) cells used for titering were grown in suspension at 27 °C in Sf900 III SFM (Cat# 12658027, Gibco), and SFX-Insect Medium (Cat# SH30278.02, Hyclone) containing 50 μg/ml G418 (Cat# 10131-035, Gibco), respectively. HEK293S and Sf9 cells were counted manually.

Chemicals and reagents

CIL56 was synthesized by Acme Bioscience. TOFA (Cat# T6575) was from Sigma-Aldrich. SYTOX Green was from Life Technologies (Cat# S7020). All compounds were prepared as stock solutions in dimethyl sulfoxide (Cat# 276855, Sigma-Aldrich) and stored at −20 °C until use.

Nuclear mKate2-expressing cell lines

Cell lines stably expressing nuclear-localized mKate2 (denoted by the superscript “N”) were generated by lentiviral transduction. Previously established (34) HEK 293T CRISPR/Cas9 control cells and HEK 293T GOLGA7 ^KO cells were seeded at a density of 3 × 10⁴ cells per well in a twelve-well dish the day before transduction. Noninfected cells were also seeded separately for a simultaneous 2-fold puromycin dose response. Twenty-four hours later, the medium was replaced with 1 ml medium containing polybrene (8 μg/ml, Cat# H9268-5G, Sigma-Aldrich) and viral particles (Essen BioScience NLR puromycin red, Cat# 4625) at a multiplicity of infection of 3 TU/cell, calculated based on the manufacturer’s titer. Twenty-four hours later, the medium was removed and replaced with fresh medium. After 48 h, puromycin (Cat# A11138-03, Life Technologies) was added at a concentration determined by the puromycin dose response with noninfected cells (between 1–10 μg/ml). All nontransduced control cells were dead by day 3, and remaining infected cells were then transferred to larger plates in selection medium, expanded, and frozen down. HEK 293T CRISPR control and HEK 293T ZDHHC5^KO1/2 were seeded and mKate2 stably expressing cell lines were generated with a bleomycin-selectable lentiviral construct (Cat# 4627, Essen BioScience). After infection with bleomycin-selectable lentivirus with the above protocol, cell lines were sorted to isolate mKate2-positive cells using a FACSAria II Fluorescence Activated Cell Sorter (BD Biosciences) at the Stanford Shared FACS Facility.

Sequence alignment of PAT proteins

Clustal O v1.2.4 online (35) was used to generate a multiple sequence alignment of all mouse Zdhhc-family amino acid sequences using the following UniProtKB accession numbers: Q8R0N9 (Zdhhc1), P59267 (Zdhhc2), Q8R173 (Zdhhc3), Q9D6H5 (Zdhhc4), Q8VDZ4 (Zdhhc5), Q9CPV7 (Zdhhc6), Q91WU6 (Zdhhc7), Q5Y5T5 (Zdhhc8), P59268 (Zdhhc9), Q14AK4 (Zdhhc11), Q8VC90 (Zdhhc12), Q9CWU2 (Zdhhc13), Q8BQQ1 (Zdhhc14), Q8BGJ0 (Zdhhc15), Q9ESG8 (Zdhhc16), Q80TN5 (Zdhhc17), Q5Y5T2 (Zdhhc18), Q810M5 (Zdhhc19), Q5Y5T1 (Zdhhc20), Q9D270 (Zdhhc21), Q5Y5T3 (Zdhhc23), Q6IR37 (Zdhhc24), Q810M4 (Zdhhc25). Percent amino acid identity conservation between mouse and human PATs were determined by NCBI blastp (36) online multiple sequence alignment using the following accession numbers for the human enzymes: Q8WTX9 (ZDHHC1), Q9UIJ5 (ZDHHC2), Q9NYG2 (ZDHHC3), Q9NPG8 (ZDHHC4), Q9C0B5 (ZDHHC5), Q9H6R6 (ZDHHC6), Q9NXF8 (ZDHHC7), Q9ULC8 (ZDHHC8), Q9Y397 (ZDHHC9), Q9H8X9 (ZDHHC11), Q96GR4 (ZDHHC12), Q8IUH4 (ZDHHC13), Q8IZN3 (ZDHHC14), Q96MV8 (ZDHHC15), Q969W1 (ZDHHC16), Q8IUH5 (ZDHHC17), Q9NUE0 (ZDHHC18), Q8WVZ1 (ZDHHC19), Q5W0Z9 (ZDHHC20), Q8IVQ6 (ZDHHC21), Q8IYP9 (ZDHHC23), and Q6UX98 (ZDHHC24). Statistical tests were performed using the Mann–Whitney U test (two-tailed) using R v.4.4.3 through RStudio Desktop v.2024.12.1+563. To map conserved motifs unique to GOLGA7 binders, the aforementioned accession numbers were used to acquire text files of amino acid sequences for all 23 mouse Zdhhc proteins. Python v.3.12.8 was used to identify conserved sites grouped by length from these sequence files that were unique to GOLGA7 binders.

Molecular biology

The following Zdhhc mouse enzyme expression vectors (22) were the kind gift of Dr Masaki Fukata: pEF-BOS-HA, pEF-BOS-HA-DHHC1 (BC026570, Zdhhc1), pEF-BOS-HA-DHHC2 (NM_178395, Zdhhc2), pEF-BOS-HA-DHHC3 (NM_026917, Zdhhc3), pEF-BOS-HA-DHHC4 (NM_028379, Zdhhc4), pEF-BOS-HA-DHHC5 (NM_144887, Zdhhc5), pEF-BOS-HA-DHHC6 (NM_025883, Zdhhc6), pEF-BOS-HA-DHHC7 (NM_133967, Zdhhc7), pEF-BOS-HA-DHHC8 (AY668947, Zdhhc8), pEF-BOS-HA-DHHC9 (AK032233, Zdhhc9), pEF-BOS-HA-DHHC10 (AY668948, Zdhhc11), pEF-BOS-HA-DHHC11 (AY668949, Zdhhc23), pEF-BOS-HA-DHHC12 (BC021432, Zdhhc12), pEF-BOS-HA-DHHC13 (BC071194, Zdhhc24), pEF-BOS-HA-DHHC14 (BC059814, Zdhhc14), pEF-BOS-HA-DHHC15 (NM_175358, Zdhhc15), pEF-BOS-HA-DHHC16 (XM_129300, Q9ESG8, Zdhhc16), pEF-BOS-HA-DHHC17 (NM_172554, Zdhhc17), pEF-BOS-HA-DHHC18 (AY668950, Zdhhc18), pEF-BOS-HA-DHHC19 (BC049761, Zdhhc19), pEF-BOS-HA-DHHC20 (AY668951, Zdhhc20), pEF-BOS-HA-DHHC21 (NM_026647, Zdhhc21), pEF-BOS-HA-DHHC22 (NM_028031, Zdhhc13), and pEF-BOS-HA-DHHC23 (BC049767, Zdhhc25). pEF-BOS-HA-Zdhhc5^DHHS, Zdhhc5^3CS, and Zdhhc5^3YA expression vectors were described previously (2). pEF-BOS-3xHA-Zdhhc5 (2), the Q5 site-directed mutagenesis kit (Cat# E0554, NEB), and the Z5 RNFR, Z5 ANAA, Z5 ANYR, Z5 ANAR, Z5 ANYA, Z5 RNAR, Z5 RNAA, and Z5 RNYA primer sets (Table S2) were used to generate derivatives containing the following mutations in the Zdhhc5 coding sequence: Zdhhc5^RNFR, Zdhhc5^ANAA, Zdhhc5^ANYR, Zdhhc5^ANAR, Zdhhc5^ANYA, Zdhhc5^RNAR, Zdhhc5^RNAA, and Zdhhc5^RNYA (targeting the R144-R147 ‘RNYR’ domain). pCI-neo-Flag-Zdhhc5 (37) obtained from Addgene (plasmid #85812), the NEB Q5 site-directed mutagenesis kit, and the Z5ΔN-term, Z5ΔDHHC, and Z5ΔC-term primer sets (Table S2) were used to generate derivatives containing the following mutations in the Zdhhc5 coding sequence: Δ14 to 270 (“Zdhhc5^ΔN-term”), Δ104 to 154 (“Zdhhc5^ΔDHHC”), and Δ270 to 714 (“Zdhhc5^ΔC-term”). pEF-BOS-3xHA-DHHC4 and pEF-BOS-3xHA-DHHC19 plasmids, together with the Q5 site-directed mutagenesis kit and the DHHC4 RNYR and DHHC19 RNYR primer sets (Table S2), were used to generate Zdhhc4^RNYR and Zdhhc19^RNYR mutants that contained the RNYR motif and flanking residues. The Zdhhc4 mutant contained a substitution starting at A188, AWNTRYFLIY to RRNYRYFFLF, and the Zdhhc19 mutant starting at H151, HRNFRLFM to RRNYRYFF.

GOLGA7B-3xFLAG (Hs) was generated in the same backbone as our pDEST-pcDNA-GOLGA7-FLAG (Hs) construct. A gene block for use in Gibson assembly was designed from human GOLGA7B cDNA (NCBI Reference Sequence: NM_001010917.3) and ordered from Integrated DNA Technologies (IDT). The synthetized sequence contained a NotI site directly before the start codon for GOLGA7B, a tobacco etch virus protease site immediately following the final GOLGA7B codon, and a three-residue linker (Gly, Arg, Ala) which contains an AscI site, a 3xFLAG tag, a stop codon, and a HindIII restriction site, allowing for removal of the entire cassette. The gene block was flanked by an AflII site which occurs 23 bp upstream of the GOLGA7 start codon in the original vector, and XhoI site, which is immediately after the stop codon in the vector. For use as the backbone, pDEST-pcDNA5-GOLGA7-FLAG (2) was digested with AflII (Cat# R0520, NEB) and Xho1 (Cat# R0146, NEB), followed by gel purification. NEB’s Gibson assembly kit (Cat# E5510S) was used to insert the GOLGA7B gene block into the empty vector and subsequently transformed in NEB Stable Escherichia coli (Cat# C3040H). From this new GOLGA7B-3xFLAG plasmid, GOLGA7b^2CS, with mutations Cys78Ser and Cys81Ser, and GOLGA7B^RDYS (R25-S28 to AAAA) were generated using the Q5 site-directed mutagenesis kit and the GOLGA7b-C78,81S and GOLGA7b-RDYSΔAAAA primer sets, respectively (Table S2).

pDEST-pCDNA-GOLGA7-3xFLAG was generated using the pDEST-pcDNA-GOLGA7B-3xFLAG backbone. The GOLGA7 cDNA was PCR amplified from pDEST-pcDNA-GOLGA7-FLAG (2) using GOLGA7 Gibson insert primers (S1) with overlapping ends for Gibson assembly. An empty vector with complimentary ends for Gibson assembly was generated by PCR amplification of pDEST-pcDNA-GOLGA7B-3xFLAG using GOLGA7 Gibson vector primers (Table S2) and inserting the GOLGA7 PCR fragment using NEB’s Gibson assembly kit (Cat# E5510S).

pcDNA-GOLGA7^2CS-3xFLAG construct containing a tobacco etch virus protease site and three residue linker (Gly, Arg, Ala) connecting the C-terminus GOLGA7^2CS to the 3x-FLAG tag was generated using the GOLGA7^2CS-3xFLAG primer pair (Table S2), the previously described (2) GOLGA7^2CS-FLAG construct, and the Q5 site-directed mutagenesis kit (Cat# E0554, NEB). Four GOLGA7 deletion mutants (GOLGA7^Δ10-35, GOLGA7^Δ35-65, GOLGA7^ΔC-term, GOLGA7^ΔN-term) were generated from pDEST-pcDNA5-GOLGA7-FLAG (ref. (2)) using the Q5 site-directed mutagenesis kit and the G7d10-35, G7d35-65, G7dC-term, and G7dN-term primer pairs (Table S2). These primer pairs deleted segments Gly10-Glu35, Glu35-Tyr65, K91-R137, and P3-L30 from GOLGA7, respectively.

To generate N-St2-sfGFP-PPX-Zdhhc5 and N-FLAG-GOLGA7 constructs for cryo-EM, GOLGA7 (Hs) and Zdhhc5 (Mm) were cloned from pEF-BOS-3xHA-Zdhhc5 and pDEST-pcDNA5-GOLGA7-FLAG plasmids (2), into a pEG BacMam vector as described (38), to allow for baculovirus generation and large-scale expression. First, the Q5 site-directed mutagenesis kit (Cat# E0554, NEB) was used to add AscI and NotI restriction sites flanking each gene, allowing for easy transfer to BacMam plasmids. This was done using the Z5 AscI/Z5 NotI, and G7 AscI/G7 NotI primer sets, respectively (Table S2). Next, restriction digests of these two plasmids were performed with Asc I (Cat# FD1894, Thermo Fisher Scientific) and Not I (Cat# FD0593, Thermo Fisher Scientific) to obtain GOLGA7 and Zdhhc5 inserts. These inserts were gel purified and GOLGA7 was ligated into an AscI/NotI digested and the gel purified N-FLAG BacMam vector generated and described previously (39) containing an N-terminal FLAG tag. Zdhhc5 was ligated into an AscI/NotI digested and gel purified N-St2-sfGFP BacMam vector generation and described previously (39), which contained an N-terminal Strep II tag, superfolder GFP (40), and an HRV 3C protease site preceding the gene. These ligations were performed according to manufacturer protocol using Thermo Fisher Scientific T4 DNA Ligase (Cat# EL0016). All DNA gel purifications were performed using 1% agarose gel and the Qiagen QIAquick Gel Extraction Kit (Cat# 28704).

Transfection, cell lysis, and coimmunoprecipitation

For co-IP studies, 500,000 HEK 293T cells were seeded in 6-well plates 24 h prior to transfection, with two wells per construct. Each well containing 1 ml of media was transfected with 2.5 μg plasmid DNA, 6 μl Lipofectamine LTX Reagent, and 2.5 μl PLUS Reagent (Cat# 15338030, Invitrogen) in 250 μl Opti-MEM Reduced Serum Media (Cat# 31985-062, Life Technologies) according to manufacturer’s instructions, incubated overnight, and transfection media replaced with 2 ml fresh media the following day. After 24 h in fresh media, cells were washed once with HBSS (Cat# 14025-134, Life Technologies) and resuspended in 1 ml HBSS. Each pair of wells was combined and pelleted (2348g, 1 min), and the supernatant removed and discarded. Cell pellets were stored at −20 °C until lysis. Pellets were resuspended in lysis buffer (50 mM Tris–HCl pH 7.4, 150 mM NaCl, 2 mM EDTA, P8340 Protease Inhibitor Cocktail EDTA-free (100×) (Cat# 11836170001, Roche), 1% n-dodecyl-β-maltoside (DDM, Cat# D310, Anatrace), and passed through a 25-gauge needle 20 times to homogenize. Lysates were spun (18,000g, 15 min, 4 °C), supernatant removed, and the pellet discarded. Protein levels were quantified using the Pierce BCA protein assay (Cat# 23228 and 23224, Thermo Fisher Scientific). For pull-downs, 90 μg of cell lysate was set aside for SDS gel input. Lysate was added to 15 μl prewashed and equilibrated Anti-FLAG(R) M2 magnetic beads (Cat# M8823, Sigma-Aldrich, 50% slurry) or anti-HA magnetic beads (Cat# 88837, Pierce, Thermo Fisher Scientific) at 1 mg/ml in 900 ml and incubated at 4 °C overnight with end-over-end rotation. Beads were washed three times with 1 ml TBS (20 mM Tris base, 150 mM NaCl, pH 7.4) containing 0.5% DDM before elution into 30 μl Bolt LDS Sample Buffer (Cat# B0007, Life Technologies). Elution was carried out by vortexing for 5 min at 4 °C followed by incubation for 30 min at 37 °C. Eluted proteins were analyzed by immunoblotting.

Immunoblotting

Briefly, 90 μg of total protein (input) was combined with 4× Bolt LDS Sample Buffer, incubated in a 37 °C water bath for 30 min, and loaded onto a Bolt 4 to 10% Bis-Tris Plus Gel (Cat# NW04120BOX, Life Technologies). Eluates from the co-IP were loaded alongside the respective inputs. The Chameleon Duo Ladder (Cat# 928-60000, LI-COR) was used as a standard. Gels were run at 100 V for 1.5 h or until the dye front reached the end, and proteins were transferred to nitrocellulose membrane using the iBlot2 transfer stack (Cat# IB23002, Invitrogen). Membranes were blocked with Odyssey Blocking Buffer (Cat# 927-50000, LI-COR) (30 min, room temperature [RT]), and incubated in primary antibody mixture (overnight, 4 °C), except for GOLGA7 primary antibody (1 h RT). Membranes were washed three times in Tris buffered saline (20 mM Tris base, 150 mM NaCL, pH 7.4) with 0.1% Tween 20 (TBST), then incubated with secondary antibody mixture (1 h, RT). Following secondary incubation, membranes were washed three times in TBST and then scanned on a two-channel Odyssey CLx Imaging System (LI-COR). All antibodies were diluted in Odyssey Blocking Buffer. Antibodies used were rabbit α-ZDHHC5 (1:250, Cat# HPA014670, Sigma-Aldrich), mouse α-GOLGA7 (1:250, Cat# H00051125, Novus Biologicals), rabbit α-ZDHHC9 (1:250, Cat# NBP1-84499, Novus Biologicals), rabbit α-FLAG (1:1000, Cat# 14793S, Cell Signaling Technology), rabbit α-HA (1:1000, Cat# 3724S, Cell Signaling Technology) and mouse α-C4 actin (1:4000, Cat# SC47778, Santa Cruz Biotechnology), donkey α-rabbit (1:15,000, Cat# 926-32213, 926-68023, LI-COR), donkey α-goat (1:15,000, Cat# 926-32214, 926-68024, LI-COR), and donkey α-mouse (1:15,000, Cat# 926-32212, 926-68022, LI-COR).

GOLGA7 binding heatmap calculation

Using Image J, protein densitometry was calculated from the mouse Zdhhc enzyme co-IP screen for endogenous GOLGA7. Immunoblots were quantified for GOLGA7 input and pull-down, Zdhhc enzyme input and pull-down, and background. After pixel inversion and background subtraction for each band, percentage of total GOLGA7 recovered by each enzyme was calculated by dividing each GOLGA7 pull-down band by its respective input band. Percent of each enzyme occupied by GOLGA7 was also calculated by dividing GOLGA7 pull-down (G7^PD) by Zdhhc pull-down (Zdhhc^PD). This was normalized to Zdhhc5/GOLGA7 binding by dividing percent enzyme occupied by GOLGA7 (G7^PD/Zdhhc^PD) by percent Zdhhhc5 occupied (G7^PD/Zdhhc5^PD) as follows:

$$\frac{{G7}^{PD}/{Zdhhc}^{PD}}{{G7}^{PD}/{Zdhhc5}^{PD}}$$

Morpheus (https://software.broadinstitute.org/morpheus) was used to generate a heatmap representing percent GOLGA7 binding for each mammalian PAT enzyme relative to Zdhhc5 binding.

Densitometry quantification

Band intensities were quantified using ImageJ (NIH). For each band, a user-defined rectangular region of interest was drawn tightly around the visible band, and the local background was subtracted using an adjacent band-free area within the same lane to minimize variation due to blot heterogeneity. Integrated intensity values were obtained and, where applicable, co-immunoprecipitated signal was normalized to background subtracted bait protein levels from the same lane. For comparative visualization, values were expressed relative to the corresponding WT control, which was set to 1.0. Conditions yielding no visible signal above background, or yielding background-subtracted intensities of zero or less, were assigned a value of 0.01 for plotting purposes to enable visualization. Quantification was performed only on blots within the linear range of detection.

Cell death assessment using microscopy

Cell death was assayed using scalable time-lapse analysis of cell death kinetics (STACK) (26, 41). HEK 293T^N cell lines were seeded at 25,000 cells per well in 12 well plates. The next day, the medium was removed and replaced with medium containing 20 nM SYTOX Green (SG, Cat# S7020, Life Technologies) ± lethal compounds and inhibitors. Cells were imaged at 4 h intervals for 96 h using the Essen IncuCyte ZOOM live-cell analysis system (Essen BioSciences). Counts of mKate2-positive objects (mKate2⁺, live cells) and SG-positive objects (SG⁺, dead cells) were counted using IncuCyte ZOOM Live-Cell Analysis System software (Essen Biosciences). Image analysis parameter values were as follows for mKate2⁺ objects: parameter adaptive, threshold adjustment 1; Edge split on; Edge sensitivity 50; Filter area min 20 mm², maximum 8100 mm²; Eccentricity max 1.0; and SG⁺ objects: Parameter adaptive, threshold adjustment 10; Edge split on; Edge sensitivity −5; Filter area min 20 mm², maximum 750 mm²; Eccentricity max 0.9. Lethal fraction scores were computed in Microsoft Excel from mKate2⁺ and SG⁺ counts as described (26, 41). For any given timepoint n in a treatment time course (t = 0 → t = n) consisting of mKate2⁺, SG⁺, and double positive (SG⁺/mKate2⁺) cells, the lethal fraction (LF) was calculated as described (41) and by Equation 1.

$${LF}_{t-n}=1-\frac{{mKate2}_{t-n}^{+}-{SG}^{+}/{mKate2}^{+}}{\left({mKate2}_{t-n}^{+}-{SG}^{+}/{mKate2}^{+}\right)+{SG}_{{\mathit{max}}_{t=0\to t=n}}^{+}}$$

For HEK 293T ZDHHC5^KO and GOLGA7^KO cell lines that did not express mKate2, only SG-positive objects were counted. The following image extraction parameters were used: Parameter adaptive, threshold adjustment 1; Edge split on; Edge sensitivity −10; Filter area min 5 mm², maximum 800 mm²; Eccentricity max 0.9.

Immunofluorescence and confocal imaging

Subsequently, 60,000 293T ZDHHC5^KO cells were seeded per well in a 12-well plate on #1.5 glass coverslips coated with poly-D30 lysine 24 h prior to transfection. To each well containing 1 ml of medium was added 1 μg plasmid DNA, 3 μl Lipofectamine 3000 Reagent, and 6 μl P3000 Reagent (Cat# L3000001, Invitrogen) in 125 μl Opti-MEM Reduced Serum Media (Cat# 31985-062, Life Technologies) according to manufacturer’s instructions, incubated overnight, and transfection media replaced with 1 ml fresh media the following day. After 48 h in fresh media, coverslips were rinsed in 1× PBS and fixed in 4% paraformaldehyde (Alfa Aesar) for 20 min at RT. Cells were rinsed once with 1× PBS and permeabilized overnight at 4 °C in 1× PBS with 3% bovine serum albumin (Gemini Bio-Products) and 0.1% Triton X-100 (“PBS-BT”). Coverslips were transferred to a hydration chamber and covered in 30 μl PBS-BT. Thirty microliters of primary antibody mix was applied to coverslips for 1 h at RT. Primary antibodies used were rabbit α-ZDHHC5 (Cat# HPA014670, Sigma-Aldrich, 1:100) and mouse α-pan-cadherin (CH-19, ab6528, Abcam1:100). Following primary incubation, coverslips were washed three times with PBS-BT and incubated with 30 μl of secondary antibody mix (1:1000 of each secondary antibody, obtained from Life Technologies) for 1 h at RT. Secondary antibodies used were goat α-rabbit 568 (Cat# A11036, Life Technologies) and goat α-mouse 488 (Cat# A11029, Life Technologies). Coverslips were washed five times with PBS then mounted in ProLong Glass antifade reagent containing NucBlue stain (Cat# P36981, Thermo Fisher Scientific). Coverslips were cured overnight at RT in the dark before storage at 4 °C and imaging. Confocal images were obtained on a Zeiss LSM780 laser scanning confocal microscope (Carl Zeiss) with a 63×/1.4 NA plan apochromat objective.

Zdhhc5 and pan-cadherin expression was robust across every condition except the Zdhhc5 negative control which displayed no signal. Fluorescence imaging was performed with identical acquisition settings across conditions, including fixed laser power, gain, and exposure time. Because different fluorescent channels exhibited varying signal intensities, only threshold levels were adjusted for each channel to optimize visualization while maintaining the dynamic range, with no adjustment to brightness or contrast. These adjustments ensured that signal was neither saturated nor undetectable. As such, adjustments were not and should not be used for quantitative comparisons between channels. Fluorescence intensity line profile analysis was performed using FIJI v2.16.0 (42) for three cells per condition across two replicates, but signal was used to determine spatial localization and colocalization only, not to compare relative expression across conditions or between channels.

Homology modeling and structural analysis

AlphaFold 3 (27) was used to generate a model of mouse Zdhhc5 and human GOLGA7 in complex using amino acid sequences from the following UniProt accession numbers: Q8VDZ4 (Zdhhc5) and Q7Z5G4 (GOLGA7). The same was done for human ZDHHC5 (Q9C0B5) and GOLGA7. The ZDHHC9/GOLGA7 and Erf2/Erf4 cryo-EM structures (19) were imported into ChimeraX v1.8 (43) with the following PDB accession numbers: 8HF3 and 8HFC. Structural comparison and figure generation were carried out in ChimeraX and PyMOL (The PyMOL Molecular Graphics System, Version 3.0 Schrödinger, LLC). Molecular graphics and analyses performed with ChimeraX.

Protein expression and purification for cryo-EM

P3 baculovirus containing N-St2-sfGFP-PPX-Zdhhc5 (Mm) and N-FLAG-GOLGA7 (Hs) BacMam constructs were generated in Sf9 cells as described (38) and used to infect HEK293S GnTI^- cells. HEK293S GnTI^- cells were grown in suspension until they reached a density of 3 × 10⁶ cells/ml and then infected with P3 baculovirus at a multiplicity of infection of one for Zdhhc5 and four for GOLGA7. After 12 h, 10 mM sodium butyrate (Cat# B5887, Sigma-Aldrich) was added to the medium and the temperature reduced to 30 °C. The cells were harvested 48 h later by centrifugation (4800g, 20 min), supernatant discarded, and pellets washed once in PBS (pH 7.5; Cat# 10010023, Gibco), and pelleted again (4800g, 20 min). Supernatant was discarded and cell pellets were frozen in liquid nitrogen and stored at −80 °C until needed.

Cell pellets were thawed on ice and resuspended in 15 ml of lysis buffer per gram of cells. Lysis buffer was composed of 50 mM Tris/NaOH (pH 7.5), 375 mM NaCl, 1 μg/ml leupeptin (Cat# E18, Sigma-Aldrich), 1 μg/ml aprotinin (Cat# A6279, Sigma-Aldrich), 1 μg/ml pepstatin A (Cat# P5318, Sigma-Aldrich), 1 mM phenylmethylsulfonyl fluoride (Cat# 10837091001, Roche), 100× cOmplete EDTA-free protease inhibitor cocktail (Cat# 73567200, Roche), 5 mM β-mercaptoethanol (β-Me), 10% glycerol (w/v), and 10 μg/ml DNaseI (Cat# 04536282001, Roche). The homogenate was sonicated at 70% for 45 s (5 s on, 5 s off) with a probe sonicator, then the complex extracted by adding 1% (w/v) DDM with 0.2% (w/v) cholesterol hemisuccinate (CHS; Cat# C6512, Sigma-Aldrich) rotating for 2 h at 4 °C. The mixture was clarified by centrifugation at 80,000g for 30 min, and the supernatant was added to 15 μl MagStrep Strep-Tactin Type III beads (Cat# 15362006, IBA Lifesciences) per gram of cells and rotated at 4 °C for 1 h. The beads were collected on a magnetic rack, washed 3× with five column volumes (cv.) of 20 mM Hepes/HCl (pH 7.5), 150 mM NaCl (HBS) with 0.01% (w/v) glyco-diosgenin (GDN, Cat# 850525P, Avanti). After removing wash buffer, 25 μl of the GDN/TBS buffer was added to the beads along with HRV 3C protease (Cat# 71493, Novagen) at 10 U/mg of complex, and the complex was left to cut off the beads rotating overnight at 4 °C. The cut complex was removed from the beads using a magnetic rack and centrifuged at 20,000g for 20 min Zdhhc5/GOLGA7 concentration was calculated from its absorbance at 280 nm assuming an extinction coefficient (ε₂₈₀) of 57,190 M⁻¹ cm⁻¹ (calculated by ProtParam) (44). The sample was then concentrated to 2 mg/ml in a centrifugal tube (Amicon Ultra-4; 100-kDa cutoff, Millipore Sigma-Aldrich) and frozen on grids.

Protein quality was assayed by size-exclusion chromatography and SDS-PAGE. For size-exclusion chromatography, 4 μl of sample was injected onto a Superose 6 Increase 10/300 column (Cytiva) previously equilibrated with HBS with 0.01% (w/v) GDN and UV absorbance measured. For SDS-PAGE, 3 μl of cut final product and 10 μl of prep fractions collected (crude, flow-through, wash) were combined with 4× Bolt LDS Sample Buffer and 10 mM β-Me, incubated in a 37 °C water bath for 30 min, and loaded onto a Bolt 4 to 10% Bis-Tris Plus Gel. The Chameleon Duo Ladder was used as a standard. The gel was Coomassie stained overnight with SimplyBlue SafeStain (Cat# LC6060, Invitrogen), destained in water, and viewed on a Bio-Rad ChemiDoc imager.

Cryo-EM sample preparation and data acquisition

Cryo-EM grids were frozen using a Vitrobot Mark IV (Thermo Fisher Scientific) as follows: 3 μl of the sample at RT was applied to a glow-discharged Quantifoil R1.2/1.3 holey carbon 400 mesh gold grid (Quantifoil Micro Tools), blotted for 2.5 to 4 s in 100% humidity at 22 °C, and plunge-frozen in liquid ethane cooled by liquid nitrogen. Grids were screened for ice thickness and particle distribution using a Glacios transmission electron microscope (200 kV; Thermo Fisher Scientific) in the Yale School of Medicine Center for Cellular and Molecular Imaging. High-resolution data were acquired on a Titan Krios (300 kV, Thermo Fisher Scientific) operated at 300 kV in the Yale West Campus Cryo-EM Core, equipped with a Gatan K3 direct electron detector and 20 eV energy filter slit. SerialEM (version 4.1-beta) (45) was used for automated data collection. Movies were collected in super-resolution mode at a nominal magnification of 105000×, corresponding to a calibrated physical pixel size of 0.832 Å per pixel and a super-resolution pixel size of 0.416 Å. The total exposure was 80 e^-/Ų distributed over 50 frames, with an exposure time of 10 s with a defocus range of approximately −0.8 to −2.0 μm.

Cryo-EM data analysis and model building

For data processing, see Fig. S6. In brief, 16,728 dose-weighted micrographs were imported into CryoSPARC Live (v4.3, Structura Biotechnology Inc) and subjected to motion correction (46), patch-based CTF estimation, particle picking, and 2D classification in CryoSPARC (47). An initial set of 2,092,781 particles was autopicked using a blob picker and extracted with a box size of 384 pixels, binned over 2 × 2 during early processing. After 2D classification, a subset of 1,478,985 particles was selected and used to generate an initial ab initio reconstruction. This map was eventually used as a reference for template-based particle picking, followed by re-extraction without pixel binning and additional rounds of 2D and 3D classification.

From the repicked dataset of 1,383,157 particles, 694,239 particles were selected from 2D classification to remove junk particles and used for ab initio reconstruction using a three-class model with a maximum resolution of 12 Å, initial resolution of 35 Å, and minibatch size of 300 to 1000. The best resolved class (272,870 particles) was further refined in a second ab initio job with stricter parameters (max resolution 6 Å; initial resolution 9 Å), followed by a third round on 141,999 selected particles (max resolution 4.5 Å; initial resolution 6 Å). The best resolved class (93,643 particles) was subjected to nonuniform refinement (48), resulting in a 5.9 Å map, and finally local refinement using a masked receptor region with alignment priors, yielding the final 3.9 Å map.

We used AlphaFold 3 (27) to generate a reference structure of Zdhhc5-GOLGA7. In the Zdhhc5-GOLGA7 reference model, the structured regions of Zdhhc5 were predicted with very high accuracy. AlphaFold 3 predicted local distance difference test (pLDTT) scores, a per-atom confidence estimate on a 0 to 100 scale (where higher values indicate higher confidence), were high (>90) for all structured elements of the complex. Flexible loops and helical termini exhibited moderate confidence (90 > pLDTT > 70), apart from the unstructured N-terminal residues and the least structured loop region between TM2 and TM3 (residues ∼76–90), which ranged from confident scores to very low scores (<50). The C terminus of ZDHHC5 (residues 260–715) was not modeled. The predicted template modeling (pTM) and the interface predicted template modeling (ipTM) scores are additional metrics of the accuracy of protein structure. The pTM score for Zdhhc5-GOLGA7_AF3 was 0.5, where 0.5 or above indicates the overall predicted fold for the complex is a reasonable prediction of the native structure (27, 29, 49). The ipTM measures the accuracy of the predicted relative position between subunits in a complex. The ipTM for Zdhhc5-GOLGA7_AF3 was 0.89, where values > 0.8 represent confident predictions (27, 29, 49). While the AlphaFold 3 model provided a reliable starting point for modeling from our cryo-EM map, manual adjustment of backbone geometry in Coot v.0.9.8.8 (50) was required during model building, particularly for the orientation of TM helices, to better fit the density. The model was further refined using Phenix real space refinement with default parameters (51). The resolution was set to 6 Å to avoid overfitting. Images of the model were created using UCSF ChimeraX and PyMOL.

Software

Data were analyzed and graphed using Excel v.16.95.4 (Microsoft Corp.), GraphPad Prism v.10.4.0 (GraphPad Software Inc), and Adobe Illustrator (Adobe Inc. v.29.4).

Data availability

The final cryo-EM map has been deposited into the Electron Microscopy Data Bank under the accession number EMD-70274. The final model has been deposited into the PDB under accession number 9OA6. Coordinates for ZDHHC9/GOLGA7 and Erf2/Erf4, used for structural comparisons in this paper, were obtained from the PDB under accession numbers 8HF3 and 8HFC, respectively. For all other data requests, contact S. J. D.

Supporting information

This article contains supporting information.

Conflict of interest

The authors declare that they have no conflicts of interest with the contents of this article.

Reviewed by members of the JBC Editorial Board. Edited by Donita C. Brady