Local and substrate-specific S-palmitoylation determines subcellular localization of Gαo

Peripheral membrane proteins (PMPs) associate with cellular membranes through post-translational modifications like S-palmitoylation. The Golgi apparatus is generally viewed as the transitory station where palmitoyl acyltransferases (PATs) modify PMPs, which are then transported to their ultimate destinations such as the plasma membrane (PM). However, little substrate specificity among the many PATs has been determined. Here we describe the inherent partitioning of Gαo – α-subunit of heterotrimeric Go proteins – to PM and Golgi, independent from Golgi-to-PM transport. A minimal code within Gαo N-terminus governs its compartmentalization and re-coding produces G protein versions with shifted localization. We establish the S-palmitoylation at the outer nuclear membrane assay (“SwissKASH”) to probe substrate specificity of PATs in intact cells. With this assay, we show that PATs localizing to different membrane compartments display remarkable substrate selectivity, which is the basis for PMP compartmentalization. Our findings uncover a mechanism governing protein localization and establish the basis for innovative drug discovery.


Results
The minimal localization code in the N-terminus of Gαo. Early studies using metabolic labeling with [ 3 H]myristate and [ 3 H] palmitate demonstrated that Gαo membrane association is mediated by N-myristoylation at Gly2 and S-palmitoylation at Cys3 in its N-terminus [15][16][17] . A recent structural analysis identified the recognition sequence of N-myristoyltransferases (NMTs) as an N-terminal hexapeptide, excluding Met1 18 . This suggests that a minimal membrane-binding information might reside within the first seven residues of Gαo. Thus far, three crystal structures of heterotrimeric Go have been solved, showing a prominent α-helix in Gαo N-terminus that extends toward the Gβγ heterodimer. Overlay of the N-termini of these structures revealed the α-helixes to start at position 6 to 10, and to end at position 31 ( Supplementary Fig. 1a-c). Similarly, overlay of the N-termini of seven solved structures of Gαi1-a close Gαo homologue-showed the α-helix between the residues 7-8 to 31 (Supplementary Fig. 1d-f). Thus, the N-termini of Gαo and its homologs contain distinct regions: the unstructured lipidated heptapeptide to be followed by the α-helix. To study if these regions have specific roles in Gαo subcellular localization, we generated the following GFP-fusion constructs (Fig. 1a): one including Gαo N-terminal heptapeptide (Gαo-Nt 7 -GFP), another with the first 31 residues covering the α-helix (Gαo-Nt 31 -GFP), and a third containing only the α-helix (Gαo-Nt 8-31 -GFP). These constructs were expressed in the mouse neuroblastoma Neuro-2a cells (N2a) and their localization at the PM and Golgi apparatus were compared with the full-length Gαo-GFP 5 . Surprisingly, Gαo-Nt 7 -GFP was predominantly at the Golgi with a weak PM localization, whereas Gαo-Nt 31 -GFP displayed a more homogenous PM and Golgi distribution similar to Gαo-GFP ( Fig. 1b-d). Quantification of average fluorescence intensities at these compartments confirmed a much higher Golgi and a lower PM localization of Gαo-Nt 7 compared to Gαo-Nt 31 and Gαo (Fig. 1e, f; see Methods for a detailed description), despite similar expression level of the constructs ( Supplementary Fig. 1g, h). Then, we performed a crude subcellular fractionation of N2a cells expressing the constructs and showed that Gαo-Nt 7 , Gαo-Nt 31 and Gαo were similarly partitioned between the cytosolic and membrane fractions (Fig. 1g, h). On the other hand, Gαo-Nt  was spread over the cytosol and nucleus ( Supplementary Fig. 1i), indicating that the N-terminal α-helix alone is not sufficient for membrane association.
In the heterotrimeric G protein complex, the N-terminal αhelix of Gα binds to Gβ 19 . Since Gβ tightly interacts with Gγ, which in turn associates to membranes via its C-terminal prenylation 20 , we tested if the poor PM localization of Gαo-Nt 7 relates to a lack of Gβγ interaction. For this aim, we co-expressed the Gαo constructs together with mRFP-Gβ1 and mRFP-Gγ3 in N2a cells, and immunoprecipitated the GFP-fusions. Full-length Gαo-GFP strongly pulled down Gβ1γ3, whereas Gαo-Nt 7 -GFP and Gαo-Nt 31 -GFP showed a very faint co-precipitation of Gβγ with no apparent difference between them ( Supplementary  Fig. 1j). Thus, the preferential localization of Gαo-Nt 7 to the Golgi is independent from Gβγ.
Together, these results indicate that Nt 7 is sufficient for Gαo overall membrane binding, but it tends to drive Golgi rather than PM localization. ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-  Key residues in Gαo-Nt 7 . We next aimed to decode the rules of intracellular localization of Gαo-Nt 7 by a systematical point mutation analysis of its residues. Amino acid substitutions were designed using the GPS-Lipid prediction tool 21 . For instance, we introduced the Gly-to-Leu mutation at position 2 of Gαo-Nt 7 -GFP as GPS-Lipid predicted very high likelihood for S-palmitoylation at Cys3 in the mutant. However, the Gly2 mutation led to a complete loss of membrane association as Gαo-G2L-Nt 7 evenly spread over the cytosol and nucleus (Fig. 1i). This result supports the notion that N-myristoylation is a prerequisite for S-palmitoylation in Gαo 6 . Accordingly, Gαo-G2L-Nt 7 appeared almost exclusively in the cytosolic fraction (Fig. 1k,l). 7 1 7 GDo-Nt -GFP: M GCTLSA -mGFP GDo-GFP GDo-GFP GM130 7 GDo-Nt -GFP 7 GDo-Nt -GFP GM130 31 GDo-Nt -GFP 31 GDo-Nt -GFP GM130 j 7 GDo-G2L-Nt -GFP GDo-S6A-Nt -GFP 7 GDo-S6C-Nt -GFP 7 GDo-S6G-Nt -GFP 7 GDo-S6T-Nt -GFP s DDD85646 P (10 M; 15 h) 7 GDo-S6A-Nt -GFP 7 GDo-S6C-Nt -GFP 7 GDo-S6G-Nt -GFP 7 GDo-S6T-Nt -GFP We then introduced the Cys-to-Asn mutation at position 3 as GPS-Lipid predicted a high score for N-myristoylation at Gly2 in the mutant. The resulting Gαo-C3N-Nt 7 -GFP was predominantly excluded from the nucleus and showed an ER-like distribution with certain accumulation at Golgi, which might account for the high membrane density of the cisternae stack (Fig. 1j). Nevertheless, Gαo-C3N-Nt 7 appeared mostly in the cytosolic fraction (Fig. 1k, l), confirming that N-myristoylation confers only weak membrane-binding properties 7 . The distinct localization patterns of Gαo-Nt 7 mutants impaired in N-myristoylation and S-palmitoylation were not associated with significant variations in their expression levels ( Supplementary Fig. 1k, l), and were emulated by the treatment with the specific inhibitors DDD85646 22 and 2-bromopalmitate (2-BrPal) 23 , respectively (Fig. 1m, Fig. 1n-p), yet a weak labeling was still observed for the Gαo-C3N mutant (~10% of Gαo) that was additionally reduced by 2-BrPal (Supplementary Fig. 1o-q). This marginal S-palmitoylation of GFP in the G2L and C3N constructs is likely stochastic and does not contribute to their membrane association (Fig. 1k, 7 , withd enoting exclusion of residues, and X-any amino acid 18 . The authors did not exclude the possibility that peptides lacking a positive charge at position 7 might also be accepted by NMTs. Thus, N-myristoylation of Gαo-and all human N-myristoylated Gα subunits-is not in conflict with the lack of a Lys/Arg residue at position 7 ( Supplementary Fig. 1a, d). At position 6, a Ser residue is found in Gαo and its homologues, and Ser is by far the most frequent residue at this position among NMT substrates 18 . Thus, we first introduced point mutations at Ser6 in Gαo-Nt 7 based on the substrate recognition motif described by Castrec et al. and on a previous study indicating that Arg, Asn, Phe, and Val can occupy this position as well 24 . Substitution of Ser6 by Ala, Cys, Gly, or Thr showed efficient membrane binding (Fig. 1r) and expression levels similar to Gαo-Nt 7 ( Supplementary Fig. 2a, b). Interestingly, the S6C mutant displayed an increased PM localization as compared to the wild-type Gαo-Nt 7 -GFP (Fig. 1r, t) and a migration shift in SDSpolyacrylamide gels ( Supplementary Fig. 2a), suggesting that Cys3 and Cys6 both undergo S-palmitoylation and that PM targeting might be enhanced by the dual S-palmitoylation. N-myristoylation of these mutants was indirectly verified with the inhibitor DDD85646 (Fig. 1s). Conversely, we observed a largely cytosolic and nuclear localization for the constructs in which the Ser6 was substituted by Arg, Asn, Phe and Val ( Supplementary Fig. 2c), suggesting that such amino acid substitutions make the N-terminal peptide of Gαo a poor substrate for MNTs; the expression levels of the mutant constructs did not significantly differ from Gαo-Nt 7 ( Supplementary Fig. 2a, b). We conclude that the lipid modifications and subcellular localization of the Gαo-Nt 7 constructs are highly sensitive to even minor changes in this heptapeptide sequence.
( Fig. 1k, l) were not significantly different from that of Gαo-Nt 7 . On the other hand, the Cys5 construct MGNTC-Nt 7 -GFP localized at the Golgi and cytosol (Fig. 2b, d) but appeared mostly in the cytosolic fraction upon cell fractionation (Fig. 1k, l). Thus, it appears that Cys5 is not a good substrate for PATs, as MGNTC-Nt 7 closely resembles the C3N mutant ( Fig. 1j-l).
Accordingly, metabolic labeling showed a wild-type-like [ 3 H] palmitate incorporation for MGNC-Nt 7 and only the residual signal for MGNTC-Nt 7 ( Supplementary Fig. 1m). S-palmitoylation of MGNC-Nt 7 was additionally confirmed by its sensitivity to the inhibitor 2-BrPal (Fig. 1o, p and Supplementary Fig. 2f). Altogether, these data uncover a remarkable characteristic of the N-terminal localization code of Gαo: minor changes in its amino acid sequence can drastically modify the subcellular localization of the protein. Due to its homology with all duallipidated Gα subunits, and to other similarly modified PMPs, we reasoned that this principle might have a broad and evolutionary conserved relevance in biology. Resultingly, MGSLCSR-Nt 7 -GFP showed stronger membrane binging than the initially designed MGNTC-Nt 7 localizing predominantly at the Golgi with low cytosolic signal (Fig. 2h) despite similar expression levels ( Supplementary Fig. 1k, l). When MGSLCSR was introduced into the full-length Gαo, however, the cytosolic localization of this mutant was well visible (Supplementary Fig. 2j), although its expression was comparable to Gαo-GFP ( Supplementary Fig. 2g, h). This might probably relate to the fact that Gα subunits carrying a Cys at position 5 are virtually absent in Metazoa (Supplementary Notes and Supplementary Data 1).
Then, we expressed our constructs Gαo-Nt 7 -GFP, MGNC-Nt 7 -GFP and the consensus MGSLCSR-Nt 7 -GFP in the Drosophila Schneider-2 cell line. Remarkably, all constructs retained the same localization patterns as seen in N2a cells: Gαo-Nt 7 and MGSLCSR-Nt 7 targeted mainly Golgi stacks labeled by GalT-mRFP, while MGNC-Nt 7 associated largely with the PM ( Fig. 2i-k). Similar localizations were also seen in HeLa cells ( Supplementary Fig. 2k). These data point to highly conserved rules for substrate compartmentalization across cell types and species.
Direct targeting of Nt 7 variants to Golgi or PM. Our next goal was to understand the molecular mechanism behind the contrasting subcellular localization of two rather similar sequences: Gαo-Nt 7 and MGNC-Nt 7 (Figs. 1c and 2a). N-myristoylation generally occurs during protein synthesis and confers a weak membrane binding to the modified protein 23 , and is essential for the subsequent S-palmitoylation to occur 6 . Our own data showed that myristoylated, but not palmitoylated Gαo-Nt 7 and MGNC-Nt 7 are indistinguishable in their localization and only prevented from free cytosolic and nuclear diffusion ( Fig. 1n and Supplementary Fig. 2f). Therefore, we speculated that S-palmitoylation is key for subcellular compartmentalization, by itself or in combination with the myristoyl group and/or neighboring amino acids. It has been suggested that S-palmitoylation of PMPsincluding Gα subunits-occurs exclusively at the Golgi, an event followed by their transport to the PM via the secretory pathway 12 . In addition to the Golgi-to-PM trafficking, a proper steady-state localization of PMPs is controlled by their rapid and ubiquitous depalmitoylation 12 . Thus, we next explored whether the distinct localizations of Gαo-Nt 7 and MGNC-Nt 7 might involve a differential depalmitoylation and/or Golgi-to-PM transport of the constructs.
We performed live imaging of N2a cells expressing Gαo-Nt 7 -GFP or MGNC-Nt 7 -GFP during incubation with the depalmitoylation blocker Palmostatin B, an acyl protein thioesterase 1 and 2 inhibitor 26 . Localization of Gαo-Nt 7 and MGNC-Nt 7 in control cells was not affected during the 45 min of recording ( Supplementary Fig. 2l, m), whereas both constructs showed progressive changes upon Palmostatin B addition. Particularly, the signal of Gαo-Nt 7 at Golgi diffused over the surrounding area upon time, whereas its PM localization seemed not to be affected ( Fig. 2l and Supplementary Movie 1). Accordingly, quantification revealed that its PM content did not significantly change during the entire recording time ( Supplementary Fig. 2n), but its colocalization with the MannII-mRFP Golgi marker showed a significant decrease starting at 25 min of treatment (Fig. 2n). The PM signal of MGNC-Nt 7 presented no major changes during Palmostatin B treatment as well, but its overall presence in the perinuclear region slightly increased in a pattern that did not resemble the Golgi (Fig. 2m and Supplementary Movie 2). In fact, MGNC-Nt 7 co-localization with the Golgi marker and PM Fig. 2 Position of the Cys in Gαo N-terminus. a-d N2a cells expressing the Nt 7 -GFP mutants MGNC (a) or MGNTC (b) and immunostained against GM130 (a, b). Nuclei in blue stained with DAPI. Color-channels are listed vertically top-to-bottom and selected areas are magnified to the right with the channels displayed vertically in the same order. Underlined letters indicate residues substituted in Gαo-Nt 7 . Mean fluorescence intensity ratios of GFPfusions at the PM versus total cell (c); four independent experiments (Gαo-Nt 7 , n = 56; MGNC-Nt 7 , n = 50; Gαo, n = 58). Co-localization analysis of the Nt 7 -GFP constructs with GM130 (d); three independent experiments (Gαo-Nt 7 , n = 33; MGNC-Nt 7 , n = 36; MGNTC-Nt 7 , n = 39). e-g N2a cells expressing the MGNC mutant of full-length Gαo (MGNC-Gαo-GFP) and immunostained against GM130 (e). Boxed area is enlarged. Underlined letters indicate residues substituted in Gαo. Relative localization of GFP-fusions at the PM (f); four independent experiments (Gαo, n = 56; MGNC-Gαo, n = 46; MGNC-Nt 7 , n = 50). Co-localization with GM130 (g); three independent experiments (Gαo, n = 47; MGNC-Gαo, n = 38; MGNTC-Nt 7 , n = 36). h Localization in N2a cells of a Nt 7 -GFP construct comprising the consensus sequence of eukaryotic Gα-Nt 7 with the conserved Cys at position 5 (MGSLCSR). Marked region is magnified to the right. i-k Drosophila S2 cells expressing Gαo-Nt 7 -GFP (i), MGNC-Nt 7 -GFP (j) or MGSLCSR-Nt 7 -GFP (k). Co-expression of GalT-mRFP marked Golgi stacks (i-k). Selected areas are zoomed-in to the right. l-n Live imaging of N2a cells co-expressing Gαo-Nt 7 -GFP (l) or MGNC-Nt 7 -GFP (m) with the Golgi marker MannII-mRFP (bottom right insets). Representative images at the time of Palmostatin B (PalmB) addition (0 min) and after 45 min. Co-localization of Nt 7 -GFP constructs with MannII-mRFP in 5 min intervals and starting at t = 0 (n). Data represent mean ± s.d. (Gαo-Nt 7 control, n = 20; Gαo-Nt 7 PalmB, n = 20; MGNC-Nt 7 control, n = 16; MGNC-Nt 7 PalmB, n = 20). a, b, e, h, l, m Scale bars, 10 µm; i-k Scale bars, 5 µm. c, d, f, g Box plots indicate median (middle line), 25th, 75th percentile (box), and lowest, highest value (whiskers). P values were determined using one-way ANOVA Tukey test for (c, d, f, g) and one-way ANOVA Šídák test for n. ns: not significant. Source data are provided as a Source Data file. content did not significantly change during the treatment ( Fig. 2n and Supplementary Fig. 2o). Comparable changes in localization were observed for the full-length Gαo-GFP and MGNC-Gαo-GFP after 45 min incubation with Palmostatin B (Supplementary Fig.  2p, q). Metabolic labeling showed the efficacy of Palmostatin B, as [ 3 H]palmitate incorporation appears similarly augmented for all constructs (Fig. 1o, p and Supplementary Fig. 1o, p). This analysis supports the notion that the steady-state localization of PMPs results from a palmitoylation/depalmitoylation equilibrium 12 . However, the fact that the PM content of Gαo-Nt 7 was not affected by Palmostatin B argues against a constant Golgi-to-PM flow of the construct.
We next aimed at visualizing how Nt 7 constructs are trafficked in the cell. We adapted to our constructs the reverse dimerization (RD) assay, originally designed for the synchronized trafficking of secretory and integral membrane proteins 27 . This method is based on the aggregation of the F36M mutant of FKBP12 (FM) that allows protein tracking upon chemical dissociation. Thus, we intercalated four FM copies (FM 4 ) between Nt 7 and GFP and performed the RD assay in HeLa cells as their larger cell bodies grant for a better visualization. The distinct localization of Gαo-Nt 7 and MGNC-Nt 7 -still observed in HeLa cells (Supplementary Fig. 2k)-was lost by the FM 4 insertion in both cell lines, and cytosolic aggregates of different sizes were visualized instead. After chemical dissociation by the D/D solubilizer, most of the aggregates vanished and the constructs showed their characteristic localizations (Fig. 3a, b).
Then, live imaging was performed in HeLa cells co-expressing either Nt 7 -FM 4 -GFP construct together with MannII-mRFP as Golgi marker. As expected, Gαo-Nt 7 -FM 4 -GFP showed a rapid accumulation at the Golgi once the D/D solubilizer was added . This result does not support a Golgi-to-PM flow of MGNC-Nt 7 -FM 4 -GFP, as PM targeting was not preceded by its Golgi accumulation. To further prove that PM localization of MGNC-Nt 7 -FM 4 -GFP is indeed independent from transport through Golgi, we performed the RD assay under the 20°C temperature block that forces cargos to accumulate in Golgi 28 . As control, we used the GFP-FM 4 -hGH construct that aggregates in the ER lumen and is secreted via Golgi-mediated transport after chemical dissociation 29 . Incubation with D/D solubilizer for 45 min induced an almost complete secretion of GFP-FM 4 -hGH at 37°C, while at 20°C the construct strongly accumulated in the Golgi (Fig. 3f). Conversely, HeLa cells expressing MGNC-Nt 7 -FM 4 -GFP showed a comparable PM localization of the construct at both temperatures (Fig. 3g). Quantification of MGNC-Nt 7 -FM 4 -GFP mean fluorescence intensity in the MannII-mRFP region revealed no significant difference in its Golgi content at 37 and 20°C (Fig. 3h), indicating that PM targeting of MGNC-Nt 7 does not occur via transport through the Golgi.
Altogether, these results imply that the characteristic steadystate localizations of Gαo-Nt 7 and MGNC-Nt 7 are not related to Golgi-to-PM trafficking. Instead, each of the constructs goes directly to its primary destination: Golgi or PM.
Lipid binding of Nt 7 variants does not explain the preferential localization. Thus far, our data indicate that S-palmitoylation might account for the compartmentalization of Gαo-Nt 7 and MGNC-Nt 7 , and that intracellular trafficking is not a major player in their distinctive localizations. We then envisioned two further scenarios that could explain the differential Nt 7 compartmentalization: (i) PATs have no specificity for substrate recognition, and substrates concentrate at different compartments due to specific interactions, and (ii) PATs discriminate substrates, which in turn accumulate at the compartment where S-palmitoylation takes place. The first scenario involves a certain degree of promiscuity among PATs 11 , and implies that lipidations and a few surrounding amino acids confer specific binding properties, e.g., to lipids, which are known to differentially accumulate all over the endomembrane system 30 . In addition, fatty acids other than palmitate can also be attached to the available Cys and might confer different binding specificities to the S-acylated protein 31 . The second scenario requires that PATs discriminate between analogous substrates as small as Nt 7 and that a given substrate accumulates at the compartment where its specific PAT is localized.
To test the first scenario, we used membrane strips spotted with fifteen different lipids found in cellular membranes and performed a protein-lipid overlay assay 32 . As control, we employed a GFP-fusion of the pleckstrin homology (PH) domain of FAPP1 (FAPP1-PH-GFP), which specifically binds to phosphatidylinositol 4-phosphate (PI4P) 33 . N2a cells were transfected with Gαo-Nt 7 -GFP, MGNC-Nt 7 -GFP or FAPP1-PH-GFP, and then membrane strips were incubated with cleared cell extracts. As expected, FAPP1-PH showed a strong binding to the spot containing PI4P ( Supplementary Fig. 4a). On the other hand, no apparent difference was found for Gαo-Nt 7 and MGNC-Nt 7 that mainly bound to phosphatidylserine, cardiolipin, and phosphoinositides, particularly to PI4,5P2 ( Supplementary Fig. 4a). A similar lipid-binding pattern was also observed by the larger construct Gαo-Nt 31 -GFP ( Supplementary Fig. 4a). Thus, the compartmentalization of Gαo-Nt 7 and MGNC-Nt 7 is unlikely to emerge from different lipid-binding affinities, arguing against the first scenario.
A tool for visualization of local S-palmitoylation in intact cells. Next, we studied the second scenario, which implies that local S-palmitoylation drives substrate compartmentalization. In mammals, intracellular S-palmitoylation is mediated by zDHHC proteins with PAT activity; 23 zDHHCs exist in humans and 24 in mice 10 . Most of the zDHHCs have four transmembrane domains (TMDs), whereas only two and three members present five and six TMDs, respectively 11 . The majority of zDHHCs localize at the Golgi and the remaining are distributed among the ER, PM, and endosomes 34,35 . In order to confirm their subcellular localization, we expressed a collection of twenty-three HA-tagged mouse zDHHCs in N2a cells 36 . As expected, immunostainings revealed that a large number of zDHHCs localized predominantly at the Golgi: zDHHC3, 7, 9, 12, 13, 15-17, 21, 23, and 25 (Supplementary Fig. 4b). Six showed a clear PM localization but were also found on endosomes: zDHHC2, 5, 11, 14, 18, and to a lesser extent zDHHC8. The rest of the PATs mostly showed ER localization, while zDHHC4 was associated almost exclusively with the nuclear envelope ( Supplementary Fig. 4b), a pattern not described in previous studies 34,35 .
Besides a unique antibody that specifically recognizes palmitoylated PSD-95 37 , experimental approaches to study zDHHCsubstrate pair relations are based exclusively on disruptive biochemical and proteomic techniques 38 . To determine if Gαo-Nt 7 and MGNC-Nt 7 are differently lipidated at the PM vs. Golgi, we aimed at developing a method to visualize local S-palmitoylation in intact cells. Inspired by the nuclear envelope localization of zDHHC4, we engineered zDHHCs that ectopically target the ONM to detect mislocalized substrates. We took advantage of some components of the well-studied LINC (linker of nucleoskeleton and cytoskeleton) complexes. LINC complexes are built by proteins of the KASH (Klarsicht, ANC-1, and Syne Homology) family that are embedded in the ONM and interact within the perinuclear space with proteins of the SUN (Sad1 and UNC-84) family, which in turn are inserted in the inner nuclear membrane 39 . As the C-termini of all zDHHCs except zDHHC24 face the cytosol, we hypothesized that the addition of a C-terminal KASH-domain might induce their targeting to the ONM (Fig. 4a). To test this hypothesis, we first generated a zDHHC5 construct carrying an N-terminal mRFP for visualization and the KASH-domain of syne-1/nesprin-1 at its C-terminus (mRFP-zDHHC5-KASH). In HeLa cells, the KASH-domain alone (mRFP-KASH) is efficiently targeted to the ONM (Fig. 4b), and mRFP-zDHHC5 localized at the PM and endosomes as expected (Fig. 4c). The addition of the KASH domain strongly impaired PM localization of zDHHC5 (Fig. 4d), but a robust ONM targeting was only achieved by the co-expression of the KASH-interacting protein SUN2 (Fig. 4e).
Next, we analyzed if known zDHHC-substrate pair relations can be recapitulated in HeLa cells and chose the PMPs SNAP23, caveolin-1 and flotillin-2. The core of SNAP23 contains 6 Cys residues that are palmitoylated by zDHHC13 and 17 40 . zDHHC7 appears as the main PAT for caveolin-1, which is palmitoylated in 3 Cys residues at its C-terminus 41 . Flotillin-2 is myristoylated at . a-d, f, g Scale bars, 10 µm. P value was determined using a twosided unpaired t-test for (h). ns: not significant. Source data are provided as a Source Data file.
Remarkably, the GFP-fusions of SNAP23 ( Supplementary Fig. 5a), caveolin-1 ( Supplementary Fig. 5b) and flotillin-2 (Supplementary Fig. 5c) were efficiently recruited to the ONM by the corresponding zDHHC-KASH constructs. Together, we successfully developed a system in intact cells for the visualization of local S-palmitoylation at the ONM and named it the "SwissKASH" assay.
S-palmitoylation of Nt 7 at the ONM-the SwissKASH assay.
Remarkably, Gαo-Nt 7 -GFP and MGNC-Nt 7 -GFP were efficiently recruited to the ONM by co-expression of mRFP-zDHHC11-KASH but not by the control mRFP-KASH ( Fig. 5a-d). To quantify these effects, we measured the mean GFP-fluorescence intensity at the ONM and nearby cytosol, and used their ratio to determine the relative ONM content of the Nt 7 constructs. Quantification revealed a similar~2.5-fold accumulation of Gαo-Nt 7 and MGNC-Nt 7 at the ONM by co-expression of mRFP-zDHHC11-KASH (Fig. 5e), suggesting that zDHHC11 equally accepts both Nt 7 constructs as substrate. On the other hand, the KASH-fusions of zDHHC2, 5, 8, 14, and 18 showed no effect on the localization of Gαo-Nt 7 and MGNC-Nt 7 ( Supplementary  Fig. 5d,e). Thus, it appears that the predominant PM localization of MGNC-Nt 7 cannot be explained by major differences in substrate specificity of PM-associated zDHHCs. Our data also identifies zDHHC11 as the main PM-localized PAT for Gαo. Next, we applied the SwissKASH system to Gαo-Nt 7 and MGNC-Nt 7 using the eleven Golgi-associated zDHHCs, as all were efficiently targeted to the ONM by the C-terminal KASHdomain ( Fig. 5f-i and Supplementary Fig. 6a, b). Notably, mRFP-zDHHC-KASH constructs for zDHHC3 and 7 caused strong ONM accumulation of Gαo-Nt 7 -GFP ( Fig. 5f, g). No other zDHHCs showed any activity toward Gαo-Nt 7 ( Supplementary  Fig. 6a), suggesting a high degree of specificity in substrate recognition among Golgi-associated PATs and identifying zDHHC3 and 7 as the main Golgi-localized PATs for Gαo. MGNC-Nt 7 -GFP, in contrast, was recruited to the ONM only to a much lower extent by the KASH-fusions of zDHHC3 and 7 (Fig. 5h, i); the remaining Golgi zDHHC-KASH constructs were inefficient in ONM recruitment ( Supplementary Fig. 6b). Quantification showed that Gαo-Nt 7 accumulates at the ONM roughly 2.0 and 2.5 folds by the co-expression of zDHHC3 and zDHHC7, respectively (Fig. 5j, k). MGNC-Nt 7 presented a significantly lower ONM recruitment of~1.5 folds by zDHHC3 and 7 (Fig. 5j, k). To confirm that the PAT activity was responsible for the mislocalization of substrates in the SwissKASH assay, we generated zDHHCinactive constructs by a Cys-to-Ser mutation in the catalytic DHHC domain 36 . As anticipated, the ability of the KASH-fusions of zDHHC3, 7, and 11 to induce ONM accumulation of Gαo-Nt 7 and MGNC-Nt 7 was completely abolished when DHHS-mutants were co-expressed (Fig. 5l-o).
As the KASH-fusion of zDHHC11 was able to target both Gαo-Nt 7 and MGNC-Nt 7 to the ONM (Fig. 5b, d, e), we asked whether it acts in an indiscriminative manner or can still distinguish among different Nt 7 substrates. Answering this question, we found that mRFP-zDHHC11-KASH induced a strong ONM recruitment of MGSSCSR-Nt 7 -GFP but showed a much weaker effect on MGLLCSR-Nt 7 -GFP ( Supplementary Fig. 7a-c), both sequences commonly found in plant Gα subunits (Supplementary Notes and Supplementary Data 1).
The data above speak for the unexpected specificity of different zDHHCs to Nt 7 substrates that results in specific substrate localizations. We wondered whether these effects could simply result from physical interactions, differential among different zDHHC-substrate pairs. However, we identified that the zDHHCs do not possess high-affinity binding abilities to Nt 7 sequences, as immunoprecipitation of Nt 7 -GFP constructs did not co-precipitate any of the relevant mRFP-zDHHC-KASH constructs ( Supplementary Fig. 7d-g).
Thus, we conclude that zDHHC3, 7 and 11 all can differentiate among Nt 7 substrates with even slight variations in amino acid composition.
S-palmitoylation of Nt 7 at the ONM-the role of accessory proteins. In recent years, a few accessory proteins have been implicated in the regulation of several zDHHCs 43 . Here, we focused on two closely related PMPs: GCP16/Golga7 which has been reported to interact with the PM-localized zDHHC5 44 as well as the Golgi-localized zDHHC9 45 , and Golga7b which associates with zDHHC5 only 46 . We added a Flag-tagged GCP16/ Golga7 construct to the SwissKASH system for zDHHC5 and 9, and applied it to Gαo-Nt 7 and MGNC-Nt 7 . Interestingly, GCP16/ Golga7 was strongly recruited to the ONM by the KASH-fusion construct of zDHHC5 but not zDHHC9 ( Supplementary  Fig. 8a-d). This effect was not seen when the DHHS-mutant of zDHHC5 was used, suggesting that GCP16/Golga7 is a substrate for zDHHC5 ( Supplementary Fig. 8e-g). On the other hand, Gαo-Nt 7 and MGNC-Nt 7 were not targeted to the ONM in the SwissKASH system using GCP16/Golga7 together with zDHHC5 or 9 ( Supplementary Fig. 8a-d).
We then tested a Flag-tagged Golga7b construct in the SwissKASH assay for zDHHC5 and also observed its accumulation at the ONM (Fig. 6a, b). Surprisingly, we also observed ONM recruitment of MGNC-Nt 7 -GFP but not Gαo-Nt 7 -GFP under these conditions (Fig. 6a, b); both Golga7b and MGNC-Nt 7 were not targeted to the ONM when the DHHS-mutant of zDHHC5 was used (Fig. 6c). We were not able to detect any ONM localization of MGNC-Nt 7 without the co-expression of Golga7b ( Supplementary Fig. 5e), implying that the zDHHC5-Golga7b complex might be involved in substrate recognition similar to zDHHC9-GCP16/Golga7 45 . Quantification showed a significant ONM accumulation of MGNC-Nt 7 over Gαo-Nt 7 and control (Fig. 6d), which is not due to a differential protein-interaction with mRFP-zDHHC5-KASH, as both Nt 7 constructs did not coprecipitate the PAT (Supplementary Fig. 8h).
Then, we asked whether the higher PM targeting of Gαo-Nt 31 compared to Gαo-Nt 7 might also be the outcome of different zDHHC substrate specificities. Thus, we tested if zDHHC3, 7, and 11 target Gαo-Nt 31 to the ONM with the same efficiency than Gαo-Nt 7 . Although zDHHC11-KASH efficiently relocalized Gαo-Nt 31 -GFP to the ONM, zDHHC3 and zDHHC7 showed a weaker effect when compared to Gαo-Nt 7 -GFP ( Fig. 6e-i). In addition, we did not observe ONM targeting of Gαo-Nt 31 by the coexpression of mRFP-zDHHC5-KASH and Golga7b (Supplementary Fig. 8i). Therefore, the weaker activity of the Golgi-associated zDHHC3 and 7 toward Gαo-Nt 31 predicts an increase in its PM accumulation upon time, in agreement with the experimental findings (Fig. 1d).
Together, our data suggest that the distinct steady-state localizations of Gαo-Nt 7 and MGNC-Nt 7 are due to the differential substrate specificities of PM-and Golgi-localized enzymatically active zDHHCs, which can further be differently influenced by the accessory proteins.
Compartment-specific S-palmitoylation of Nt 7 . The notion that compartmentalization is controlled by PATs implies that the localization of Nt 7 constructs might be shifted toward the PM or Golgi by manipulating the expression level of specific zDHHCs. To test this hypothesis, we first co-transfected N2a cells with Gαo-Nt 7 -GFP and HA-zDHHC11, and determined its relative PM content and co-localization with the MannII-BFP Golgi marker. In addition, we co-expressed MGNC-Nt 7 -GFP together with HA-zDHHC3 or HA-zDHHC7. Remarkably, the PM localization of Gαo-Nt 7 strongly increased by the overexpression of zDHHC11, whereas its co-localization with MannII-BFP showed a significant decrease (Fig. 7a-d and Supplementary Fig. 9a). On the other hand, MGNC-Nt 7 presented a higher co-localization with the Golgi marker but a significant decrease in PM content by the co-expression of zDHHC3 or zDHHC7 (Fig. 7e-l Supplementary Fig. 9b). The effects of the different zDHHCs on Gαo-Nt 7 and MGNC-Nt 7 were not evident upon co-expression of the corresponding DHHS-mutants (Fig. 7b, f, j). The changes in the localization of Nt 7 constructs were not due to variations in their expression level upon co-expression of the different zDHHCs ( Supplementary Fig. 9c-f). No changes in the overall localization of Gαo-Nt 7 and MGNC-Nt 7 could be observed upon coexpression of zDHHC3/7 and zDHHC11, respectively (Supplementary Fig. 9g-i). Similarly, co-expression of zDHHC5 and Golga7b appeared not to alter the localization pattern of the Nt 7 constructs (Supplementary Fig. 9j, k). Intriguingly, [ 3 H]palmitate metabolic labeling of Gαo-Nt 7 -GFP was increased exclusively by the co-expression of zDHHC3 and 7 (Fig. 7m, n), whereas only zDHHC11 and the zDHHC5-Golga7b combination rose the labeling of MGNC-Nt 7 -GFP (Fig. 7m, o). Thus, drastic changes in the localization of the Nt 7 constructs seem not to be followed by significant variations in their overall S-palmitoylation status.
Next, we analyzed the localization of Gαo-Nt 7 and MGNC-Nt 7 in cells after depletion of zDHHCs. We employed HeLa instead of the murine N2a cells for this analysis, as siRNAs have been used successfully to downregulate the expression of human zDHHCs in this cell line 47 . We first determined if, in addition to zDHHC5 (Fig. 4c), zDHHC3, 7 and 11 show localizations in HeLa similar to those in N2a cells. Accordingly, HA-tagged zDHHC3 and 7 were predominantly localized at the Golgi, while zDHHC11 was at the PM ( Supplementary Fig. 10a-c). Then, we expressed Gαo-Nt 7 -GFP or MGNC-Nt 7 -GFP in cells pretreated with siRNAs against the pairs of zDHHC3/7 and zDHHC5/11. Gαo-Nt 7 and MGNC-Nt 7 localized normally in HeLa cells transfected with control siRNA, with Gαo-Nt 7 predominantly at the Golgi and MGNC-Nt 7 mainly at the PM (Fig. 8a, d). Remarkably, the simultaneous depletion of the Golgi-associated zDHHC3 and 7 clearly affected the localization of Gαo-Nt 7 : the construct exhibited a much weaker presence at the Golgi whereas its targeting to the PM appeared enhanced (Fig. 8b). MGNC-Nt 7 , however, seemed not affected by this treatment (Fig. 8e), but its localization was notoriously shifted toward the Golgi in HeLa cells upon the concomitant downregulation of the PM-associated zDHHC5 and 11 (Fig. 8f). On the other hand, the distribution of the Gαo-Nt 7 construct did not show major changes in cells depleted of zDHHC5 and 11 (Fig. 8c). Quantification showed a significant reduction in the relative Golgi localization of Gαo-Nt 7 by the downregulation of zDHHC3/7, and a significant increase in the Golgi content of MGNC-Nt 7 upon zDHHC5/11 depletion (Fig. 8g). Accordingly, [ 3 H]palmitate metabolic labeling of Gαo-Nt 7 and MGNC-Nt 7 was strongly reduced by the simultaneous downregulation of zDHHC3/7 and zDHHC5/11, respectively (Fig. 8h, i). A lower but significant reduction in labeling was additionally observed for Gαo-Nt 7 by the combined depletion of zDHHC5/11, and for MGNC-Nt 7 by zDHHC3/7 (Fig. 8h, i), demonstrating that these closely related sequences are differentially palmitoylated by PM-and Golgi-associated zDHHCs.
Collectively, these data confirm the conclusion that local S-palmitoylation plays a pivotal role in the compartmentalization of Gαo-Nt 7 and MGNC-Nt 7 , and that the localization pattern can be shifted upon the up-or downregulation of different PATs.
Targeting full-length Gαo to the ONM. We analyzed next if the SwissKASH assay might reveal local S-palmitoylation of the fulllength Gαo. Thus, we expressed Gαo-GFP together with the KASH-fusions of zDHHC3, 7, and 11, as they showed activity toward Gαo-Nt 7 . Remarkably, full-length Gαo was targeted to the ONM by the zDHHC constructs (Fig. 9a-d). Alternatively, the Cys4 mutant of Gαo, MGNC-Gαo-GFP, localized at the ONM primarily when the KASH-fusion for zDHHC11 was coexpressed ( Supplementary Fig. 10d), similar to what we observed for MGNC-Nt 7 (Fig. 5d, h, i). MGNC-Gαo was additionally targeted to the ONM by co-expression of mRFP-zDHHC5-KASH and Golga7b, an effect not seen for Gαo (Fig. 9e, f). Overall, our data show that the distinct steady-state localization of Gαo is controlled by the PM-and Golgi-localized zDHHCs.
To explore a potential role of ER-resident PATs in Gαo localization, we co-expressed Gαo-GFP with the HA-tagged  Fig. 10e). We noticed that zDHHC1 and to a lesser extend zDHHC20 were also present at the PM in HeLa cells ( Supplementary Fig. 10e), a pattern that was not obvious in N2a cells (Supplementary Fig. 4b).
Thus, we generated mRFP-KASH fusions for these PATs and tested their ability to affect Gαo-Nt 7 and MGNC-Nt 7 localization in the SwissKASH assay. The KASH-fusion efficiently targeted zDHHC1 and 20 to the ONM, but they were not able to recruit Gαo-Nt 7 or MGNC-Nt 7 ( Supplementary Fig. 11a-d).
Finally, we wondered whether the pool of Gαo targeting the ONM would keep its functionality. To address this, we first tested  the ability of Gαo to interact with Gβ1γ3. As expected, Gαo-GFP and mRFP-Gβ1γ3 strongly co-localized at the PM and endomembranes in control HeLa cells (Supplementary Fig. 11e). When these G proteins were expressed together with zDHHC11 in the SwissKASH assay, Gβ1γ3 relocalized to the ONM in a pattern identical to Gαo (Fig. 9g). We tested next if a downstream signaling molecule might be co-recruited to the ONM by Gαo, and selected RGS19 which preferentially interacts with the active, GTP-loaded Gαo 48 . In addition to Gαo wild-type, we used the GTPase-deficient mutant of Gαo (Q205L) that is present in the GTP-loaded state in cells. Accordingly, a clear ONM accumulation of a His 6 -tagged RGS19 construct was observed when the SwissKASH system was applied to the Q205L mutant ( Fig. 9h) but not the wild-type Gαo (Supplementary Fig. 11f). Altogether, our SwissKASH system shows that local S-palmitoylation of Gαo can cause its in-situ accumulation in the signaling-competent form, indicating that zDHHC selectivity and subcellular localization are key players in the substrate compartmentalization of the G protein, and potentially PMPs in general.

Discussion
The fact that Gα subunits can be found at the Golgi apparatus is known for 30 years 49 . The Golgi localization of Gα subunits was assumed to be part of their trafficking pathway to the cell periphery, although some non-canonical functions of G-proteins were described at compartments other than the PM 50 . Only recently, the Golgi-pool of Gα subunits has been implicated in the downstream signaling of typical GPCRs 3 and the KDELR 4,5 . Thus, subcellular compartmentalization has emerged as an important player in G protein functions. The present study provides an in-depth characterization of key elements controlling N-myristoylation, S-palmitoylation and localization of Gαo. More importantly, we developed a system-SwissKASH-for the rapid detection of zDHHC-substrate pair relations in intact cells, opposing to the disruptive biochemical and proteomic techniques currently used to study S-palmitoylation 38 . Altogether, this work supports a model in which local S-palmitoylation by distinct zDHHCs is crucial for Gαo compartmentalization, and most likely for other PMPs (Fig. 10). In this model, N-myristoylation allows the nascent protein to reach the whole endomembrane system. Subsequently, specific PM-and Golgi-localized zDHHCs catalyze S-palmitoylation of PMPs, which in turn accumulate at these compartments. Our model is in agreement with the local S-palmitoylation described for substrates such as PSD-95 at dendritic spines 37 and calnexin at the ER 47 , but adds the Golgi as a compartment in which PMPs can be locally modified and retained. To our knowledge, S-palmitoylation via the large number of Golgi-localized zDHHCs has been exclusively discussed in the context of Golgi-to-PM transport of integral 51 and PMPs 12 . We hereby challenge this notion and show instead that PMP localization to one compartment or another (including Golgi and PM) is not a consequence of sequential trafficking of the PMPs but rather a direct localization driven by the substratespecific PAT activities at different compartments. The fact that zDHHCs can modify multiple substrates has led to the notion that S-palmitoylation is nonspecific and proximitybased. Accordingly, a recent study demonstrated the stochastic S-palmitoylation of engineered Cys residues in membrane proteins, proposing that catalysis is determined by substrate accessibility and not specific sequences 52 . Several zDHHCs, however, contain PDZ-binding motifs, SH3 domains or ankyrin-repeats outside the catalytic DHHC core that have been implicated in enzyme-substrate interactions 11 . Importantly, our study uncovered a striking substrate specificity for some zDHHCs toward peptides as small as Gαo-Nt 7 . We additionally showed that minor sequence modifications might result not only in drastic changes in substrate recognition by zDHHCs, but also in substrate localization (Fig. 10). Another interesting implication of this model is that cells and tissues could shift the localization-and associated function -of PMPs toward one specific compartment to another by controlling the expression of zDHHCs. The large number of zDHHC genes and their distinct tissue-specific expression patterns in humans is thus not surprising 34 . This property might be widespread among Metazoa as Nt 7 constructs showed remarkably similar localization in Drosophila and mammalian cells, and Drosophila contains also a large number of zDHHC genes that are differentially expressed in embryonic and adult tissues 53 . Regulation of compartmentalized PMP functions might also be acutely regulated by controlling the action of PATs locally. This seems true for zDHHC5, as recent studies point to an intricated regulation of its activity and localization by posttranslational modifications and accessory proteins (see below) 43,54 .
unstructured and is most likely not involved in the binding and presentation of substrates 11 , implying that the fusion of a KASH domain does not interfere with substrate specificity. The SwissKASH pointed to zDHHC3, 7, and 11 as the major PATs for Gαo. This result agrees with a previous study showing zDHHC3 and 7 as the main PATs for Gαi2, Gαq and Gαs 56 , while zDHHC11 activity toward Gα subunits has not been detected. Similarly, we were unable to detect an increase in [ 3 H]palmitate labeling of Gαo-Nt 7 by the overexpression of zDHHC11, although it significantly changed Gαo-Nt 7 localization pattern. This suggests that the low starting level of S-palmitoylation of Gαo-Nt 7 at the PM makes it particularly sensitive to overexpression of zDHHC11, sufficient to relocalize significant quantities of Gαo-Nt 7 to this compartment without a measurable increase in [ 3 H]palmitate. Reciprocally, the low starting levels of MGNC-Nt 7 at the Golgi make this construct to significantly relocalize to this compartment upon overexpression of zDHHC3/ 7-again without accompanying measurable [ 3 H]palmitate increase. In contrast, siRNA-mediated downregulations of zDHHCs result in stronger changes in the overall S-palmitoylation, thus confirming that these closely related Nt 7 sequences are differentially modified by PM-and Golgi-associated PATs. The SwissKASH system, therefore, appears more sensitive than the traditional metabolic labeling approach, as even small amounts of a substrate are visible at the ONM.
Several recent studies have revealed that various zDHHCs are regulated by accessory proteins which control their activity, stability, and/or localization 43 . We focused here on GCP16/Golga7 and Golga7b, and introduced them into the SwissKASH approach. We found that the zDHHC5-Golga7b complex efficiently targeted MGNC-Nt 7 to the ONM, an effect not observed for Gαo-Nt 7 or in the absence of Golga7b. Interestingly, our MGNC-Nt 7 construct resembles the N-terminus of flotillin-2 (also known as reggie-1) that is myristoylated at Gly2 and palmitoylated at Cys4 by zDHHC5 42 , and it is similarly localized at the PM but excluded from Golgi 57 . These data indicate that the position of the palmitoylatable Cys within a myristoylated Nt 7 peptide might be relevant for substrate specificity by zDHHC5. Moreover, zDHHC5 efficiently targeted both GCP16/Golga7 and Golga7b to the ONM although only the zDHHC5-Golga7b complex co-recruited MGNC-Nt 7 , pointing to striking different functions for these complexes. Thus, the SwissKASH system might also be applied for functional studies of the complexes formed by zDHHCs and accessory proteins. A polybasic stretch present in the N-terminal region of all Gα subunits has been implicated in PM binding 58 . The Gαo-Nt 31 construct containing the basic region within the α-helix showed indeed a greater steady-state localization at the PM compared to Gαo-Nt 7 . In the SwissKASH system, the presence of the α-helix in Gαo-Nt 31 appears to impact substrate recognition by the Golgiassociated zDHHC3 and 7, but not by the PM-localized zDHHC11. Thus, these data point to a second level of complexity in substrate affinity by zDHHCs, where regions distant from the palmitoylatable Cys might subserve regulatory functions. How the polybasic region in Gαo impacts the substrate specificity by zDHHCs remains to be clarified.
S-palmitoylation has gained much attention lately as a potential target for drug discovery to combat pathologies mediated by proteins such as the oncogenic Ras and infectious diseases like malaria caused by Plasmodium falciparum or the respiratory syndromes caused by coronaviruses 59-61 . Our SwissKASH system could enable screens for molecules that perturb zDHHC-specific S-palmitoylation of pathological proteins. This approach may provide an attractive alternative to directly target the desired oncogene (or protein of another pathogenic activity), focusing instead on developing tools to target the zDHHCs controlling the proper (or pathological) localization of the protein. In this regard, zDHHCs can be targeted by the drug discovery efforts. The large number of these enzymes in humans and their distinct expression patterns provide the reasoning to expect the power of such an approach in obtaining the favorable pharmacodynamics with limited side effects. Another approach might aim at targeting the specific zDHHC-substrate interaction pairs; this avenue is expected to deliver even more specific targeted therapies.
To sum up, we have delineated the code describing the Golgi vs. PM localization of the G protein Gαo, showed that these two localizations are independent from the previously assumed Golgito-PM trafficking, and discovered a general mechanism controlling PMP localization to its target membrane through the substrate selectivity of differently localized PATs. This major insight into the cell biology of protein subcellular localization may pave the way to drug discovery principles. introducing a stop codon in frame with the SUN2 sequence. The SUN2-GFP was generated by replacing the SalI/NotI CFP fragment with the XhoI/NotI GFP fragment from pEGFP-N1. The mRFP-zDHHC5 wild-type and DHHS-mutant were cloned by cutting the BamHI/NotI sequence of pCI-neo-Flag-DHHC5 and pCI-neo-Flag-DHHS5 66 , and ligating in frame into the BglII/PspOMI sites of pmRFP-C1. For the generation of mRFP-zDHHC5-KASH wild-type and DHHSmutant, the sequences mRFP-zDHHC5 and mRFP-zDHHS5 were PCR-amplified from mRFP-fusions, cut with NheI/KpnI and used to exchange the GFP sequence cut with the same enzymes from GFP-KASH. Masaki Fukata (National Institutes of Natural Sciences, Japan) generously provided a collection of 23 mouse zDHHC isoforms cloned into the pEF-Bos-HA plasmid 36 . Slightly different cloning strategies were used for the generation of all mRFP-zDHHC-KASH constructs. Specifically, the zDHHC2, 11 and 18 sequences were PCR-amplified from the corresponding pEF-Bos-HA plasmids, cut with AgeI/KpnI and ligated in frame into the BspEI/KpnI sites of mRFP-KASH and also BFP-KASH for zDHHC11. The sequence for zDHHC1, 3,7,12,13,15,16,20,21, and 25 were PCR-amplified from the original pEF-Bos-HA plasmids, cut with BglII/EcoRI and inserted into the same sites of mRFP-KASH. The zDHHC8 sequence was PCR-amplified from pEF-Bos-HA-DHHC8, cut with XhoI/EcoRI and ligated into the same sites of mRFP-KASH. zDHHC9 was PCR-amplified from pEF-Bos-HA-DHHC9, cut with SacI/BamHI and ligated into the same sites of mRFP-KASH. The zDHHC14 sequence was PCRamplified from pEF-Bos-HA-DHHC14, cut with AgeI/BamHI and ligated into the BspEI/BamHI sites of mRFP-KASH. The zDHHC17 and 23 were PCR-amplified from the corresponding pEF-Bos-HA plasmids, cut with BglII/HindIII and inserted into the same sites of mRFP-KASH. All inactive DHHS-mutants for the relevant mRFP-zDHHC-KASH and HA-zDHHC constructs were obtained by point mutagenesis. The GFP-fusion of reggie-1/flotillin-2 was previously described 67 . Tamas  Immunofluorescence and microscopy. For microscopy, N2a and HeLa cells were transfected for 7 h, trypsinized, and seeded on poly-L-lysine-coated coverslips in complete MEM for an additional 15 h before fixation. When indicated, cells were seeded in complete MEM supplemented with 10 µM DDD85646 or 100 µM 2-bromopalmitate. S2 cells were transfected for 24 h, washed one time with PBS, resuspended in complete media, and seeded on poly-L-lysine-coated coverslips for 30 min before fixation. All cells were fixed for 20 min with 4% paraformaldehyde in PBS. For immunostaining, cells were permeabilized for 1 min using ice-cold PBS supplemented with 0.1% Triton X-100, blocked for 30 min with PBS supplemented with 1% BSA, incubated with the primary antibody in blocking buffer for 2 h at room temperature (RT), washed and subsequently incubated with secondary antibodies and DAPI in blocking buffer for 2 h at RT. Coverslips were finally mounted with VECTASHIELD on microscope slides. Cells were recorded with a Plan-Apochromat 63x/1.4 oil objective on a LSM800 Confocal Microscope using the ZEN 2.3 software (all Zeiss). When required, mean fluorescence intensity was determined from confocal images using ImageJ v1.53c (National Institutes of Health). Images were not recorded using the same confocal settings, therefore ratio fluorescence values, such as Golgi fluorescence vs. total fluorescence or PM vs. total, were used for quantifications (see below). As a proof of validity of this approach, we recorded N2a cells expressing Gαo-Nt 7 -GFP or Gαo-Nt 31 -GFP under identical confocal settings, and found that to a similar Golgi content of both constructs, Gαo-Nt 31 showed consistently higher PM values than Gαo-Nt 7 (Supplementary Fig. 11g). Taken values together, the Golgi signal from Gαo-Nt 7 is~1.5fold higher, and its PM signal is~2-fold lower, than those of Gαo-Nt 31 (Supplementary Fig. 11h), recapitulating the ratio representation we adopted in Fig. 1e, f and further elsewhere. All images were finally edited using ZEN lite 3.3 (Zeiss) and CorelDRAW 2020 (Corel).

PM
PM and Golgi accumulation. N2a cells expressing the GFP-fusion constructs were immunostained as indicated above using a mouse monoclonal antibody (mAb) against GM130 to visualize the Golgi apparatus, and a DAPI staining for the nucleus (both not displayed in all final images). Alternatively, co-expression of the MannII-BFP construct was used to label the Golgi (see below under "co-localization analysis" for plasmid ratio used for transfection). To avoid interferences due to different expression levels of GFP-constructs among cell population, mean fluorescence intensity was measured at the GM130-or MannII-BFP-positive Golgi region as well as at the total cell area, and ratio values were used to determine the relative Golgi accumulation of the constructs. Simultaneously, mean fluorescence intensity was determined at an unbroken region of the PM lacking membrane protrusions, and the ratio over total cell fluorescence was used to define relative PM content for each GFP-construct.
Co-localization analysis. N2a cells were transfected and immunostained against GM130 as described above. To determine co-localization of the various GFP constructs and the GM130 signal, an area covering the whole perinuclear region but excluding the PM was selected in confocal images and the Pearson's correlation was calculated using the co-localization tool of ImageJ v1.53c. The same analysis was done in N2a cells co-expressing the phosphatidylserine biosensor mRFP-Lact-C2 25 and the different GFP-fusions, transfected at equal plasmid ratio. The Golgi marker GalT-GFP was used as control. This analysis was additionally done in N2a cells cotransfected with GFP-fusions, HA-tagged zDHHCs or empty pcDNA3.1(+), and MannII-BFP as the Golgi marker (2:4:1 plasmid ratio).
siRNA knockdown experiments. For siRNA transfection, 1.2 × 10 5 HeLa cells were seeded on culture plates, and transfected 24 h later with 40 pmol of control siRNA (1027281) or 20 pmol of each siRNAs against human zDHHC3 (SI02777642), zDHHC5 (SI04159694), zDHHC7 (SI00766850), and zDHHC11 (SI04365914) (all Qiagen) using TransIT-X2 transfection reagent (Mirus, MIR 6004). siRNAs against the different zDHHCs were previously validated in HeLa cells 47 . After 48 h, cells were transfected with GFP-fusion constructs, and 24 h later cells were fixed and immunostained against GM130. Golgi accumulation of the constructs was determined from confocal images (all steps as described above). The relative PM localization of GFP-constructs was not determined for HeLa cells due to their heterogenous morphology, which does not allow for the clear visualization of a PM monolayer in the majority of the cells within the population.
Co-immunoprecipitation. The recombinant GST-tagged Nanobody against GFP 69 expressed in Escherichia coli Rosettagami (Novagen, 71351) was purified with Glutathione Sepharose 4B beads according to the manufacturer's instructions. Protein purity was assessed by SDS-PAGE and Coomassie blue staining. N2a cells were co-transfected with the different Gαo-GFP constructs and mRFP-Gβ1/Gγ3 at 1:1:1 plasmid ratio. HeLa cells were transfected with the constructs used for the SwissKASH assay at the same plasmid ratio described below. SUN2-GFP was used as positive control for the interaction with mRFP-zDHHC-KASH constructs. After 24 h transfection, cells were resuspended with icecold GST-lysis buffer (20 mM Tris-HCl, pH 8.0, 1% Triton X-100 and 10% glycerol in PBS) supplemented with a protease inhibitor cocktail (Roche) and 50 µM Palmostatin B (for SwissKASH-derived samples), and passed 10 times through a 25 G needle. Extracts were cleared by centrifugation at 15,000 × g for 15 min at 4°C, and supernatants were incubated with 2 µg of purified GST-tagged GFP-Nanobody for 30 min on ice. Then, 20 µL of Glutathione Sepharose 4B beads were added and samples rotated overnight at 4°C. Beads were repeatedly washed with GST-lysis buffer, prepared for SDS-PAGE, and finally analyzed by Western blot using antibodies against GFP, mRFP, and Flag-tag when needed, as well as HRPconjugated secondary antibodies for ECL detection. All co-immunoprecipitations were done in duplicate with very similar outcomes.
Crude subcellular fractionation. N2a cells were transfected for 24 h and cell extracts were prepared using non-denaturing conditions and fractionated by highspeed centrifugation 67 . Cells were rinsed twice with PBS and resuspended in hypoosmotic buffer (20 mM HEPES, pH 7.0) supplemented with a protease inhibitor cocktail. The cell suspension was passed 20 times through a 25 G needle, and nuclei and unbroken cells were removed by centrifugation at 700 × g for 10 min at 4°C. The supernatant was centrifuged at 100,000 × g for 60 min at 4°C. The new supernatant (cytosolic fraction) was directly prepared for SDS-PAGE, and the pellet (membrane fraction) was gently washed with hypo-osmotic buffer and resuspended in GST-lysis buffer supplemented with 0.5% SDS and protease inhibitors. The membrane fraction was cleared by centrifugation at 15,000 × g for 20 min at 4°C, and prepared for SDS-PAGE. Western blot analysis was done as above using the antibodies against GFP, and flotillin-2 and GAPDH as endogenous membrane and cytosolic markers, respectively.
Metabolic radiolabeling with [ 3 H]palmitate. [ 3 H]palmitate radiolabeling was performed as previously reported 70 . Transfected HeLa cells were starved for 1 h in MEM supplemented with 10 mM HEPES, pH 7.4, and subsequently radiolabeled for 3 h with 70 μCi/ml of [9,10-3 H]palmitate (American Radiolabeled Chemicals, Inc., ART-0129-25). Alternatively, cells were starved and radiolabeled in the presence of 100 µM 2-bromopalmitate, or incubated in complete media with 50 μM Palmostatin B for 4 h previous to the starvation and radiolabeling, both done also in the presence of Palmostatin B. Cells were then washed, lysed in a 0.5 M Tris-HCl, pH 7.4 buffer containing 0.5% Nonidet P-40, 20 mM EDTA, 10 mM NaF, 2 mM benzamidine and a protease inhibitor cocktail (Roche), and cleared by centrifugation. Supernatants were incubated overnight with anti-GFP agarose beads (Chromotek, GTA-100), beads were then washed several times, and incubated for 10 min at room temperature with 1 M hydroxylamine, pH 7.4 (Sigma-Aldrich, 159417) or 2 M Tris-HCl, pH 7.4. Beads were finally prepared for and loaded into a 4-20% gradient SDS-PAGE. One third of the immnunoprecipitate was analyzed by WB using antibody against GFP, and two thirds were used for fluorography on film. Radiolabeled products were analyzed using a Typhoon Scanner v1.1.0.7 and the ImageQuant TL v8.1.0.0 software (Amersham). Metabolic labeling was done in triplicate with similar outcomes.
Protein-lipid overlay assay. After 24 h of transfection, N2a cells were washed twice with TBS (50 mM Tris-HCl, pH 8.0, 150 mM NaCl, and 10% glycerol) and cell extracts were prepared as explained above in TBS supplemented with 0.5% Tx-100 and protease inhibitors. Membrane lipid strips were blocked in 3% BSA in TBS supplemented with 0.1% Tween-20 (TBS-T) for 1 h at 4°C. Then, strips were incubated overnight at 4°C with cleared cell extracts previously diluted 1/10 in TBS-T supplemented with 1% BSA. Strips were incubated first with a pAb against GFP, and then with a secondary HRP-conjugated pAb to detect GFP-constructs bound to lipid dots. The protein-lipid overlay assay was done in duplicate for each condition with similar results.
Reverse dimerization assay. HeLa cells were co-transfected with the Golgi marker GalT-mRFP (not displayed in all figures) and Gαo-Nt 7 -FM 4 -GFP, MGNC-Nt 7 -FM 4 -GFP or GFP-FM 4 -hGH (at a 1:3 plasmid ratio), and seeded on poly-L-lysine-coated coverslips for 15 h as indicated above. Then, cells were incubated for 2 h at normal culture conditions in HBSS supplemented with 20 mM HEPES, pH 7.4, and 50 μM Cycloheximide to block de novo synthesis of proteins. The reverse dimerization was induced by adding fresh HBSS supplemented as above plus 1 µM D/D solubilizer for the time indicated in the corresponding figures. For the temperature block of Golgi transport, the last two steps were simultaneously performed at 37 and 20°C. Then, cells were fixed and prepared for microscopy. Quantification of fluorescence intensity at the MannII-mRFP Golgi region was done as described above.
For the live imaging of reverse dimerization, HeLa cells were transfected as above but seeded on µ-Slide-4-wells coverslips (Ibidi). Cell were first incubated for 30 min at 37°C in Hank's Balanced Salt Solution (HBSS; Gibco) supplemented with 20 mM HEPES, pH 7.4 and 50 μM cyclohexamide. Slides were then mounted on a temperature-controlled stage in a VisiScope CSU-X1 spinning disk confocal system (Visitron Systems) equipped with a Plan-Apochromat 63x/1.4 oil objective on an AxioObserver.Z1 microscope (Zeiss), a Evolve 512 EMCCD Camera (Photometrics), and the VisiView v4.00.10 Imaging software (Visitron Systems). Reverse dimerization was induced by adding D/D solubilizer to reach a 1 µM final concentration, and cells were immediately recorded at one image per 5 s for 10 min. For analysis, movies were generated from stacks using ImageJ v1.53c, and the mean fluorescence intensity of an area at the center of the MannII-mRFP Golgi region was measured from stacks.
Palmostatin B treatment. For live imaging, N2a cells co-expressing the GFPfusion constructs and the MannII-mRFP Golgi marker were seeded on µ-Slide-4wells coverslips (Ibidi). Cells were first incubated for 30 min at 37°C in HBSS supplemented with 20 mM HEPES, pH 7.4 and 50 μM cycloheximide, then Palmostatin B in DMSO was added to a 50 µM final concentration, and cells were immediately recorded at one image per 30 s for 45 min in the spinning disk confocal system described above. Same volume of DMSO was added for control cells. PM content and co-localization with MannII-mRFP was done as above. For fixed samples, N2a cells expressing only GFP-fusions were treated as above but fixed after 45 min of Palmostatin B addition. Cells were then immunostained against GM130 to visualize the Golgi and DAPI for nuclei (not displayed in the figures).
S-palmitoylation at the outer nuclear membrane (SwissKASH assay). HeLa cells were co-transfected as above with plasmids encoding the GFP-fusion, SUN2, and control mRFP/BFP-KASH or mRFP/BFP-zDHHC-KASH at a 1:3:3 plasmid ratio, at a 1:2:2:1 when GCP16/Golga7-Flag, Golga7b-Flag or His 6 -RGS19 was included, or at 1:2:2:1:1 when mRFP-Gβ1 and mRFP-Gγ3 were co-transfected. Cells were then seeded on poly-L-lysine-coated coverslips, fixed and prepared for microscopy. For quantification, mean GFP-fluorescence intensity at the ONM region labeled by the mRFP-KASH fusion was determined using ImageJ 1.53c. Simultaneously, GFP-fluorescence was measured at a nearby cytosolic region as well, and the ratio ONM over cytosol was used to define the relative ONM content of each GFP-construct. Ratio values were used due to different expression levels of the GFP-constructs among the cell population.
Statistics and reproducibility. Statistical parameters, including the exact values of n are reported in figures and figure legends. Results in Box plots indicate median (middle line), 25th, 75th percentile (box), and lowest, highest value (whiskers), other data are shown as the mean, and error bars represent the s.e.m. or s.d. as indicated in the corresponding figure legend; ns, not significant, *p ≤ 0.01; **p ≤ 0.005; ***p ≤ 0.001; ****p ≤ 0.0001 using one-way or two-way ANOVA test with Tukey or Šídák corrections (for multi-sample groups), two-sided unpaired ttest (for two-sample comparison), and one-sample t-test (for one-sample comparison to control). Prism 9 (GraphPad) was used to determine statistical significance. No statistical methods were used to predetermine the sample sizes. All replicates successfully reproduced the presented findings. No data were excluded from the analyses. The experiments were not randomized and investigators were not blinded to allocation during experiments and outcome assessment.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.