In previous work, we used protein constructs comprised solely of transmembrane domains (TMDs) to interrogate raft-dependent recycling in the endocytic system (Diaz-Rohrer et al., 2023; Diaz-Rohrer et al., 2014b; Lorent et al., 2017). The preference of these probes for raft domains can be quantified directly by measuring their partitioning between ordered (raft) and disordered (non-raft) phases of isolated GPMVs (Levental and Levental, 2015a; Sezgin et al., 2012), as shown in Fig 1A-B. As previously shown (Diaz-Rohrer et al., 2014b; Lorent et al., 2017), the TMD of a single-pass transmembrane adapter called Linker for Activation of T-cells (LAT) preferentially partitions into the raft phase of GPMVs, away from the non-raft phase lipid marker (F-DiO) (Fig 1A, left). A model raft-excluded TMD is a 22-Leu construct (i.e. allL) which strongly prefers the non-raft phase (Fig 1A, right). This behavior can be quantified via the raft partition coefficient (K_p,raft) defined as the ratio of background-subtracted intensities for various constructs in the raft versus non-raft phases (Fig 1B). The LAT TMD (open symbols) retains the raft affinity of full-length LAT, while allL-TMD, either on its own or inserted into full-length LAT (LAT-allL), has very low raft affinity (Fig 1B).

To quantitatively assay the effects of raft affinity on the kinetics of secretory trafficking, we inserted these probes into the RUSH system, which allows synchronized release and tracking through the secretory system (Fig 1C). This assay is based on the reversible interaction of a target protein (fused to streptavidin-binding peptide, SBP) with a selective “hook”; e.g., streptavidin stably anchored in the ER via a KDEL motif. The strong interaction between the hook and SBP retains the protein of interest (POI) in the ER. Introduction of biotin causes rapid release of the POI, which is then tracked via a fluorescent tag. SBP-tagged versions (i.e. RUSH) of all constructs assayed here had indistinguishable raft affinity from non-SBP versions (RFP-only) (Fig 1B).

The isolated TMD of LAT was sufficient to recapitulate the steady-state plasma membrane (PM) localization of its full-length protein (Fig S1), consistent with previous reports (Diaz-Rohrer et al., 2014b; Lorent et al., 2017). In clear contrast, LAT-TMD did not recapitulate the ER exit kinetics of full-length LAT (Fig 1D). 45 mins after biotin addition, LAT had completely exited the ER and was localized almost exclusively to a bright perinuclear region, likely the Golgi. After 90 min, most full-length LAT achieved its steady-state localization at the PM. In contrast, LAT-TMD expressed in the same cell was still in the ER after 90 min and clearly accumulated at the PM only after several hours. Such slow kinetics were a challenge for live-cell imaging, so ER exit was quantified on a population level, by fixing cells at various time points after biotin addition and quantifying the fraction of cells with observable ER localization. The temporal reduction in the fraction of ER-positive cells was well described by a single-exponential fit for all constructs (e.g. Fig 1E), revealing that the half-time for ER exit was >4-fold faster for LAT (0.7 h) than for LAT-TMD (3.1 h) (Fig 1E-G).

We next tested whether raft affinity affected ER exit kinetics. To this end, we generated a RUSH construct of full-length LAT whose TMD was replaced with one that has minimal raft affinity (LAT-allL, Fig 1A). The general kinetics of ER exit of this non-raft LAT were comparable to LAT, with clear Golgi accumulation after 60 min of biotin. Post-Golgi accumulation was prominent at 90 min for both LAT and LAT-allL (the steady state localization of LAT-allL is in endolysosomes, Fig S1, as previously explained (Diaz-Rohrer et al., 2023; Diaz-Rohrer et al., 2014b)). In contrast, the isolated allL-TMD had much slower ER exit kinetics (Fig 1F-G). This trend was generalizable to several other proteins, with full-length versions having >3-fold faster ER exit kinetics than the TMD-only versions, regardless of their raft affinity (Fig 1F-G, Table S1). It is worth noting that isolated TMDs still exited the ER and reached a steady-state localization after several hours (Fig S1), despite no direct interactions with cytosolic trafficking machinery (e.g. COPII or clathrin). Altogether, we conclude that features present in cytosolic domains play a dominant role over TMD-determined raft affinity in ER exit.

LAT is comprised of a minimal N-terminal ectodomain (<5 residues), a single-pass TMD, and a largely disordered cytosolic domain (CTD). Since LAT-TMD showed slow ER exit, we inferred that the determinant of rapid ER exit is likely located in the CTD and created a series of C-terminal truncations to locate the signal (Fig 2A). The two smaller C-term truncations (ΔCt1 and ΔCt2) had no effect on ER exit kinetics, behaving like full-length LAT (Fig 2B-C). In contrast, both larger truncations (ΔCt3 and ΔCt4) were significantly slower, behaving like LAT-TMD (Fig 2B-C). Thus, the motif facilitating fast ER exit of LAT is located within residues 140-185 (Fig 2A). We analyzed this fragment for possible COPII binding motifs (Barlowe, 2003b; Mikros and Diallinas, 2019) and identified ¹⁴⁶AAPSA¹⁵², which corresponds to a ΦxΦxΦ motif (Φ = hydrophobic residue, x = spacer) (Otsu et al., 2013) (Fig 2A-inset). Point mutations in this putative motif confirmed that P148 and A150 were essential for fast ER exit, while other neighboring residues were not (Fig 2D-E and S2). We note that these two residues are highly conserved in LAT from 30 species (Fig 2A). Finally, inserting this AaPsA motif into LAT-TMD signficantly accelerated its ER exit, nearly recapitulating that of full-length LAT (Fig 2F). Thus, we conclude that fast ER exit of LAT is mediated by a ΦxΦxΦ motif in the cytosolic CTD, which likely mediates COPII association.

The RUSH system can be adapted to measure exit rates from Golgi by changing the “hook” to Golgin84, a resident protein of cis-Golgi (Diao et al., 2003). As expected, all RUSH constructs (LAT, LAT-TMD, LAT-allL, allL-TMD) cotransfected with this hook were localized almost exclusively in the Golgi without biotin (Fig 3A). Within 90 min of biotin introduction, LAT exited the Golgi, achieving the expected steady-state PM localization (Fig 3A). LAT-allL was clearly slower, with abundant Golgi signal remaining after 90 min. Tracking the kinetics by quantitative imaging, LAT exited the Golgi ∼2.5-fold faster than LAT-allL (Fig 3B-C). Strikingly, the TMD-only versions of these constructs behaved nearly identically to the full-length (Fig 3B-C). LAT-TMD, which has almost no residues that could interact with cytosolic trafficking machinery, had similar Golgi-exit kinetics as full length VSVG (Fig S3), a model transmembrane secretory cargo. The kinetics of VSVG traffic measured here with RUSH are consistent previous measurement by an orthogonal method (Hirschberg et al., 1998), validating our approach.

To validate these conclusions in a different experimental paradigm, we measured Golgi exit kinetics after a temperature block. It was previously shown that culturing mammalian cells at 15-20°C leads to accumulation of secretory cargo in the Golgi (Milgram and Mains, 1994; Venditti et al., 2019). To accumulate our probes in the Golgi, we used the ER-RUSH constructs to initially restrict them to the ER (Fig 3D, left). Biotin was then added at 17°C for 3 h to release the protein from the ER and allow sufficient time to accumulate in the Golgi. The cells were then shifted to 37°C to synchronize Golgi export. As with Golgi-RUSH, LAT largely exited the Golgi within 90 min, whereas LAT-allL was significantly slower, with notable Golgi accumulation remaining after 2 h (Fig 3D). The kinetics of the temperature-block experiment were similar to Golgi RUSH, with raft-preferring LAT exiting Golgi ∼2-fold faster than non-raft LAT-allL. Based on these observations, we conclude that TMD-encoded raft affinity is an essential determinant of Golgi exit kinetics for LAT.

We attempted to rationalize our observations of ER and Golgi efflux using a kinetic model wherein the secretory kinetics of a TMP are determined by equilibrium partitioning between coexisting compartments in the ER and Golgi (Fig 4A). In the model, partitioning to an ER-exit compartment (ERex) (analogous to a cellular ER exit site) allows ER efflux with first-order rate constant k_a. Full-length and TMD-only proteins have different partition coefficients into this ERex compartment, represented by two free fit parameters K_{p,ERex_full} and K_{p,ERex_TMD}.

Analogously in the Golgi, we model coexisting compartments that enrich raft or non-raft proteins, again with partition coefficients (i.e. K_{p,raft_LAT} and K_{p,raft_allL}) describing the ratio of a protein’s concentration in the two compartments. These values were set by measurements of K_p,raft for the various constructs in GPMVs (Fig 1A-B), with the assumption that raft affinities measured at the PM are representative of those in the Golgi. A first-order rate constant k_b is used to describe the transfer of proteins from the Golgi raft sub-compartment to post-Golgi compartments. This conceptual model can be represented via a series of coupled differential equations describing the temporal evolution of abundance of various constructs in three compartments (ER, Golgi, post-Golgi). We emphasize that this model is highly simplified and does not include many features relevant to trafficking of endogenous proteins. Rather, the goal was to identify the minimal set of features that could describe the secretory behavior of our defined set of probes.

This simple model (four free parameters) was used to simultaneously fit experimental observations of eight independent data sets, i.e. release from the ER or from the Golgi for four different protein constructs (analogous to experiments described in Fig 1 and 3, respectively). For comparison to the models, experimental datasets were re-measured via live-cell imaging (see Supp Fig 5) to directly quantify relative construct abundance in the ER or Golgi. Model global fits nicely reproduced the general experimental features of all constructs (Fig 4B), though with notable divergences. The general shape of both ER and Golgi efflux curves were well described by the model and the specific Golgi efflux kinetics for all four proteins were modeled quite accurately (Fig 4B, filled symbols). Consistent with expectation, the fitted ER exit partition coefficient (K_{p,ERex_full}) for full-length constructs (i.e. those containing the putative COPII binding motif) was ∼20-fold greater than for TMD-only constructs. In contrast, the specific ER release kinetics could not be completely reproduced, with the model underestimating ER efflux rate for both constructs with raft-preferring TMDs, and overestimating those of the non-raft proteins (i.e. allL) (Fig. 4B, open symbols).

These fits could not be improved by adding a Golgi-to-ER retrieval path, consistent with none of our constructs possessing a known retrieval motif (i.e. KDEL) (not shown). However, excellent agreement between the model and all experimental observations could be obtained by allowing the ERex partition coefficients of LAT and LAT-allL to independently vary (Fig 4C). Put another way, in this 5-parameter fit, the TMD of these constructs was allowed to influence their affinity for the ER exit compartment. The best-fit was obtained when LAT with a wild-type TMD had ∼6-fold greater affinity for the ERex than LAT-allL.

The general agreements between the model and observations support the plausibility of an underlying hypothesis that partitioning between coexisting membrane domains could explain the inter-organelle transfer kinetics in our study. A key aspect of the model is that selective Golgi membrane domains are important for Golgi exit kinetics of raft-associated cargo. To test this inference, we used inhibitors to block the production of raft-forming lipids (Lasserre et al., 2008; Levental et al., 2017) and tested their effects on trafficking rates of our probes. Specifically, sphingolipid and cholesterol synthesis were inhibited by treatment with 25 µM myriocin and 5 µM Zaragozic acid (MZA), respectively, for 2 days. This treatment was previously shown by us and others to reduce ordered membrane domain stability (Lasserre et al., 2008; Levental et al., 2017; Wang et al., 2023). We observed significantly reduced Golgi exit kinetics of LAT-TMD in MZA-treated cells, evidenced by clear Golgi localization up to 2 h after release of Golgi-RUSH, in contrast to control cells (Fig 4D). Quantifying exit kinetics by imaging Golgi residence, we observed that the half-time for Golgi exit was significantly increased (by ∼60%) for the raft-preferring TMD but not for the non-raft allL-TMD (Fig 4E-F).

To directly image separation between raft and non-raft probes in the Golgi, we co-transfected RUSH versions of LAT-TMD and allL-TMD with the Golgin84-hook to reversibly accumulate both constructs in the Golgi. Their colocalization was then imaged using super-resolved Structured Illumination Microscopy (SIM) (Fig 5A-B). Prior to introduction of biotin, the two constructs showed near-perfect colocalization, as expected from their association with the same “hook”. However, 5 minutes after releasing the proteins with biotin, their colocalization was significantly reduced (Fig 5C), with areas of LAT-TMD and allL-TMD enrichment apparent within the general morphology of the Golgi (Fig 5B). Similar segregation of LAT-TMD from allL-TMD could be observed when both were accumulated in the Golgi via a 20°C temperature block (Fig 5D). Notably, such segregation was not observed when LAT-TMD was co-accumulated in the Golgi with either the full-length LAT or with raft-preferring GPI-GFP, which colocalized nearly completely with each other (Fig 5D).

To support these observations without relying on temperature modulation, we treated cells with a short-chain Ceramide (D-Ceramide-C6, Cer-C6) which was previously reported to disrupt Golgi organization and post-Golgi transport (Capasso et al., 2017; Duran et al., 2012; van Galen et al., 2014). We combined this treatment with the Golgi-RUSH (Golgin84-hook) constructs to determine the fate of raft- and non-raft probes when their Golgi exit is blocked. Trapping both constructs in Golgi via RUSH produced very high co-localization, as expected (Fig 5F). When the constructs were released with biotin in cells treated with Cer-C6, LAT-TMD and allL-TMD both remained in a perinuclear compartment with the general morphology of the Golgi, but their colocalization was notably reduced, with raft-TMD-rich and -depleted areas visible under confocal imaging (Fig 5F). This reduced colocalization was also observed without biotin present, suggesting that Cer-C6 led to a remodeling of the Golgi and associated spatial segregation of raft from non-raft probe proteins, as previously proposed (van Galen et al., 2014).

These observations are consistent with previous reports that proteins can segregate in Golgi (Chen et al., 2017). Specifically, TGN46, a resident protein in the vesicular trans-Golgi network (TGN), was shown to physically segregate from a trans-Golgi-resident enzyme Sialyltransferase (ST) under Cer-C6 treatment (van Galen et al., 2014). To determine whether this physical segregation was related to the behavior of our raft domain probes, we co-transfected these Golgi-resident proteins with our probes and imaged them under Cer-C6 treatment. Intriguingly, in Cer-C6-treated cells, the raft probe (LAT-TMD) colocalized well with TGN46 but not with ST (Fig 5G-H). Conversely, the non-raft-preferring probe (allL-TMD) colocalized with ST, but not TGN46. These colocalization were consistent with the raft affinity of the Golgi markers evaluated in GPMVs: TGN46 was enriched in the raft phase approximately at parity with LAT-TMD, while ST was depleted from the raft phase (like allL-TMD). These observations support the hypothesis that proteins can segregate in Golgi based on their affinity for distinct membrane domains.

HEK-293NT (HEK) and HeLa cells were purchased from ATCC and cultured in medium containing 89% Eagle’s Minimum Essential Medium (EMEM), 10% FCS, and 1% penicillin/streptomycin at 37 °C in humidified 5% CO2. COS-7 cells were cultured in medium containing the same formulation but with Dulbecco Modified Essential Medium (DMEM) instead of EMEM and under the same conditions. Transfection was done by Lipofectamine 3000 using the protocols provided with the reagents. 4–6 h after transfection, cells were washed with PBS and then incubated with serum-free medium overnight. To synchronize the cells, 1 h before biotin addition, the cells were given full-serum medium. Lipid synthesis inhibitors (25 uM Myriocin and 5 uM Zaragozic Acid) were added for to the medium and given to the cells 2 days before transfection and during the whole experiment. D-Ceramide-C6 was added to the medium alone or in combination with biotin and given to the cells for 4 h, and then cells were fixed.

All single pass protein constructs from the ER were based on the bicistronic GPI RUSH backbone previously described. We replaced the protein and fluorophore by the amino acid sequence of LAT-TMD, which is NH2-MEEAILVPCVLGLLLLPILAMLMALCVHCHRLP followed by a short linker (GSGS) and monomeric RFP (mRFP). A full length (LAT, LATallL, LAX) and TMD library (all-TMD, allA8L-TMD, LAT-TMD6Dendo, TfR-TMD, LAX-TMD, VSVG-TMD, LATDC1-4, P148E, P148S, P148A, A150S, P151S, P151A, A183S) were generated by amplifying the sequence of interest by PCR and subsequent cloning of the mutant sequence into the LAT-TMD RUSH construct with KDEL hook to synchronize from the ER. Several constructs were purchased from Addgene: full length RUSH version of VSVG (#65300). All the single pass RUSH constructs from Golgi were based on the bicistronic VSVG RUSH backbone kindly donated by Jennifer Lippincott-Schwartz lab. We cut the construct with BamHI and EcoRI and replaced the protein with LAT, LATallL, LAT-TMD or all-TMD by amplifying the sequence of interest by PCR. TGN46 and ST plasmids were provided by Felix Campelo lab.

Unless specified, imaging was performed on a confocal microscope using appropriate filters for GFP/RFP fluorescence for transfected plasmids. Single-cell tracking live imaging was performed using u-dish Grid-50 glass bottom plates (Ibidi GmbH) to relocate the selected cells at each acquisition after incubating at 37°C in the incubator when not imaging. When needed, cells were fixed using 4% paraformaldehyde (PFA), 10 min at RT. Anti-Giantin (1:1000 Rabbit polyclonal, Abcam ab80864) was used to create Golgi mask.

In brief, transfected cells with different plasmids containing the RUSH retention system were incubated overnight at 37 °C, and then incubated in the presence of 100 µM biotin (Sigma-Aldrich) to release the cargo proteins. Cells were incubated at 37°C for various chase times and either directly imaged or fixed with PFA as described above.

Cell membranes were stained with 5 μg/ml of FAST-DiO (Invitrogen), green fluorescent lipid dye that strongly partitions to disordered phases(Levental and Levental, 2015b). Following staining, GPMVs were isolated from transfected HEK-293, or HeLa cells as described (Sezgin et al., 2012)(cell type had no effect on results). Briefly, GPMV formation was induced by 2 mM N-ethylmaleimide (NEM) in hypotonic buffer containing 100 mM NaCl, 10 mM HEPES, and 2 mM CaCl2, pH 7.4. To quantify protein partitioning, GPMVs were observed on an inverted epifluorescence microscope (Nikon) at 4°C after treatment with 200 μM DCA to stabilize phase separation; this treatment has been previously demonstrated not to affect raft affinity of various proteins (Zhou et al., 2013). The partition coefficient (K_p,raft) for each protein construct was calculated from fluorescence intensity of the construct in the raft and non-raft phase for >10 vesicles/trial (e.g. Fig. 1), with multiple independent experiments for each construct.

After discarding uncharacterized proteins and taking just the first 100 results from 246 hits in the Blast for LAT sequence extracted from Uniprot database, we have used Unipro Ugene to analyze the results of 30 species with > 80% similarity.

Residence fraction data were analyzed globally by numerically solving a first-order, homogeneous system of equations that accounts for intra- and inter-compartment transport. A schematic of the trafficking model is shown in Fig. 4A. Briefly, the ER was modeled with two sub-compartments representing bulk and exit sites, with an intra-ER partition coefficient K_p,ERex, defined as the ratio of concentrations between the exit and bulk compartments. Similarly, the Golgi was modeled with two sub-compartments representing raft and non-raft sites, with an intra-Golgi partition coefficient K_p,raft,, i.e. the ratio of concentrations between the raft and nonraft compartments. These values are set in the model via measurements of K_p,raft in GPMVs (Fig 1)(Diaz-Rohrer et al., 2023; Lorent et al., 2017). Inter-compartment trafficking from ER to Golgi initiates from ER exit sites and terminates at the Golgi with a rate constant k_a, while unidirectional post-Golgi traffic originating from Golgi raft sites was modeled with a rate constant k_b. In total, the trafficking model for a given construct is a function of two rate constants and two partition coefficients, one of which is fixed experimentally.

Eight data sets representing the ER and Golgi efflux kinetics of LAT, LAT-TMD, AllL-LAT, and AllL-TMD were fit to the trafficking model. We performed a single global analysis (i.e. simultaneous fit of all eight data sets) using k_a, k_b, and two K_p,ERex (one for full-length and one for TMD-only) as free fit parameters. A second analysis included a third K_p,ERex, allowing LAT and LAT-allL to be different.

All analysis was performed with custom code written in Mathematica v.12.2 (Wolfram Research Inc., Champaign, IL). Model parameters were optimized with a Levenberg-Marquardt algorithm implemented in the built-in Mathematica function NonlinearModelFit. The target function for minimization was the sum of squared residuals for the combined data sets, , where i indexes the eight data sets, j indexes the data points within a data set, y_ij is the observed residence fraction, and is the predicted residence fraction. To improve the probability of finding the global minimum, the optimization was repeated 10³ times with different random initial values for the adjustable parameters; the best-fit parameters associated with the overall lowest 𝒳² value is reported as the solution.