At the plasma membrane of mammalian cells, major histocompatibility complex class I molecules (MHC-I) present antigenic peptides to cytotoxic T cells. Following the loss of the peptide and the light chain beta-2 microglobulin (β2m, encoded by B2M), the resulting free heavy chains (FHCs) can associate into homotypic complexes in the plasma membrane. Here, we investigate the stoichiometry and dynamics of MHC-I FHCs assemblies by combining a micropattern assay with fluorescence recovery after photobleaching (FRAP) and with single-molecule co-tracking. We identify non-covalent MHC-I FHC dimers, with dimerization mediated by the α3 domain, as the prevalent species at the plasma membrane, leading a moderate decrease in the diffusion coefficient. MHC-I FHC dimers show increased tendency to cluster into higher order oligomers as concluded from an increased immobile fraction with higher single-molecule colocalization. In vitro studies with isolated proteins in conjunction with molecular docking and dynamics simulations suggest that in the complexes, the α3 domain of one FHC binds to another FHC in a manner similar to that seen for β2m.
Major histocompatibility class I molecules (MHC-I) fulfil central tasks of the adaptive immune response against infections and malignancies by presenting antigenic peptides to T cells (Comber and Philip, 2014; Kaufman, 2018; Townsend and Bodmer, 1989). MHC-I heterotrimers consist of the polymorphic transmembrane heavy chain (HC), the non-polymorphic soluble light chain beta-2 microglobulin (β2m, encoded by B2M), and the peptide (Townsend et al., 1990, 1989). In addition to this trimer, two more states of MHC-I occur at the cell surface, the ‘empty’ HC–β2m heterodimer that lacks peptide (Ljunggren et al., 1990; Montealegre et al., 2015), and the monomeric ‘free’ heavy chain (FHC) (Edidin et al., 1997; Geng et al., 2018). Since the binding of peptide and β2m is cooperative (Elliott et al., 1991; Gakamsky et al., 1996), HC–β2m heterodimers are conformationally unstable, and loss of peptide leads to the rapid formation of FHCs (schematic in Fig. 1A) and to the subsequent endocytic removal of FHCs by a sorting mechanism that is not understood (Montealegre et al., 2015). For FHCs present at the cell surface, important regulatory functions mediated by homo- and hetero-meric interactions in cis and trans have been proposed (Arosa et al., 2021, 2007; Campbell et al., 2012), which suggest defined spatiotemporal organization and dynamics of FHC in the plasma membrane. Indeed, clustering and covalent dimerization of MHC-I have been identified using a variety of approaches including recombinant proteins and live cells (Allen et al., 1999; Antoniou et al., 2011; Armony et al., 2021; Baía et al., 2016; Blumenthal et al., 2016; Bodnar et al., 2003; Capps et al., 1993; Chakrabarti et al., 1992; Fassett et al., 2001; Ferez et al., 2014; Lu et al., 2012; Makhadiyeva et al., 2012; Matko et al., 1994; Triantafilou et al., 2000); these represent complexes of different and largely unclear composition, size and type of intermolecular bonding.
Recently, we have achieved direct detection of homomeric FHC interactions in the intact plasma membrane by means of a live-cell two-hybrid micropattern assay (Fig. 1B,C; Dirscherl et al., 2018), where micrometer-sized patterns of anti-hemagglutinin tag (HA) monoclonal antibody are printed onto glass coverslips (Schwarzenbacher et al., 2008; Sevcsik et al., 2015). Onto these micropatterns, cells that express two different MHC-I HC constructs are seeded: one construct has an N-terminal (extracellular) HA tag and thus, while diffusing laterally in the plasma membrane, is captured into the printed antibody pattern. The other has no HA tag but a C-terminal (intracellular) green fluorescent protein (GFP) fusion domain (Fig. 1B). Interaction of GFP-tagged HC with micropatterned HA–HC is detected by an increased GFP fluorescence of the pattern elements (Fig. 1C). The system allows the observation of different defined conformational states of class I molecules – with transporter associated with antigen processing 2 (TAP2)-deficient fibroblasts [which cannot transport peptide into the endoplasmic reticulum (ER); Zimmer et al., 1999], empty HC–β2m heterodimers are present at the cell surface. By adding peptide, shifting cells to 25°C or incubating at 37°C, we therefore can cause the accumulation of trimers, HC–β2m heterodimers or FHCs, respectively, at the cell surface (Fig. 1B and described below). These experiments revealed that formation of homomeric MHC-I only takes place in the absence of β2m, that is, only between FHCs.
We now have uncovered the molecular principles that govern such homomeric MHC-I FHC association in the plasma membrane. Cell micropatterning in conjunction with fluorescence recovery after photobleaching (FRAP) was used to probe the dynamics, stability and prominence of FHC complexes. We directly demonstrate FHC association in the plasma membrane under physiological conditions by using real-time single-molecule tracking (SMT) and co-tracking (SMCT). Surprisingly, we find that FHCs transiently associate into non-covalent dimers with lifetimes in the sub-second range. Based on our findings that the HC–HC complexes contain no β2m and that the α3 domain of the HC is sufficient for dimerization in vitro and in cells, we propose a molecular model structure of MHC-I HC–HC dimers supported by in silico docking and molecular dynamics (MD) simulation. Our findings clearly differentiate the cell surface dynamics and properties of empty FHCs from the peptide-loaded trimers of MHC-I, pointing to the formation of structurally well-defined HC–HC homodimers that might be responsible for distinct endosomal trafficking and other biological functions previously ascribed to FHCs.
A micropattern assay reveals non-covalent association of MHC class I FHCs
In STF1 cells, which are mouse fibroblasts that cannot load MHC-I with peptides due to a deficiency in the TAP peptide transporter, incubation at 25°C leads to the accumulation of murine HC–β2m heterodimers at the plasma membrane, since at that temperature, dissociation of β2m and subsequent endocytosis are inhibited (Day et al., 1995; Ljunggren et al., 1990; Montealegre et al., 2015). When the temperature is shifted to 37°C, β2m dissociates to yield FHCs, and lateral in cis interactions between HA–Kb and Kb–GFP (both hybrids of the HC of the murine MHC-I molecule H-2Kb, encoded by H2-K1) become visible in the micropattern two-hybrid assay, where the GFP fluorescence arranges in the shapes of the antibody micropattern (Fig. 1C). When cognate peptide is added to the cells, HC–β2m–peptide trimers are stable and do not associate with each other (Fig. 2B) (Dirscherl et al., 2018).
Up to six different MHC-I allotypes are present at the cell surface of human and murine cells. To date, though, an interaction of different MHC-I allotypes in the same plasma membrane has not been shown. We therefore tested for such heterotypic interactions between Kb and H-2Db (Db, encoded by H2-D1). Just as for Kb–Kb, Kb and Db FHCs (at 37°C) also interacted with each other, whereas β2m-bound heterodimers at 25°C did not (Fig. 1D). Hence, our micropattern assay confirmed that formation of heterotypic FHC interactions is possible.
We next asked which molecular characteristics are required for HC–HC interactions, and we first tested whether the HCs are held together by cytosolic disulfide bonds, as previously suggested for certain allotypes under oxidizing conditions (Baía et al., 2016; Capps et al., 1993; Makhadiyeva et al., 2012). We replaced the single cysteine in the cytosolic tail of Kb–GFP (residue 332) with a serine and tested for association between this mutant and wild-type HA-Kb (Fig. 1E). Kb(C332S)–GFP and wild-type HA–Kb interacted in over 90% of cells, just as with the wild-type Kb-GFP, demonstrating that cysteine 332 is not required for HC–HC interactions.
To test more generally for any disulfide bonding in MHC-I interactions, we immunoprecipitated HA–Kb molecules from STF1 cell lysate with an anti-HA antibody. As determined by non-reducing gel electrophoresis, covalent homodimers of Kb were not formed, while the well-described disulfide-linked homodimers of the human MHC-I allotype HLA-B*27:05 (Dangoria et al., 2002) were readily detected (Fig. 1F). We conclude that the Kb FHCs are non-covalently associated and that they do not undergo intramolecular disulfide bonding under the conditions of our assay.
Since the cysteine residues in the cytosolic tail of Kb are not required for HC–HC interaction, we hypothesized that FHCs non-covalently associate via their extracellular domains, and that the loss of β2m is a prerequisite for HC–HC interaction. As anticipated, a disulfide-stabilized variant of Kb–GFP (Y84C/A139C), in which β2m dissociation is dramatically decreased (Hein et al., 2014), did not interact with HA–Kb (Fig. 1G). In agreement with our earlier finding that covalent attachment of β2m to the HC also prevents HC–HC association (Dirscherl et al., 2018), this result demonstrates that HC–β2m heterodimers do not interact with other HC–β2m heterodimers, nor with FHCs. These findings demonstrate that the MHC class I complexes observed in our system are non-covalent in nature and comprise free heavy chains, devoid of β2m and peptide.
Free heavy chains associate on TAP-proficient cells
So far, we used TAP2-deficient STF1 cells to obtain homogeneous populations of FHCs at the cell surface. Since MHC-I HC–β2m heterodimers and FHCs both exist at the cell surface of wild-type cells (Day et al., 1995; Ljunggren et al., 1990; Ortiz-Navarrete and Hämmerling, 1991), we next tested whether FHCs also associate in cells with wild-type TAP function (schematic in Fig. 2A). Following the 25°C to 37°C temperature shift that triggers FHC formation, we performed the same anti-HA antibody two-hybrid micropatterning assay with HA–Kb and Kb–GFP as above (in Fig. 1B,C). Again, the punctate GFP fluorescence signifies the recruitment of the Kb–GFP fusion to the printed patterns of the anti-HA antibodies, mediated by the HA–Kb fusion. We quantified the Kb HC–HC interaction over time in STF1 TAP2-deficient and STF1 TAP2-transduced cells, the latter of which are able to load class I molecules with peptides, by GFP fluorescence contrast analysis between pattern elements and interspaces. Interactions increased in TAP2-proficient and in TAP2-deficient cells with the same dynamics and reached a maximum after ∼60 min of incubation at 37°C (Fig. 2B,C), with the kinetics likely governed by the dissociation of β2m from the HC–β2m heterodimers, which occurs on this timescale (Montealegre et al., 2015). As expected, HC–β2m–peptide trimers, stabilized by cognate SIINFEKL peptide, did not show any interaction (Fig. 2B). This experiment confirms that HC–HC interactions indeed take place at the surface of TAP2-proficient cells. In addition, these experiments again confirm our earlier observations (Dirscherl et al., 2018) that the HC–HC interaction occurs only between FHCs; when dissociation of β2m is inhibited by low temperature (Fig. 1D,E,G) or by added peptide (Fig. 2B,D), the interaction does not occur. The same experiments also strongly suggest that the recruitment of the Kb–GFP fusion to the pattern elements specifically reports on the interaction between MHC class I molecules (since it depends on their conformation) and is not due to an attraction between Kb–GFP and other surface proteins, for example, adhesion molecules, that might have been attracted to the plasma membrane above the pattern elements.
FHC association slows down cell surface diffusion of MHC-I
Since dimers and oligomers of FHCs have more transmembrane domains (TMDs) than single HC–β2m–peptide complexes, we hypothesized that they should diffuse more slowly in the plasma membrane (Gambin et al., 2006; Wilmes et al., 2015). We therefore carried out total internal reflection fluorescence (TIRF) fluorescence recovery after photobleaching (FRAP) experiments on STF1 cells with HA–Kb and Kb–GFP on surfaces with or without micropatterns (Fig. 2D) and quantified the recovery dynamics (Fig. 2E). Diffusion constants of FHCs (– peptide) were significantly decreased as compared to HC–β2m–peptide complexes (+ peptide; Fig. 2F), suggesting that FHCs form complexes. The moderate decrease of ∼40% is in line with the formation of dimers rather than clustering into larger complexes. Since this effect occurred both in cells seeded on the pattern elements (on pattern) and in cells seeded off the micropatterns (off pattern), we conclude that the micropatterns are not required for HC–HC interactions, that is, the interactions are not an artifact of the two-hybrid micropattern assay. This statement is also supported by the co-immunoprecipitation of MHC-I molecules in the absence of micropatterns (Dirscherl et al., 2018; Triantafilou et al., 2000). Diffusion constants of HC–β2m–peptide trimers on pattern and off pattern were not significantly different, which demonstrates that the antibody micropatterns themselves do not impede the diffusion of plasma membrane proteins.
In the FRAP experiments, a portion of Kb–GFP appeared to be immobile on a timescale of seconds as evidenced by the incomplete fluorescence recovery (Fig. 2E). This immobile fraction was significantly higher for FHCs than for HC–β2m–peptide trimers (Fig. 2G). This observation is readily explained for Kb–GFP bound to HA–Kb molecules immobilized within micropatterns, but it is remarkable for the FHCs on cells outside of the micropattern (gray column in Fig. 2G). There, it might be ascribed to the formation of large oligomers, and/or to the association of FHCs with immobile structures, such as the cytoskeleton and/or due to sequestration in endocytic membrane compartments (Bondar et al., 2020; Ibach et al., 2015; Mylvaganam et al., 2018; Vámosi et al., 2019). From the FRAP curves, the exchange rate of the freely diffusing pool of Kb–GFP into and out of the bleached regions of interest (ROIs) is obtained through a bi-exponential fit as the slow recovery rate (kslow) (Fig. 2H, solid bars); the fast recovery rate kfast represents free diffusion (Sprague and Mcnally, 2005) (see the Materials and Methods). On the pattern elements, the kslow of FHCs was much smaller than in cells outside the patterns, suggesting a half-time of dissociation from the pattern-bound immobile associations of ∼140 s. Likewise, the fluorescence signal of pattern elements in the immediate vicinity of the bleached region remained unaltered, indicating that no detectable exchange of Kb–GFP between the enriched HA–Kb regions occurs on the second timescale (Fig. S2D,E). This very slow exchange of Kb–GFP molecules associated on pattern elements suggests either multiple association and dissociation events in a small radius due to densely immobilized binding partners, or a very stable association of the immobile FHCs (see the Discussion). Taken together, the data from Fig. 2 show reduced diffusion rates of Kb FHCs at the cell surface compared to HC–β2m–peptide complexes, which is most easily explained by homotypic association of the former.
Transient FHC dimerization directly observed at single molecule level
We therefore turned to directly visualizing FHC diffusion and interaction in the plasma membrane under physiological conditions by single-molecule tracking (SMT) and co-tracking (SMCT) (Moraga et al., 2015; Sevcsik et al., 2015; Wilmes et al., 2020). STF1 cells transiently expressing Kb with its N-terminus fused to monomeric GFP (GFP–Kb) were imaged by TIRF microscopy. The GFP tag was labeled with photostable fluorescent dyes by using equal concentrations of anti-GFP nanobodies conjugated to either ATTO Rho11 (Rho11NB) or ATTO 643 (ATTO643NB), with one nanobody binding one GFP molecule, ensuring selective imaging of Kb in the plasma membrane and tracking with high fidelity (Fig. 3A). Imaging was performed at 37°C to induce formation of FHC in the absence of the peptide, while replenishment of (GFP–Kb)–β2m dimers from the ER was inhibited by treatment with brefeldin A (BFA).
After labeling with Rho11NB and ATTO643NB, individual Kb subunits were observed at densities of <1 molecule/µm² diffusing randomly in the plasma membrane (Movie 1, Fig. S3A). Co-tracking analysis revealed homomeric interaction of Kb FHCs in the absence of peptide, whereas these events were very rare for peptide-loaded Kb (Fig. 3B–D). As a positive control for the formation of homodimers, a crosslinker based on the tandem anti-GFP nanobody LaG16 (Fridy et al., 2014), which recognizes a different epitope than the labeled nanobodies, was added to the medium. Whereas dimerizing of Kb via the GFP tag resulted in a nominal fraction of ∼38% (median) colocomoting molecules, only ∼3% of the FHCs were found to be associated, indicating weak interaction of FHCs at these low cell surface expression levels of Kb (∼1 molecule/µm² in total, corresponding to <5000 molecules/cell).
From the trajectory analysis, we determined the diffusion coefficients of individual MHC-I molecules by mean square displacement analysis (Fig. 3E; Fig. S3B). In the absence of peptide, diffusion of FHCs was significantly slower than in the presence of the peptide. This decrease in single molecule diffusion coefficients supports interaction of FHCs in the plasma membrane. A similar decrease in the diffusion coefficient was observed upon dimerization of peptide-loaded Kb with the tandem nanobody, suggesting that FHCs associate into dimers. Indeed, largely identical diffusion coefficients were found for the fraction of molecules identified as dimers by SMCT (Fig. S3C, Table 1). Strikingly, the diffusion constants obtained by SMT for peptide-loaded Kb and for FHC from SMT are consistent with those obtained by FRAP experiments under the same conditions (Fig. 2E, Table 1), highlighting that similar phenomena are being probed by these complementary techniques. Tracking analysis revealed a slightly elevated immobile fraction of ∼20% for Kb in the absence of the peptide compared to ∼17% for peptide-loaded Kb (Fig. 3F). Again, a similar effect was observed for artificially dimerized peptide-loaded Kb, corroborating the dimeric stoichiometry of associated FHCs. Importantly, the ‘immobile fraction’ in SMT refers to much shorter time and length scales as compared to FRAP (see details in the Materials and Methods section), and therefore, the absolute numbers are not comparable. Within the immobile fraction, a higher level of FHC was found to be associated, as compared to within the mobile fraction (Fig. 3G). Overall, the single-molecule diffusion analyses suggest that FHC dimers have an increased propensity to become immobilized at the plasma membrane, probably through clustering that might be related to endocytosis. Interestingly, this feature is reproduced by artificial dimerization of intact MHC-I by a crosslinker, suggesting that FHC dimerization is a switch regulating its cell surface dynamics.
SMCT analysis moreover revealed dissociation of FHC homomers, confirming its transient nature (Fig. 3B, Movies 2–5). We estimated the lifetime of FHC complexes from co-trajectory length histograms. Fitting of an exponential decay revealed a significantly shorter lifetime for FHC co-trajectories compared to the average co-tracking lifetime determined for stably NB-crosslinked Kb, which is limited by co-tracking fidelity and photobleaching (Fig. 3H). Taken together, the SMT and SMCT results of Fig. 3 directly confirm a transient homotypic HC–HC interaction in the plasma membrane that leads to reduced diffusion velocity and immobilization, in line with the FRAP experiments. Since the degree of association was low, and higher-order oligomerization was not observed, we propose that HC–HC interactions are weak and have dimeric stoichiometry.
The α3 domain of Kb forms dimers and is sufficient for FHC association
We next explored the molecular mechanism of HC–HC interaction. Since cytosolic cysteine residues are not involved (Fig. 1E,F) and dissociation of β2m is necessary (Fig. 1G), we hypothesized that HC–HC interactions involve the extracellular portion of the HCs. To determine more precisely which domains are involved, we used STF1 cells that expressed constructs of Kb that lacked the α1-α2 domain in the micropattern assay. Remarkably, α3–GFP showed excellent co-patterning with HA–Kb–RFP (fused to red fluorescent protein; Fig. 4A) and with HA–α3 (Fig. 4B), demonstrating that isolated α3 domains can bind to each other on the surface of live cells. Indeed, the isolated, soluble α3 domain of Kb, when refolded in vitro, showed (after testing of its folded state by fluorescence spectroscopy, Fig. S4C) significant formation of homodimers in size exclusion chromatography (SEC; Fig. 4C), while no prominent higher-order complexes were observed. These α3 homodimers were not linked by disulfide bonds as shown by nonreducing gel electrophoresis (Fig. 4D), just like the interaction of full-length FHC in the plasma membrane (Fig. 1F). Given the 1:1 peak ratio observed at 10 µM protein concentration, a binding affinity of 10 to 20 µM can be estimated for this interaction. The non-covalent α3 homodimers were also detectable by native mass spectrometry (MS) in a concentration-dependent manner, and they easily separated into the monomers when the collision cell voltage was increased in an tandem MS (MS/MS) experiment, suggesting a low-affinity interaction (Fig. 4E–G; Fig. S4A). These biochemical results, together with the observations on the α3 constructs above, suggest that FHC dimerization in the plasma membrane is at least partly based on the intrinsic affinity between the α3 domains. To investigate whether the α3 domain is indeed required for FHC dimerization, we performed a co-patterning experiment with an HA–Kb fusion protein that lacked the α3 domain [HA–Kb(Δα3)–RFP] and Kb–GFP and found the interaction significantly reduced, but still measurable (Fig. S4B), which suggests that the α1-α2 domain is also involved in the interaction.
A molecular model of the Kb HC dimer
To image a possible arrangement of the Kb heavy chains in a dimer, we used molecular docking. A starting geometry was obtained by placing the α3 domain of one Kb HC in the same position as the β2m in the original Kb HC–β2m heterodimer and adjusting the α1-α2 superdomain to avoid sterical overlap. The structure was then energy-minimized and further refined by MD simulations (Fig. 5A). During MD simulations of 400 ns, no signs of dissociation were observed, and a stable root-mean-square deviation (RMSD) from the start structure was reached (Fig. S5). In the model, the α3 domain of one FHC substitutes for β2m, binding both the α3 domain and the α1-α2 superdomain of another FHC with a buried interface area of 2480 Å2 (1 Å=0.1 nm), which is comparable to the buried interface area of 2740 Å2 between β2m and the HC in heterodimers (Fig. 5B, bottom), and to other stable protein complexes (Bahadur and Zacharias, 2008). Similar results were obtained with other arrangements of the two heavy chains (not shown).
Interactions of MHC-I FHC with FHC of the same allotype, of different allotypes, and even with other cell surface proteins have been proposed to play an important role in regulating adaptive and innate immune responses (Arosa et al., 2021, 2007; Campbell et al., 2012), but the molecular principles that govern FHC interactions have remained unclear. By combining live-cell interaction and diffusion analysis using cell micropatterning and FRAP as well as SMT and SMCT, we have shown here that, after losing β2m, murine H-2Kb MHC-I molecules at the cell surface interact in a homotypic and heterotypic manner to form dimers, which are transient with a stability on the second timescale. The α3 domains of the FHCs alone are sufficient for such interactions, but it is conceivable that the α1-α2 domain is also involved: in addition to the α3-α3 interaction, the α3 domain might also bind to the α1-α2 domain of the other protein, or the α1-α2 domains of the two proteins might interact with each other. We cannot currently distinguish these possibilities.
Although we observed a somewhat increased tendency of FHCs to form higher oligomers, our data surprisingly identify non-covalent FHC dimers as the prevalent species at the cell surface. We analyzed the diffusion properties of FHCs and HC–β2m–peptide trimers at different cell surface expression levels by FRAP and SMT. FRAP experiments were carried out at high surface densities, which probably exceeded the typical endogenous cell surface expression levels of 105 copies/cell (Spack and Edidin, 1986). By contrast, SMT was performed at densities of ∼1 molecule/µm², that is, <5000 molecules/cell. Despite these differences in expression levels, we found highly consistent diffusion coefficients D by both FRAP and SMT, which revealed moderately decreased mobility upon dissociation of β2m (Table 1). However, under both conditions, the decrease in D was ∼40–50%, which perfectly agrees with the change in D upon dimerizing intact MHC-I through the tandem nanobody (Fig. 3E). Likewise, we and others have previously observed very similar decreases of 30–50% upon dimerization of cell surface receptors (Ho et al., 2017; Low-Nam et al., 2011; Moraga et al., 2015; Richter et al., 2017; Váradi et al., 2019; Wilmes et al., 2015, 2020), corroborating that the mobile HC–HC complexes are mostly dimers.
Direct detection of FHC interaction by SMCT corroborated transient non-covalent dimerization and only minor clustering into higher oligomers. Estimated lifetimes of the non-covalent FHC–FHC dimer (t1/2=220±150 ms, Fig. 3G,H) were well within the measurable range, that is, substantially below the apparent half-life obtained for the quasi-irreversibly nanobody-crosslinked control (t1/2>10 s), which defined the limit of co-tracking fidelity. Together with the immunoprecipitation experiments (Fig. 1F) and micropatterning of FHC lacking free cysteine residues (Fig. 1E), these observations clearly establish that the FHC molecules at the plasma membrane are not covalently linked. The short half-life and the non-covalent monomer-dimer equilibrium observed by size exclusion chromatography at higher micromolar concentration for the isolated α3 domain (Fig. 4C,D), as well as the MS data (Fig. 4E–G), point to a low-affinity interaction with a dimerization Kd of perhaps 10 to 20 µM. Previous quantitative studies on affinity-dimerization correlation of heterodimeric cytokine receptors (Wilmes et al., 2015) would predict efficient FHC dimerization at physiological densities above 10 molecules/µm², which is in line with the endogenous MHC-I expression level. The relatively low level of dimerization we observed by SMCT can be rationalized with the high background of endogenous MHC-I molecules that are unlabeled and invisible to us. In line with this observation, the decrease of the diffusion coefficient in SMT was much more prominent than the dimer fraction identified by SMCT. This interpretation is in line with the observation that more efficient interaction was observed at the elevated cell surface expression levels used in micropatterning experiments.
The weak and transient nature of HC–HC dimerization seen by SMT is similar to the nature of cell surface protein–protein association measured in other systems (Lin et al., 2014). It does not conflict with the clear and stringent patterning of the GFP fusion in the micropattern and FRAP experiments. In the latter, the concentration of HA-tagged bait HCs is considerably higher due to their immobilization by the antibodies, which prevents their endocytosis, and this probably creates an affinity matrix for the GFP fusions that can retain them for many seconds due to rebinding events. This also suggests that in endosomes, whose internal volume is very small, HC–HC interaction might be potentiated due to the increase in the concentration of the monomers compared to the plasma membrane.
In line with the observation of non-covalent dimerization in the plasma membrane, we obtained a robust structural model of a self-contained Kb FHC homodimer. The atomistic model (Fig. 5) was derived from the experimental findings that dissociation of β2m is required, that the α3 domains are sufficient, and that a direct α3–α3 interaction exists (Figs 1E and 4A–C). We propose that β2m dissociation exposes a binding site on the FHC for the α3 domain of another FHC. However, other arrangements of the two FHCs in a dimer are theoretically possible, and only experimental data will give a definitive answer.
Several findings in the literature are consistent with the formation of complexes of MHC class I HCs in the absence of β2m and peptide; this applies both to the ‘classical’, or class Ia, proteins HLA-A, HLA-B and HLA-C (and in mouse H-2D, H-2K and H-2L) as well as to the ‘non-classical’, or class Ib, protein HLA-F (Armony et al., 2021; Bodnar et al., 2003; Chakrabarti et al., 1992; Matko et al., 1994; Triantafilou et al., 2000). Still, it is important to differentiate these HC–HC dimers from class I associations described elsewhere (partially reviewed in Arosa et al., 2021, 2007; Campbell et al., 2012), namely homo- and hetero-typic HC–β2m–peptide trimers that are covalently dimerized via disulfide bonds in their cytosolic tails (Capps et al., 1993; Makhadiyeva et al., 2012) and that may play a role in binding the LILRB natural killer (NK) cell receptor (Baía et al., 2016); the non-covalent nano- and micro-scale clusters of HC–β2m–peptide trimers (not detected in our system) that might stem from the fusion of exocytic vesicles with the plasma membrane and that might play a role in T cell receptor (TCR) recognition (Blumenthal et al., 2016; Ferez et al., 2014; Fooksman et al., 2006; Lu et al., 2012); the macroscopic ‘clusters’ of class I molecules at the signaling interface between cells (Fassett et al., 2001); and the covalent dimers of the HLA-B*27:05 heavy chain that are linked by disulfide bonds through Cys67 (Chen et al., 2017), though our non-covalent HC–HC dimers may be a precursor to the formation of the latter.
The unexpected discovery that FHCs non-covalently associate into defined dimers allows exciting hypotheses of their distinct functional properties. FHC dimers might be responsible for the immunomodulatory functions of cell surface heavy chains, that the stabilization of MHC-I trimers to assist T cell activation (Geng et al., 2018; Schell, 2002) and direct binding of FHCs to receptors on other cells [for example, FHCs of HLA-F binding to activating receptors on NK cell (Dulberger et al., 2017; Goodridge et al., 2013)]. Furthermore, dimerization of FHCs might enhance endocytosis in order to remove the non-functional FHCs, which themselves cannot activate T cells and are known to be short-lived (Mahmutefendic et al., 2011; Montealegre et al., 2015); alternatively or additionally, such associated HCs may bind to other proteins in cis and promote their removal from the plasma membrane. Such endocytic removal might be achieved by altered endosomal routing, since the local density of membrane proteins in endosomes is higher than at the plasma membrane, and thus, efficient dimerization of MHC-I FHCs is expected. In such a scenario, even transient oligomerization in endosomes might prevent the return of internalized MHC-I FHCs to the cell surface (Montealegre et al., 2015). Taken together, the presence of non-covalently bound transient FHC dimers points to exciting new aspects in the regulation of MHC-I functions with much potential for further investigation.
MATERIALS AND METHODS
Cells and cell lines
TAP-deficient human STF1 fibroblasts (kindly provided by Henri de la Salle, Etablissement de Transfusion Sanguine de Strasbourg, Strasbourg, France) were cultivated at 37°C and 5% CO2 in Earle's minimum essential medium (OptiMEM, #11058-021, Invitrogen Life Technologies) with stable glutamine supplemented with 10% fetal bovine serum (FCS-Superior, Merck Millipore #S0615), non-essential amino acids (PAA Laboratories, #M11-003, Cytiva Life Sciences) and HEPES buffer without addition of antibiotics. The generation of the TAP2-transduced STF1 cells is described in Hein et al. (2014).
Genes, vectors, and gene expression
HA–Kb and Kb–GFP constructs were described previously (Dirscherl et al., 2018). HA–Kb carries an influenza hemagglutinin (HA) tag at the N-terminus of the full-length murine H-2Kb, whereas Kb–GFP carries a GFP domain at the C-terminus of H-2Kb. The Db–GFP construct is analogous to the Kb–GFP construct. The α3–GFP construct consists of the H-2Kb signal sequence and residues 204–369 of H-2Kb, including the transmembrane and cytosolic domains. The HA–Kb(Δα3)–RFP construct, compared to HA–Kb, lacks residues 204–294 and carries an additional RFP taken from pcDNA3-mRFP (see Table S1). The GFP–Kb construct used in single-molecule imaging consists of a signal sequence and GFP fused to the N-terminus of H–2Kb itself lacking a signal sequence. Amino acid numbering of Kb in this paragraph includes the signal sequence.
Stable cell lines were generated by lentiviral transduction as described (Hein et al., 2014), and transient transfection was achieved by electroporation (Garstka et al., 2007) or by calcium phosphate precipitation (Graham and van der Eb, 1973) as described previously.
Silicon master molds were prepared by semiconductor photolithography as described previously (Dirscherl et al., 2017).
PDMS stamps and antibody patterns
PDMS stamps were generated from basic elastomer and curing agent (Sylgard 184 Silicone Elastomer Kit) as described previously (Dirscherl et al., 2017).
Patterning cell surface proteins
Coverslips with antibody pattern were placed into sterile six-well plates. Cells were immediately seeded as indicated at a concentration of ∼50,000 cells per well. Usually, cells were incubated for 4–6 h at 37°C for adhesion and then shifted to 25°C to accumulate MHC-I molecules at the cell surface. Samples were then kept at 25°C to increase cell surface heterodimer levels or shifted back to 37°C for 3–4 h to induce FHCs by dissociation of β2m.
Purified antibodies were labeled with Alexa Fluor 647 NHS ester (Thermo Fisher Scientific, Darmstadt, Germany) according to the manufacturer's protocol.
The Kb-specific peptide SL8 (SIINFEKL in the single-letter amino acid code) was synthesized by GeneCust (Ellange, Luxemburg) and EMC microcollections (Tübingen, Germany) and purified by HPLC (90% purity). Peptides were added to the cells at a final concentration of 2 µM for 15–30 min at 37°C to induce peptide binding (Dirscherl et al., 2018).
Washing and fixation
Cells were washed with phosphate-buffered saline (PBS; 10 mM phosphate pH 7.5, 150 mM NaCl), fixed with 3% paraformaldehyde (PFA) and observed by confocal laser scanning microscopy (cLSM).
We used a confocal laser scanning microscope (LSM 510 Meta, Carl Zeiss Jena GmbH, Germany) equipped with argon and helium-neon lasers at 488, 543 and 633 nm. Images were recorded with a 63× Plan Apochromat oil objective (numerical aperture 1.4) at a resolution of 1596×1596 pixels. Data acquisition was performed with the LSM 510 META software, release 3.2 (Carl Zeiss Jena). During image acquisition, patterns and cells were imaged in the same focal plane at a pinhole of 1 Airy unit. Image analysis and processing were performed using ImageJ (National Institutes of Health, Bethesda, USA). Image processing comprised cropping, rotation and adjustment of brightness and contrast levels. Experiments in Fig. 1 were repeated at least three times each.
Recombinant ɑ3 domain of H-2Kb
The ɑ3 domain of H-2Kb (residues 205–295) was cloned into pET3a (Novagen Merck Biosciences; preceded by the residues MAIQR and followed by DRDM) and expressed in Escherichia coli BL21(DE3) pLysS, refolded in vitro as described, and isolated by size exclusion chromatography (SEC) on a Cytiva Hiload Superdex 200 16/600 column (Anjanappa et al., 2020). Molecular masses of the peaks were determined by comparison to SEC protein standards (Cytiva), namely bovine thyroglobulin (670 kDa), bovine γ-globulin (158 kDa), chicken ovalbumin (44 kDa), horse myoglobin (17 kDa), and vitamin B12 (1.35 kDa). The D5 fraction, corresponding to the elution peak at ∼20–30 kDa, was boiled with or without dithiothreitol (DTT; 0.6 M final concentration) in sample buffer (LSB) (350 mM Tris-HCl pH 6.8, 10.28% SDS, 36% glycerol 0.012% Bromophenol Blue). Inclusion body extract boiled without DTT (non-reducing) served as the positive control for the formation of covalent oligomers. Protein quality control after refolding and SEC was performed by nanoscale differential scanning fluorimetry (nanoDSF) runs (see Fig. S4D) acquired with a Nanotemper Prometheus NT.48 fluorimeter (Nanotemper, Munich) controlled by PR.ThermControl (version 2.1.2).
Precipitation of surface class I
TAP2-deficient STF1 cells expressing either E3–HA-Kb (Dirscherl et al., 2018) or E3–HLA-B*27:05 (Hein et al., 2018) were kept overnight at 25°C and then pretreated with tris(2-carboxyethyl)phosphine (TCEP; 1 mM, 10 min), labeled with 400 nm of Bio-MPAA-K3 (5 min) at room temperature (Reinhardt et al., 2014). Lysis was performed in native lysis buffer (50 mM Tris-HCl pH 7.4, 150 mM NaCl, 5 mM EDTA and 1% Triton X-100) for 1 h at 4°C. Biotinylated surface proteins were then isolated with neutravidin-coated agarose beads (Thermo Fisher Scientific, Darmstadt, Germany). The isolates were boiled at 95°C for 7 min in the presence (reducing) or absence (non-reducing) of 10 mM dithiothreitol (DTT) in sample buffer as described above. Samples were separated by SDS-PAGE and transferred onto PVDF membranes. MHC molecules were visualized on the membranes with polyclonal rabbit anti-HA antibody as primary antibody (1:1000, ab9110, Abcam, Cambridge, UK) and alkaline phosphatase-conjugated anti-rabbit-IgG serum from goat as secondary antibody (1706518, Biorad, Munich, Germany). The signals were visualized by treating the blot with BCIP/NBT substrate (B1911, Sigma-Aldrich, St Louis, MI, USA).
For verification of cell surface levels of MHC-I, flow cytometry was performed with anti-HLA class I antibody W6/32 (Barnstable, 1978) and anti-HA antibody (12CA5, described in key resources table). Antibody–antigen complexes were labeled with goat secondary antibody against mouse IgG conjugated to allophycocyanin (APC) (115-135-164, Dianova, Hamburg, Germany). Fluorescent signal was recorded by a CyFlow1Space flow cytometer (Sysmex-Partec, Norderstedt, Germany) and analyzed by Flowjo, LLC software.
Microcontact printing and antibody patterning for TIRF microscopy
The microcontact printing and antibody patterning for TIRF microscopy was performed as described previously (Lanzerstorfer et al., 2020). In short, a field of a large-area PFPE elastomeric stamp (1 µM grid size), obtained by the EV-Group (St. Florian am Inn, Upper Austria, Austria), was cut out, and washed by flushing with ethanol (100%) and distilled water. After drying with nitrogen, the stamp was incubated in 50 ml bovine serum albumin (BSA) solution (1 mg/ml) for 30 min. This step was followed by washing the stamp again with PBS and distilled water. After drying with nitrogen, the stamp was placed with homogeneous pressure onto the clean epoxy-coated glass bottom of a 96-well plate and incubated overnight at 4°C. The next day, the stamp was stripped from the glass with forceps, and the glass bottom was bonded to a 96-well plastic casting with adhesive tape (3 M) and closed with an appropriate lid. For the live-cell experiments, a reaction chamber was incubated with 100 µl streptavidin solution (50 µg/ml) and incubated for 30 min at room temperature. After washing two times with PBS, 100 µl biotinylated antibody solution (10 µg/ml) was added for 30 min at room temperature resulting in an antibody surface coverage of >85% (Fig. S1A,B). Finally, the incubation wells were washed twice with PBS, and cells were seeded at defined cell density for the live-cell microscopy analysis. The cells were allowed to attach to the surface for at least 3–4 h prior to imaging to ensure a homogeneous cell membrane–substrate interface, which is a prerequisite for quantitative TIRF microscopy. Deformation of the plasma membrane on top of the pattern elements was excluded, since control cells that were stained with the lipophilic tracer 1,1′-dioctadecyl-3,3,3′,3′-tetramethylindodicarbocyanine (DiD), which stains the plasma membrane uniformly, showed no patterning (Fig. S1C). For negative control to test the adhesion of the antibodies, anti-HA antibody (Abcam, ab26228) was labeled with a Zenon Alexa Fluor 488 IgG Labeling Kit (Thermo Fisher Scientific, Z25102), printed and bleached, and fluorescence recovery was quantified as described below (Fig. S2A,B). In a second control experiment, binding and dissociation of a construct with both tags (HA–Kb–GFP) from the antibody micropattern was tested, which only accounted for less than 20% of the mobility of the Kb fraction (Fig. S2C). Dissociation of Kb–GFP from micropattern elements was negligible (Fig. S2D,E). Thus, in our experiments, kslow was determined by the binding events between Kb–GFP and other Kb molecules.
Live-cell TIRF microscopy
The detection system was set up on an epi-fluorescence microscope (Nikon Eclipse Ti2). A multi-laser engine (Toptica Photonics, Munich, Germany) was used for selective fluorescence excitation of GFP at 488 nm and RFP at 568 nm. The samples were illuminated in total internal reflection (TIR) configuration (Nikon Ti-LAPP) using a 60× oil immersion objective (NA=1.49, APON 60XO TIRF). After appropriate filtering with standard filter sets, the fluorescence was imaged onto a sCMOS camera (Zyla 4.2, Andor, Northern Ireland). The samples were mounted on an x-y-stage (CMR-STG-MHIX2-motorized table, Märzhäuser, Germany), and scanning of the larger areas was supported by a laser-guided automated Perfect Focus System (Nikon PFS).
TIR-FRAP experiments and calculation of diffusion coefficients
Diffusion coefficients were obtained using the initial image recordings and the simFRAP plugin for ImageJ (Blumenthal et al., 2015).
Temperature-induced Kb associations
Temperature-dependent experiments were carried out on an epi-fluorescence microscope as described above further equipped with a cage incubator (Okolab, Shanghai, China). Cells were grown at 25°C overnight on antibody-patterned surfaces and treated with SIINFEKL peptide as indicated. For induction of Kb FHC association, cells were mounted on pre-warmed microscopy stage, and imaging of the GFP signal was started when the medium reached 37°C.
Fluorescence contrast quantification
Cells were transferred 48 h post transfection onto glass coverslips coated with a poly-L-lysine-graft-(polyethylene glycol) copolymer functionalized with RGD tripeptide to minimize non-specific binding of fluorescent nanobodies (You et al., 2014). Cells imaged in presence of SL8 peptide (sequence SIINFEKL) were pre-incubated with 1 µM of SL8 peptide 12 h before imaging. Single-molecule imaging experiments were conducted by TIRF microscopy with an inverted microscope (Olympus IX83) equipped with a motorized 4-line TIR illumination condenser (Olympus) and a back-illuminated electron multiplying (EM) CCD camera (iXon Ultra 897, Andor Technology). A 100× magnification objective with a numerical aperture of 1.45 (UPLAPO 100× HR, NA 1.5, Olympus) together with a 1.6× magnification changer was used for TIR illumination of the sample. Imaging was conducted with or without 1 µM of fresh SL8 peptide. The sample was preincubated with 10 µg/ml of brefeldin A (BFA) for 15 min in order to inhibit protein transport of (GFP–Kb)–β2m heterodimers to the plasma membrane. All experiments were carried out at 37°C in medium without Phenol Red supplement with an oxygen scavenger and a redox-active photoprotectant to minimize photobleaching (Vogelsang et al., 2008) and penicillin and streptomycin (PAA). For cell surface labeling of GFP–Kb, a 1:1 mixture of anti-GFP nanobodies (2 nM each) site-specifically conjugated to ATTO 643 and ATTO Rho11 (Wilmes et al., 2020), respectively, were added to the medium, thus ensuring >90% binding given the 0.3 nM binding affinity (Kirchhofer et al., 2010). After incubation for at least 5 min, image acquisition was started with the labeled nanobodies kept in the bulk solution during the whole experiment in order to ensure high equilibrium binding. Dimerization of the positive control was induced by applying 0.3 nM of a tandem nanobody crosslinker (LaG16V2) binding to an orthogonal epitope (Fridy et al., 2014). For single-molecule colocalization and co-tracking experiments, orange (ATTO Rho11)- and red (ATTO 643)-emitting fluorophores were simultaneously excited by illumination with a 561 nm laser (MPB Communications) and a 642 nm laser (MPB Communications). Fluorescence was detected with a spectral image splitter (QuadView QV2, Photometrics) with a dichroic beam splitter (Chroma) combined with the bandpass filter 600/37 (BrightLine HC) for detection of ATTO Rho11 and 685/40 (Brightline HC) for detection of ATTO 643 dividing each emission channel into 256×256 pixels. Image stacks of 150 frames were recorded for each cell at a time resolution of 32 ms/frame. Diffusion constants were determined by mean square displacement analysis within a time window of 320 ms (10 frames).
Immobile molecules were classified by their appearance within a radius described by the localization precision and molecular observation probability.
Native mass spectrometry
In advance of native MS measurements, a small amount (<1%) of covalent dimers of the α3 domain, which had formed as side product during refolding, were removed by size exclusion chromatography on Superdex 75 10/300 GL (Cytiva). Amicon Ultra 0.5 ml centrifugal filter units (molecular mass cut-off 3 kDa; Merck Millipore) were used at 14,000 g and 4°C to exchange purified protein samples to 150 mM ammonium acetate (99.99%; Sigma-Aldrich), pH 7.2. The final concentration of the α3 domain monomer was 5 µM, 10 μM or 20 µM. Native MS analysis was implemented on a Q-Tof II mass spectrometer in positive electrospray ionization mode. The instrument was modified to enable high mass experiments [Waters and MS Vision, (van den Heuvel et al., 2006)]. Sample ions were introduced into the vacuum using homemade capillaries via a nano-electrospray ionization source in positive ion mode (source pressure: 10 mbar). Borosilicate glass tubes (inner diameter: 0.68 mm, outer diameter: 1.2 mm; World Precision Instruments) were pulled into closed capillaries in a two-step program using a squared box filament (2.5 mm×2.5 mm) within a micropipette puller (P-1000, Sutter Instruments). The capillaries were then gold-coated using a sputter coater (5.0×10−2 mbar, 30.0 mA, 100 s, 3 runs to vacuum limit 3.0×10−2 mbar argon, distance of plate holder: 5 cm; CCU-010, safematic). Capillaries were opened directly on the sample cone of the mass spectrometer. In regular MS mode, spectra were recorded at a capillary voltage of 1.45 kV and a cone voltage of 100 V to 150 V. Protein species with quaternary structure were assigned by MS/MS analysis. These experiments were carried out using argon as collision gas (1.2×10−2 mbar). The acceleration voltage ranged from 10 V to 100 V. Comparability of results was ensured as MS quadrupole profiles and pusher settings were kept constant in all measurements. A spectrum of cesium iodide (25 g/l) was recorded on the same day of the particular measurement to calibrate the data.
All spectra were evaluated regarding experimental mass (MassLynx V4.1, Waters) and area under the curve (AUC; UniDec; Marty et al., 2015) of the detected mass species. The values of the shown averaged masses and AUC of the different species as well as the corresponding standard deviation result from at least three independent measurements. Exact experimental masses are presented in Fig. S4A.
Molecular model of the Kb heavy chain dimer
The crystal structure of Kb in complex with a chicken ovalbumin epitope (PDB 3P9M) served as template to generate a model of the heavy chain dimer. Using the Pymol program (DeLano, 2002), it was possible to superimpose a copy of the 3P9M heavy chain on the β2m subunit of a second 3P9M structure (FHC dimer). The superposition involved only the matching of the α3 domain backbone of the Kb molecule (residues 181–277) onto the β2m subunit resulting in a small root-mean-square deviation (RMSD) of 1.25 Å. Very little steric overlap of the superimposed α3 domain with the second heavy chain was detected. This minimal overlap of the α1-α2 domain of the superimposed heavy chain was removed by adjusting backbone dihedral angles of residues 179–181 in the linker between the α1-α2 domain and the α3 domain. The initial structural model was energy-minimized to remove any residual steric overlap and was prepared for MD simulations using the Amber18 package (Case et al., 2018).
For comparison of FHC homodimers and Kb HC–β2m heterodimers, MD simulations were performed starting from the coordinates of the Kb structure in PDB 3P9M. Proteins were solvated in octahedral boxes including explicit Na+ and Cl– ions (0.15 M) and explicit TIP3P water molecules keeping a minimum distance of 10 Å between protein atoms and box boundaries (Jorgensen et al., 1983). The parm14SB force field was used for the proteins and peptides (Maier et al., 2015). The simulation systems were again energy minimized (5000 steps) after solvation followed by heating up to 300 K in steps of 100 K with position restraints on all heavy atoms of the proteins. Subsequently, positional restraints were gradually removed from an initial 25 kcal mol−1·Å−2 to 0.5 kcal mol−1·Å−2 within 0.5 ns followed by a 1 ns unrestrained equilibration at 300 K. All production simulations were performed at a temperature of 300 K and a pressure of 1 bar. The hydrogen mass repartition option of Amber was used to allow a time step of 4 fs (Hopkins et al., 2015). Unrestrained production simulations for up to 400 ns were performed. The interface packing was analyzed by calculation of the buried surface area using the Shrake method (Shrake and Rupley, 1973); analysis of root-mean-square deviations (RMSD) was performed using the cpptraj module of the Amber18 package.
High-speed atomic force measurements for antibody density estimation
We employed high-speed atomic force microscopy (HS-AFM), capable of resolving individual antibodies on surfaces (Preiner et al., 2014; Strasser et al., 2020) to estimate the antibody surface density in our pattern elements. Antibody micropatterns were prepared as described above. Antibody solution (10 µg/ml) was finally incubated for 15–1200 s, followed by washing and imaging in PBS. HS-AFM (RIBM, Japan) was operated in tapping mode at room temperature with free amplitudes of 1.5–2.5 nm and an amplitude set point of larger than 90%. Silicon nitride cantilevers (USC-F1.2-k0.15, Nanoworld AG, Neuchâtel, Switzerland) with nominal spring constants of 0.1–0.15 N/m, resonance frequencies of ∼500 kHz, and a quality factor of ∼2 in liquids were used.
Statistics and statistical analysis
For Fig. 3D–H, each data point represents the analysis from one cell, with ≥14 cells measured per experiment and many trajectories analyzed per cell (3D, 238–2513; 3E, 238–2513; 3F, 314–2732; 3G, 41–354 tracked immobile particles and 362–3141 mobile particles; 3H, 2–301 co-trajectories). Statistical significances were calculated by two-sample Kolmogorov–Smirnov test, Kruskal–Wallis test with multiple comparisons, unpaired t-test and two-way ANOVA with multiple comparisons as indicated in the figure legends, using version Prism 8.4.0 for MacOS (GraphPad, San Diego, USA). Box plots were used for visualization and indicate the data distribution of second and third quartile (box), median (line), mean (square) and complete data range (whiskers).
We thank the donors of reagents as mentioned in the Materials and Methods, Venkat Raman Ramnarayan for comments on the manuscript, Christian P. Richter for support with SMT/SMCT evaluation, Ankur Saikia and Christian Guenther for performing protein chromatography, the iBiOs staff for technical support with single-molecule microscopy, the SPC facility at EMBL Hamburg for technical support, and Uschi Wellbrock for excellent technical assistance. The Heinrich-Pette-Institute, Leibniz Institute for Experimental Virology is supported by the Free and Hanseatic City Hamburg and the German Federal Ministry of Health.
Conceptualization: S.S.; Investigation: C.D., S.L., Z.H., J.-D.K., A.R.H., A.K., J. Preiner, M.Z., P.L.; Writing - original draft: C.D., S.L., J.-D.K., N.L.; Writing - review & editing: M.G., C.U., M.Z., J. Piehler, P.L., S.S.; Supervision: J.W., M.G., C.U., J. Piehler, S.S.; Funding acquisition: J.W., M.G., C.U., M.Z., J. Piehler, P.L., S.S.
This work was supported by Deutsche Forschungsgemeinschaft (DFG, SP583/7-2 and SP583/18-1), Bundesministerium für Bildung und Forschung (BMBF, 031A153A); Tönjes Vagt Foundation (XXXII), iNEXT-Discovery (11911), Jacobs University (all to S.Sp.); DFG (SFB 944, projects P8 and Z, Facility iBiOs, PI 405/14-1) to J. Piehler, P.L. and J.W. acknowledge funding from the province of Upper Austria as a part of the FH Upper Austria Center of Excellence for Technological Innovation in Medicine (TIMed Center), the Austrian Science Fund (FWF, project I4972–B) and the Christian Doppler Forschungsgesellschaft (Josef Ressel Center for Phytogenic Drug Research). C.U. acknowledges funding from the Leibniz Association through grant SAW-2014-HPI-4.
The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE (Perez-Riverol et al., 2022) partner repository with the dataset identifier PXD033485.
Peer review history
The peer review history is available online at https://journals.biologists.com/jcs/article-lookup/doi/10.1242/jcs.259498.
The authors declare no competing or financial interests.