Cysteine residues in positions 77 and 95 of FGF2 play a role in its unconventional secretion from cells. (A) Representative wide-field and TIRF images of real-time single molecule TIRF recruitment assay conducted on stable CHO K1 cell lines overexpressing either wild-type (WT) or mutant (C77A, C95A, C77/95A) FGF2-GFP in a doxycycline-dependent manner. Beyond cell lines expressing various forms of FGF2-GFP, a GFP-expressing cell line was used to subtract GFP background. Wide-field images show the overall FGF2-GFP (or GFP) expression levels. Single FGF2-GFP (or GFP) particles recruited at the inner plasma membrane leaflet and detected within the TIRF field are highlighted with a pink circle. (B) Quantification of real-time single molecule TIRF recruitment assay conducted on the cell lines shown in panel A. Recruitment efficiency at the inner plasma membrane leaflet of FGF2-GFP wild-type was set to 1. Mean recruitment efficiency values are shown in brackets. Data are shown as mean with standard deviations (n = 4). Statistical analysis was based on a one-way ANOVA test performed in Prism (version 9.4.1), *** P ≤ 0.001. (C) FGF2-GFP recruitment efficiency following GFP background subtraction. For each independent biological replicate, background signal coming from GFP in proximity of the inner plasma membrane leaflet was subtracted from the four variant forms of FGF2-GFP shown in panel A and B (WT, C77A, C95A, C77/95A). Within each replicate, recruitment efficiency of each cell line was normalized to the one of wild-type FGF2-GFP, which was set to 100%. Mean recruitment efficiency values are shown in brackets. Data are shown as mean with standard deviations (n = 4). Statistical analysis was based on a one-way ANOVA test performed in Prism (version 9.4.1), not significant (ns) P > 0.05, ** P ≤ 0.01. (D) Representative Western Blot of cell surface biotinylation assay conducted on stable CHO K1 cell lines overexpressing either wild-type (WT) or mutant (C77A, C95A, C77/95A) FGF2-GFP in a doxycycline-dependent manner. Total cellular proteins and biotinylated surface proteins were analyzed. The analysis was conducted against GFP, to detect the various FGF2-GFP mutant forms, and GAPDH, both as a loading and a cellular integrity control. (E) Quantification of cell surface biotinylation assay conducted on the cell lines shown in panel D. Secretion efficiency of FGF2-GFP wild- type was set to 100%. Mean secretion efficiency values for each cell line are shown in brackets. Data are shown as mean with standard deviations (n = 4). Statistical analysis was based on a one-way ANOVA test performed in Prism (version 9.4.1), not significant (ns) P > 0.05, *** P ≤ 0.001).

Formation of higher FGF2 oligomers on the membrane surface of GUVs depends on C95 (A) Oligomeric size distribution of FGF2-GFP variants. Giant unilamellar vesicles (GUVs) with a plasma membrane-like lipid composition containing 2 mol% PI(4,5)P2 were incubated with variant forms of His-tagged FGF2-Y81pCMF-GFP as indicated. The oligomer size was determined by brightness analysis as described in detail in Materials and Methods. Each dot corresponds to a data-point measured on a single GUV with the number of GUVs analyzed n (WT; n = 68, C77A; n = 42, C95A; n = 31, C77/95A; n = 9). Mean values with standard deviations are shown. One-way ANOVA with Tukey’ post hoc test was performed in Prism (version 9.4.1). Mean values are shown in brackets, not significant (ns) P >0.05, **** P ≤ 0.0001. Data distribution was assumed to be normal, but this was not formally tested. (B) SDS-PAGE analysis of FGF2-Y81pCMF-GFP variant forms indicated. Purified proteins were analyzed for homogeneity using Coomassie staining.

FGF2 dimer formation in cells depends on C95 as revealed by chemical cross-linking Using cellular lysates, FGF2 dimer formation was analyzed by chemical cross-linking. The FGF2 variants (WT, C77A, C95A and C77/95A) were transiently expressed in HeLa S3 cells as constructs connecting the FGF2 open reading frame with GFP via a P2A site, producing stoichiometric amounts of untagged FGF2 and GFP, the latter used to label transfected cells. The corresponding cellular lysates were treated with three different crosslinkers: PMPI [N-p-maleimidophenylisocyanate; bifunctional crosslinker with a spacer length of 8.7 Å targeting sulfhydryl groups at one end (maleimide) and hydroxyl groups at the other end (isocyanate), Fig. 3A and D], BMOE [(bismaleimidoethane; bifunctional maleimide-based crosslinker with a short 8 Å spacer length targeting sulfhydryl groups), Fig.3B and E] or BMH [(bismaleimidohexane; bifunctional maleimide-based crosslinker with a long 13 Å spacer length targeting sulfhydryl groups), Fig.3C and F], respectively. Cross-linking products were analyzed by SDS-PAGE followed by Western blotting using polyclonal anti-FGF2 antibodies. (A, B, C) Representative examples of the Western analyses for each of the three crosslinkers described above. FGF2 monomers (18 kDa) are labeled with “♦ ”, FGF2 dimers (36 kDa) with “♦♦ ” and small amounts of FGF2-P2A-GFP (∼50 kDa) with “◊”. (D, E, F) Quantification of FGF2 dimer to FGF2 monomer ratios. Signal intensities were quantified using a LI-COR Odyssey CLx imaging system. The FGF2 dimer to monomer ratios were determined in four independent experiments with the standard error of the mean shown, not significant (ns) P >0.05, * P ≤ 0.05, **≤ 0.01, *** P ≤ 0.001. Statistical analyses were based on two-tailed, unpaired t test using GraphPad Prism (version 9.4). Data distribution was assumed to be normal, but this was not formally tested.

Membrane pore formation triggered by FGF2 oligomers depends on C95 Carboxyfluorescein was sequestered in large unilamellar liposomes containing a plasma-membrane-like lipid composition including 2 mol% PI(4,5)P2. (A) Liposomes were incubated with the His-tagged FGF2-Y81pCMF (2 µM)-based variant forms indicated (WT, C77A, C95A and C77/95A). Membrane pore formation was analyzed by measuring the release of luminal carboxyfluorescein quantified by fluorescence dequenching as detailed in Materials and Methods. The results shown are representative for three independent experiments. (B) Quality of the various recombinant proteins was analyzed by SDS PAGE and Coomassie staining.

Full membrane translocation of FGF2 across GUV lipid bilayers depends on C95 Reconstitution of FGF2 membrane translocation with purified components. Giant unilamellar vesicles with a plasma membrane-like lipid composition containing 2 mol% PI(4,5)P2 were prepared in the presence or absence of long-chain heparins as described in detail in Materials and Methods. In brief, Rhodamine-PE (Rhod.-PE) was incorporated into the lipid bilayer during GUV preparation as membrane marker. After removal of excess heparin by low-speed centrifugation, GUVs were incubated with His-tagged FGF2-Y81pCMF-GFP (200 nM) variants as indicated and a small fluorescent tracer (Alexa647). Following 180 min of incubation luminal penetration of GUVs by FGF2-Y81pCMF-GFP and small tracer molecules was analyzed by confocal microscopy. (A) Radial intensity profiles of representative examples quantifying GFP fluorescence in the GUV lumen versus the exterior for FGF2-Y81pCMF-GFP wt (sub-panel a), C77A (sub-panel b), C95A (sub-panel c) and C77/95A (sub- panel d). (B) Quantification and statistical analysis of FGF2 membrane translocation and membrane pore formation. Gray bars indicate the percentage of GUVs with membrane pores with a ratio of Alexa647 tracer fluorescence in the lumen versus the exterior of ≥0.6. Green bars indicate the percentage of GUVs where membrane translocation of GFP-tagged proteins had occurred with a ratio of GFP fluorescence in the lumen versus the exterior of ≥1.6 being used as a threshold value. Each dot represents an independent experiment each of which involved the analysis of 20–120 GUVs per experimental condition. Mean values with standard deviations are shown. Statistical analyses are based on two-tailed, unpaired t-test performed in Prism (version 9.4.1), not significant (ns) P > 0.05, * P ≤ 0.05). Data distribution was assumed to be normal, but this was not formally tested. For details see Materials and Methods. (C) Representative confocal images of plasma membrane-like GUVs containing PI(4,5)P2 and long-chain heparins in the lumen after 180 min incubation with His-tagged FGF2-Y81pCMF-GFP (200 nM) variants as indicated and a small fluorescent tracer (Alexa647; scale bar = 10 µm).

FGF2 membrane translocation across GUV lipid bilayers is abrogated when PI(4,5)P2 is substituted by a Ni-NTA lipid used to recruit His-tagged FGF2 fusion proteins Giant unilamellar vesicles with a plasma membrane-like lipid composition containing 2 mol% Ni-NTA-lipid anchor were prepared in the presence or absence of long-chain heparins. Luminal penetration of GUVs by FGF2- Y81pCMF-GFP was analyzed by confocal microscopy as described in the legend to Fig.5. (A) Quantification and statistical analysis of FGF2 membrane translocation and membrane pore formation. Gray bars indicate the percentage of GUVs with membrane pores with a ratio of Alexa647 tracer fluorescence in the lumen versus the exterior of ≥0.6. Green bars indicate the percentage of GUVs where membrane translocation of GFP-tagged proteins had occurred with a ratio of GFP fluorescence in the lumen versus the exterior of ≥1.6 being used as a threshold value. Each dot represents an independent experiment each of which involved the analysis of 20–120 GUVs per experimental condition. Mean values with standard deviations are shown. Statistical analyses are based on two-tailed, unpaired t-test performed in Prism (version 9.4.1), not significant (ns) P > 0.05. Data distribution was assumed to be normal, but this was not formally tested. (B) Representative confocal images of plasma membrane-like GUVs containing Ni-NTA-lipid-anchor and long-chain heparins in the lumen after 180 min incubation with His-tagged FGF2-Y81pCMF-GFP (200 nM) variants as indicated and a small fluorescent tracer (Alexa647; scale bar = 10 µm).

C77 is a component of the protein-protein interaction surface between FGF2 and the α1 subunit of the Na,K-ATPase Kinetic analysis of the direct interaction of FGF2 with α1-subCD3 (Legrand et al., 2020). (A) FGF2 binds to α1 with K54, K60 and C77 being part of the protein-protein interaction interface [Figure adapted from (Legrand et al., 2020)]. (B) FGF2 directly binds to α1 in a dose-dependent manner. Biolayer interferometry (BLI) allows temporal resolution of association and dissociation. Biotinylated Hisα1-subCD3-WT protein was immobilised on Streptavidin sensors followed by incubation with His-tagged FGF2 wild type protein (HisFGF2-WT) at concentrations indicated. The data shown is representative of three independent experiments. Data were analyzed with Data Analysis HT 12.0 software (Sartorius) using a 1:1 binding model. See Material and Methods for details. Mean with standard deviations of KD, ka, and kd values (n = 3) are given. (C, D, E) Comparison of FGF2 variants. (C) BLI measurements were conducted using immobilized Hisα1-subCD3 with FGF2 variants (1000 nM concentration) as indicated. The data shown is representative for four independent experiments. (D) The quality of HisFGF2 proteins was analyzed by SDS-PAGE and Coomassie staining. 3 µg of each variant were loaded as indicated. (E) Phase-shift at time point 600 s. Mean values with standard deviations of four independent experiments are shown. One-way ANOVA with Tukey’ post hoc test was performed in Prism (version 9.4.1). Mean values are shown in brackets, not significant (ns) P > 0.5, **** P ≤ 0.0001. Data distribution was assumed to be normal, but this was not formally tested.

Characterization of C95 disulfide-bridged FGF2 via Molecular Dynamics Simulations. (A) The AlphaFold2 model dimeric interface’s stability (subpanel a) was tested by conducting 1-microsecond- long MD simulations in water, which revealed the interface’s high flexibility (subpanel b). (B) Unbiased all-atom MD simulations were used to sample the FGF2 Dimer-membrane interaction pathway, mediated via the experimentally known PI(4,5)P2 binding pocket (K127, R128, K133). The free energy profile of FGF2 Dimer- membrane interaction was determined from biased (umbrella sampling) MD simulations and plotted against the center of the mass distance of FGF2 dimer from phosphate atoms of the interacting membrane surface. Subpanels (a-c) show the interaction pathway’s initial, intermediate, and final states, while subpanel (d) shows the free energy profile.

Cryo-electron tomography visualization of FGF2 dimer in proteo-liposomes. (A) Example slices of cryo-electron tomograms (sub-panels a-c; 2.3 nm thickness) showing His-FGF2-Y81pCMF- Halo bound to PI(4,5)P2-containing liposomes acquired at nominal defocus -3 µm. Magnified views from FGF2- Halo monomers, dimers and higher oligomers (sub-panels e-f). (B) Subtomogram average of the V-shaped FGF2- Halo dimer interacting with the membrane of PI(4,5)P2-containing liposomes (sub-panels a-b). Subtomogram average of “V-shaped” FGF2 dimers were manually picked using a dipole model in Dynamo (number of particles = 186). Top and side views are shown in subpanel a and b); c-d) 3D map with manually fitted crystal structures of two Halo domains (PDB:4KAJ) and two FGF2 (PDB:1BFF) monomers; e-f) Atom-scale molecular dynamics simulation model of V-shape C95 disulfide bridged FGF2-Halo dimer stable over 500 nanoseconds.

Cryo-electron tomography visualization of FGF2 dimer in proteo-liposomes. From a top and side view perspective, a comparison between the AlphaFold2 Multimer v3 model (panels A-B) and the one used for molecular dynamics simulations (panels C-D). The AlphaFold model accurately predicted the orientation of FGF2 dimer, with the PI(4,5)P2 binding residues correctly positioned for a membrane-bound state (blue residues). However, the two cysteines 95 (red residues), although located at the interface, were observed to be distant. To improve the model, we replaced the FGF2 dimer with the 95 disulfides bridged dimer interface characterized with MD simulations (Fig. 8). Furthermore, we used the “Fit in Map” command provided by the ChimeraX software to locally optimize the fit of one of the two Halo domains’ atomic coordinates into the density map.