Recombinant production, purification, crystallization, and structure analysis of human transforming growth factor β2 in a new conformation

Transforming growth factor β is a disulfide-linked dimeric cytokine that occurs in three highly related isoforms (TGFβ1–TGFβ3) engaged in signaling functions through binding of cognate TGFβ receptors. To regulate this pathway, the cytokines are biosynthesized as inactive pro-TGFβs with an N-terminal latency-associated protein preceding the mature moieties. Due to their pleiotropic implications in physiology and pathology, TGFβs are privileged objects of in vitro studies. However, such studies have long been limited by the lack of efficient human recombinant expression systems of native, glycosylated, and homogenous proteins. Here, we developed pro-TGFβ2 production systems based on human Expi293F cells, which yielded >2 mg of pure histidine- or Strep-tagged protein per liter of cell culture. We assayed this material biophysically and in crystallization assays and obtained a different crystal form of mature TGFβ2, which adopted a conformation deviating from previous structures, with a distinct dimeric conformation that would require significant rearrangement for binding of TGFβ receptors. This new conformation may be reversibly adopted by a certain fraction of the mature TGβ2 population and represent a hitherto undescribed additional level of activity regulation of the mature growth factor once the latency-associated protein has been separated.

complex. These complexes are disulfide-linked to a third protein, one of three latent TGFβ binding proteins, in the large latent complexes [21][22][23] . After secretion, these complexes are targeted to fibrillin-rich microfibrils as inactive species that are covalently bound by tissue transglutaminase to the extracellular matrix for storage. Finally, the large latent complexes are transformed into biologically active GFs by thrombospondin 1, reactive oxygen species, integrins, and/or peptidases such as furin and other related pro-protein convertases. Proteolytic cleavage by these enzymes severs the linker between GF and LAP, but the two proteins remain noncovalently associated until physically separated for function 2,10,13,24,25 . In this way, most TGFβ in the body is latent and sequestered within the extracellular matrix, although it is also found on the surface of immune cells or in granules of platelets and mast cells 2 . Thus, localization and compartmentalization ensure tight spatial and temporal regulation of the GFs 22 .
Although TGFβ GFs are very similar (70-82% sequence identity in humans) and have partially overlapping functions, they also have distinct roles. This is reflected in their differential expression during embryogenesis and specific roles in renal fibrogenesis and regulation of airway inflammation and remodeling 26 . In particular, human TGFβ2 is biosynthesized as a latent 394-residue precursor (pro-TGFβ2) arranged as a glycosylated homodimer of 91 kDa containing three glycan chains attached to each LAP. Peptidolytic activation at bond R 302 -A 303 (see UniProt [UP] entry P61812 for residue numbers of human TGFβ2 in superscripts 27,28 ) produces the GF, which spans 112 residues and is arranged as a disulfide-linked 25-kDa homodimer noncovalently associated with the respective LAP moieties. Given the importance of this cytokine and its potential in therapeutic applications 29,30 , here we developed a new high-yield human expression system for tagged pro-TGFβ2. We report the X-ray crystal structure of its GF, which was obtained in a different crystallographic space group and deviates from current structures.

Results and Discussion
A new recombinant overexpression system for pro-TGFβ2. After its discovery, TGFβ2 GF was initially purified from human and porcine platelets 31,32 and from bovine bone 33 . However, to obtain large amounts, recombinant expression systems are generally the option of choice as the resulting homogeneity and purity is greater than that of proteins purified from tissues or fluids. Mature human and mouse TGFβ2 GFs were obtained in high yields (~8-10 mg/liter of cell culture) from Escherichia coli systems 34 . However, these proteins lacked glycosylated LAP, which has (glycan-mediated) functions beyond latency maintenance [35][36][37][38] . In addition, these systems produced insoluble protein in inclusion bodies that had to be refolded 34,39 . From a functional perspective, mammalian expression systems are preferred for human cytokines as they provide native environments for folding and disulfide formation in the endoplasmic reticulum, glycosylation in the Golgi, and subsequent quality control in the endoplasmic reticulum. These factors ensure that only correctly folded proteins are secreted. However, the development of such systems has proven difficult for TGFβs 29,34,40,41 . Most initial trials-based on murine CHO and human HEK293 cells-outputted well below ~1 mg of pure protein per liter of cell culture, owing to low expression levels and multiple purification steps [42][43][44][45] . It was not until 2006 that Zou and Sun reported expression of pro-TGFβ2 from stable CHO-lec cell lines with a significantly higher yield 40 . However, this system was based on stable cell lines, which generally require long selection processes for establishment, are prone to contamination, require more time for preparation and for protein expression, and have little flexibility with respect to the constructs tested. In addition, the lack of equivalence of recombinant proteins produced in CHO and HEK cells owing to differences in the glycosylation patterns between hamsters and humans has been widely documented [46][47][48] .
Here, we developed a new homologous production procedure for recombinant full-length human pro-TGFβ2 based on transient transfection of human Expi293F cells, a HEK293 cell-line variant that was adapted to high-density suspension growth and selected for high transfection efficiency and protein expression. Such transient expression systems are flexible, the constructs can be changed and retested fast, they have comparable yields to the stable systems, and they can be much more easily shared among scientists by just sending the plasmid 49 . We obtained ~2.7 and ~2.3 mg of N-terminally octahistidine-tagged and Strep-tagged forms, respectively, per liter of cell culture in only three days ( Fig. 1A-D). The protein was in a disulfide-linked dimeric state, and a major fraction was cleaved at the inter-domain linker due to endogenous proteolytic processing, as previously reported for this and other TGFβs 34,45 . However, the LAP and the GF remained associated in size-exclusion chromatography and native polyacrylamide gel electrophoresis (PAGE). Our efforts to separate them by size-exclusion chromatography in the presence of high salt contents (1 M sodium chloride), chaotropic agents (1 M/4 M urea), detergents (0.05% sodium dodecylsulfate [SDS]), reducing agents (tris [2-carboxyethyl]phosphine or 1,4-dithiothreitol), and low-pH buffers (glycine pH 3.0) failed. We could not isolate them either by ionic exchange chromatography with 20 mM sodium acetate pH 4.0 as buffer and a sodium chloride gradient. This is consistent with previous reports on purified and recombinant pro-TGFβ1 45,50 and with the very high affinity of GF and LAP for each other, with dissociation constant values in the low nanomolar range 51 . In contrast, other publications reported activation of TGFβs in vitro but only under harsh conditions including extreme pH values, high temperature, presence of peptidases and glucosidases, and presence of SDS and urea 34,40,52,53 . Separation and crystallization of mature TGFβ2. During studies of the interactions between induced human α 2 -macroglobulin (hα 2 M), a ~720-kDa homotetrameric pan-peptidase inhibitor purified from blood, and recombinant human pro-TGFβ2 (Marino-Puertas, del Amo-Maestro, Taulés, Gomis-Rüth & Goulas, manuscript in preparation), a mixture of both proteins was set up for crystallization. Diffraction-grade protein crystals appeared after five days with 20% isopropanol, 0.2 M calcium chloride, and 0.1 M sodium acetate pH 4.6 as reservoir solution (Fig. 1E). Reducing SDS-PAGE (Fig. 1F) and peptide mass fingerprinting of carefully washed and dissolved crystals indicated that the crystallized species was the GF. The crystals belonged to a hitherto undescribed, tightly-packed tetragonal space group, diffracted to 2.0 Å resolution, and contained half a GF disulfide-linked dimer per asymmetric unit. Diffraction data processing statistics are provided in Table 1. These www.nature.com/scientificreports www.nature.com/scientificreports/ crystals also appeared when pro-TGFβ2 alone was subjected to similar crystallization conditions. In this case, the crystals diffracted to lower resolution, i.e. hα 2 M apparently played a favorable role as an additive for crystallization. To pursue this further, we collected the supernatant from crystallization drops that had given rise to crystals, purified it by size-exclusion chromatography, and assayed a fraction that migrated according to the mass of hα 2 M by gelatin zymography. We detected gelatinolytic activity (Fig. 1G), which points to a contaminant present in the purified hα 2 M sample. Accordingly, we conclude that the low pH of the crystallization assay, together with the crystallization process, achieved the separation of the LAP and GF moieties that we could not obtain by chromatography (see above). This process was probably facilitated by a peptidolytic contaminant present in the purified hα 2 M sample, as peptidases trapped within the induced tetrameric hα 2 M cage are known to still possess activity [54][55][56] . www.nature.com/scientificreports www.nature.com/scientificreports/ Structure of mature TGFβ2. The crystal structure was solved by maximum likelihood-scored molecular replacement, which gave a unique solution for one GF per asymmetric unit at final Eulerian angles and fractional cell coordinates (α, β, γ, x, y, and z) 348.4, 21.1, 119.9, 0.127, 0.696, and −0.899. The initial values for the rotation/translation function Z-scores were 6.0/14.0 and the final log-likelihood gain was 132. Taken together, these values indicated that P4 1 2 1 2 was the correct space group, and the final model after model rebuilding and refinement contained residues A 303 -S 414 and 23 solvent molecules. Segments P 351 -W 354 and N 371 -S 377 were flexible but clearly resolved in the final Fourier map for their main chains. Table 1 provides refinement and model validation statistics.
Human TGFβ2 GF is an elongated α/β-fold molecule consisting of an N-terminal helix α1 disemboguing into a fourfold antiparallel sheet of simple up-and-down connectivity ( Fig. 2A). Each strand is subdivided into two, β1 + β2, β3 + β4, β5 + β6, and β7 + β8, and the sheet is slightly curled backwards with respect to helix α1. The most exposed segment of the moiety is the tip of hairpin β6β7, and β-ribbon β2β3 is linked by a second α-helix (α2) as part of a loop, whose conformation is mediated by the cis conformation of residue P 338 . Finally, a long loop segment connects strands β4 and β5 and contains helix α3, whose axis is roughly perpendicular to the β-sheet. TGFβ2 is internally crosslinked by four disulfides forming a cysteine knot ( Fig. 2A) and a further intermolecular disulfide by symmetric C 379 residues links two crystallographic symmetry mates to yield the functional dimer. Here, helix α3 of one protomer nestles into the concave face of the β-sheet of the other protomer (Fig. 2B). The respective tips of hairpins β6β7, as well as β-ribbons β2β3 with connecting helices α2, are exposed for functional interactions.
Comparison with previous TGFβ2 structures. The structure of isolated human TGFβ2 GF was solved in 1992 by two groups simultaneously. They obtained the same trigonal crystal form with half a disulfide-linked dimer per asymmetric unit (Protein Data Bank access codes [PDB] 1TFG 57 and 2TGI 58 , see Table 2). In 2014, the complex of the GF with the Fab fragment of a neutralizing antibody was reported in an orthorhombic crystal form (PDB 4KXZ; 26 ). These crystals contained two GF dimers per asymmetric unit, each bound to two Fab moieties. Three more structures of TGFβ2 GF were reported in 2017 59 (Table 2). These corresponded to engineered monomeric forms from human (PDB 5TX4; in complex with TGFR-II ectodomain) and mouse (PDB 5TX2 and 5TX6) that lacked helix α3 and encompassed several point mutations.
Superposition of the GF structures of PDB 2TGI, 1TFG, and 4KXZ (Fig. 2C) revealed very similar conformations of bound and unbound protomers and dimers, despite distinct chemical and crystallographic environments, which can generally contribute to different conformations owing to crystallographic artifacts 60  www.nature.com/scientificreports www.nature.com/scientificreports/ www.nature.com/scientificreports www.nature.com/scientificreports/ Cα atoms, respectively. Only minor variations occurred at the tips of hairpins β6β7 and at helix α2 (Fig. 2C). Interestingly, also the engineered monomeric variants kept the overall structure of the native TGFβ2 protomer, with the exception of a loop that replaced helix α3 59 . This close similarity was also found with the unbound human ortholog TGFβ3 (PDB 1TGJ 61 ).
Inversely, the present GF structure (PDB 6I9J; hereafter the current structure) showed deviations from the reference structure, which are reflected by an rmsd of 1.90 Å for 107 common Cα atoms. The central β-sheets nicely coincide and only minor differences are found within the deviating regions of the above GF structures. However, a major rearrangement is found within segment Y 352 -C 380 , which encompasses helix α3 plus the flanking linkers to strands β4 and β5 (Fig. 2D,E). Owing to an outward movement of segment W 354 -D 357 , α3 is translated by maximally ~5.5 Å (at I 370 ). This, in turn, causes downstream segment E 373 -C 379 to be folded inward, with a maximal displacement of ~6.7 Å (at A 374 ). This displacement cascades into a slight shift (~2 Å) of C 379 and, thus, the intermolecular disulfide. These differences within protomers further result in variable arrangement of the dimers (Fig. 2F). The second protomer is rotated by ~10°, so that K 396 at the tip of hairpin β6β7 is displaced by ~9 Å. In addition, the central β-sheets overlap but are laterally shifted with respect to each other.
Finally, inspection of the respective crystal packings of the current and reference structures (Fig. 3A,B) reveals that while the latter is loosely packed in a hexagonal honeycomb-like arrangement, with large void channels of > 40 Å diameter and high solvent content (61%), the former is tightly packed (solvent content of 43%). In particular, the tip of hairpin β6β7 is not engaged in significant crystal contacts in either crystal form, which supports that both conformations are authentic and do not result from crystallization artifacts.

Potential implications of a new dimeric arrangement. Structural information on TGFR binding is
available for human TGFβ1 and TGFβ3, whose wild-type forms were crystallized in complexes with the ectodomains of TGFR-I and -II (PDB 3KFD 62 ; PDB 1KTZ 63 ; and PDB 2PJY 12 ). In the binary complex of TGFβ3 with TGFR-II (PDB 1KTZ), the cytokine was observed in a unique dimeric arrangement, termed "open", in which the respective helices α3 were completely disordered. This led the second protomer to be rotated by ~180° and the tip of hairpin β6β7 to be displaced by ~36 Å when compared with the reference structure 63 . The functional significance of this structure, which differs from any other mature TGFβ dimer, is not clear.
By contrast, the other two complex structures, which contained both receptors, showed that the TGFβ1 and TGFβ3 isoforms dimerize very similarly to the unbound reference structure when bound to receptors (rmsd of 1.43 Å and 1.61 Å for 222 and 219 common Cα atoms for PDB 3KFD and PDB 2PJY, respectively). In contrast, they differed from the current structure (Fig. 2G), which shows a rmsd of 2.94 Å for 216 common Cα atoms when compared with the reference. In addition, both cytokine isoforms-and by similarity probably also TGFβ2-bound their receptors in an equivalent manner. TGFR-II was contacted by the tip of hairpin β6β7 and the linker between α2 and β3 of one protomer, and TGFR-I was liganded by the interface between protomers created by strands β5-β8 from one protomer and α1 plus the region around α3 of the other protomer. In addition, the TGFRs further contacted each other through the N-terminal extension of TGFR-II (Fig. 2H). Notably, the binding mode of TGFR-II was shared with the engineered monomeric variant of TGFβ2 (PDB 5TX4 59 ). In contrast, superposition of the cytokine dimers of the current structure and the triple complexes evinced that among other structural elements hairpin β6β7 is displaced away from interacting segment S 52 -E 55 of TGFR-II (see PDB 2PJY) by ~2.7 Å. This hairpin rearrangement also impairs interaction of the β5β8 ribbon with TGFR-I segment I 54 -F 60 . Thus, TGFRs could not bind the current structure in the same way they bind TGFβ1 and TGFβ3 without rearrangement.  www.nature.com/scientificreports www.nature.com/scientificreports/ Conclusion. We developed a recombinant human overexpression system based on Expi293F cells grown in suspension, which produces high yields of well-folded human N-terminally histidine-or Strep-tagged pro-TGFβ2 with native post-translational modifications.
We further crystallized mature TGFβ2 in a different crystallographic space group, which deviates mainly in the region around helix α3 from current functional TGFβ1, TGFβ2 and TGFβ3 structures. Importantly, this region is the segment that shows the highest variability in sequence among TGFβs 26 . Although crystal packing artifacts cannot be completely ruled out, the fact that several different crystal forms of the three TGFβ GFs had produced highly similar structures to date suggests that the new conformation may be authentic and have functional implications.
The divergent conformation of the region around helix α3 led to differences in the dimer, which could not bind cognate TGFR-I and -II in the same way as TGFβ1 and TGFβ3 do. Moreover, as revealed by pro-TGFβ1 structures from pig (PDB 5VQF 64 ) and human (PDB 5VQP 65 and PDB 6GFF 66 ), this region differed substantially between latent and mature moieties owing to the presence of the respective LAPs. Hence, the GF protomers associated differently within the respective dimers. Large differences were also found between the mature form and the pro-form of the more distantly related TGFβ family member activin A (PDB 2ARV 67 and PDB 5HLY 68 ). In contrast, bone morphogenetic factor 9 showed deviations in hairpin β6β7 but kept the dimer structure (PDB www.nature.com/scientificreports www.nature.com/scientificreports/ 5I05 69 and PDB 4YCG 70 ). Overall, we conclude that our mature structure may represent an inactive variant or be one of an ensemble of conformational states, which may still undergo an induced fit or selection fit mechanism to form a functional ternary ligand-receptor complex.

Materials and Methods
Protein production and purification. The coding sequence of human pro-TGFβ2 without its signal peptide, i.e. spanning the LAP (L 21 -R 302 ) and the GF (A 303 -S 414 ), was inserted in consecutive PCR steps into the Gateway pCMV-SPORT6 vector (ThermoFisher) in frame with the Kozak sequence. The signal peptide from the V-J2-C region of a mouse Ig κ-chain was used instead of the native leader sequence, as this was reported to be more efficient for expression 40 . This vector attached an N-terminal octahistidine-tag to the protein of interest. For the five PCR steps, oligonucleotides 5′-TCACCACCACCATCATCTCAGCCTGTCTACCTGCAGCA-3′; 5′-GGTTCCACTGGTGACCACCACCATCACCACCACCATC-3′; 5′-GGGTACTGCTGCTCTGGG TTCCAGGTTCCACTGGTGAC-3′; 5′-GACAGACACACTCCTGCTATGGGTACTGCTGCTC-3′; and 5′-CAATCCCGGGGCCACCATGGAGACAGACACACTCC-3′ were used as forward primers, respectively, and 5′-CAATCTCGAGCTAGCTGCATTTGCAAGACTTTAC-3′ was employed as the reverse primer. The intermediate PCR products were purified with the EZNA Cycle Pure Kit (Omega Bio-tek, USA) prior to the next step. The final PCR product was digested twice with SmaI and XhoI restriction enzymes (1 h at 37 °C and o/n at 37 °C), with an intercalated PCR product purification step. This product was ligated into pre-digested pCMV-SPORT6 by adding 2 μL of T4 DNA Ligase (ThermoFisher) per 20 μL of reaction (o/n at r.t.). The resulting plasmid (pS6-TG-FB2-H8) was verified for its sequence (GATC Biotech) and transformed into competent Escherichia coli DH5α cells for vector storage and production.
Prior to transfection into mammalian cells, pS6-TGFB2-H8 was produced in E. coli DH5α, purified with the GeneJET Plasmid Maxiprep Kit (ThermoFisher) according to the manufacturer's instructions, and stored in Milli-Q water at 1 mg/mL. Expi293F cells (ThermoFisher), which had been kept in suspension in FreeStyle F17 Expression Medium (Gibco) plus 0.2% Pluronic F-68 at 150 rpm in a humidified atmosphere with 8% CO 2 in air at 37 °C, were transfected at a density of 1 × 10 6 cells/mL with a mixture of 1 mg of purified pS6-TGFB2-H8 and 3 mg of linear 25-kDa polyethylenimine (Polysciences) in 20 mL of Opti-MEM Medium per liter. The mixture of the reagents was incubated for 15-20 min at room temperature and then added dropwise to the cells in 1 L-disposable Erlenmeyer flasks (Fisherbrand) after proper mixing. Cells were harvested for 72 h and then centrifuged at 2,800 × g in the JLA9.1000 rotor of an Avanti J-20XP centrifuge (Beckman Coulter) for 20 min, and the supernatant was subjected to single-step Ni-NTA affinity chromatography (washing buffer: 50 mM Tris·HCl pH 8.0, 250 mM sodium chloride, 20 mM imidazole; elution buffer: 50 mM Tris·HCl pH 8.0, 250 mM sodium chloride, 300 mM imidazole). Elutions were pooled, concentrated and subjected to size-exclusion chromatography in a Superdex 75 10/300 column (GE Healthcare) attached to an ÄKTA Purifier system (GE Healthcare) at r.t. with 20 mM Tris·HCl pH 8.0, 150 mM sodium chloride as buffer. Purified protein was routinely concentrated with Vivaspin centrifugal devices (Sartorius) with a 10-kDa cutoff, and concentrations were determined by A 280 in a NanoDrop Microvolume spectrophotometer (ThermoFisher). Protein identity was confirmed by peptide mass fingerprinting and purity was assessed by 10% SDS-PAGE using Tris-Glycine buffer and Coomassie Brilliant Blue or silver staining, as well as by Western-blot analysis with a histidine antibody horse radish peroxidase conjugate (His-probe Antibody H-3 HRP, Santa Cruz Biotechnology) at 1:5,000 in PBS buffer further 0.1% in Tween 20.
To produce Strep-tagged pro-TGFβ2, plasmid pS6-TGFB2-H8 was modified by PCR to replace the histidine-tag with a twin Strep-tag with an intermediate G/S spacer (sequence WSHPQFEKGGGSGGGSGGSAWSHPQFEK) by using forward primer 5′-GGTGGAGGTTCTGGAGGTGGAAGTGGAGGTAGCGCATGGAGCCATCCACA ATTCGAAAAGCTCAGCTGTCTACCTGC-3′ and reverse primer 5′-CTTTTCGAATTGTGGATGGCTCCAG TCACCAGTGGAACCTGGAACCCAGAGCAG-3′. The resulting PCR product was purified with the EZNA Cycle Pure Kit, phosphorylated with 1 μL of T4 polynucleotide kinase (ThermoFisher) per 20 μL of reaction (o/n at 37 °C), and ligated as above to generate plasmid pS6-TGFB2-STREP. This plasmid was further processed and transfected into Expi293F cells for protein production as above. Cells were harvested and centrifuged as aforementioned, and the supernatant was extensively dialyzed against 100 mM Tris·HCl pH 8.0, 150 mM sodium chloride and purified by single-step affinity chromatography using Strep-Tactin XT Superflow Suspension resin (iba) according to the manufacturer's instructions (washing buffer: 100 mM Tris·HCl pH 8.0, 150 mM sodium chloride; elution buffer: 100 mM Tris·HCl pH 8.0, 150 mM sodium chloride, 50 mM biotin). Final polishing by size-exclusion chromatography, concentration, and purity assessment followed as above, except that Western-blot analysis was performed with a Streptavidin antibody horse radish peroxidase conjugate (Streptavidin-Peroxidase Antibody from Streptomyces avidinii; Sigma-Aldrich) at 1:1000 in PBS, supplemented with 0.1% Tween 20 and 1% BSA.
Crystallization of mature TGFβ2 and diffraction data collection. Octahistidine-tagged pro-TGFβ2, mostly cleaved in the linker between LAP and GF, at ~5 mg/mL in 10 mM Tris·HCl pH 8.0, 150 sodium chloride was incubated o/n at 4 °C with human induced α 2 -macroglobulin (hα 2 M) at ~7 mg/mL in 10 mM Tris·HCl pH 8.0, 150 mM sodium chloride. The hα 2 M was previously purified from blood and reacted with methylamine as reported 71 . The hα 2 M/pro-TGFβ2 reaction mixture was subjected to crystallization assays by the sitting-drop vapor diffusion method at the Automated Crystallography Platform of Barcelona Science Park. Reservoir solutions were mixed by a Tecan robot and crystallization drops of 100 nL were dispensed by a Cartesian Microsys 4000 XL (Genomic Solutions) robot or a Phoenix nanodrop robot (Art Robbins) on 96 × 2-well MRC nanoplates (Innovadyne). Crystallization plates were stored at 20 °C or 4 °C in Bruker steady-temperature crystal farms, and successful conditions were scaled up to the microliter range in 24-well Cryschem crystallization dishes (Hampton Research).
www.nature.com/scientificreports www.nature.com/scientificreports/ Diffraction-grade crystals of ~20 microns maximal dimension were obtained at 20 °C with protein solution and 20% isopropanol, 0.2 M calcium chloride, 0.1 M sodium acetate pH 4.6 as reservoir solution from 1 or 2 μL: 1 μL drops. Carefully washed and pooled crystals revealed the presence of a species of ~12 kDa in SDS-PAGE, which was identified as TGFβ2 by peptide mass fingerprinting. Crystals were cryoprotected by immersion in reservoir solution plus 15% glycerol and flash cryocooled in liquid nitrogen. Diffraction data were collected at 100 K on a Pilatus 6 M pixel detector (Dectris) at beam line XALOC 72 of the ALBA synchrotron in Cerdanyola (Catalonia, Spain). These data were processed with programs XDS 73 and XSCALE 74 , and transformed with XDSCONV to MTZ format for the CCP4 suite of programs 75 . Crystals belonged to space group P4 1/3 2 1 2 based on the systematic extinctions, contained one mature TGFβ2 molecule per asymmetric unit (V M = 2.14; 42.6% solvent contents according to 76 ), and diffracted to 2.0 Å resolution.
Proteolytic assay of crystallization-drop supernatant. Supernatant from crystallization drops that had produced crystals of mature TGFβ2, i.e. which contained methylamine-induced hα 2 M, was pooled and subjected to size-exclusion chromatography in a Superose6 10/300 column (GE Healthcare) at r.t. with 10 mM Tris·HCl pH 8.0, 150 mM sodium chloride as buffer. Fractions corresponding to hα 2 M (~720 kDa) were concentrated with Vivaspin centrifugal devices (Sartorius) with a 30-kDa cutoff and employed for zymography studies. To this aim, 10% SDS-PAGE gels were prepared containing 0.2% (w/v) gelatin. Protein samples were subjected to electrophoresis at 4 °C with equal volume of SDS-PAGE sample buffer without β-mercaptoethanol. Gels were washed twice in 20 mM Tris·HCl pH 7.4, 150 mM sodium chloride, 10 mM calcium chloride, 2.5% (v/v) Triton X-100 for 15 min and then incubated in the same buffer without detergent for 16 h at 37 °C under gentle shaking. Gels were stained with Coomassie Brilliant Blue.
Structure solution and refinement. The structure of mature TGFβ2 was solved by likelihood-scoring molecular replacement with the PHASER 77 program and the coordinates of a GF protomer crystallized in a different space group and unit cell (PDB 2TGI 58 ). Subsequently, an automatic tracing step was performed with ARP/ wARP 78 , which outputted a model that was completed through successive rounds of manual model building with the COOT program 79 and crystallographic refinement with the PHENIX 80 and BUSTER/TNT 81 programs. The latter included TLS refinement.
Miscellaneous. Structure figures were made with the CHIMERA program 82 . Structures were superposed with the SSM program 83 within COOT. A search for structural similarity against the PDB was performed with DALI 84 . The final model of human TGFβ2 GF was validated with the wwPDB Validation Server (https://www. wwpdb.org/validation) 85 and is available at the PDB at https://www.rcsb.org (access code 6I9J).