Dimerization-dependent Folding Underlies Assembly Control of the Clonotypic αβT Cell Receptor Chains*

Background: Assembly of heteromeric membrane protein complexes is monitored prior to their transport to the cell surface. Results: Each of the αβTCR clonotypic chains comprises one domain that folds upon assembly. Conclusion: Proper αβTCR biogenesis is scrutinized by assembly-coupled folding steps. Significance: This study shows how the assembly of heteromeric membrane proteins can be monitored and indicates limitations of the possible T cell repertoire. In eukaryotic cells, secretory pathway proteins must pass stringent quality control checkpoints before exiting the endoplasmic reticulum (ER). Acquisition of native structure is generally considered to be the most important prerequisite for ER exit. However, structurally detailed protein folding studies in the ER are few. Furthermore, aberrant ER quality control decisions are associated with a large and increasing number of human diseases, highlighting the need for more detailed studies on the molecular determinants that result in proteins being either secreted or retained. Here we used the clonotypic αβ chains of the T cell receptor (TCR) as a model to analyze lumenal determinants of ER quality control with a particular emphasis on how proper assembly of oligomeric proteins can be monitored in the ER. A combination of in vitro and in vivo approaches allowed us to provide a detailed model for αβTCR assembly control in the cell. We found that folding of the TCR α chain constant domain Cα is dependent on αβ heterodimerization. Furthermore, our data show that some variable regions associated with either chain can remain incompletely folded until chain pairing occurs. Together, these data argue for template-assisted folding at more than one point in the TCR α/β assembly process, which allows specific recognition of unassembled clonotypic chains by the ER chaperone machinery and, therefore, reliable quality control of this important immune receptor. Additionally, it highlights an unreported possible limitation in the α and β chain combinations that comprise the T cell repertoire.

In eukaryotic cells, proteins of the secretory pathway are produced in the endoplasmic reticulum (ER). 5 In higher eukaryotes, these proteins usually enter the ER co-translationally as unfolded polypeptide chains and must acquire their native structure before traversing further along the secretory pathway. This often includes formation of disulfide bonds and glycosylation as well as subunit assembly, processes that are aided and scrutinized by a comprehensive ER-resident protein folding and quality control (QC) machinery (1,2). To avoid detrimental effects to the organism, the cell has to distinguish between native proteins, partially folded proteins, and ones that are unable to fold. However, in some cases, this discrimination fails, allowing either the ER exit of potentially harmful molecules or the degradation of proteins that could perform their biological functions (3,4). Therefore, which features of partially folded or incompletely assembled proteins the ERQC machinery recognizes is a key question. Shortcomings in our understanding are partially due to the limited resolution with which protein folding can be monitored in the cell (5,6).
To provide molecular insights into processes of ERQC, we focused on the ␣ and ␤ chains of the T cell receptor (TCR). The ␣␤TCR is a longstanding model for the quality control of oligomeric membrane proteins, and a great deal is understood about cellular checkpoints in its assembly (7)(8)(9)(10)(11)(12)(13)(14)(15). It is composed of eight polypeptide chains, the clonotypic ␣ and ␤ chains and the invariant co-receptor chains (CD3␥, ␦, ⑀, and ; Fig. 1A), all of which must assemble correctly for the ␣␤TCR to be transported to the cell surface (16 -18). First, the ␣ and ␤ chains pair with their designate CD3 co-receptor subunits, giving rise to ␣-CD3␦⑀ and ␤-CD3␥⑀ trimers, respectively (19). In each case, complementary basic/acidic residues in the transmembrane segments of these chains guide assembly of the trimers and, as has been dissected in detail for the ␣ chains, provide a means to identify chains that do not assemble with CD3 components (7,8,10,12,20). Retention motifs in the CD3 subunits appear to play an important role in the final step of ␣␤TCR assembly: interaction of ␣␤-CD3␥␦⑀ 2 hexamers with 2 homodimers to form the complete ␣␤TCR (21,22). However, it has remained unclear how assembly of the two ␣-CD3␦⑀ and ␤-CD3␥⑀ trimers is monitored and what role the various lumenal domains of the clonotypic ␣␤ chains play in ERQC. Combining the individual strengths of in vitro and in vivo experimental approaches, we set out to study the molecular events occurring upon heterodimerization of the TCR ␣/␤ chains with a view to deriving general insights into how the ERQC system monitors protein assembly.

Experimental Procedures
Protein Production and Purification-For in vitro studies, individual constructs were amplified from synthetic TCR genes optimized for Escherichia coli expression (Geneart, Regensburg, Germany) and cloned into the pET28a expression vector (Novagen, Gibbstown, NJ). Expression of the various constructs was performed overnight at 37°C, and resulted in inclusion bodies. Inclusion bodies were solubilized in 100 mM Tris/ HCl (pH 8.0), 10 mM EDTA, 10 mM ␤-mercaptoethanol, and 8 M urea. Solubilized inclusion bodies were centrifuged (20,000 ϫ g, 30 min, 10°C). The supernatant was applied to a Q-Sepharose column equilibrated in 100 mM Tris/HCl (pH 8.0), 10 mM EDTA, and 5 M urea. Subsequently, the flowthrough was applied to an SP-Sepharose column equilibrated in the same buffer. None of the proteins of interest bound to either column. Accordingly, the flowthrough of the SP-Sepharose column was collected and dialyzed overnight against 100 mM Tris/HCl (pH 8.0), 3 M guanidinium chloride, 5 mM EDTA, and 1 mM DTT. Subsequently, the proteins were applied to a Superdex 200pg (26/60) gel filtration column (GE Healthcare) equilibrated in the same buffer. Proteins were then diluted to 0.1 mg/ml in 250 mM Tris/HCl (pH 8.0), 250 mM L-arginine, 5 mM EDTA, 0.25 mM GSSG, and 0.25 mM GSH and refolded overnight at 4°C via dialysis against the same buffer. The refolded proteins were concentrated and applied to a Superdex 200pg (26/60) gel filtration column equilibrated in PBS and, if necessary, additionally to a Superdex200 10/300 GL HPLC column (GE Healthcare) equilibrated in PBS. All plasmids were sequenced, and protein identities were verified by mass spectrometry. For NMR samples, proteins were expressed in M9 minimal medium containing 15 N-labeled ammonium chloride.
Spectroscopic Techniques-All CD spectra were recorded with a Jasco J-710 spectropolarimeter (Jasco, Grossumstadt, Germany) at 25°C in PBS in 0.2-mm quartz cuvettes at a protein concentration of 35 M. Spectra were recorded 16 times, averaged, and buffer-corrected. Gradient-selected 1 H-15 N TROSY HSQC spectra of samples in PBS and 90% H 2 O/10% D 2 O were recorded at 298 K on a Bruker Avance 800 MHz NMR spectrometer with a TCI triple-resonance cryogenic probe with 1024 ϫ 128 complex data points and varying number of scans to account for the different protein concentrations and sensitivity. Spectra were processed with nmrPipe (23).
Assessment of Disulfide Bridge Formation-To monitor disulfide bridge formation between and within different proteins in vitro, the respective proteins were (co-)incubated at 25°C in PBS supplemented with 1 mM GSH/GSSG each. The concentration of each protein was 35 M. After 3 h, samples were supplemented with 20 mM N-ethylmaleimide and Laemmli buffer (with or without ␤-mercaptoethanol) and boiled for 3 min. Proteins were separated on 14% SDS-PAGE gels.
Partial Proteolysis and Mass Spectrometry-Stability against proteolytic digestion was assessed by partial proteolysis experiments. Before the addition of protease, proteins were (co-)incubated for 3 h at 25°C in PBS at a protein concentration of 35 M each. If present, GSH and GSSG were used at a concentration of 1 mM each, and any reduction/oxidation reactions were quenched by the addition of 20 mM N-ethylmaleimide prior to addition of the protease. After the preincubation step, trypsin (Trypsin Gold, Promega, Madison, WI) was added at a concentration of 1:80 (w/w). Aliquots were withdrawn after different time points, and the proteolysis was terminated by the addition of Roche complete protease inhibitor without EDTA (Roche Applied Science) and Laemmli buffer (without ␤-mercaptoethanol) and boiling for 3 min. Proteins were separated on 14% SDS-PAGE gels. Gels were quantified with ImageQuant TL software (GE Healthcare). Mass spectrometry experiments were performed by the St. Jude Proteomics core facility using tryptic digestion combined with LC-MS/MS.
Cell Culture Experiments-Constructs were either amplified from synthetic TCR genes optimized for human expression (A6 TCR) (Geneart) or from synthetic genes with the authentic cDNA sequence of the proteins (HA TCR) and cloned into the pSVL vector. BiP was used in a pMT vector (24). All sequences were verified. Experiments were performed in COS-1 cells that were grown in DMEM supplemented with 10% (v/v) fetal bovine serum, 2 mM L-glutamine, and a 1% (v/v) antibioticantimycotic solution (25 g/ml amphotericin B, 10,000 g/ml streptomycin, and 10,000 units of penicillin; Cellgro/Mediatech, Manassas, VA) (complete DMEM) at 37°C and 3% CO 2 . COS-1 transfections were carried out for 24 h in p60 dishes using GeneCellin (BioCellChallenge, Toulon, France) according to the protocol of the manufacturer. For metabolic labeling, cells were starved for 30 min in complete DMEM without Met and Cys and subsequently supplemented with 100 Ci/p60 dish of EasyTag TM EXPRESS35S protein labeling mixture (PerkinElmer Life Sciences), with labeling times indicated in the respective figures. Labeling was carried out in the presence of 10 mM DTT for DTT washout experiments. Prior to lysis, cells were washed twice in ice-cold PBS supplemented with 20 mM N-ethylmaleimide when samples were to be run on nonreducing SDS-PAGE gels. Cells were lysed with radio-immunoprecipitation assay buffer (50 mM Tris/HCl (pH 7.5), 150 mM NaCl, 1.0% Nonidet P40 substitute, 0.5% sodium deoxycholate, 0.1% SDS, 0.1 mM PMSF, and 1ϫ Roche complete protease inhibitor without EDTA) or Nonidet P-40 lysis buffer in the case of chaperone co-immunoprecipitation experiments (50 mM Tris/HCl (pH 7.5), 150 mM NaCl, 0.5% Nonidet P40 substitute, 0.5% sodium deoxycholate, 0.1 mM PMSF, and 1ϫ Roche complete protease inhibitor without EDTA, supplemented with 10 units/ml Apyrase for BiP interaction studies (Sigma-Aldrich, St. Louis, MO)). Immunoprecipitations were performed with antibodies against the TCR C␣ domain (TCR1145) or C␤ domain (TCR1151) (Thermo Fisher Scientific, Rockford, IL), the FLAG tag (F7425, Sigma), BiP (25), or the HA tag (produced in our laboratory). 3 g of antibody or 3 l of serum were used for each cell lysate of one p60 dish and rotated overnight at 4°C, and, subsequently, CaptivA TM PriMAB protein A-agarose (Repligen Bioprocessing, Waltham, MA) was added for 2 h at 4°C under rotation. Immunoprecipitated proteins were washed three times with radio-immunoprecipitation assay buffer or Nonidet P-40 washing buffer (for chaperone co-immunoprecipitations 50 mM Tris/HCl (pH 7.5), 400 mM NaCl, 0.5% Nonidet P40 substitute, and 0.5% sodium deoxycholate) and eluted with Laemmli buffer for 5 min at 95°C. For Endo H/PNGase F (New England Biolabs, Ipswich, MA) deglycosylation experiments, proteins were eluted with glycoprotein-denaturing buffer and subsequently treated according to the protocols of the manufacturer. For pulse-chase experiments, cells were washed twice with ice-cold PBS after the labeling step, and then complete DMEM supplemented with additional 2 mM of cold Cys and Met was added for the chase times indicated in the figures.

Results
Heterodimerization of the TCR ␣ and ␤ Chains Is a Prerequisite for ER Exit-To provide insights into the assembly process of the clonotypic TCR ␣and ␤-chains, we used the structurally well characterized human A6 ␣␤TCR (26,27). Experiments were performed in COS-1 cells, which allow for complete TCR assembly and transport (28). Because this study focused on quality control steps that monitor assembly of the TCR ␣/␤ chains, the membrane-integrated basic residues of these chains ( Fig. 1A) were mutated to Leu (designated ␣ RK-ϾLL and ␤ K-ϾL , respectively). This allows stable membrane integration of the TCR ␣ chain in the absence of CD3 assembly (10,29) and, therefore, permitted us to focus on the quality control checkpoints for the clonotypic TCR ␣/␤ chains downstream of this critical step.
The fate of ␣ RK-ϾLL and ␤ K-ϾL was monitored by pulse-chase metabolic labeling experiments in which neither ␣ RK-ϾLL nor ␤ K-ϾL acquired Endo H resistance during a 3-h chase, indicative of their retention in the ER (Fig. 1B, top panel). In contrast, when ␣ RK-ϾLL and ␤ K-ϾL were co-expressed, ␣ chains could be co-precipitated with anti-C␤ antibodies (Fig. 1B, bottom panel, black asterisk), and an Endo H-resistant, PNGase F-sensitive species was observed after the chase (Fig. 1B, bottom panel, red asterisk), an indication of further sugar modification in the Golgi. When analyzed under non-reducing conditions, the formation of an ␣␤ heterodimer was evident (Fig. 1C, black arrow). This heterodimer became larger in size over time, also suggesting further glycan modification in the Golgi (Fig. 1C, red arrow). Therefore, when their basic transmembrane (TM) residues are mutated and CD3 co-receptor interactions are abolished, the clonotypic TCR ␣ and ␤ chains are retained in the ER when unassembled and can continue to the Golgi upon heterodimerization. This indicates retention motifs that become inactive upon heterodimerization, which are different from the intramembrane basic residues or CD3 association.
The ER Chaperones Calnexin and BiP Recognize Unassembled TCR ␣ and ␤ Chains-As glycoproteins, the TCR ␣ and ␤ chains are substrates of the ER Calnexin (Cnx)/Calreticulin (Crt) lectin chaperone system (30,31). The action of Cnx, however, seems to be limited to early stages of ␣␤TCR biogenesis (32) and is not essential for the correct assembly of ␣␤TCRs (33). Additionally, the ER Hsp70 orthologue BiP (10,34) and the ER lumenal large Hsp70 Grp170 (35) bind to TCR ␣ and ␤ chains. The precise role of these chaperones in TCR assembly and their binding sites within the clonotypic TCR chains have remained unclear.
When chaperone interactions were analyzed, we observed stable interaction of Cnx with unassembled ␣ and ␤ chains ( Fig.  2A). These interactions were reduced significantly when proteins devoid of their TM regions were analyzed (Fig. 2B). Interaction with the ER-lumenal lectin chaperone Crt was significantly less pronounced and not even detectable for the ␣ chain ( Fig. 2, A and B), in agreement with earlier studies (30). The ␣ and ␤ chains can each be structurally divided into two Ig-like ectopic domains (V␣ and C␣ or V␤ and C␤, respectively), a flexible stalk that connects the constant domains to the TM regions, the TM regions themselves, and a short cytoplasmic tail (Fig. 2C). To assess the role of these different elements in Cnx interaction and ERQC, we created mutants of ␣ RK-ϾLL and ␤ K-ϾL where we individually replaced either the flexible stalk or the cytoplasmic tail with poly-GS sequences of similar length (denoted as ␣ RK-ϾLL/stalk-ϾGS and ␣ RK-ϾLL/tail-ϾGS or ␤ K-ϾL/stalk-ϾGS and ␤ K-ϾL/tail-ϾGS , respectively). None of these mutants became Endo H-resistant over a chase period of 5 h, and all mutants showed similar half-lives (Fig. 2, D and E) and wildtype-like interaction with Cnx ( Fig. 2F), arguing that the lumenal domains themselves are key factors in ERQC.
Next we analyzed how the other major chaperone system of the ER, the Hsp70 system centered around BiP (36), recognized the TCR ␣ and ␤ chains. In agreement with data published previously (10,34,35) we found both ␣ RK-ϾLL and ␤ K-ϾL to be BiP substrates (Fig. 3A). Constructs devoid of their TM regions (␣ ⌬TM or ␤ ⌬TM , respectively; Fig. 1A) showed interactions with BiP that were comparable with their membrane-integrated counterparts (Fig. 3A), revealing the presence of BiP-binding sites within the ectodomains of both the TCR ␣ and ␤ chains. Conversely, Cnx interactions were reduced significantly when proteins devoid of their TM regions were analyzed (Fig. 2, A and  B), indicating that membrane proximity or the TM regions themselves are important for the interaction of TCR ␣/␤ chains with the integral membrane protein Cnx (37,38). To further dissect BiP interactions with the TCR clients, we individually analyzed its interaction with each of the isolated ␣/␤ chain domains. We reproducibly observed binding, albeit weak, of BiP to the ␣ chain C␣ domain but essentially no binding to V␣ (Fig. 3B). The opposite was true in the case of the ␤ chain. The V␤ domain bound to BiP very strongly, whereas we observed very minimal interaction of C␤ with BiP (Fig. 3B). Of note, BiP preferentially bound the less glycosylated form of the ␤ chain ( Fig. 3, A and B), whereas Cnx did not show such a preference ( Fig. 2A). The absence of glycosylation may facilitate BiP binding by exposing binding sites or because of the absence of competition with lectin chaperone binding (37)(38)(39).
The Isolated TCR ␣ and ␤ Chains Contain Incompletely Folded Domains-To understand the observed chaperone binding characteristics in more structural detail, we assessed the folding status of the ␣/␤ chains and their individual domains in the ER by monitoring intradomain disulfide bond formation in the various ␣/␤ chain constructs. To do so, cells were treated with DTT during metabolic labeling to inhibit the formation of disulfide bonds in newly synthesized proteins. Subsequent reculturing in DTT-free media allows oxidation to occur. The presence of a disulfide bond can lead to a more compact state of a protein unfolded in SDS, which may be detected by a modest increase in its mobility on non-reducing SDS-PAGE gels (40). If native disulfide bonds do not form, then it is a clear indication of incomplete folding, whereas, if oxidation occurs, this most often correlates with folding. The ␣ and ␤ chains ran slightly faster on non-reducing SDS-PAGE gels after the removal of DTT (Fig. 4A), indicative of the formation of at least one intradomain disulfide bond. The shift was very small, however, likely because of the large size of the proteins, and it was more readily detectable for constructs devoid of their TM regions (Fig. 4A). When the individual domains were assessed similarly, V␣ and both bands of C␤, which correspond to different glycospecies, showed increased mobility on non-reducing gels after removal of DTT, indicative of oxidative folding. In contrast, C␣ had an unaltered mobility, and V␤ actually migrated slightly slower, arguing neither of these domains formed an internal disulfide bond (Fig. 4A). Therefore, in agreement with our BiP binding data, the isolated C␣ and V␤ domains appeared to remain incompletely folded.
Genetic recombination and selection give rise to a different set of variable domains in each TCR, which might affect their residues (␣ RK-ϾLL and ␤ K-ϾL , respectively). Only the C␤ antibody is able to immunoprecipitate the ␣␤ heterodimer (the co-precipitating ␣ chain is marked with a black asterisk). Where indicated, proteins were deglycosylated with Endo H or PNGase F. A red asterisk indicates the Endo H-resistant species. Chase times are shown above the lanes. 2.5 g of each ␣ chain construct and 1.5 g of each ␤ chain construct were (co-)transfected, and COS-1 cells were metabolically labeled for 1 h. C, pulse-chase experiments were performed as in B, except that the samples were analyzed under non-reducing conditions. The black arrow indicates the ␣␤-heterodimer, whereas its Golgi-modified form is marked with a red arrow.
folding, whereas C␣ and C␤ are conserved between different ␣␤TCRs. We therefore assessed the folding properties of the V␣ and V␤ domains derived from another, anti-HA human ␣␤TCR (41). Interestingly, for the anti-HA ␣␤TCR, the V␣ domain seemed to be unable to form its disulfide bond, whereas V␤ showed a small shift upon DTT washout, indicative of it forming an intradomain disulfide bond (Fig. 4B). Furthermore, in contrast to the A6 V␣ domain, the HA TCR V␣ domain strongly bound to BiP (compare Figs. 3B and 4C), whereas binding of the HA V␤ domain was of similar strength as that Lysates were immunoprecipitated with the indicated antibodies, and their association with the lectin chaperones is indicated. m, mock. B, the same as in A, except that constructs devoid of their TM regions (␣ ⌬TM and ␤ ⌬TM ) were used. C, schematic of the ␣␤TCR with the flexible stalks, TM basic residues, and cytoplasmic tails indicated. D, to assess the impact of the flexible stalk and the cytoplasmic tail on ER retention of the TCR ␣ chain, COS-1 cells were transfected with the indicated ␣ chain constructs, metabolically labeled for 1 h, and subsequently either lysed directly or chased for the indicated times. Where indicated, proteins were treated with Endo H. E, experiments similar to those described in D were conducted for the TCR ␤ chain. F, influence of the stalk or the cytoplasmic tail on the interaction of either the TCR ␣ chain or the ␤ chain with Cnx. The antibodies used for immunoprecipitation are indicated below the gel (the anti-Cnx antibody is denoted with Cx). Note that the C␣ antibody precipitates an unspecific band the size of Cnx, which is also observed in cells transfected with empty pSVL vector (m, mock). 4 g of each construct were transfected. OCTOBER  To provide structural insights into the TCR ␣␤ assembly process that are not amenable to in vivo analyses, we performed in vitro studies with purified proteins. To this end, we recombinantly expressed the A6 TCR ␣ and ␤ chains (using soluble constructs devoid of their TM regions, denoted ␣ ⌬TM or ␤ ⌬TM , respectively; Fig. 1A) as well as their individual ectodomains and assessed the folding status by CD and NMR spectroscopy. As expected from the ␣␤TCR crystal structure (Fig. 1A) (26,27), the far-UV CD spectrum of ␤ ⌬TM displayed typical characteristics of a protein dominated by ␤ sheet secondary structure; namely, a minimum around 220 nm and a maximum around 200 nm (Fig. 5A, blue tracing). However, ␤ ⌬TM displayed low signal intensities around 200 nm, which is indicative of the presence of unstructured regions. This was even more pronounced for ␣ ⌬TM (Fig. 5A, red tracing). To further dissect this finding, we individually analyzed the variable and constant domains. In keeping with our in vivo studies, the far-UV CD spectrum of V␣ (Fig. 5B, green tracing) was that of a well folded ␤ sheet protein, whereas C␣ showed the spectrum of a predominantly unfolded protein (Fig. 5B, red tracing). We observed a similar behavior for the constituents of the ␤ chain but with the domains switched. C␤ was well folded (Fig. 5B, blue tracing), whereas V␤ was not amenable to structural studies because of its tendency to aggregate. Therefore, it most likely accounts for the less well folded parts within ␤ ⌬TM (compare the spectra in Fig. 5, A and B). In agreement with our CD data, the 1 H-15 N TROSY HSQC NMR spectrum of the C␣ domain had limited chemical shift dispersion and narrow line widths, indicating fast interconversion of different conformations. It was therefore typical of a disordered protein lacking a defined structure (Fig. 5C, red spectrum). Of note, the resonances observed for the C␣ domain were almost completely superimposable with a subset of signals of the 1 H-15 N TROSY HSQC spectrum obtained for ␣ ⌬TM (Fig. 5C, black spectrum), and the remaining, mostly well dispersed resonances superimposed well with the spectrum of V␣ (Fig. 5C, green spectrum). Therefore, also in the context of ␣ ⌬TM , C␣ appears to be unstructured, whereas V␣ is well folded.

Mechanism of ␣␤T Cell Receptor Assembly Control
Assembly-dependent Folding of the TCR ␣ and ␤ Chains-Taken together, our in vitro data show that the A6 TCR ␣ and ␤ chains each comprise one well folded domain (V␣ and C␤, respectively) and one less well folded domain in isolation. C␣ appeared to be almost completely unfolded both in isolation and within the ␣ ⌬TM chain. In contrast, V␤ seemed to possess some ␤ sheet structure, at least in the context of the complete ␤ ⌬TM chain, as indicated by the presence of more ␤ sheet structure in the CD spectrum of ␤ ⌬TM compared with that of ␣ ⌬TM (Fig. 5A)

FIGURE 3. Interactions of the TCR ␣/␤ chains and domains with BiP in vivo.
A, association of BiP with the A6 TCR ␣ and ␤ chains in the presence (␣ RK-ϾLL or ␤ K-ϾL , respectively) or absence of their TM regions (␣ ⌬TM or ␤ ⌬TM , respectively) was detected by co-immunoprecipitation. 3 g of the TCR constructs were co-transfected with 2 g of BiP. Cells were metabolically labeled for 30 min. Chase times and precipitating antibodies are indicated. m, mock transfection. B, BiP binding experiments for the individual A6 TCR ␣ and ␤ chain domains were performed as in A. Variable domains were epitope-tagged as indicated. The two bands observed for the ␤ chain/C␤ domain represent two glycospecies (56).  our in vitro and in vivo data, we used the recombinant proteins to analyze the assembly process of the ␣ and ␤ chains and the concomitant structural changes in more detail. To this end, we performed partial proteolysis experiments on the recombinant proteins. Products were analyzed by both SDS-PAGE and mass spectrometry. Under non-reducing conditions, ␣ ⌬TM migrated as four distinct bands (Fig. 6A, left panel) that resolved into a single band under reducing conditions (Fig. 6D), indicative of disulfide bonding heterogeneity. Partial proteolysis with trypsin rapidly degraded these four species into a single one that we identified by mass spectrometry to be the V␣ domain (Fig. 6A,  left panel, and B), arguing that C␣ causes the disulfide bonding heterogeneity. C␣ contains two cysteines, and the flexible stalk of the ␣ chain contains an additional cysteine that forms an interchain disulfide bond with the ␤ chain in the mature receptor (26,27) (Fig. 1A). If all redox species of a three cysteinecontaining protein are populated, four species are expected, exactly what we observed. In vivo, only a single redox species could be detected (Fig. 4A), arguing that this potential for heterogeneity is suppressed in cells. In comparison to the ␣ chain, the ␤ chain was degraded more slowly by trypsin (Fig. 6A, center  panel). All stable fragments contained the C␤ domain (Fig. 6B), in agreement with the finding that V␤ is the less well folded domain. When assembled, the ␣␤ heterodimer only populated a single redox species and was very resistant to proteolytic digestion (Fig. 6, A-C), which indicates assembly-induced folding events in the TCR ␣ and ␤ chains.
To identify minimal elements needed for ␣␤ heterodimerization and concomitant folding events, we assessed the interaction of the ␣ chain with the C␤ domain and, vice versa, of the ␤ chain with the C␣ domain. No covalent dimers could be detected in either case (Fig. 6D), arguing that interaction of the constant domains depends on the additional presence of the variable domains. NMR spectroscopy supported this finding (Fig. 6E). Interestingly, when we used a ␤ chain in which the cysteine residue that forms the interchain disulfide bridge with the ␣ chain was deleted (Fig. 1A, ␤ ⌬TM, ⌬Cys ), it was still capable of reshuffling the various ␣ chain redox species and led to the predominant population of a single ␣ chain species in ␣ ⌬TM (Fig. 6D). This strengthens our conclusion that the ␤ chain induces folding of C␣ and that assembly-dependent folding does not depend on the presence of the interchain disulfide bond, which is in agreement with the fact that the ␣␤ interchain disulfide bond is dispensable for TCR transport to the cell surface (42).
Taken together, our data suggest a mechanism of reciprocally induced domain folding upon heterodimerization of the TCR ␣/␤ chains. Unfolded domains are specifically recognized by ER chaperones, and their folding is a prerequisite for release from the designated chaperones in the ER environment. Together, this allows ERQC to monitor the proper assembly of functional ␣␤TCRs.

Discussion
How comparable data obtained from in vitro analyses are to those from in vivo folding studies is a matter of considerable debate. For all proteins analyzed in this study, a very good agreement existed. For domains that appeared well folded in vitro (A6 V␣ and C␤), we found evidence of disulfide bridge formation and the absence of binding to the Hsp70 chaperone BiP in vivo. Similarly, in vitro patterns signifying unfolded/unstructured domains correlated with stable BiP binding and the absence of disulfide bond formation in vivo (A6 C␣ and V␤). Binding of calnexin was strongly dependent on membrane integration and could therefore not be assessed easily on a singledomain basis. Of note, even though disulfide bond formation can occur in unfolded proteins in vitro, in vivo folding and oxidation seem to be coupled more tightly (24,43,44).
The good agreement between our in vitro and in vivo folding data allowed us to use the complementary strengths of these approaches to analyze the structural details underlying ERQC of TCR ␣/␤ chain assembly. Our data show that the constant domain of the ␣ chain, C␣, is unfolded and gains structure upon ␣␤ heterodimerization. Before assembling with a ␤ chain, the C␣ domain therefore provides interaction sites for ER chaperones, BiP and Cnx, allowing recognition of unassembled ␣-CD3␦⑀ trimers. In the case of the ␤ chains analyzed in this study, the V␤ domain is a strong client of BiP, providing a means for retaining ␤-CD3␥⑀ trimers. Of note, even though the HA TCR V␤ domain formed its disulfide bond in vivo, it remained a strong BiP substrate. Therefore, BiP binding sites are present even in an oxidized V␤ domain. This implies that, unlike antibodies, where free light chains (LCs) can be secreted whereas unassembled heavy chains (HCs) are retained, both clonotypic chains of ␣␤TCRs provide a focus of ERQC. This difference becomes biologically meaningful when one keeps in mind that antibody LCs do not have any effector functions whereas antibody HCs do, and TCR ␣-CD3␦⑀ and ␤-CD3␥⑀ trimers might each trigger aberrant signaling when transported to the cell surface (45). ERQC therefore acts on all antibody and TCR components that could be potentially harmful when transported while assembled incompletely. Taken together, only the combined reciprocal folding of both V and C domains upon assembly produces an ␣␤-CD3␥␦⑀ 2 hexamer that no longer interacts with ER chaperones and can be transported to the Golgi, where 2 homodimers are added to complete the ␣␤TCR. However, each V domain is unique. Our data show that this may cause different folding properties, which indicates some previously unappreciated restrictions on compatible TCR V␣-V␤ pairs and, therefore, the possible T cell repertoire.
The data presented in this study reveal interesting similarities and differences in terms of ERQC mechanisms between the ␣␤TCR and antibodies. For IgG antibodies, the first constant domain (C H 1) of the HC is unfolded while HCs are unpaired and gains structure upon HC-LC assembly. This allows HCs to be retained by BiP in developmentally immature pre-B cells that are not producing LCs and for BiP to be released and antibody secretion to occur in more mature cells that produce LCs (24,46,47). Our data indicate, for ␣␤TCRs, that the C␣ domain behaves as a similar target of ERQC. Of note, although C H 1 adopts a regular Ig fold when assembled, C␣ lacks an entire ␤ strand in its Ig fold and remains flexible even in the ␣␤ heterodimer (26,27). This might explain why assembly of the TCR constant domains in vitro depends on the additional presence of the variable domains, whereas C H 1 can assemble with its cognate partner domain C L in vitro in the absence of the variable regions (46). Another distinction is that IgG assembly primarily relies on the BiP-centered Hsp70 system of the ER (48), whereas the lectins Crt and, in particular, Cnx are major chaperones for the ␣␤TCR (30 -32, 49). C␣ is the most heavily glycosylated of all ␣␤TCR domains (Fig. 1A), which might facilitate Cnx recruitment to this incompletely folded domain and contribute to the relatively weak BiP binding we observed (38,39).
Interestingly, in the case of antibodies, the unfolded C H 1 domain is found in HCs, which are expressed first during development. In contrast, the incompletely folded C␣ domain is found in the TCR ␣ chain, which developmentally corresponds to antibody LCs, and not in the ␤ chain, which is expressed first (50,51). For both arms of the adaptive immune response, a mechanism exists for transporting a small amount of the chain that is produced first to the surface for signaling. In the case of pre-B cells, this is achieved by the expression of a surrogate light chain (52) that is able to assemble with HCs to relieve them from ER retention (50,52,53). Conversely, immature T cells make a pre-TCR␣ (54). Even though the pre-TCR␣ lacks a variable region, it is still able to interact with both the C␤ and the V␤ domain (55) to allow surface expression of the ␤ chain. Our data suggest that C␤ is well folded, which might provide a docking site for pre-TCR␣, allowing it to induce further folding or shielding of hydrophobic surfaces in V␤ (55) so that BiP is released and the ␤ chain can be expressed on the cell surface of pre-T cells. This provides a rationale for why pre-TCR␣ needs to interact with both domains to allow the TCR ␤ chain to leave the ER. 7. A model for ␣␤TCR assembly control in the cell. Unassembled clonotypic ␣ and ␤ chains are both Cnx substrates. The ␣ chain comprises either one (C␣) or two (C␣ and V␣) incompletely folded domains. The ␤ chain C␤ domain is well folded whereas V␤, in at least some ␣␤TCRs, is not. All incompletely folded domains provide BiP interaction sites, with particular strong binding to incompletely folded variable domains. Of note, V␤ binds to BiP independently of its capability to form its intradomain disulfide bond. Over time, because of their hydrophilic TM region, unpaired ␣ chains completely enter the ER lumen and are degraded by ER-associated degradation (ERAD) (14,15,29,46). Note that ␣␤ heterodimerization should retain the ␣ chain at the membrane but might fail to integrate its TM region if the ␣-chain is not assembled with the CD3 complex (7,8,10). Correctly assembled ␣-CD3␦⑀ and ␤-CD3␥⑀ pairs en route to the native receptor, whose interaction with Cnx might be reduced or abolished (32), still provide BiP binding sites in V␤ and, likely, C␣, even though the temporal relationship of Cnx and BiP binding is not entirely clear. Upon ␣␤ heterodimerization, the incompletely folded domains in the clonotypic chains become structured and are therefore released from chaperones. Finally, assembly with dimers in the Golgi allows transport to the cell surface to occur (13,21,22,57).
Taken together, our study allows us to propose a detailed model for ␣␤TCR assembly control in the cell (Fig. 7). Even though important differences are observed in ERQC mechanisms overseeing the maturation of the clonotypic chains in ␣␤TCRs and antibodies, the functions of both molecules in the immune response demand highly reliable quality control checkpoints prior to transport to the cell surface. To fulfill these requirements, antibodies and TCRs have evolved quality control steps relying on assembly-induced folding. The choice of domains and even on which chains they are found, in terms of the developmental sequence of expression, argues that this critical step in ERQC for immune receptors has arisen twice through convergent evolution.