Each T cell receptor (TCR) recognizes a peptide antigen bound to a major histocompatibility complex (MHC) molecule via a clonotypic αβ heterodimeric structure (Ti) non‐covalently associated with the monomorphic CD3 signaling components. A crystal structure of an αβ TCR‐anti‐TCR Fab complex shows an Fab fragment derived from the H57 monoclonal antibody (mAb), interacting with the elongated FG loop of the Cβ domain, situated beneath the Vβ domain. This loop, along with the partially exposed ABED β sheet of Cβ, and glycans attached to both Cβ and Cα domains, forms a cavity of sufficient size to accommodate a single non‐glycosylated Ig domain such as the CD3ϵ ectodomain. That this asymmetrically localized site is embedded within the rigid constant domain module has implications for the mechanism of signal transduction in both TCR and pre‐TCR complexes. Furthermore, quaternary structures of TCRs vary significantly even when they bind the same MHC molecule, as manifested by a unique twisting of the V module relative to the C module.
T cells recognize peptides derived from cell‐associated pathogens of viral, bacterial or fungal origin, as well as tumor antigens bound to MHC molecules through a multi‐subunit transmembrane surface complex termed the T cell receptor (reviewed in Meuer et al., 1984; Marrack and Kappler, 1986; Clevers et al., 1988; Davis and Bjorkman, 1988; Ashwell and Klausner, 1990). The monomorphic CD3 chains (CD3γ, δ and ϵ) within this complex were the first TCR components to be identified serologically and their functional significance was suggested by the ability of CD3ϵ‐specific monoclonal antibodies (mAbs) to inhibit antigen‐specific T cell recognition (Reinherz et al., 1980, 1982). Subsequently, through the development of clone‐specific (anti‐clonotypic) mAbs (Allison et al., 1982; Haskins et al., 1983; Meuer et al., 1983a,b,c), disulfide‐linked αβ heterodimers (Ti α‐β) were defined in non‐covalent association with the CD3 components (Meuer et al., 1983a,b,c) including CD3ζ (Samelson et al., 1985; Manolios et al., 1991). In contrast to the CD3 molecules, the Ti α and β subunits are biochemically distinct on each T cell clone analyzed, including multiple clones derived from the same individual (Acuto et al., 1983a,b; Kappler et al., 1983; Reinherz et al., 1983). Peptide mapping of α and β subunits identified conserved as well as variable peptides, implying the existence of constant and variable domains for both Ti α and β, a notion later confirmed by protein microsequencing (Acuto et al., 1984; Fabbi et al., 1984; Sim et al., 1984) and DNA cloning (Chien et al., 1984; Gascoigne et al., 1984; Hedrick et al., 1984a,b; Patten et al., 1984; Saito et al., 1984; Yanagi et al., 1984; Hayday et al., 1985). The anti‐clonotypic mAb‐mediated crosslinking of TCRs on a given T cell clone selectively activated clonal proliferation and effector function, thereby substituting for the specific combination of MHC and antigen recognized by that clone (Meuer et al., 1983d). Moreover, gene transfer experiments indicated that both the antigen and the MHC specificity of a T cell clone could be conferred to another cell by transfection of an appropriate pair of Ti α and β cDNAs (Dembic et al., 1986; Saito et al., 1987; Kaye and Hedrick, 1988).
Several features of TCRs predicted that they would share with antibodies a common structural basis of ligand recognition (reviewed in Novotny et al., 1986; Chothia et al., 1988; Davis and Bjorkman, 1988). Over the last few years, direct evidence has been provided to support this idea. Crystal structures of αβ TCR components have now been reported, including the β chain (Bentley et al., 1995; Fields et al., 1996) and a Vα‐Vα homodimer (Fields et al., 1995). Very recently, an intact murine αβ TCR (Garcia et al., 1996), and a complex between a human TCR, a viral peptide and a human class I MHC molecule (Garboczi et al., 1996), have also been described. As anticipated, the three‐dimensional structure of the TCR resembles an antibody Fab fragment in that each of the α and β chains consists of Ig‐like variable and constant domains, with the hypervariable loops from the two variable domains (Vα and Vβ) forming the antigen‐combining motif. However, there are some clear differences between the TCR and Fab structures. Although each of the four TCR α and β domains has been assigned an Ig fold, deviations are notable in both TCR constant domains (Bentley et al., 1995; Garcia et al., 1996), as well as the Vα domain (Fields et al., 1995). We have now performed a structural determination of a TCR complexed with an anti‐TCR Fab, offering direct analysis of T cell and B cell immune receptors in the same crystal lattice. For this purpose, we employed the murine N15 TCR (Vα8Jα5Cα; Vβ5.2D2Jβ2.6Cβ2) specific for the vesicular stomatitis virus nuclear protein octapeptide (RGYVYQGL) (VSV8) bound to the murine H‐2Kb class I MHC molecule (Shibata et al., 1992; Witte et al., 1997; Chang et al., 1997). We formed a complex between N15 and an Fab fragment derived from the hamster anti‐mouse Cβ region‐specific mAb H57 (Kubo et al., 1989).
Structural analysis of the N15 TCR was made possible by crystallization of an endoglycosidase H‐treated derivative of N15 secreted from Lec22.214.171.124 cells, which synthesize homogeneous GlcNAc2‐Man5 glycans (Liu et al., 1996). The three‐dimensional structure determined at 2.8 Å provides considerable insight into the architecture of the TCR versus Fab. This complex also delineates precisely the H57‐binding site on the unusual TCR Cβ FG loop. In addition, because the asymmetric unit of the crystal contains two TCR‐Fab complexes in very different crystallographic environments, we can infer characteristics about the intrinsic TCR flexibility versus rigidity within, and between, the individual domains. Our results highlight the overall rigidity of the C domain module and suggest that signaling is unlikely to result from a global conformational change within this part of the αβ heterodimer. Lastly, we have been able to identify a cavity within the C module, bounded by the Cβ FG loop, partially exposed Cβ domain strands, and conserved glycans from both constant domains, which can accommodate a single Ig‐like domain, probably representing the CD3ϵ‐binding site.
Results and discussion
Overall structure of the N15 TCR‐Fab complexes
The structure of the N15 TCR‐Fab complex was determined by the combination of molecular replacement and MAD phasing. The crystallographic data and refinement statistics are reported in Table I. There are two TCR‐Fab complexes per asymmetric unit of the crystal. Between the two TCR‐Fab complexes, the relative orientation of the Fab with respect to the TCR differs by ∼19° (see the TCR‐H57 Fab description below). This is essentially a relative rotation of the Fab molecule around the FG loop of the TCR Cβ domain. In fact, there is no significant conformational difference in the binding region of the TCR loop, or major conformational variation within either Fab between these two complexes. This rotation is a feature of antigen interaction which has not previously been reported and indicates that differences in receptor‐ligand orientation may exist even when the antibody and antigen are chemically identical in two complexes. The two TCR molecules are generally very similar. For example, the r.m.s. value for superposition of their Cβ domains is only 0.3 Å. Therefore, unless otherwise noted, in comparisons below we shall be referring to one of the two TCR molecules in the asymmetric unit, termed N15A.
The overall structure of one of the two complexes in the asymmetric unit is shown in Figure 1. The complex consists of a single N15 TCR and a single H57 Fab molecule. A striking feature of the structure is that the unusually long FG loop of the β chain TCR Cβ domain pokes into the Fab's antigen combining pocket in a remarkably complementary way. The TCR and Fab molecules are similar in length (80 versus 78 Å) and are both ∼40 Å in depth. However, the width of the TCR is substantially greater than that of the Fab (58 versus 46 Å). In addition, there is an obvious difference between the ligand binding surfaces of the V domain modules of the two receptors, with a relatively flat antigen binding surface in the case of the TCR, versus a concave surface in the case of the Fab. Presumably this TCR surface offers a better interaction with the elongated peptide‐MHC ligand, which is itself relatively flat on the face contacting the TCR (Fremont et al., 1992; Zhang et al., 1992). Another difference between immunoreceptors is the symmetry of the Fab molecule compared with the asymmetry of the TCR molecule. This TCR asymmetry results from the unusual arrangement of the Cα and Cβ domains relative to one another, as discussed in detail below.
Assuming the view in Figure 1 represents the disposition of the TCR molecule on T cells, as seen from the side with the TCR lying roughly perpendicular to the cell membrane, then the Fab fragment of the antibody molecule reaches the TCR with its long dimension lying parallel to the membrane. One can envision how TCRs on a T cell surface might be crosslinked by the Fab arms of individual antibody molecules. This is the case for the H57 mAb, which selectively activates all αβ TCR‐bearing T cells but not γδ T cells (Kubo et al., 1989).
All four domains of the N15 TCR, in both complexes, are well ordered in the crystal. Figure 2 is a stereoview of the N15 TCR molecule. Although each of the four domains is Ig‐like, they all differ in a significant way from a canonical Ig domain. The fold of the Cα is the most extreme in its deviations. While the A, B, E and D β strands, which form the interface with the Cβ domain, are similar to the corresponding sheet in Ig C regions, the C, F and G strands are highly divergent from the analogous components. In fact there is no outer β sheet per se. Rather, there is a loosely tethered connection of structural elements. In particular, after the EF turn, which contains a conserved glycosylation site at Asn185α, there is an F segment which lacks H‐bonds to the neighboring strands. This segment contains a mini α‐helix, including Cys191α which forms the intrachain disulfide bond with Cys141 of the Cα domain. This Cα structure is consistent with that reported by Garcia et al. (1996). The Cβ domain is remarkable for containing a 13 amino acid insertion within the FG loop, resulting in its striking elongation compared with Ig constant domains, as initially reported by Mariuzza and co‐workers in the structure of an isolated TCR β subunit (Bentley et al., 1995). Sequence comparisons show that this insertion is not found in other Ig‐related structures, except for TCR Cγ domains (Kabat et al., 1991). The translocation of the C″ strand of the Vα8 domain from the GFCC′ face to the BED face, as first recognized in the structure of a Vα‐Vα homodimer (Fields et al., 1995), is also observed in the N15 TCR. In addition, the C″ strand of the N15 Vβ5.2 domain deviates from the canonical Ig V domain position. It shifts away from the BED face and hence in the opposite direction compared with the movement of the corresponding C″ strand in the Vα domain.
In the electron density maps there are clear densities for all seven of the potential glycosylation sites, three in the α chain and four in the β chain. Note that of the seven TCR glycans, five are located in the C module, whereas only two are found in the V module. Since the N15 TCR protein was produced in Lec126.96.36.199 cells, which synthesize homogeneous glycans of the type GlcNAc2‐Man5, and was subjected to endo‐H digestion before crystallization, only one N‐acetylglucosamine residue (GlcNAc) remains attached to the Asn at each of the seven N‐linked sites (Liu et al., 1996): Asn21α, Asn185α and Asn203α of the TCRα subunit, and Asn78β, Asn121β, Asn186β and Asn236β of the TCRβ subunit (Figure 2).
Owing to the large relative orientation difference of the Fab with respect to the TCR, and small rotational variation between individual domains, there is no detectable simple non‐crystallographic symmetry for the two complexes. Nonetheless, these two complexes are roughly related by a pseudodyad that is slightly tilted away from the crystal 2‐fold. The major contacts between the two TCRs are made by the Vβ domain, and there is no contact between the two Vα domains. This is in contrast to the structure of the Vα‐Vα homodimer, reported previously (Fields et al., 1995). Based on this result, it seems unlikely that TCRs will dimerize physiologically through their Vα domains.
Deviations in quaternary structure of TCRs
To mediate their specific antigen‐binding functions, antibody molecules assume highly divergent shapes, both locally, as a consequence of unique CDR loop configurations, and more globally, through quaternary structural rearrangement among their domains (Stanfield et al., 1993). To assess the quaternary structure variation among TCRs, we have chosen the N15 TCR molecule A as a reference molecule and compared it with the N15 TCR molecule B within the same asymmetric unit, and with the other crystallographically defined TCR αβ heterodimers, the mouse 2C TCR (Garcia et al., 1996) and the human A6 TCR (Garboczi et al., 1996). Since the Cβ domain is highly conserved in all three structures, it was used for the superposition of TCR pairs. The FG loop was excluded in the superposition. The r.m.s. value for the Cβ domain superposition between N15A and N15B molecules is 0.3 Å and between N15 and 2C molecules is 0.5 Å (107 residues used). The r.m.s. value for the N15‐A6 superposition is slightly greater at 0.7 Å (75 residues used). Relative orientation differences of the other domains with respect to the Cβ domain were then calculated, as listed in Table II. Several points emerge from this analysis. First, it is evident that the Cα domain has essentially the same orientation, with respect to the Cβ domain, in all different TCR molecules (⩽2.4° rotation), implying a fixed Cα‐Cβ domain configuration. We will discuss this relatively rigid C module later. Second, we note a large rotational deviation (18°) between the Vα domains of the N15 and 2C TCRs. By contrast, there is only a very small rotation difference (4.4°) between the two N15 Vα domains, despite their location in significantly different crystallographic environments. An intermediate deviation (10.9°) is noted between the Vα domains of N15 and A6. This is a striking observation given that the N15 (Shibata et al., 1992; Chang et al., 1997; Witte et al., 1997) and 2C TCRs (Garcia et al., 1996) interact with the same murine class I MHC molecule, namely Kb, but loaded with different peptides (VSV8 versus dEV8), whereas A6 recognizes the human HLA‐A2 molecule loaded with a Tax peptide (Garboczi et al., 1996). Third, note also that deviation exists in the relative rotation of the Vβ domain in the pairwise comparisons (3.6‐10.6°), but to a lesser extent than observed with Vα. Collectively, these findings indicate that quaternary differences in structures result from variations in at least the disposition of Vβ and Cβ domains as well as Vβ and Vα domains.
Figure 3 shows the superpositions of N15A‐N15B and N15A‐2C TCR pairs. On the left side is the superposition of the two N15 TCRs in the asymmetric unit, whereas on the right side is the overlay between N15 and 2C. The similarities in the C module are evident in both, while the difference between N15 and 2C quaternary structures involving the V module is most notable in this side view (Figure 3, bottom right). When viewed from the top of the molecules (i.e. V module toward C module), the V module of the 2C TCR has an anticlockwise twist with respect to N15 (Figure 3, top right). The same quaternary variation, although to a lesser extent, is even evident when the two N15 TCRs are compared with each other (Figure 3, top left). This twist is another novel feature of the TCR, reflecting its basic asymmetry. In the case of Fabs, the movement of the V module with respect to the C module is defined by the elbow angle, as the VH and VL domains rotate in the same direction. In contrast with the TCR, there is a Vα‐Vβ twist rather than a concerted movement in the same direction. This orientational variation affects how a given TCR molecule docks onto the peptide‐MHC assuming the latter complex is fixed.
In the TCR‐peptide‐MHC complex structures reported by Garboczi et al. (1996) and Garcia et al. (1996), a common docking feature is that the long dimension of the TCR and MHC binding surface is not parallel but tilted. Garboczi and co‐workers argue that this relative orientation is due to the fact that the two highest points on the MHC molecule binding surface impose a restriction on how the TCR can approach the deeply buried peptide. Since the two high points appear to be a topographical feature for both class I and class II MHC molecules, Garboczi et al. (1996) speculate further that the ‘diagonal binding’ mode may be a general one. We have noticed that the fundamental structural basis for having the ‘two high points’ lies in the inherent twist of the large eight‐stranded β sheet that forms the platform of the antigen binding groove. The two helical MHC segments run across the entire sheet diagonally. They must break at the two highest points, defining the relatively flat antigen binding surface in between. Our observation of the V module twist, derived from comparison of 2C with N15 TCRs, raises the possibility that there can be some angular variation in the docking of the TCR during immune recognition events. The result is especially noteworthy in the case of 2C and N15 TCR complexes, as they have identical C modules and recognize the same MHC molecule (Kb), but complexed with different peptides. It should be noted that this statement assumes that no rearrangement in the relative orientation of the variable domains occurs upon antigen binding, which may not be the case.
The TCR V domain module
The hydrophobic core found in VH‐VL interfaces, and noted in the 2C TCR (Garcia et al., 1996), is largely conserved in N15. It consists of F43α, L87α, Y89α, Y104α, F106α, L43β, F45β, F91β, Q106β and F108β. As shown in Table III, the Vα and Vβ interface in N15 comprises 1350 Å2 of buried surface area (∼690 Å2 for Vα and 660 Å2 for Vβ). This value is close to an average value of 1420 Å2 found in Fabs (Stanfield et al., 1993; Wilson and Stanfield, 1993; Padlan, 1994), slightly larger than the value reported for 2C and ∼150 Å2 less than that calculated for A6. This difference in buried surface area correlates with the length of the CDR3β residues such that the Vα‐Vβ buried surface increases with 2C < N15 < A6 which have CDR3β loops of 7, 9 and 12 residues, respectively. The contribution of all CDR residues to the buried surface is ∼40% in the case of N15, again near the average observed for an Fab (Padlan, 1994). The finding that the 2C Vα‐Vβ domain buried surface area value was at the extreme low end of the range observed for antibodies, suggested the possibility of large Vα‐Vβ displacement, which might provide better complementarity for the TCR and peptide‐MHC interaction (Garcia et al., 1996). However, based on our observations with N15, this postulated mechanism may be highly dependent on the V module of an individual TCR. Another interesting feature of the interface of the N15 V module is the presence of two interdomain salt bridges which are not conserved in 2C. The first is between K48α and D100β. While the K48α residue is found in both N15 and 2C, the D100β residue is absent in 2C. This bond serves to connect CDR2α with CDR3β regions, stabilizing the loop configuration. The second salt bridge is between K103α and D56β and functions in a similar manner, to connect CDR3α with CDR2β. In 2C, the lysine at position 103 is replaced by alanine.
Overall, it is clear from Figure 4 that CDR1α, CDR2α and CDR1β are similar in all three known αβ TCR heterodimeric structures. In fact, many of the interactions that maintain the conformations of CDR1α and CDR2α loops are preserved among these TCRs. Although the V domain of N15 belongs to the Kabat β subgroup I, while that of 2C and A6 belong to subgroup II (Kabat et al., 1991), the conformation of CDR1β of N15 is still very similar to other TCRs. There is one salt bridge, however, between R64β of the C″D loop and D86β of the EF loop in the N15 TCR, which is characteristic of the Kabat subgroup I and which is also a hallmark for the variable domains of antibodies. As expected, the CDR3α and CDR3β loops of each receptor are idiosynchratic. We will discuss these two CDR3 loops in the next section. While the CDR2β loops of 2C and A6 are similar to each other, they differ markedly from CDR2β of N15, largely due to the forward displacement and twist of the Vβ C″ strand noted above. There is a well‐defined salt bridge between D56β of CDR2β and K103α of CDR3α, which appears to play a role in pulling this C″ strand away from its canonical position. It is interesting to note that in the N15 A molecule, this C″ strand makes a contact with the D strand of the Fab VH domain, from a symmetry related complex in an anti‐parallel way. There is at least one main chain H‐bond between R55β of the TCR molecule and R71 of the VH domain. However, in the N15 B molecule, the C″ strand of Vβ is free of any crystallographic contact. Therefore, it is safe to conclude that the shift of this C″ strand is an inherent feature of the N15 TCR molecule. Although the C″ strand moves away from its canonical position, it remains well integrated into the β sheet by forming main chain H‐bonds between the T54β‐F59β segment in the C″ strand and the Y50β‐F45β segment in the C′ strand. The net consequence of this movement is that the β sheet has a lesser degree of left‐handed twist at the C″ edge. This may be a unique feature of the Kabat Vβ subgroup I, and will probably impact the antigen recognition process. The shift of the Vβ C″ strand in an opposite direction, toward the BED face, has recently been reported for another TCR. Therefore, considerable variability in structure may exist among TCRs in this region (Housset et al., 1997).
CDR3 residues critical for MHC‐restricted antigen recognition map centrally on the molecular surface of the V module
The importance of CDR3 residues in α and/or β subunits for T cell recognition has been well documented experimentally (Engel and Hedrick, 1988; Danska et al., 1990; Lai et al., 1990; Taylor et al., 1990; Kabat et al., 1991; Casanova et al., 1993; Ehrich et al., 1993; Kasibhatla et al., 1993; Kelly et al., 1993; Luescher et al., 1995). To examine the role of individual CDR3α and CDR3β residues on T cell recognition by the N15 TCR, we constructed a series of alanine‐scanning point mutants which differed by a single amino acid at one or another of the CDR residues. Upon transfection with CD8αβ co‐receptors into 58α−β− T cells, the mutant TCRs were tested for their ability to alter peptide‐triggered IL‐2 production (Chang et al., 1997). Critical residues which diminished peptide antigen recognition by 1000‐10 000, in molar terms, were identified in both N15 Vα (E94A, Y102A and K103A) and Vβ (R97A, W98A and D100A) CDR3 loops. The Vα CDR3 residue, N101A, reduced recognition 100‐fold. Figure 5 gives a GRASP surface representation of the N15αβ heterodimer, as viewed from the perspective of the peptide‐MHC. As is readily apparent, these mutations, which dramatically reduce immune recognition by N15, are centrally placed at the surface interface of the V domain module. Note that not all surface exposed residues alter such recognition function, given that Vβ residue E105 has no effect on the ability of N15 to recognize VSV8 in the context of Kb. It is tempting to speculate that the depression on the N15 surface around E94α will accommodate the arginine residue at the p1 position of VSV8 peptide, the hydrophobic W98β and Y102α will interact with the p4 valine residue, and the upward pointing R97β will H‐bond to glutamine at p6. The distances of 13‐16 Å between the α‐carbons of E94α and R97β and the α‐carbons of Rp1 and Qp6 are consistent with this notion.
Rigidity of the C domain module at the Cα‐Cβ interface
As mentioned earlier, the Cα domain adopts a peculiar fold, deviating substantially from the conventional Ig‐like domain. However, its relative orientation to the Cβ domain remains the same in all three TCR structures studied to date, two in the N15‐Fab crystal and one in the 2C crystal as shown in Table II. This is largely because Cα maintains a regular ABED β sheet, facing the Cβ domain's ABED surface to engage in extensive sheet‐sheet packing. Table III lists the calculated buried surface areas for each domain‐domain interface in all of the known TCR structures. As is evident from this analysis, the buried surface area of the Cα‐Cβ TCR interface is the largest (2 400‐2 375 Å2) when compared with the other domain‐domain interfaces.
The qualitative nature of contacts is also distinctive. Except for a few hydrophobic contacts at the one corner of this C module, involving a cluster of aromatic rings (CαF161, CαW183, CβF130 and CβY179), the Cα‐Cβ interface is predominantly composed of hydrophilic interactions, as predicted by Mariuzza and co‐workers (Bentley et al., 1995) and noted by Garcia et al. (1996), with a skewed distribution of acidic residues in Cα and basic residues in Cβ. We further observe that these interactions cover a very broad surface area. Multiple polar interactions distributed over a large contact area will serve as a stabilizing force to rigidify the C module with respect to the Cα‐Cβ interface. This is in striking contrast to the V module where Vα‐Vβ dimerization is mediated largely by hydrophobic interaction, as discussed in the preceding section. The non‐polar nature of these latter interactions makes the disposition of the Vα and Vβ modules more malleable. From the perspective of antigen recognition, an adaptable V module, which can better undergo Vα‐Vβ displacement, may be more able to accommodate ligands of varying characteristics, as has been shown in the case of Fab VH‐VL interactions upon ligation by antigen (Stanfield et al., 1993; Wilson and Stanfield, 1993). On the other hand, the much more rigid C module may function to maintain the integrity of the heterodimer as well as to aid in physically complexing with the other components of the TCR machinery.
Conservation of Cα residues in pTα interacting with the Cβ interface
One extremely interesting feature of the Cα‐Cβ domain pairing relates to certain critical Cα residues which are preserved in pTα, a 30 kDa glycoprotein whose expression is restricted to early CD4−CD8− double negative (DN) T lineage cells (Saint‐Ruf et al., 1994; Bruno et al., 1995; Wilson and MacDonald, 1995; Ramiro et al., 1996). pTα is thought to function as a surrogate for the mature TCR α chain. In such progenitor cells, the pre‐TCR, a disulfide‐linked heterodimer of a functionally rearranged TCR β chain and pTα non‐covalently associated with the CD3 components, is expressed. During development, the DN to double positive (DP) (CD4+CD8+) thymocyte transition is induced by an as yet unknown ligand, which binds to the cell surface β‐pTα pre‐TCR complex. This differentiation then results in formation of mature type αβ TCR heterodimers on DP thymocytes (Shinkai et al., 1993; Fehling et al., 1995; Levelt et al., 1995; Xu et al., 1996; Koyasu et al., 1997). The pTα ectodomain in man and mouse is 79% identical, consisting of one Ig‐like domain with a short connecting peptide containing the cysteine that forms the interchain disulfide with the TCR β chain (Del Porto et al., 1995).
Although there is only 12% identity between pTα and Cα in man and mouse, the Cα residues involving important polar interactions with the Cβ domain are all conserved in the pTα sequences (Figure 6A). These include a complex salt bridge between CαD145 and CβR197, and H‐bonds between CαD164 N and CβY179 OH, CαT166 O and CβR195 NH2, CαS177 O and CβR197 NH2, and CαW183 NE1 and CβE181 OE2 (Figure 6B). Given that the conserved structural elements account for close to half of the pTα‐Cα amino acid sequence identities, it is very obvious that the heterodimer interface of pTα‐Cβ will be extremely similar to that of Cα‐Cβ. More importantly, this wide conservation from mouse TCR to human TCR and from pre‐TCR to mature TCR, in both species, indicates that the rigid C module is a structural element important for immature, as well as mature, T cell function.
TCR‐H57 Fab interaction site involves the well‐structured Cβ FG loop
The binding of the H57 Fab to the TCR surface is largely restricted to an epitope residing within the FG loop of the Cβ domain. This result is consistent with molecular mutational studies (K.Karjalainen, manuscript in preparation). Collectively, four structure determinations have been made of this loop; two in our N15 TCR‐Fab crystal, one in the β chain structure (Bentley et al., 1995) and one in the 2C TCR structure (Garcia et al., 1996). In every case, the sequence of the loop is identical and every conformation of the 12 residue loop is surprisingly well defined, lacking the floppiness which might be anticipated in a loop of this size. A comparison of the 2C and N15 loops is given in Figure 7A. Intricate internal interactions within the loop limit its flexibility. These include a mini hydrophobic patch formed by W225β at the tip of the loop as well as L219β and P232β at the ‘root’ of the loop (Figure 7B). L219β and P232β are the last H‐bond‐paired residues on the F and G strands, respectively. In this way, the mini hydrophobic patch links to the hydrophobic core of the domain. Moreover, there are multiple H‐bonds within the loop, most notable among which are the bifurcated H‐bonds between the side chain atom NE1 of W225β and two carbonyl oxygens, one preceding and the other following the rigid P230β. Apparently, these polar interactions further anchor the W225β at the base of the loop. When the unligated 2C TCR, and the H57‐bound N15 TCR are superimposed based on the Cβ domain framework, the FG loop shows several major differences between the ligated and unligated states (Figure 7A). The striking feature is that the residue W225β maintains the same conformation, while the two ‘front’ corners of the loop differ substantially. On one side, E227β flips almost 180°, from being pointed inward to the loop in the unligated state to outward in the ligated conformation. In complex with the Fab, its carboxyl group reaches deeply into the bottom of the Fab‐binding pocket, forming a charged H‐bond to R50 within CDR2 of the Fab‘s heavy chain (Figure 7C). This binding mode is somewhat reminiscent of that of a lysozyme‐Fab complex (Amit et al., 1986) in the sense that the side chain of one glutamine (Q121) from the antigen penetrates deeply into the antibody combining site. On the other corner, the main chain around E221β also deviates quite substantially in our ligated structure (Figure 7A). There are two salt bridges between the N15 and Fab here: one involving K224β in the N15 TCR and D51 in the H57 light chain's CDR2 loop, and the other involving E221β in the N15 TCR and K30 in the light chain's CDR1 loop (Figure 7C). Note also the very different conformation of the E221β side chain in the Fab‐ligated N15 TCR (Figure 7B). There is one additional salt bridge outside the FG loop of the Cβ domain, which involves D14 of the N15 Vβ domain and K53 in the CDR2 loop of the Fab light chain. Since in the 2C TCR structure only residues 218‐220 are involved in crystal packing (Garcia et al., 1996), and no FG loop residues are so affected in the N15 structure, we conclude that the conformational change described here upon N15‐Fab interaction may reflect an induced‐fit mode of binding. As listed in Table IV, there are also multiple hydrophobic contacts between the TCR epitope and the H57 Fab including a number of aromatic rings such as those of F32 and Y34 from the light chain and W33, Y35 and F98 from the heavy chain. This set of hydrophobic contacts strengthens the aforementioned salt bridges and is consistent with the frequency of aromatics observed by others in antigen combining sites (Mian et al., 1991). The total buried molecular surface area in this complex A TCR‐Fab interaction is 1460 Å2 (720 Å2 for N15 and 740 Å2 for H57), consistent with the range of values observed for other intact protein antigen‐Fab interactions (Stanfield et al., 1993; Wilson and Stanfield, 1993; Padlan, 1994). One unusual feature of the N15‐H57 Fab interaction is the preponderance of contacts made by the light chain (430 Å2 buried surface) rather than heavy chain (310 Å2 buried surface).
Analysis of the two complexes within a single asymmetric unit shows that each Fab molecule binds to the N15 TCR using essentially the same interacting residues in a given pair, but with significantly different overall orientation, as described above. Figure 7D depicts how the two H57 Fab fragments orient differently relative to N15 when the two TCRs are overlaid on top of each other based on Cβ domain superposition. As can be seen, there is a 19° rotation around the FG loop of the TCR Cβ domain. While the major binding surface between the TCR Cβ FG loop and the Fab remains essentially the same, there are differences in the number of the atomic contacts in complex A versus complex B, as defined in Table IV. This finding implies that the same immune receptor may dock somewhat variably to a single ligand and that slight differences in the binding interaction can result in rather large rotations at the distal C region of the Fab ‘lever arm’. Antibodies use several mechanisms to adopt substantially different shapes in order to recognize antigens. These include utilization of diverse CDR loops varying in sequence and size, alteration of the elbow angle between the V module and C module, variation of VL‐VH pairing, and exploitation of hinge flexibility between Fab and Fc modules (Huber and Bennett, 1987; Stanfield et al., 1993; Wilson and Stanfield, 1993; Padlan, 1994). Compared with the unligated Fab H57 structure, the elbow angle of H57 Fab does change ∼20° upon binding to the N15 TCR (data not shown). Nevertheless, both Fab molecules in the two complexes undergo roughly the same elbow angle change, despite their distinctive orientation with respect to the TCR epitope.
A putative CD3ϵ‐binding region
The overall shape of the TCR C domain module is remarkably asymmetric. The Cβ domain is ∼55 Å in overall length while that of the Cα is only 40 Å, as viewed in Figure 2. The Cβ domain bends more acutely towards the Vβ domain compared with the angle formed between the Cα and Vα domains. Cβ also has an unusually long and well‐structured FG loop protruding down from the Vβ‐Cβ interface. About half of the Cβ domain's ABED sheet is surface exposed, yet uninvolved in contact with the corresponding Cα domain. This asymmetry creates a cave‐like structure or cavity underneath the β chain as shown in the right lower corner of Figure 2. This cavity measures ∼25 Å in depth, 20 Å in height and 25 Å in width (Figure 8). The partially exposed ABED β sheet of the Cβ domain forms an extensive ceiling (Figure 2). The CD loop and EF loop of the Cα domain, as well as the glycans attached to CαN185, CαN121 and CβN186, form one side wall, while the FG loop of Cβ and the glycan emanating from CβN236 form the other side wall of the cavity. The glycans appear to project outward and hence, probably will not occlude the cavity. In contrast, the AB loop of the Cα domain projects into this cave. The floor of the cave is presumably formed by the plasma membrane on the T cell surface. It is noteworthy how the interchain disulfide bond between Cα 213 and Cβ 247 is positioned below the Cα domain (Figure 2), leaving the cavity unobstructed as the TCR αβ heterodimer projects from the T cell surface membrane. Since the TCRγ subunit is the analogue of the TCRβ subunit in γδ TCRs (Bentley et al., 1995; Krangel et al., 1987), and is predicted to contain an equivalent insertion in the corresponding constant domain loop (Bentley et al., 1995), a similar cavity must exist in the TCR γδ heterodimer as well.
What might be the utility of this cavity for TCR structure and function? It seems most likely that it serves to physically link the CD3 components with the TCR αβ heterodimer. The CD3 subunits are known to function as both structural components of the TCR as well as mediators of signal transduction. The importance of their structural role is emphasized by the fact that in the absence of the CD3ϵ subunit, the TCR cannot be expressed on the cell surface (Sussman et al., 1988). The signaling role of the CD3 subunits was first identified in studies showing that anti‐CD3ϵ mAb could either trigger (van Wauwe et al., 1980) or inhibit activation (Reinherz et al., 1980), depending on the form of the antibody utilized and its ability to cluster receptors. Subsequent studies showed that the various CD3 subunits contain immunoreceptor tyrosine‐based activation motifs (ITAMs), which become phosphorylated by TCR triggering and allow recruitment of SH2‐containing tyrosine kinases to the TCR (Reth, 1989; Irving and Weiss, 1991; Letourneur and Klausner, 1991; Romeo et al., 1992; Weiss, 1993). The αβ TCR can be viewed as four associated pairs of dimers: TCR αβ, CD3 γϵ, CD3 δϵ and CD3 ζζ (Koning et al., 1990; Manolios, 1991). The TCR αβ and CD3 ζζ dimers are the only ones that are disulfide‐linked, and both CD3 δϵ and CD3 γϵ are stable in non‐ionic detergents. With regard to interaction between TCR αβ heterodimers and CD3 components, CD3ϵ is by far the strongest.
Several general models of the functional TCR have emerged (Blumberg et al., 1990; Koning et al., 1990; de la Hera et al., 1991; Kappes and Tonegawa, 1991). Most of these offer a view in which two CD3ϵ molecules are present: one associated with TCRβ and one associated with TCRα. Our structure shows that if two CD3ϵ molecules are present per TCR, then only one of these occupies the cavity within the TCR C domain module, leaving the other exposed. While chemical crosslinking studies imply that CD3γ in man and mouse is in close proximity to the TCRβ subunit, the position of CD3δ relative to the αβ heterodimer is less certain (Brenner et al., 1985; Koning et al., 1990). We predict that the CD3γϵ dimer is oriented such that its CD3ϵ molecule is in the cave.
Although, as yet there is no direct structural information on the localization of CD3ϵ, we propose that this cavity within the C module accommodates the CD3ϵ subunit for the following reasons. First, binary complex assembly studies demonstrate that the TCRβ‐CD3ϵ association is mediated by their ectodomain whereas, in contrast, detectable associations involving TCRβ‐CD3δ, TCRα‐CD3ϵ and TCRα‐CD3δ are all mediated through their transmembrane interactions (Manolios et al., 1994). Second, CD3ϵ is an essential component of both T cell and pre‐T cell receptors (Malissen et al., 1995). The nature of the interaction site between CD3ϵ and TCR αβ proteins, predicted from our structural analysis, would anticipate this central structural role for CD3ϵ. As discussed earlier, the pre‐TCR retains the same Cβ domain, and pTα is homologous to Cα. Hence, the pTα‐Cβ module is likely to be similar in structure to the Cα‐Cβ module, particularly at the dimer interface. In addition, the predicted size of the Cα domain's AB loop (five residues), with a single positively charged side chain in all homologues, as well as the conservation of six residues in the neighboring A and B strands (Pro, Leu and Asp; Cys, Leu and Asp, respectively), make it likely that these structural features will be maintained. Consistent with this notion is the observation that the conserved Cα glycosylation site at Asn185 is replaced by a glycosylation site at the predicted end of the C strand of human and mouse pTα, which should occupy space adjacent to that occupied by the Cα Asn185 glycan. In principle then, the Vβ‐Cβ‐Cα TCR module should be very similar to the Vβ‐Cβ‐pTα pre‐TCR module in so far as making association with the CD3ϵ signaling component. Third, the murine CD3ϵ subunit consists of 87 residues in its extracellular segment and has Ig‐like characteristics (Gold et al., 1987), which can be readily folded into a small C2 type Ig domain, a typical example being domain 2 of CD4 (Wang et al., 1990). CD3ϵ is also the only CD3 subunit with an Ig‐like ectodomain in the TCR machinery of man or mouse that is non‐glycosylated (Gold et al., 1987). This makes the CD3ϵ subunit the only subunit that could readily fit into a cavity of this size. Fourth, the ectodomain of the CD3ϵ subunit has twice as many acidic residues as basic ones and is therefore negatively charged (pI = 4.5). The cavity, by way of contrast, is filled with more basic residues including the conserved R134α, which protrudes from the AB loop of the Cα domain. Hence, charge complementarity favors this notion. Note that the predicted pIs of CD3γ and CD3δ extracellular segments are more typical of ectodomains, being 8.76 and 5.82, respectively.
Implications for signal transduction
The precise functional role of the FG loop in Cβ is far from clear. However, it is likely to have an important impact on TCR structure and signaling for two reasons. First, this extensive loop will prevent lateral movement of the Vβ domain relative to the Vα domain, and thereby further rigidify this domain‐domain interaction. Hence, perturbation of the V domain CDR loops by peptide‐MHC binding can more easily transfer information to the CD3 signaling subunits via this rigid body. Second, although the mini hydrophobic patch of this loop (W225β, L219β and P226β) fixes the overall loop orientation relative to the Vβ and Cβ domains, local movement of the main chain atoms around residues E227β, E221β and D223β are permitted. Such perturbation may affect the interaction of Cβ with CD3ϵ or other CD3 components. The unique FG loop of the Cβ domain is widely conserved among sequences of mouse, rat, human and rabbit. Although in some species (fish, chicken and shark), the loop is much shorter (4‐7 residues versus 17 residues), a glycosylation site can be found in these foreshortened FG loops which may serve a similar function to that of the peptide in the homologues with longer FG loops (Rast and Litman, 1994).
One additional observation argues that V domain ligation may alter the quaternary structure of CD3ϵ within the C domain module or, at least, the strength of association between CD3ϵ and the Ti αβ heterodimer. Antibodies to CD3ϵ coprecipitate the CD3 components γ, δ and ϵ as well as the Ti αβ heterodimer in mature T cells (Meuer et al., 1983c; Reinherz et al., 1983). In contrast, under the same experimental conditions as used to immunoprecipitate with anti‐CD3ϵ mAb, all anti‐αβ clonotypic antibodies fail to detect the CD3 components (Meuer et al., 1983b,c). Assuming that physiological ligation of TCRs by peptide‐MHC, like the anti‐clonotypic mAbs, alters the interaction of CD3ϵ with the TCR cavity, this change may be a basis for initiation of signal transduction. Clearly, future mutational studies guided by the atomic detail provided herein will test the validity of this speculation.
Materials and methods
Data collection and processing
Protein production and crystallization were as previously described except that a mono Q column was used to further purify the H57 Fab (Liu et al., 1996). The crystals belong to the space group P21 (unit cell a = 74.7 Å, b = 122.3 Å, c = 115.8 Å and β = 108.0°) containing two complexes in one asymmetric unit with 50% solvent. Each of the data sets was collected using an MAR research imaging plate system at the Brookhaven National Synchrotron Light Source on beamline X12C, using one single frozen crystal under −165°C. The mosaicity is only 0.2°‐0.3° for native crystals. Raw data were integrated and merged using DENZO and SCALEPACK (Otwinowski and Minor, 1997). Further processing was performed with the CCP4 suite (Science and Engineering Research Council collaborative computer project 4; Daresbury Laboratory, Daresbury, UK) (CCP4, 1994).
The program packages AMoRe (Navaza, 1994) and X‐PLOR (Brünger, 1992) were used to carry out molecular replacement calculations. The search models were the 2C TCR (Garcia et al., 1996) and the Fv module of the partially refined H57 Fab structure (Lim et al., manuscript in preparation). Surface side chains were truncated where the sequence differed between 2C and N15 TCR. A two‐type (TCR and Fv)‐four‐body search protocol approach (Navaza, 1994) was taken and the normalized structure factors were used as experimental data. The solution was quite clear at 4 Å resolution. The correlation coefficient for the correct solution after fitting was 22.8, with the next highest peak value 16.3. Multiple domain rigid body refinement using X‐PLOR (Brünger, 1992) reduced the R‐factor from 54 to 50.1%. The result became immediately convincing when the models were displayed on graphics where the crystal packing was observed to be reasonable and the Fab‐TCR binding mode made biological sense. The constant module of the Fab was found at this stage by systematically changing the elbow angle in rigid body refinement. When a 20° elbow angle change was applied to the original unligated Fab model, after rigid body refinement the R‐factor dropped from 0.442 to 0.412 at 4 Å and the map began to show reasonable density.
MAD data were collected using a crystal in which all six TCR methionine residues were replaced by selenomethionine. Cell growth and crystallization conditions were as previously described (Liu et al., 1996) except that cells were preincubated for 10 h in GMEM medium containing 50 μg/ml seleno‐l‐methionine (Sigma #3132), followed by fresh medium replacement containing 2 mM Na butyrate and supernatant harvested after four days for protein processing. One single frozen crystal was used to collect three data sets at the inflection point, peak point and remote point respectively, with ‘Friedel flip’ geometry. The integrated data sets were local‐scaled together carefully with the program DSCALEAD (Rould, 1997) to obtain accurate dispersive differences. Difference‐Fourier maps were calculated using the initial phases from the preliminary refined model derived with molecular replacement. Nine of 12 Se sites showed up in both anomalous and dispersive difference maps, and the sites agree with the methionine positions in the model. The MAD data were treated as a special case of multiple isomorphous replacement (Ramakrishnan et al., 1993) and the nine Se sites were refined and used for phasing with MLPHARE in the CCP4 package (CCP4, 1994).
Model building and refinement
Initially the experimental electron density maps phased by the MAD data at 4 Å were used to rebuild the model to avoid any bias from the molecular replacement. The maps were first subjected to the density‐modification procedure incorporated into the program DM (Cowtan and Main, 1993). Multiple domain averaging, solvent flattening, and histogram matching were applied. Care was taken to gradually improve domain by domain transforms relating the two complexes so that the DM map was of quite high quality. Model building and refinement were alternated using the programs O (Jones et al., 1991) and X‐PLOR (Brünger, 1992). An alternative position and individual temperature factor refinement protocol was used. There were 10 rounds of model building and refinement. At the fourth round, the model phases started to be incorporated and at the sixth round, only model phases were used. During the refinement, a bulk solvent correction and the overall B factor were applied on the Fo. The resolution was extended gradually to 2.8 Å, and 2 Fo‐Fc as well as Fo‐Fc difference maps were used for further model building. The simulated annealing omit maps were used for checking and rebuilding. The seven glycans were seen clearly on the maps and could be built easily. NCS restraint for each individual pair of domains was applied in the refinement until the last several cycles. In the final model, four residues of the appended linker sequence following the interchain disulfide of the TCRα and β chains at the C‐termini are absent due to disorder. This final model contains 13 892 atoms and has Rfree 0.309 and Rwork 0.243 at 15‐2.8 Å, with good geometry (r.m.s. values for bond length and angle are 0.006 Å and 1.2°, respectively, and there is only 0.6% of the structure in the disallowed region on the Ramachandran plot generated from PROCHECK) (Laskowski et al., 1983). The statistical data are listed in the Table I. The coordinates have been deposited in the Protein Data Bank (access code 1NFD).
The authors thank Drs Ian Wilson and Don Wiley for providing 2C and A6 TCR coordinates, respectively, for purposes of comparison and molecular replacement. We thank Drs Stephen C.Harrison and Linda K.Clayton for helpful comments on the manuscript. We also thank Drs E.Dodson and K.Cowtan for helpful advice. This work was supported by NIH grants to E.L.R., J.C.S., S.G.N. and H.‐C.C.
- Copyright © 1998 European Molecular Biology Organization