A number of advances in classifying CDR-H3 structure have been made recently. These are detailed on this page.
Note that the residue numbering differs from the "AbM" numbering used in the input and alignment page; it is of the format ZZ:N+x or ZZ:C-y where ZZ is the loop identity, x is the displacement from the loop N terminus and y the displacement backwards from the C terminus.
The "kinked" and "extended" classes
Shirai et al and Morea et al have observed that the C-terminal conformation fell into two groups; kinked , in which residue H3:C-2 points inwards and H3:C-1 outwards and extended in which a standard beta-strand extended conformation is assumed, with residue H3:C-2 pointing outwards and H3:C-1 inwards, in contrast to the kinked group. The majority of structures fell into the kinked group.
A number of rules were formulated for deciding which conformation would be adopted:
a) If residue H3:C-1 is not Asp, a kinked structure is formed due to a hydrogen bond between the carbonyl O of residue H3:C-1 and the ring N of Trp H3:C+1
b) If residue H3:C-1 is Asp, but residue H3:N-1 is not Arg or Lys, the carboxyl of Asp H3:C-1, rather than the carbonyl of residue H3:C-2, forms the hydrogen bond with Trp H3:C+1, forming the extended structure.
c) If residue H3:C-1 is Asp, and residue H3:N-1 is Arg or Lys, the carboxyl of Asp H3:C-1 forms a salt bridge with residue H3:N-1 rather than a hydrogen bond with residue H3:C+1. This, together with the H3:C-2/H3:C+1 interaction in a), forms the kinked structure.
d) If residue H3:C-1 is Asp, and both residue H3:N-1 and H3:N-2 are Arg or Lys, the carboxyl of Asp H3:C-1 forms a salt bridge with residue H3:N-2 rather than H3:N-1, and the extended structure is formed.
The majority of structures conformed to these rules. There were two exceptions: 1mam and 1igi. The former has an Ala rather than Asp at residue H3:C-1, and so would be expected to form the kinked structure, according to a). However, it is extended. This is because there are two prolines in the 8-residue loop, which distort the structure. In 1igi, residue H3:C-1 is Asp and residue H3:N-1 is not Arg or Lys, so the extended structure would be expected. However, the carboxyl of Asp H3:C-1 forms a salt bridge with a lysine within the H3 loop, rather than with Trp H3:C+1, and the kinked structure is thus formed by the H3:C-2/H3:C+1 interaction in a).
The latest paper by Shirai et al describes
three sub-types of kink. No defining rules can be found for the most common,
trans or T, but for the other types, G or C, sequence-structure rules have
been postulated. The rules are:
a) If there is a Gly at H3:C-3, the G type kink will be formed. This
is true for all 7 examples where Gly is at H3:C-3.
b) If there is a Trp at H3:C-3, the C type kink will be formed. This
is true for all 5 examples. Using steric considerations they have also
postulated that this rule appies if there is any large hydrophobic at
H3:C-3 and a Gly at H3:C-2 (in the original
paper ) but this is based on only one observed case.
However, the converse - that these rules must be satisfied in order to
form the G or C type kink - is not true.
The kinked/extended rules generally work for predicting C-terminal structure.
However, they don't tell us anything about the centre of the loop.
Investigations were carried out here to try and work out a set of rules for
the centre of the loop. No clear rules were found; but instead, a conclusion
was reached that H3 loops prefer to form standard hairpin shapes unless they
are perturbed by interactions resulting from unusual residues at certain
positions in the other loops.
The "hairpin" group is populated by all 3 7-residue kinked H3s, 3 out of
the 5 8-residue kinked H3s, 3 out of the 5 9 residue kinked H3s, 8 of the
13 10-residue kinked H3s and 3 of the 7 12 residue kinked H3s. These
observations are used in the RMSD screen for
the final conformation. With two or three notable
exceptions , the 11-residue H3s do not show a well defined class,
but many are moderately (2.0-3.0A RMSD) similar to each other so the RMSD
screen is still of reasonable value.
The details are given below.
Due to the short length, 5-residue CDR-H3s do not divide clearly into
kinked and extended groups, but they have their own, quite well-defined, set
of sequence/structure rules. These have been used to refine screening of
5-residue H3s in WAM and so are described separately
here .
These fall into three classes, a kinked group (1cgs, 1tet, 1mlb, 1ghf),
an extended group (4fab, 1mrc) and an irregular structure, 3hfl. The
kinked/extended rules apply to the first two groups. In 3hfl, the irregular
structure (where a kink would be expected) is caused by the Asp H3:C-1
residue forming a salt-bridge with residue L2:N-4, unusually an Arg. This
forces the carbonyl of H3:C-2 away from Trp H3:C+1, preventing the kink
from being formed. The kinked group is fairly conserved, as the RMSD
table below shows:
Kink subtype
Further classification of H3 structure
5-residue loops
7-residue loops
| 1cgs | 1tet | 1mlb | |
| 1cgs | - | - | - |
| 1tet | 1.6 | - | - |
| 1mlb | 1.6 | 1.8 | - |
| 1ghf | 1.1 | 1.4 | 1.7 |
The kinked structures are shown below.
8-residue loops
There are again two broad classes for 8-residue loops: a kinked class, covering most structures (1vfa, 1fgn, 1mim, 1a6t, 1kem, 1b2w), a single non-kinked, irregular example (1mam) and two structures with a weak kink (1plg, 1cic). All should form the kink by the Shirai rules: however, the two prolines in 1mam distort the structure (Shirai et al).Kinked
The table below shows that all kinked examples except 1kem and 1b2w are highly conserved in conformation:
| 1vfa | 1fgn | 1mim | 1a6t | 1kem | 1b2w | 1plg | |
| 1vfa | - | - | - | - | - | - | - |
| 1fgn | 1.4 | - | - | - | - | - | - |
| 1mim | 1.2 | 0.8 | - | - | - | - | - |
| 1a6t | 0.9 | 1.5 | 1.3 | - | - | - | - |
| 1kem | 2.3 | 2.1 | 2.5 | 2.7 | - | - | - |
| 1b2w | 2.1 | 2.7 | 2.3 | 1.9 | 4.0 | - | - |
| 1plg | 3.7 | 4.2 | 3.6 | 3.4 | 5.5 | 2.0 | - |
| 1cic | 2.9 | 3.2 | 2.7 | 2.6 | 4.4 | 1.4 | 1.2 |
1kem has its loop pointing more upright than the others. Examination on Insight has revealed that this is due to the Trp residue at H3:N which blocks the normal conformation. Owing to backup information published by the Shirai group, it has been decided to incorporate this feature into the RMSD screen. 1b2w also has a slightly differing conformation in the middle of the loop (residues Pro H3:N+3 and Trp H3:N+4). There are two possible reasons - examination on Insight reveals that the Trp H3:N+4 sidechain occupies the place normally occupied by the backbone of H3:N+3; and the Pro H3:N+3 may force the backbone to adopt a certain conformation.
Two structures, 1plg and 1cic, are very similar in conformation (global backbone RMSD 1.2 angstroms), and show a "weak" kink. Both show Gly at the N terminal. Again, this is a quite well defined interaction and so it has been decided to incorporate it into WAM (see the sequence/structure page) for more details).
Within the close group, 1fgn and 1mim, as well as the new structure (not shown) 1jpt, are particularly close. All show Asp at the N terminal. Once again, this is a quite well defined interaction and so it has been decided to incorporate it into WAM (see the sequence/structure page) for more details).
The conformations are illustrated below. 1kem is in yellow, 1plg in green, 1b2w in magenta and the rest in red.
9-residue loops
Again, the majority are kinked. 1bbd forms an extended structure by the rules of Shirai et al, while 2fbj, 1igt, 1jhl, 2jel, 7fab and 1nfd all form true kinks. In addition, one structure, 1mfb, forms perturbed kink-like C- terminals.
Main kinked group
The table below shows that three of the structures, 2fbj, 1igt and 1jhl, are all conserved in structure. The H3 forms a standard U-shaped loop shape (see fig), more open than hairpin loops.
| 2fbj | 1igt | 1jhl | 1ar1 | 2jel | 7fab | 1nfd | |
| 2fbj | - | - | - | - | - | - | - |
| 1igt | 1.4 | - | - | - | - | - | - |
| 1jhl | 1.3 | 1.3 | - | - | - | - | - |
| 1ar1 | 1.6 | 1.5 | 1.6 | - | - | - | - |
| 2jel | 2.4 | 2.6 | 2.4 | 3.0 | - | - | - |
| 7fab | 2.3 | 2.6 | 2.4 | 2.7 | 2.4 | - | - |
| 1nfd | 4.9 | 5.3 | 4.6 | 4.8 | 3.2 | 4.5 | - |
| 1mfb | 1.8 | 2.1 | 2.2 | 2.0 | 2.4 | 2.6 | 4.5 |
Kinked structures with differences in the centre
2jel and 7fab both form kinks too, but show differences in the centre of the loop. 7fab, unusually, has an Arg in the L3 loop at position L3:C-1, which forms a salt-bridge with the sidechain of Asn H3:N and a hydrogen- bond with the carbonyl of residue H3:N+2, forming an unusual bulge in the N-terminus which subsequently affects the centre.
In 2jel, the difference is not so great but a salt bridge between the Lys at residue L2:N and the Glu at residue H3:N+3 perturbs the centre of the loop. This is the only H3 for which this is possible in terms of sequence. 2jel and 7fab are also rather tighter loops than the main group (fig).
Of all the true kinked structures, 1nfd is the most unusual. This has some similarities with 1kem in that the histidine at position H3:C-3 occupies the position that the apex would normally occupy, forcing a more upright conformation. This is further stabilised by hydrogen bonds between the histidine ring and the backbone.
The various kinked structures are illustrated below: the main group is in red, 2jel in yellow, 7fab in green, 1nfd in cyan and 1mfb in magenta.
Perturbed kinks
One structure, 1mfb, shows a perturbed kink, forming a rather shallower kink than the true kinks. This could well be due to a Gly at the first residue ( see here ), or alternatively it could be due to the sidechain of residue H3:C-3 pointing towards L3 rather than L2, something which Chothia has speculated may be important (Morea et al ). They suggested that this would happen if the L3/H3 interface was not sterically hindered, though there was no single determining residue. In this case, the L3 has a unique, non-canonical structure, which is likely to be the cause.
10-residue loops
There are a large number of 10-residue H3 loops, with the majority being kinked. Two (1eap and 1aif) form extended C-terminals, in accordance with the rules of Shirai et al. The kinked structures break up into different groups depending on the conformation of the centre of the loop, as shown by the RMSD table below.
| 1for | 1dba | 1igi | 1a4j | 1ay1 | 1cfv | 1clz | 1nbv | 1kb5 | 1clo | 1igf | 1rmf | 1bln | |
| 1for | - | - | - | - | - | - | - | - | - | - | - | - | - |
| 1dba | 1.4 | - | - | - | - | - | - | - | - | - | - | - | - |
| 1ig i | 1.1 | 1.4 | - | - | - | - | - | - | - | - | - | - | - |
| 1a4 j | 1.3 | 0.9 | 1.3 | - | - | - | - | - | - | - | - | - | |
| 1ay 1 | 1.7 | 2.0 | 1.9 | 1.3 | - | - | - | - | - | - | - | - | - |
| 1cf v | 1.8 | 1.8 | 1.7 | 1.4 | 1.8 | - | - | - | - | - | - | - | - |
| 1cl z | 2.0 | 2.3 | 1.8 | 1.9 | 2.4 | 1.6 | - | - | - | - | - | - | - |
| 1n bv | 2.5 | 2.3 | 2.2 | 2.1 | 2.4 | 2.0 | 2.0 | - | - | - | - | - | - |
| 1k b5 | 3.0 | 2.3 | 2,9 | 2.8 | 3.4 | 3.4 | 3,9 | 4.1 | - | - | - | - | - |
| 1cl o | 2.7 | 2.7 | 2.6 | 2.8 | 3.2 | 3.2 | 3.5 | 4.1 | 2.1 | - | - | - | - |
| 1ig f | 3.0 | 2.7 | 2,7 | 3.1 | 3.5 | 2,9 | 3.4 | 3.8 | 2.3 | 2.1 | - | - | - |
| 1r mf | 2.9 | 3.1 | 3.2 | 2.6 | 2.2 | 2.8 | 2.7 | 2.3 | 4.9 | 4.8 | 5.0 | - | - |
| 1bl n | 1.7 | 1.8 | 1.5 | 1.5 | 1.9 | 1.7 | 1.2 | 1.7 | 3.4 | 3.2 | 3,2 | 2.6 | - |
U-shaped structures
This is the most frequent group, with four definite members (1for, 1igi, 1dba, and 1a4j) and four less certain members (1ay1, 1bln, 1clz and 1cfv). The conformation is a U-shaped loop, rather like hairpin loops but lacking the even beta-structure (fig), although 1bln, 1clz and 1cfv are closer to a hairpin structure than the remainder. All satisfy the Shirai kink rules except for 1igi and 1a4j, which have a D at H3:C-1 but no R or K at H3:N-1. In both cases, the R or K at H3:N+4 takes the place of the N-terminal residue in forming the salt bridge.
Within the five close structures, the RMSD values suggest a further separation into three groups: 1dba, 1a4j and the new structures 1c1e and 1jgu (see the sequence-structure rules page for more); 1for and 1igi; and 1ay1. In the latter, the bulky W at H3:C-4, and the large number of aromatics in general, mean there is more of a steric hindrance problem, restricting the space available to the conformation.
Lid and finger structures
The three structures of 1igf, 1kb5 and 1clo are all fairly, though not very, similar (RMSD from each other 2 to 2.5 angstroms). They have a more perturbed structure, with a lid-like centre with a finger projecting upwards. Examination of the interactions, however, did not reveal a single defining interaction: in 1igf, the finger appeared to be held in place by residue Tyr H3:C-3 hydrogen bonding with Ser H3:N+1 and the carbonyl of Asp H3:N+3; in 1kb5 it appeared to be the sidechains of Asp H3:C-1 with Arg H3:N+2 and Tyr H3:C-3; and in 1clo it appeared to be the sidechain of Asp H3:N with the N-H groups of H3:N+2 and H3:C-3.
Each of these three structures also form unusual salt bridges with other CDRs, which could stabilise the structures. In 1kb5, Arg H3:N+1 interacts with Glu L2:C; a Glu at this position is unusual. In 1igf, Asp H3:N+3 interacts with Lys L2:N; this is the only instance in the 10 residue H3 loops where this combination is found. In 1clo, Arg H3:N+4 (unusual) interacts the Glu H2:C.
Irregular structures
Two kinked structures did not fit well into either class. 1nbv was found to form unusual contacts with H1 (H1 charged bonds to R?), while the cause of the unusual structure in 1rmf is due to a "sandwich" hydrophobic interaction between the Trp at H3:N+2, a Tyr at H2:N+4 and the His at L1:C. This is the only instance of a 10-residue loop with this combination, and also perturbs the H2 structure from the norm.
The kinked 10-residue CDR-H3 loops are shown below. The U-shaped loops are in red, the "lid and finger" structures in yellow and the irregular structures in green.
All known structures have a kinked C-terminal. However, as the RMSD table below shows, the 11-residue loops in general do not show the same degree of conservation found in smaller loops. There is a notable exception: 1frg, 1fpt and 1ejo, which most resemble a hairpin shape, are very close in conformation and have common sequence features, and so are dealt with separately as these rules are used in WAM when screening 11 residue loops.
As for the others, only three pairs (1fvc/1igc, 1igc/1mcp, and 2h1p/1nqb) have an RMSD from each other of less than 2 angstroms, and then only just (except for the 1.5 angstrom RMSD between 2h1p and 1nqb). There are three broad groups; firstly a group with a "twisted" loop conformation (1a6v, 1fvc, 1igc, 1ad9), and secondly a group with a finger at the apex (2h1p and 1nqb). There are also some irregular structures; 1axs, 1ncb and 12e8. One common interaction reveals itself; the low RMSD pair 1fvc/1igc share similar sequences (WGGDGFYAMDY, WGNYPYAMDY) and form some common hydrophobic interactions not observed in other 11-residue H3 loops. The unusual Trp at H3:N packs against a His at H1:C in both cases (His has moderate frequency of occurrence here), and also against a Tyr at H3:C-4. A further possible stabilising interaction is between the Tyr at H3:N+5 and the Tyr at L2:C-1; Tyr is fairly unusual at this L2 position. However, these interactions do not seem as well-defined as those observed for 1frg, 1fpt and 1ejo, so they have not yet been incorporated into the WAM sequence-structure rules.
| 1a6v | 1fvc | 2h1p | 1frg | 1igc | 1mcp | 1fpt | 1ncb | 1axs | 12e8 | 1ad9 | |
| 1a6v | - | - | - | - | - | - | - | - | - | - | - |
| 1fvc | 2.3 | - | - | - | - | - | - | - | - | - | - |
| 2h1p | 2,6 | 2.1 | - | - | - | - | - | - | - | - | - |
| 1frg | 2.3 | 2.1 | 2.1 | - | - | - | - | - | - | - | - |
| 1igc | 2,6 | 1.8 | 2.6 | 2.4 | - | - | - | - | - | - | - |
| 1mc p | 2.5 | 2.1 | 2.3 | 2.6 | 1.8 | - | - | - | - | - | - |
| 1fpt | 2.2 | 3.0 | 3.0 | 1.8 | 3.7 | 3,7 | - | - | - | - | - |
| 1ncb | 5.3 | 5,2 | 4.2 | 3.2 | 5,9 | 6.0 | 4.5 | - | - | - | - |
| 1axs | 4,2 | 5.0 | 5,9 | 5.2 | 5.4 | 5.7 | 4.2 | 7.4 | - | - | - |
| 12e8 | 3.0 | 3.2 | 2.3 | 4.5 | 3.8 | 3.6 | 2.4 | 3.3 | 5.9 | - | - |
| 1ad9 | 2.0 | 2.1 | 3.3 | 2.5 | 2.5 | 3.0 | 2.4 | 6.4 | 3.8 | 4.1 | - |
| 1nqb | 2.5 | 2.3 | 1.5 | 2.0 | 2.9 | 2.3 | 3.0 | 4.1 | 5.5 | 2.4 | 3.2 |
| 1ejo | 3.0 | 2.3 | 2.3 | 1.6 | 3.1 | 3.1 | 1.3 | 4.6 | 4.6 | 2.6 | 3.1 |
One structure, 1hil, has an ill-defined kink. However, upon binding antigen the true kink is formed, so the rules of Shirai et al are not violated
The 11-residue loops are shown below.
12-residue loops
The RMSD between the regular kinked examples is shown below:
| 1igm | 1vge | 6fab | 1iai | 1jrh | 1osp | 1dee | |
| 1igm | - | - | - | - | - | - | |
| 1vge | 1.7 | - | - | - | - | - | - |
| 6fab | 4.9 | 5.7 | - | - | - | - | - |
| 1iai | 5.8 | 6.7 | 1.7 | - | - | - | - |
| 1jrh | 4.6 | 5.3 | 2.1 | 2.2 | - | - | - |
| 1osp | 2.5 | 3.6 | 3.5 | 4.1 | 3.0 | - | - |
| 1dee | 2.1 | 2.9 | 3.7 | 4.4 | 3.0 | 1.6 | - |
| 1dsf | 5.7 | 6.1 | 4.7 | 4.3 | 3.2 | 6.5 | 4.4 |
Sequence / structure patterns With loops of this length it has been found that there are rules which define the conformation for just part of the loop, in a similar manner to the kinked/extended rules. This applies to the structures 1igm, 1vge, 1osp and 1dee. Like certain 5, 8 and 11 residue loops, these rules have been used within WAM and so are dealt with separately .
The remainder of 12-residue loops do not show well-defined rules and are described below.
a) U-shaped kinked
Two more structures, 6fab and 1iai, form a more hairpin-like U- shaped loop shape. Both of these have a Tyr at H3:C-4, which may, due to steric constraints, force the conformation. No other 12 residue H3s have a Y here. Alternatively it may simply be the default conformation formed when no other interactions are made, as 1jrh (no Y) also forms a conformation similar to these two, with the difference of a bulge at residue H3:N+2. This structure is predicted not to be kinked as the Arg at H3:N rather than H3:N-1 is predicted to form a salt bridge with Asp H3:C-1 (Shirai et al, 1996) ; however, the Arg at H3:N instead hydrogen bonds with His H3:C-4, leaving Asp H3:C-1 to bond with Arg H3:N-1, leading to a kinked conformation. This Arg/His interaction also occupies the space that a 6fab/1iai like conformation would occupy, forcing the bulge at H3:N+2.
b) 1ap2 (irregular structure)
This structure is predicted to be kinked by the Shirai rules, but forms an irregular C-terminus. This appears to be due to the Arg H3:N-1 residue forming a salt bridge with Asp H1:C-3 instead of Asp H3:C-1, the H1 residue being, unusually, charged. Additionally, the H3 loop is in close contact with the H3 loop of the second molecule in the unit cell.
c) 1mpa
This is extended, and is predicted to be so by the Shirai rules.
The 12 residue loops are shown below. 6fab and 1iai are in red; 1igm and 1vge in yellow; 1osp in green and 1ap2 in magenta.
Longer loops
There is more chance of flexibility in solution in the centre of the H3 for loops of lengths greater than 12, so it does not seem worth attempting to classify these structures.
Conclusions
For H3 lengths 7-10, the kinked H3 loops by default appear to have a preference for a U-shaped, fairly unperturbed, hairpin-like loop, with exceptions being caused by specific interactions caused by unusual residue types. A preference for hairpin-like loops was suggested in Shirai et al; they suggested that aromatics at certain loop positions (H3:N+2 and H3:C- 3) sharpened the hairpin definition; however, no correlation with aromatics at those positions could be found in the current investigation, and, indeed, in contrast to the results of Chothia and Lesk (1987), no sequence determinants could be found for these conserved structures. Thus, the rule appears to be default U-shaped loop, unless there are unusual interactions. It may be that, once more structures become available, other examples of the unusual structures for each length will be found which will make up canonical classes.
The 11-residue loops were varied in conformation so only broad classifications could be made; the 12-residue loops, however, did form two distinct conformations with two members each, with a possible rule for determination, and one outlier. It may become clear that the two conformations will make up canonical classes once more structures are released.
Last updated 21/10/02