Factors contributing to decreased protein stability when aspartic acid residues are in β-sheet regions

Abstract

Asp residues are significantly under represented in β-sheet regions of proteins, especially in the middle of β-strands, as found by a number of studies using statistical, modeling, or experimental methods. To further understand the reasons for this under representation of Asp, we prepared and analyzed mutants of a β-domain. Two Gln residues of the immunoglobulin light-chain variable domain (V_L) of protein Len were replaced with Asp, and then the effects of these changes on protein stability and protein structure were studied. The replacement of Q38D, located at the end of a β-strand, and that of Q89D, located in the middle of a β-strand, reduced the stability of the parent immunoglobulin V_L domain by 2.0 kcal/mol and 5.3 kcal/mol, respectively. Because the Q89D mutant of the wild-type V_L-Len domain was too unstable to be expressed as a soluble protein, we prepared the Q89D mutant in a triple mutant background, V_L-Len M4L/Y27dD/T94H, which was 4.2 kcal/mol more stable than the wild-type V_L-Len domain. The structures of mutants V_L-Len Q38D and V_L-Len Q89D/M4L/Y27dD/T94H were determined by X-ray diffraction at 1.6 Å resolution. We found no major perturbances in the structures of these Q→D mutant proteins relative to structures of the parent proteins. The observed stability changes have to be accounted for by cumulative effects of the following several factors: (1) by changes in main-chain dihedral angles and in side-chain rotomers, (2) by close contacts between some atoms, and, most significantly, (3) by the unfavorable electrostatic interactions between the Asp side chain and the carbonyls of the main chain. We show that the Asn side chain, which is of similar size but neutral, is less destabilizing. The detrimental effect of Asp within a β-sheet of an immunoglobulin-type domain can have very serious consequences. A somatic mutation of a β-strand residue to Asp could prevent the expression of the domain both in vitro and in vivo, or it could contribute to the pathogenic potential of the protein in vivo.

Aspartic acid generally would appear to be acceptable anywhere on a protein surface. However, statistical analysis of the distribution of amino acids in protein structures has shown that Asp residues are under represented substantially in β-chain regions (McGregor et al. 1987; Muñoz and Serrano 1994). Experimental studies of Minor and Kim (1994) and of Smith et al. (1994), though slightly different from each other, showed that the amino acid Asp has the least propensity to be found in β-sheets, when Gly and Pro residues are not considered. Similar results were obtained by theoretical calculations in a study that took only the effect of a residue's side chain steric clashes with its own backbone into consideration (Street and Mayo 1999) and when β-sheet propensities were modeled by a complex energy function (Finkelstein 1995). To explore in more detail the reasons for this biased distribution of Asp residues, we constructed and studied Q→D mutants of a β-domain, namely, an immunoglobulin light-chain variable domain (V_L) of human κ4 protein Len (Huang et al. 1997). In agreement with the above observations, the stability of the V_L-Len domain decreased significantly when Gln residues Q38 or Q89, located in a β-sheet, were replaced by an Asp residue (Fig. 1 ▶). To understand the reasons for the decline of stability upon introduction of the above Asp residues into a β-sheet of the V_L domain, we determined the structures of mutants that contained Asp38 or Asp89 and compared them to known structures of parent molecules. As a control, we also determined the influence on stability of aspargine substitution at positions 38 and 89.

Fig. 1.

Two orientations of the β-sheet from V_L-domain Len that contains residues 38 and 89 are shown in stereo. Only the main-chain atoms of the polypeptide are pictured along with the hydrogen bonds between them. The residues 38 and 89 are shown as alanines for clarity. Panel A focuses on residue 38 and panel B focuses on residue 89.

Results

Protein expression

Glutamine residues 38 and 89 in human κ4 V_L Len (Huang et al. 1997) were mutated to aspartic acid and aspargine residues in both the wild-type V_L-Len background (mutants Q38D, Q38N and Q89D, Q89N) and in a stabilized V_L-Len variant (Raffen et al. 1999), in which residue 27d had been changed from tyrosine to aspartic acid (mutants Q38D/Y27dD, Q38N/Y27dD and Q89D/Y27dD, Q89N/Y27dD). The Q89D and Q89N mutations also were introduced into a highly stabilized triple mutant V_L-Len, M4L/Y27dD/T94H (Pokkuluri et al. 2002), yielding quad mutants Q89D/M4L/ Y27dD/T94H and Q89N/M4L/Y27dD/T94H. While soluble protein was obtained from the Q38D, Q38N and Q89N single-mutant clones, no soluble protein expression could be observed from the Q89D single-mutant clone. This observation suggested that protein Q89D was too unstable to be expressed in a soluble form. Diminished expression or lack of expression of soluble proteins has been observed previously for unstable proteins (Schiffer et al. unpubl.). Expression of soluble V_L-Len-derived domains that contain the Q89D mutation was possible in the Y27dD and the M4L/Y27dD/T94H backgrounds.

Protein stability

The thermodynamic stability values of the wild type, Y27dD, M4L/Y27dD/T94H, Q38D, Q38N, Q38D/Y27dD, Q38N/Y27dD, Q89D/Y27dD, Q89N/Y27dD, Q89D/M4L/Y27dD/T94H, and Q89N/M4L/Y27dD/T94H proteins were determined by the guanidine hydrochloride (GdnHCl) denaturation method; the results are listed in Table 1. The stability value for the V_L-Len Q89D mutant, which was not possible to measure directly, as this mutant clone did not express soluble protein, was estimated based on how much the Q89D mutation reduced the stability values of the Y27dD and M4L/Y27dD/T94H variants. To estimate the stability value for the Q89D mutant, we assumed that the components of the ΔG and C_m values that are contributed by specific mutations are largely additive, a reasonable assumption because these substitutions are not at interacting sites of the molecule.

Table 1.

Stability of the mutants

	Cm [GdnHCl]	m (kcal/mol/M)	ΔGunf (kcal/mol)	ΔΔGaunf (kcal/mol)
Wild-type Lenb	1.76	−4.4 ± 0.4	7.7 ± 0.6	0
Y27dDb	2.37	−4.0 ± 0.2	9.4 ± 0.5	−2.7
M4L/Y27dD/T94Hc	2.71	−4.5 ± 0.1	12.1 ± 0.2	−4.2
Q38Dd	1.30	−3.0 ± 0.3	3.8 ± 0.3	2.0
Q38N	1.51	−3.7 ± 0.1	5.6 ± 0.1	1.1
Q38D/Y27dD	2.10	−3.5 ± 0.2	7.4 ± 0.4	−1.5
Q38N/Y27dD	2.20	−3.5 ± 0.1	7.6 ± 0.1	−1.9
Q89D/Y27dD	1.17	−3.9 ± 0.2	4.5 ± 0.3	2.6
Q89N/Y27dD	1.98	−4.0 ± 0.1	7.9 ± 0.1	−1.0
Q89D/M4L/ Y27dD/T94H	1.48	−4.4 ± 0.1	6.5 ± 0.2	1.2
Q89N/M4L/ Y27dD/T94H	2.52	−4.6 ± 0.2	11.5 ± 0.5	−3.3
Q89De	(0.55)	—	—	(5.3)
Q89N	1.39	−4.9 ± 0.2	6.8 ± 0.3	1.6

^a ΔΔG_unf = m^Len × C_m^mut − m^Len × C_m^Len.

^b From Raffen et al. 1999.

^c From Pokkuluri et al. 2002.

^d From Raffen et al. 1998.

^e No protein expression: C_m and ΔΔG_unf are calculated values, as explained in the text.

The Q38D mutation reduces the stability of both the wild type and Y27dD mutant proteins (Table 1). The C_m value for the V_L-Len Q38D and V_L-Len Q38N mutants are 1.3 M and 1.51, respectively. They are 0.46 M and 0.25 M less than the Cm value of the wild type. The introduction of the Q38D and Q38N mutations into the Y27dD construct (C_m = 2.37 M) reduces its C_m by 0.27 M and 0.17 M, respectively as is reflected in the C_m values of 2.10 M and 2.20 M for the double mutants Q38D/Y27dD and Q38N/Y27dD. The stability is affected more dramatically by the introduction of the Q89D mutation than by introduction of the Q38D mutation. The C_m value of the Q89D/Y27dD mutant is 1.17 M; therefore the introduction of Q89D into the Y27dD mutant reduces its C_m by 1.20 M. The introduction of Q89D mutation into the triple mutant (Cm = 2.71 M) reduces its C_m value by a similar amount, 1.23 M; C_m of the quad mutant is 1.48 M. If the C_m of the wild type were to be reduced by the same amount, the calculated C_m for the Q89D mutant protein would be 0.53–0.56 M, a value reflective of an unstable protein. The C_m values of Q89N/Y27dD and Q89N/M4L/Y27dD/T94H are 1.98 M and 2.52 M, respectively. Therefore, the introduction of Q89N reduces the C_m values of the parent molecule by 0.39 M and 0.19 M, respectively. This compares with the introduction of Q89N mutation into the wild-type protein, which reduces the wild-type value by 0.37 M to a C_m of 1.39 M.

Protein structures

The crystal structure of V_L-Len Q38D was determined in the presence of UO₂²⁺ ions. No crystals were obtained without UO₂²⁺ ions. Q38D was crystallized in the P6₃ (Table 2) space group with one monomer in the asymmetric unit. The protein also has a low association constant in solution (Raffen et al. 1998). In the Q38D structure, the "40 loop" (consisting of residues 38–43) is wider than it is in the native V_L-Len domain. The UO₂²⁺ interacts with the side-chain carboxyl group of Asp38 and the carbonyl oxygen of residue 42. The closest distance between Asp38 and the carbonyl oxygen of residue 42 is 2.84 Å. The χ1 and χ2 values for the Asp38 side chain in the Q38D mutant are −170° and −166°. With these χ values used for Asp38 modeled in the native Len structure (PDB code 1lve), the above closest distance is reduced by ∼0.3 Å, from 2.84 Å to 2.51 Å. Therefore it appears that the widening of the 38–43 loop in protein Len is required to accommodate an Asp residue at position 38.

Table 2.

Crystallographic data

	Q89D/M4L/Y27dD/T94H	Q38D
Space group	P212121	P63
a (Å)	62.9	65.8
b (Å)	104.8	65.8
c (Å)	42.4	48.1
VM (Å3/dalton)	2.8	2.4
R merge (last shell) %	6.1 (26.3)	7.4 (20.4)
Highest resolution (Å)	1.6	1.6
No. of reflections used with │Fo│>3σ│Fo│	34586	14410
Final R-factor	0.195	0.230
Free R-factor (10% data)	0.222	0.289
No. of refined protein atoms	1729	858
No. of solvent molecules	336a	105b
Rmsd bond length (Å)	0.005	0.009
Rmsd bond angle (°)	1.3	1.6
Mean temperature factor
For protein atoms (Å2)	17.6	14.3
For solvent molecules (Å2)	32.7	23.3

^a Includes two 2-propanol molecules.

^b Includes one Zn²⁺ ion and one UO₂²⁺ ion.

The decrease in stability was less when Asp was introduced at the equivalent residue 38 in κ1 Rei V_L domain (Chan et al. 1996); ΔΔG was 1.2 kcal/mol in Rei compared to 2 kcal/mol in the Len protein. A possible reason for the difference is that in protein Rei, residue 43 is Ala while in Len it is Pro. With the more flexible Ala residue instead of Pro, a wider loop is possible, that is, the wider loop can better accommodate the Asp side chain. Actually, the conformations of the "40 loop" in Q38D crystallized with UO₂²⁺ and the "40 loop" in wild type V_L-Rei are similar, as shown in Figure 2 ▶.

Fig. 2.

The conformations of the "40 loop" in Len Q38D (black) and in κ1 protein Rei (gray) from PDB code 1rei are shown. The hydrogen bonds are indicated by dashed lines.

The V_L-Len Q89D/M4L/Y27dD/T94H quad mutant is isomorphous with its parent triple mutant (Pokkuluri et al. 2001). The crystallographic data are shown in Table 2. In both structures, a noncovalent dimer is in the asymmetric unit. Both molecules are also dimers in solution. For the individual domains, the root mean square deviation of α carbons between the two mutants is 0.14 Å and 0.11 Å, respectively. Since modeling of Asp89 in the wild-type Len protein suggested some close contacts, we expected that the segment with Asp89 might not be well defined. Contrary to our expectations, the overall temperature factor of the Q89D/M4L/Y27dD/T94H structure is slightly lower than that of the parent triple mutant, and the segment that includes residues 86 to 103 is well defined.

To understand the details of the structural changes introduced by the Q89D mutation, we examined the φ, ξ, and χ angles in both monomers of the dimer for both the Q89D/M4L/Y27dD/T94H structure and the parent triple-mutant structure. Table 3 lists the significant differences in φ, ξ, and χ angles between the two structures near the mutation site. Changes occur in both monomers of the quad mutant compared to the triple mutant in the ξ values of residues Tyr96, and Ser97 and in the φ values of Phe98 and residue 89. Differences in χ2 values of both monomers occur for Gln90 and Tyr91 and for χ1 values of Phe98. However, there are no significant changes in the φ, ξ, or χ angles for residues in the other neighboring β-strand segment, namely residues 32, 33, and 34. The main-chain hydrogen bonds involving the above segments are not altered.

Table 3.

Differences in dihedral angles^adue to the introduction of Asp at residue 89

Dihedral angle	Corrb	Residue	mon1c	mon2c	diffd	diffe
φ	✓	89	6°	10°	3°	1°
ξ		89	8°	(4°)	1°	3°
φ		Tyr91	(1°)	6°	2°	3°
φ		Tyr92	8°	(3°)	2°	3°
ξ		Ser93	11°	(3°)	15°	7°
φ		Ser93	(4°)	6°	4°	2°
φ		Thr94	8°	(2°)	11°	5
ξ	✓	Tyr96	8°	7°	5°	4°
ξ	✓	Ser97	12°	9°	3°	0°
φ	✓	Phe98	19°	15°	2°	2°
χ2	✓	Gln90	8°	8°	1°	0°
χ2	✓	Tyr91	8°	11°	5°	2°
χ2		Tyr92	10°	(1°)	8°	3°
χ1		Ser97	(1°)	6°	2°	7°
χ1	✓	Phe98	16°	6°	1°	9°
χ1		Gln100	9°	(4°)	2°	3°
χ3		Gln100	13°	(2°)	2°	13°

^a The average difference between individual monomers within one structure is 3° for residues 86 to 103.

^b Significant changes observed in both monomers are marked by "✓"; values in brackets are not considered significant.

^c Difference between Q89D/M4L/Y27dD/T94H and triple mutant.

^d Difference between monomers in Q89D/M4L/Y27dD/T94H.

^e Difference between monomers in M4L/Y27dD/T94H.

Discussion

As expected from the previously reported statistical, experimental, and modeling studies, both of the Gln to Asp mutations at residues 38 and 89 located at the end and middle of β-strands respectively of V_L-Len reduced the stability of the mutant proteins relative to their parent molecules. Clearly the effects on stability of the Gln to Asp mutations at positions 38 and 89 of the V_L domain are different. Although both Q→D mutations reduce the stability of the V_L, the Q89D mutation in the middle of a β-strand reduces the stability by significantly more (ΔΔG = 5.3 kcal/mol) than does the Q38D mutation at the end of a β-strand (ΔΔG = ∼2.0 kcal/mol). The influence of Asn substitution is less; the Q38N and Q89N mutations reduce the stability of the wild-type protein by 1.1 kcal/mol and 1.6 kcal/mol, respectively. The above values exceed the ΔΔG value of 1.2 kcal/mol for Gln to Asp, and <0.3 kcal/mol for Gln to Asn substituted protein G variants reported by Minor and Kim (1994) and Smith et al. (1994). The differences on mutating residues at the 38 and 89 positions also are evident when the influences on stability of Asp and Asn substitutions are compared. The average reduction in stability is ∼0.6 kcal/mol at residue 38 and ∼4 kcal/mol at residue 89 (see ΔΔG values on Table 1). Our crystallographic investigation of the very destabilizing mutation Q89D of V_L-Len was only possible because the mutation could be introduced into a very highly stabilized triple mutant background, a construct previously designed.

Statistical studies on residue locations by McGregor et al. (1987) on 61 proteins found only 67 Asp residues in β-strands. Of these 67 Asp residues, 16 were located in the central region of a β-strand and 51 were located at the ends, where end is defined as the last two residues of a β-strand. Swindells et al. (1995) examined residues in 85 proteins and found only 112 Asp residues in β-sheet regions. Whereas McGregor et al. (1987) found that the rotomers of the Asp residues in β-sheet regions were about equally divided between t and g⁺ with a small proportion in g⁻ conformation, the t rotomer conformation was the most frequent (58%) in the data analyzed by Swindells et al. (1995). In our studies of V_L-Len, we found that both Asp38 and Asp89 are in the t rotomer conformation. With the t rotomer conformation, in general, the closest main-chain atoms are within the same chain segment as the carbonyl oxygen of the residue following the aspartic acid; and are on the adjacent chain segment from the carbonyl oxygen of the hydrogen bonded partner of the Asp+1 residue.

Residues 38 and 89 are both located in the two neighboring extended central strands of a four-strand, antiparallel β-sheet (Fig. 1 ▶). Residue 38 is at the end of a β-strand, residue 89 is in the central region of its β-strand; they are on opposite sides of the domain. Asp89 is relatively confined compared to Asp38. The peptide nitrogen of residue 90 is hydrogen bonded to the carbonyl oxygen of residue 97 on the neighboring β-strand, the peptide nitrogen of residue 39 is not hydrogen bonded because this residue is part of a turn. As shown below, the side chain of Asp38 does not get close to the protein backbone of the neighboring chain, while the geometry of the β-sheet forces the side chain of Asp89 to close proximity.

The closest approach of the Asp38 side chain is to the carbonyl oxygen of residues 39 and 42; but because this loop is not a tight β-turn, the modeled distances are relatively large: 2.5 Å and 2.9 Å, respectively. In the V_L-Len Q38D structure, the above distances measured 2.84 and 3.19 Å. Because of the size of the loop, there are no close contacts with neighboring side chains. The φ and ξ values of residue 38 differ in the various mutant Len protein structures we previously determined; the φ and ξ angles range from −118°, 93° in wild-type V_L-Len (Pokkuluri et al. 1998) to −145°, 127° in the triple mutant of V_L-Len (Pokkuluri et al. 2001). In the V_L-Len Q38D mutant, φ and ξ angles of −123°, 103° were found, which fall into the above range. The most likely χ1 and χ2 angles found in the Backbone-Dependent Rotomer Library (Dunbrack et al. 1997) at φ and ξ angles of −120°, 100° are −178° and −166°, close to the χ1 and χ2 values of −170° and −166°, respectively, observed in the Q38D mutant. It has to be pointed out that the conformation of Asp38 as determined in the Q38D structure is influenced by its coordination to the UO₂²⁺ ion.

The unfavorable steric interactions of Asp89 in the triple mutant background were modeled to occur with carbonyl oxygens of residues 90 and 97 and the side chain of Phe98. Without the change in φ and ξ angles as a result of the Asp89 substitution, the close contact in the triple mutant would be 2.5 Å for the distance from Asp89 OD1 to the carbonyl of residue 97, given the same χ angles for the Asp side chain. The distance between Asp89 OD1 and the carbonyl of residue 90 is 3.3Å in the modeled structure. Examination of the structures of the mutant and parent molecules showed that with the Asp89 mutation, the structure adjusted to accommodate this residue, with acceptable φ and ξ angles and nonbonded contacts with no obvious strain visible in the molecule. Table 3 shows that there are correlated changes in both monomers in the φ angles of residues 89 and 98 and in the ξ angles of residues 96 and 97. Correlated changes also were observed in the χ angles of residues Gln90, Tyr91, and Phe98, summarized in Table 3.

The φ, ξ values of residue Gln89 varied within restricted ranges (−149°, 129° to −143°, 142°) in previously determined structures of variants of the V_L-Len domain. The average values found for Asp89 (−150°, 133°) for the two independent monomers are close to the above ranges found for Gln89. The average observed χ1 and χ2 values are 170° and 49° respectively (Table 4). These χ values compare with −171.7°, 61.2° observed at φ, ξ values of −150°, 130° in the Backbone-Dependent Rotomer Library (Dunbrack et al. 1997) where they represent 20% of the conformations in the above φ, ξ interval.

Table 4.

Average dihedral angles of the quad and triple mutants near residue 89

	Quad		Triple		Quad			Triple
	φ	ξ	φ	ξ	χ1	χ2	χ3	χ1	χ2	χ3
Asp/Gln89	−150	133	142	140	170	49	—	170	−162	—
Gln90	−109	136	−113	138	71	83	22	70	75	29
Tyr91	−128	36	−131	35	48	−77	—	48	−87	—
Tyr96	−65	153	−66	160	−66	−13	—	−66	−14	—
Ser97	−152	148	−152	158	53	—	—	57	—	—
Phe98	−99	155	−116	156	−88	67	—	−77	70	—

In the quad mutant, the average φ and ξ values for residue Phe98 are −99°, 155°; these values are comparable to values observed in other Len structures. The χ1 and χ2 values for residue Phe98 are −88°, 67°, which are not favorable; these values compare to previously observed χ1 and χ2 values −77°, 70 in the triple mutant. In the Backbone-Dependent Rotomer Library at φ and ξ values of −100°, 150°, 74% of the residues have χ values of −66°, 98°.

The close contacts made by Asp89 in both monomers are with atom OD1 of Asp89, which is 3.05 Å, 2.94 Å (distances for first and second monomers) from residue 97 carbonyl oxygen and 3.10 Å, 3.12 Å from Phe98 side-chain atom CD1. The ∼3 Å distance between two oxygen atoms and the 3.1 Å distance between an oxygen and a carbon atom are close, but within reasonable limits of nonbonded interactions. However, the interaction of the negative charge on the carboxyl oxygen of Asp89 with the partial negative charge on the carbonyl oxygen of residue 97 is unfavorable. An unfavorable interaction of the carboxyl oxygen of Asp89 with the partial negative charge of the carbonyl oxygen of residue 90 also occurs; the distance between Asp89 OD1 and the carbonyl oxygen of residue 90 is 3.67 and 3.59 in the two monomers.

Although the introduction of Asp at position 89, in the middle of a β-strand of an immunoglobulin variable domain had a major effect on the stability of the protein, it did not cause a major structural perturbance. We do not know the effect of these substitutions on the stability of the unfolded state, therefore we have to discuss the lowering of stability based on the folded state. We find that the observed change in stability is influenced by several factors that include changes in dihedral angles, rotomers, close contacts between atoms, and mainly by the unfavorable charge-charge interactions between the negative charge of the Asp and the partial negative charge of the carbonyl oxygens. The importance of the charge-charge interaction is shown by comparison of the Asp and Asn mutants, the Asp residues are clearly more destabilizing (see Table 1). The neutral Asn residues have similar steric constraints resulting in similar rotomer preferences as the Asp residues, but do not carry a negative charge. Depending on the distance to a neighboring backbone carbonyl the negative charge on the Asp residue could result in varying degrees of destabilization.

A negative charge can alter stability by having both local and global effects. Though Glu affects the global charge distribution of the protein, its longer side chain places the negative charge further from the backbone carbonyls. The Q89D mutation appears to introduce a local repulsion with the protein backbone carbonyl oxygen because the introduction of Glu, a negatively charged residue with a longer side chain, at position 89 has only a relatively minor effect on stability (for the Q89E mutant ΔG = 6.3 kcal/mol; X. Cai, unpubl.). In a structurally related protein, myelin protein zero (MPZ), negatively charged Asp99 residue appears to be required at an equivalent position to Asp89 in Len. Asn substitution at this position leads to a disease state (Marques et al. 1999). In MPZ, positively charged residues Lys101 and Arg38 are in hydrogen bonding distances from Asp99 (Shapiro et al. 1996) and therefore compensate for the negative charge.

Our data are in agreement with the statistical work of McGregor et al. (1987) in which fewer Asp and Asn residues were found in the middle than in the ends of β-strands. We demonstrated that the introduction of Asp in the middle of a β-sheet (at residue 89) is greater than twofold more destabilizing than the substitution of Asp at the end of the β-sheet (at residue 38), the difference in the destabilizing effect of Asn at residues 89 and 38 is less, about a factor of 1.5.

The detrimental effect of Asp within a β-sheet on the stability of an immunoglobulin-type domain could be very serious. A somatic mutation to Asp could prevent the expression of the domain both in vitro and in vivo, or could contribute to the fibril or amorphous precipitate formation of the expressed mutant protein leading to a disease state (Stevens 2000).

Materials and methods

V_L mutagenesis, expression and purification

Construction of the vectors producing recombinant V_L protein Len and mutants derived from it has been described (Wilkins Stevens et al. 1995). Site-specific mutations of these V_L domains were produced with the MORPH Site-Specific Plasmid DNA Mutagenesis Kit (5 Prime to 3 Prime, Inc.) by recombinant PCR (Higuchi et al. 1988) or by the method of Kunkel et al. (1987). All mutations were verified by dideoxy sequencing (Sanger and Coulson 1975). Expression and purification of V_Ls were performed as described (Wilkins Stevens et al. 1995) except that the Escherichia coli host strain JM83 (Vieira and Messing 1982) was used to improve expression of certain V_Ls. Extinction coefficients for purified V_Ls were calculated from the amino-acid sequence with the Wisconsin Sequence Analysis Package (Version 8, September 1994; Genetics Computer Group). Light-chain residues are numbered and the complementarity determining regions (CDRs) are defined according to Kabat et al. (1991).

V_L stability measurements

Equilibrium guanidine hydrochloride (GdnHCl) denaturation of V_Ls was measured by following the increase in tryptophan fluorescence that occurs upon exposure of Trp35, which is highly quenched in the native form. Samples containing 1.5 μM V_L, 10 mM sodium phosphate, pH 7.5, and various concentrations of GdnHCl were incubated overnight at 25°C. Fluorescence was measured at 25°C with an SLM Aminco SPF-500C spectrofluorometer (SLM Instruments) with excitation at 295 nm and emission at 350 nm. Raw data were corrected for buffer fluorescence and the denaturation curves were analyzed by the linear extrapolation method using the equation derived by Santoro and Bolen (1988). Nonlinear least-squares fitting, performed with the program KaleidaGraph (Synergy Software), yielded values for ΔG_unf, the free energy of unfolding in the absence of denaturant; m, the free energy change per mole of denaturant added; and the statistical error associated with these parameters. The concentration of GdnHCl at the midpoint of the denaturation curve (C_m) was given by ΔG_unf/m. The decrease in stability, ΔΔG°_unf, was calculated relative to the wild-type protein by multiplying ΔC_m, the difference between the mutant and wild-type C_m, by the m value for the wild-type protein (Cupo and Pace 1983).

Structure determination

The Q38D mutant of Len was crystallized, by the hanging drop method, from 25% PEG monomethyl ether 550, 0.01 M zinc sulfate, and 0.1 M MES, pH 6.5 in presence of 1 mM uranyl acetate. Protein concentration was 10 mg/mL. Initially, diffraction data to 2.6 Å were collected using R-Axis IIc and later to 1.6 Å at Argonne National Laboratory Structural Biology Center's beamline 19ID at the Advanced Photon Source. The diffraction data were processed with Denzo (Otwinowski and Minor 1997) and merged and scaled using Scalepack (Otwinowski and Minor 1997).

The unit cell of Q38D contains one monomer per asymmetric unit (V_M = 2.4). The structure of Q38D was solved by molecular replacement with the program AMoRe (Navaza 1994) using the model of wild-type Len (PDB code: 1lve; atoms beyond CB removed for residue 38). A solution was obtained with a correlation coefficient of 42% and an R-factor of 44.2%. Rigid body refinement of the resulting model improved the correlation coefficient to 55% and R-factor to 38.7% for 10.0–4.0 Å data. This model was refined further with the program X-PLOR (Brünger 1992) using simulated annealing and positional and restricted individual B-factor refinement. The program CHAIN (Sack 1988) was used for manual rebuilding between cycles. The structure solution and initial refinement were performed using the R-Axis data. Synchrotron data were used in later stages of refinement. Data to 1.6 Å resolution were introduced in small steps. Data collection and final refinement statistics are presented in Table 3.

The Q89D/M4L/Y27dD/T94H mutant of protein Len was crystallized by the hanging drop method from 20% PEG 4000, 20% 2-propanol, 0.1 M sodium citrate, pH 5.6. The protein concentration was 10 mg/mL. Diffraction data to 1.6 Å were collected at Argonne National Laboratory Structural Biology Center's beamline 19ID at the Advanced Photon Source. Before flash cooling for data collection, crystals were soaked briefly in a cryoprotectant solution containing 30% PEG 4000 in addition to all other constituents (except the protein) of the mother liquor in which the crystals were grown. The diffraction data were processed with Denzo (Otwinowski and Minor 1997) then merged and scaled using Scalepack (Otwinowski and Minor 1997).

The unit cell of Q89D/M4L/Y27dD/T94H was found to be isomorphous to that of M4L/Y27dD/T94H (Pokkuluri et al. 2001). The model of M4L/Y27dD/T94H (PDB code:1eeq) was subjected to rigid body refinement against Q89D/M4L/Y27dD/T94H data; atoms beyond CB were removed for residues 89 in both monomers. The R-factor and R-free were 25.9% and 27.3%, respectively for 8.0–3.0 Å data. This model was refined with the program CNS (Brünger et al. 1998) using a maximum likelihood target for refinement; B-factor and bulk solvent corrections were applied. Manual rebuilding was performed using the program CHAIN (Sack 1988). Resolution of the data used in refinement was increased in several small steps to 1.6 Å. Data collection and refinement statistics are shown in Table 3. We checked if the refinement program forced good stereochemistry on the Q89D/M4L/Y27dD/T94H structure and found that the refined structure agreed with the electron density calculated with 3F_o−F_c coefficients after rigid body refinement but before refinement.

Ramachandran plot of the two structures contained 90% of residues in most favored regions with only Ala51 of both monomers in the disallowed region as observed previously (Steipe et al. 1992). Both the structures were deposited in the Protein Data Bank (codes: Q38D, 1efq; Q89D/M4L/Y27dD/T94M, 1eeu).

Fig. 3.

Stereo figure of the conformation of Asp89 showing the short contacts of its carboxyl group with the carbonyl oxygen of residue 97 and side chain of Phe98; these might contribute to the loss of stability.