protein Science (1995), 4:2073-2081. Cambridge University Press. Printed in the USA.
`Copyright 0 1995 The Protein Society
`
`Destabilizing loop swaps in the CDRs
`of an immunoglobulin VL domain
`
`_ _ _ ~ _ _ _______”~~
`LARRY R. HELMS AND RONALD WETZEL
`Macromolecular Sciences Department, SmithKIine Beecham Pharmaceuticals, King of Prussia. Pennsylvania 19406
`(RECEIVED May 15, 1995; ACCEPTED August 3, 1995)
`
`Abstract
`It is generally believed that loop regions in globular proteins, and particularly hypervariable loops in immuno-
`globulins, can accommodate a wide variety of sequence changes without jeopardizing protein structure or stabil-
`ity. We show here, however, that novel sequences introduced within complementarity determining regions (CDRs)
`1 and 3 of the immunoglobulin variable domain REI VL can significantly diminish the stability of the native state
`of this protein. Besides their implications for the general role of loops in the stability of globular proteins, these
`results suggest previously unrecognized stability constraints on the variability of CDRs that may impact efforts
`to engineer new and improved activities into antibodies.
`Keywords: complementarity determining regions; hypervariable loops; immunoglobulin stability; protein stabil-
`ity; Stern-Volmer plots; unfolding
`
`of “canonical structures” appear to be available for each CDR.
`For example, in an examination of the structural database avail-
`able in 1989, only four different loop conformations were iden-
`tified in the light chain crystallographic database for CDRl, and
`
`only three for CDR3 (Chothia et al., 1989). If these patterns d o
`suggest constraints on the sequence variability of CDRs, how-
`ever, it remains difficult from examination of evolved antibodies
`to either gauge the severity of these constraints or to understand
`their structural or functional basis.
`In this report we describe the preparation of a number of mu-
`tants of the immunoglobulin light chain variable domain REI,
`in which wild-type CDR sequences were replaced by loops of dif-
`fering sequence and length. These replacements cause dramatic
`reductions - and in some cases apparent elimination - of the net
`folding stability of the protein. The results have implications for
`the role of loops in determining the stabilities of immunoglob-
`ulins and other globular proteins and also suggest a folding sta-
`bility constraint on the evolution of CDR structures.
`
`In the structures of antibody heavy and light chain variable
`domains, complementarity determining region (CDR) loops of
`diverse sequence and conformation project from P-sandwich
`frameworks to determine the affinity and specificity of antigen
`binding. These diverse sequences arise in the development of an
`antibody response via a complex combination of gene rearrange-
`ments and somatic mutations (Branden & Tooze, 1991). Pro-
`tein engineers have transferred antigen binding activity from one
`framework to another by swapping hypervariable CDR loops
`between frameworks (Jones et al., 1986; Riechmann et al.,
`1988). In addition, totally artificial loop sequences have been
`introduced into variable domain frameworks to generate novel
`binding functions (Barbas et al., 1993; Fisch et al., 1994). This
`apparent tolerance of the immunoglobulin fold for both
`se-
`quence diversity and loop swaps is consistent with the view that
`the major determinants of structural stability of globular pro-
`teins lie in the sequence-specific formation and packing of reg-
`ular secondary structure elements (Shortle,
`1992; Matthews,
`1993), whereas surface loops are more forgiving of sequence
`changes (EL Hawrani et al., 1994).
`Although the CDR loops differ greatly in sequence, they are
`
`not infinitely variable in either sequence or conformation. Po-
`sitions with highly conserved amino acids are found within many
`CDR sequences (Kabat et al., 1991). Further, X-ray crystallo-
`graphic studies have demonstrated that only a limited number
`
`Reprint requests to: Ronald Wetzel, Macromolecular Sciences Depart-
`ment, SmithKline Beecham Pharmaceuticals, 709 Swedeland Road, King
`of Prussia, Pennsylvania 19406; e-mail: wetzelrb%phvaw.dnet@sb.com.
`
`Results
`Figure 1 shows a ribbon diagram of the REI V, domain high-
`lighting its three CDRs. Figure 2 lists the mutants examined in
`this work and shows the CDR sequence changes carried out to
`produce the proteins. The RGD-containing sequences inserted
`
`into the REI CDRs are either derived from loop regions of the
`snake venom disintegrins kistrin (Dennis et a]., 1989), barbou-
`rin (Scarborough et al., 1991), and echistatin (Can et al., 1988)
`or (in REI-RGD18 and REI-RGD26) were designed, unnatural
`2073
`
`Ex. 1044 - Page 1 of 9
`
`AMGEN INC.
`Exhibit 1044
`
`

`

`2074
`
`CDR 1
`
`Fig. 1. Ribbon diagram of the REI V,. domain from the X-ray struc-
`tural coordinates (structure 1 REI from the Brookhaven protein struc-
`ture database [Abola et al., 19873) with the CDRs highlighted. Drawing
`produced using MOLSCRIPT (Kraulis, 1991).
`
`sequences. Loop replacements were constructed in both CDRl
`and CDR3, yielding domains that by several criteria appeared
`to be stably folded proteins. Thus, each mutant was stably ex-
`pressed in Escherichia coli by secretion into the periplasm. In
`addition, the intradomain disulfide bonds of these mutants-
`like that of the wild type- form readily in the E. coli periplasm.
`Despite these indications of stability, however, and despite the
`
`L.R. Helms and R. Werzel
`
`solubility of these proteins in native buffer, further examination
`revealed significant destabilization in most of the mutants.
`The folding integrity of an immunoglobulin domain can be
`assessed (Tsunenaga et al., 1987) by the intrinsic fluorescence
`of a conserved tryptophan, which is dramatically quenched in
`the folded state by the close proximity of the intradomain di-
`sulfide bond (Epp et al., 1974). Thus, the fluorescence of the
`wild-type domain in native buffer at RT is quite low compared
`to the fluorescence in 4 M guanidine hydrochloride (Gdn-HCI),
`where the protein is unfolded. Table 1 shows the fluorescence
`yield for WT REI V,, in native buffer is only I 1 Yo of the fluo-
`rescence in denaturant, and that some of the mutants (RGDI2,
`RGDI, RGD34, RGD23, RCD18, RGD35, RGD32) alsoexhibit
`low fluorescence and are therefore also essentially completely
`folded in native buffer. In contrast, fluorescence ratios in Ta-
`ble l for the other mutants suggest they are substantially un-
`folded in native buffer under these conditions.
`For the mutants that d o exhibit folding stability in native
`buffer, stability was quantified by Gdn-HCI denaturation ex-
`periments to determine AAG,,,",,'s, the difference in free energy
`of stabilization between the wild-type sequence and each mu-
`tant. The wild-type domain at pH 7 exhibits an unfolding tran-
`sition (Fig. 3), which gives a midpoint for Gdn-HCI unfolding
`(C,,,) of 1.55 M and a corresponding free energy of stabilization
`of -6.8 kcal/mol (Table 1) (Hurle et al., 1994). These values
`are typical of globular proteins in this size range (Pfeil, 1981)
`and are also similar to values obtained for other V, domains
`(Ahmad & Bigelow, 1986; Tsunenaga et al., 1987).
`All of the loop insertion mutants examined prove to be sub-
`stantially destabilized compared to the wild type (Fig. 3; Ta-
`ble l). Some mutants, such as RGDI, RGD12, RGD23, and
`RGD34, are stably folded at pH 7 and RT in native buffer, but
`at the same time require significantly lower concentrations of
`Gdn-HCI, compared to wild type, to induce unfolding. Other
`mutants, such as RGD14, RGD17, and RGD22, are even more
`unstable, exhibiting significant fluorescence at pH 7 in native
`buffer at RT. Although it is difficult to rigorously determine the
`standard free energies of unfolding from the incomplete unfold-
`ing curves obtained for many of these proteins, because the na-
`
`A REI-VL WT
`REI -RGD17
`RE I - RGD4
`REI -RGD18
`REI-RGD12
`REI -RGD2 6
`REI-RGD23
`REI -RGD2 1
`REI -RGD2 2
`REI-RGD14
`REI-RGD15
`
`..YYCQOY------OSLP------YTFGQ . . .
`. . YYCQQYGTVSRVAKGDWNDDTSYTFGQ . . .
`. . YYCQ--GKISRIPRGDMPDDRS--FGQ . . .
`.. YYCQQ-----KGGRGDSGGK--YTFGQ . . .
`. . YYCQQY----RIPRGDMP----YTFGQ . . .
`.. YYCQ---KGGGGGRGDSK------FGQ . . .
`. . YYCQQ-----RIPRGDMP-----TFGQ . . .
`. . YYCQQ-----RIPRGDMP------FGQ . . .
`.. YYCQ------RIPRGDMP-----TFGQ . . .
`. . YYCQ------RIPRGDMP------FGQ ...
`.. YYCQQY------PRGD------YTFGQ . . .
`
`20
`
`B REI-VL WT
`in
`..TITCQA---SOD---IIKYLNWYQQTP..
`..TITCQASRIPRGDMPIIKYLNWYQQTP..
`REI -RGD3
`5
`..TITCQAKRARGDDM-IIKYLNWYQQTP..
`2
`REI -RGD3
`..TITCQARIPRGDMP-IIKYLNWYQQTP..
`REI-RGD34
`..TITCQAKIGRGDLV-IIKHLNWYQQKP..
`REI-RGD37
`..TITCQA---RGD---IIKYLNWYQQTP..
`REI -RGD1
`
`Fig. 2. Sequences of the loop swap mutants in this study.
`Residues underlined in the wild-type sequence indicate the
`CDR as defined by sequence homology within the immu-
`noglobulin family (Kabat et al., 1991). Most of the RGD
`sequences introduced are derived from the active site loops
`of the snake venom disintegrins kistrin (Dennis
`et al.,
`1989) (RGD4, 12, 14, 15,21,22,23,34, and 3 9 , echistatin
`(Gan et al., 1988) (RGD32). and barbourin (Scarborough
`et al., 1991) (RGDI7). REI-RGDI and RGD4 were de-
`scribed previously (Lee et al., 1993). Hyphens are in-
`cluded to accommodate the longest sequence in the lineup.
`A: Sequences of mutants derived from
`loop swaps in
`CDR3. B: Sequences of mutants derived from loop swaps
`in CDRl.
`
`Ex. 1044 - Page 2 of 9
`
`

`

`Destabilizing loop swaps
`
`Table 1. Properties of REI V,-RGD hybrid proteins
`
`
`
`
`
`Structurea Residuesb Fluorescence'
`
`
`
`
`
`2075
`
`A
`
`___
`
`c,,
`T n
`
`(M)C AAGUnff
`( oC)d
`
`
`
`+I2
`+8
`+5
`+4
`+3
`+2
`+ 1
`+ I
`0
`0
`+6
`+5
`+5
`0
`
`25
`
`37
`
`33
`
`34
`
`Name
`
`RGDI7
`RGD4
`RGDl8
`RGD 12
`RGD26
`RGD23
`RGD21
`RGD22
`RGD14
`RGDI5
`RGD35
`RGD32
`RGD34
`RGD 1
`WT REI
`
`0.78
`0.75
`0.25
`0.12
`0.85
`0.25
`0.66
`0.82
`0.84
`0.54
`0
`0.21
`0.20
`0.22
`0.32
`0.13
`0.42
`0.12
`0.5
`55
`I .55"
`0.11
` - - ~" - -~ -
`_
`_
`_
`_
`~
`~
`_
`_
`~
`-~ -
`- -~
`a Structural nomenclature for loop insertion mutants is as described by Wetzel (1988). Deleted wild-type residues are indi-
`cated on the left side of the slash, and the new sequence introduced into the gap is indicated to the right of the slash.
`Difference in REI length introduced by the loop swap.
`Intrinsic fluorescence at 350 nm upon excitation at 295 nm by a 4 pM protein solution at 25 "C relative to the signal in
`4 M Gdn-HCI.
`Temperature for 50% unfolding as described in Figure 4.
`for Gdn-HCI unfolding in 10 mM sodium phosphate, pH 7. The C,, value for REI WT is from Hurle et al. (1994).
`e C,
`'AGUnf for the mutant REI VL minus AGUnf for the WT, calculated at the C,, of the WT; see the Materials and methods.
`
`<O
`0
`0.13
`0.40
`
`0.28
`
`>6.8
`>6.8
`6.2
`5.1
`>6.8
`5.6
`>6.8
`>6.8
`>6.8
`6.8
`5.9
`5.4
`5.0
`4.6
`
`tive baseline is short or nonexistent, reasonable estimates for the
`V, defined by
`tions. In addition, the unfolded states of REI
`amount of destabilization by the loop swaps range from 5 to
`fluorescence dequenching are nonnative when analyzed by other
`greater than 7 kcal/mol (Table 1).
`spectroscopic probes. CD (Fig. 4) of REI-RGD4, a mutant that
`is about 50% unfolded in native buffer at pH 7 and RT by flu-
`Although fluorescence can be sensitive to even small structural
`changes, the observed tryptophan fluorescence dequenching ap- orescence measurements, shows that the
`molecule's conforma-
`pears to be associated with a radical disorganization of the
`folded structure of REI. Unfolding curves such as those shown
`in Figure 3 are normally associated with global folding transi-
`
`0
`
`-0.2 !
`
`0.0
`
`0.5
`
`I .o
`I .s
`[Gdn-HCI], M
`
`2.0
`
`I
`
`2.5
`
`Fig. 3. Mole fraction of unfolded protein (Fap) versus Gdn-HCI con-
`centration for denaturant unfolding of selected REI V, sequence vari-
`ants. 0 , wild type; H, RGD17; 0, RGDl8; A, RGD32; A, RGD34. The
`data set for wild-type REI V, is reproduced from Hurle et al. (1994).
`
`200
`
`220
`
`240
`260
`Wavelength, nrn
`) and RGD4 (- - - - -) REI V,.
`Fig. 4. CD spectra of WT (-
`Measurements were made at 25 "C in IO mM sodium phosphate, pH 7.4,
`on a Jasco model J-500-C spectropolarimeter equipped with a Macin-
`tosh computer-based data acquisition system. The wild-type data were
`collected on a sample of 0.7 mg/mL in a 0.02-cm-pathlength cell. REI-
`RGD4 mutant data were collected on a sample of 0.13 mg/mL in a 0.02-
`cm-pathlength cell. Each spectrum is presented as an average of 20 scans
`over the 265-190-nm range.
`
`Ex. 1044 - Page 3 of 9
`
`

`

`/
`
`2076
`
`"'1
`
`2.5
`
`2.3
`
`2. I
`
`% 1.9
`
`b
`
`I .7
`
`I
`
`3
`
`
`
`I .3
`
`1.1
`
`0.15
`
`0.x
`
`0 . m
`
`0.0s
`
`0.10
`Nal, M
`Fig. 5. Stern-Volmer plots of iodide quenching of the tryptophan in var-
`ious REI V,. sequence variants. All samples were at 10pg/mL in 10 mM
`sodium phosphate, 50 mM NaCI, pH 7.4. The wild-type sample also
`contained Gdn-HCI at 4 M. Fluorescence was read in a Perkin-Elmer
`MPF-66 spectrofluorometer in a thermostatted cell equilibrated at 22 "C.
`The Ks.v values obtained for the REI VI, variants plotted are: 0 , WT,
`10.1; A, RGD4, 1.5; 0, RGD14.4.6; 0 , RGDZI, 5.5; W. RGD22, 5 . 5 .
`
`L.R. Helms and R. Wetzel
`
`of Trp3S to the large, charged iodide quencher supports the in-
`these REI VI, variants are significantly un-
`terpretation that
`folded or misfolded in native buffer.
`In principle, inability to achieve a highly quenched fluores-
`cent (native-like) state might be attributable to the absence of
`the intradomain disulfide bridge. Figure 6 shows an SDS-PAGE
`gel of a wild-type REI VI. and series of loop-swap mutants. The
`gel shows that, in general, a nonreduced, alkylated sample mi-
`grates more rapidly through the gel than a reduced, alkylated
`
`sample of the same protein. This is consistent with the existence
`of a disulfide in the nonreduced sample (Pollitt & Zalkin, 1983).
`Such mobility shifts can be difficult to observe in small proteins
`like REI V,, and examination of Figure 6 shows the shifts to
`be small and variable. The absence of a shift in RGDI4 proba-
`bly simply reflects the limit of detection for this mutant. Other
`means were also used to confirm the existence of the disulfide
`in some of the mutant V, domains. An alkylated sample of
`REI-RGD4 gives a parent ion in mass spectrometry consistent
`with molecular weight expected for the oxidized mutant and in-
`consistent with the alkylated molecule (L. Helms & D. McNulty,
`unpubl. result).
`Although these proteins were purified by reverse-phase
`HPLC, in which they required acetonitrile concentrations for
`elution similar to that found for wild type, it is formally possi-
`ble that some of the mutants exist as disulfide-linked dimers or
`oligomers that would likely exhibit alternate fluorescence char-
`acteristics. However, Figure 6 shows there are no
`disulfide-
`linked dimers or trimers in these preparations. Neither was there
`any Coomassie blue-stained material at higher molecular weights
`in the gel (not shown), consistent with the absence of larger co-
`valent aggregates.
`Noncovalent aggregation could in principle also compromise
`proper interpretation of the fluorescence data, by offering a
`competing "folding" pathway for the proteins and thus exagger-
`ating their apparent instability as monitored by fluorescence de-
`quenching. In fact, some REI VI. point mutants associated with
`light chain deposition disease
`exhibit substantial unfolding-
`related aggregation by gel-filtration analysis (L.R. Helms & R.
`Wetzel, unpubl.). However, gel filtration analysis indicates no
`large aggregates to be present in solutions of RGD 15 and RGD
`21 (data not shown).
`The destabilization of REI VI- variants assessed by denatur-
`ant unfolding was confirmed for some of the mutants by thermal
`unfolding studies. Figure 7B shows that the wild-type domain
`at pH 7 undergoes a cooperative thermal unfolding transition,
`as monitored by fluorescence, with a T, of 55 "C. In contrast,
`RGD12, RGD23, and RGD34, three of the more stable of the
`
`tion under the same conditions, although not in statistical coil,
`
`is significantly different from the &sheet conformation exhib-
`ited by the wild type (the spectrum of the E. coli-produced wild
`type matches well with the published spectrum of the REI VI,
`derived from Bence-Jones protein [Brahms & Brahms, 19801).
`I t is possible that the observed fluorescence dequenching
`in
`is due to misfolding into an alternatively packed structure
`which Trp" is no longer proximal to the disulfide bond. Stern-
`Volmer analysis was conducted on several of the REI VI. mu-
`tants to determine the accessibility of the Trp to a small molecule
`quencher. Figure 5 shows Stern-Volmer plots for WT REI VI.
`unfolded in 4 M Gdn-HCI and a number of apparently unfolded
`
`mutants in native buffer. (Because the Trp in the native folded
`structure is already quenched by the disulfide, it is not possible
`to derive a Ks.,, for Trp in the native state of REI.) This anal-
`ysis shows that the tryptophan of the unfolded WT is fully ac-
`cessible (Eftink & Ghiron, 1981) to iodide quenching, giving a
`Ks.,, of 10. I . The mutants RGDl4, RGD21, and RGD22 are
`also sensitive to quenching, with Ks.v values around 5 . RGD4
`is more resistant to iodide quenching, with a Ks.v value of 1.5.
`tryptophan would be expected to yield a Ks.v
`A fully buried
`value of 0.2 or less (Eftink & Ghiron, 1981). The accessibility
`
`M
`
`RElwt
`RCDl
`R C W RCDM RGDIS RCD2l RGD34
`O R O R O R O R O R O R O R
`
`Fig. 6. Coomassie brilliant blue-stained, nonreducing
`SDS-polyacrylamide gel of various mutants of REI V,..
`Reduced (R) samples were treated 30 min at 37 "C with
`1 mM dithiothreitol in SDS-PAGE gel loading buffer, af-
`ter which iodoacetic acid was added to 5 mM and incu-
`bated an additional 30 min. Nonreduced (0) samples
`were treated only with 5 mM iodoacetic acid. Samples
`were run on a 12% acrylamide gel (Laemmli, 1970).
`
`Ex. 1044 - Page 4 of 9
`
`

`

`Destabilizing loop swaps
`
`I *
`
`0-
`
`v) Y
`
`20
`
`30
`
`40
`50
`Temperature, "C
`
`60
`
`70
`
`1.2 i
`
`I
`
`a 0.6
`Lr
`
`0
`a
`
`. .
`
`0.0
`20
`
`30
`
`50
`40
`Temperature, C
`
`60
`
`.,
`
`t
`
`70
`
`Fig. 7. Temperature dependence of binding and folding.
`WT REI
`V, ; 0 , REI-RGD23; A, REI-RGD34; 0, SKF#106760; A, echistatin.
`A: Temperature dependence of IC5,,'s for inhibition of binding of bio-
`tinylated fibrinogen to allbP3. Assays were conducted as described (Lee
`et al., 1993), with all incubations done at 30 "C, except for the cornpe-
`tition step, which was done at the temperatures indicated. Data points
`are the mean values of three or four independent determinations. The
`lack of significant temperature dependence to the inhibitory action of
`the peptide antagonist SKF#106760 (Samanen et al., 1991) and the snake
`venom disintegrin echistatin (Can et al., 1988) show that the tempera-
`ture dependence for RGD23 and RGD34 derives from a thermal effect
`on these REI mutants and not on the receptor. The wild-type REI VL
`exhibits no binding at any temperature (Lee et al., 1993). Other desta-
`bilized RGD-containing mutants also exhibit thermal dependent binding
`(data not shown). B: Thermal unfolding curves for REI VL domains.
`
`loop replacement mutants, have Tm's that are at or below phys-
`iological temperature, about 20 "C lower than the WT (Fig. 7B;
`Table 1). Interestingly, the thermal unfolding transitions of these
`mutants are mirrored in the thermal sensitivity of their recep-
`tor binding. Figure 7A shows that these mutants are good an-
`tagonists of fibrinogen binding to the platelet receptor allb&
`by virtue of their installed RGD sequences (Lee et al., 1993;
`Helms & Wetzel, 1994), but that binding affinity depends on as-
`say temperature. The transitions in the temperature dependence
`of the IC,, values exactly follow the independently measured
`unfolding transitions. Because unrelated receptor antagonists do
`not exhibit temperature-dependent activity (Fig. 7A), the ther-
`mal dependence is not due to effects on the receptor.
`The sequence RGD37 listed in Figure
`2 , containing the
`26-28 with the sequence
`replacement of wild-type residues
`
`2077
`
`KIGRGDLV, was generated in an REI VL sequence that con-
`tained two stabilizing mutations, Y32H (1.1 kcal/mol) and T39K
`(1.3 kcal/mol) (Frisch et al., 1994; H.-J. Fritz, pers. comm.).
`REI VL with the two stabilizing mutations exhibited a C, of
`2.25 M Gdn-HC1, indicative of a AAG with respect
`to WT
`of -3.1 kcal/mol. RGD37 exhibited a C, of 0.55 M Gdn-HCI,
`for a AAG with respect to REI(Y32H/T39K) of 7.0 kcal/mol
`destabilization. Thus, a sequence lacking prolines inserted into
`CDRl is actually more destabilizing than
`is RIPRGDMP in
`both REI-RGD34 and REI-RGD35. Table 1 shows that another
`CDRl insertion lacking prolines, KRARGDDM in REI-RGD32,
`is also highly destabilizing.
`
`Discussion
`We have been developing the immunoglobulin VL domain REI
`for use as a "presentation scaffold" (Lee et al., 1993; Helms &
`Wetzel, 1994)-a small
`domain expected to hold an installed
`peptide sequence into a defined conformation and thus help elu-
`cidate the receptor-binding conformations of peptides. We chose
`the integrin receptor-binding motif Arg-Gly-Asp (RGD) for in-
`stallation and study because of its small size and because of spec-
`ulations on the importance
`of conformation in its
`receptor
`binding selectivity (Ruoslahti & Pierschbacher, 1987). We chose
`an immunoglobulin VL domain as a scaffold, in part, based on
`the expectation that its CDRs should be particularly adept at ac-
`commodating sequence changes without incurring significant
`losses in folding stability.
`The data presented here, however, suggest that sequence re-
`placements in the CDR loops of immunoglobulin domains can
`be significantly destabilizing to the native structure of the do-
`main. The destabilizing effects are observed for insertions in two
`separate CDRs and are seen for such changes as a single point
`mutation, a short replacement that leaves the loop length un-
`changed, and insertions that increase loop size to lengths equal
`to or greater than those known to be accommodated at these po-
`sitions in immunoglobulins. The results have important impli-
`cations for antibody engineering experiments as well as for the
`general perception that loops are particularly resilient parts of
`proteins that are capable of accommodating many sequence
`changes.
`
`The nature of the nonnative state generated
`in destabilized REI VL mutants
`A number of experiments were conducted to investigate the de-
`gree of unfolding in the fluorescence dequenched state moni-
`tored in these studies. These proteins contain no large covalent
`or noncovalent aggregates, as assessed by nonreducing SDS-
`PAGE and native gel-permeation chromatography. The proteins
`contain the expected intramolecular disulfide bond, so that the
`absence of quenching of Trp in apparently unstable mutants can-
`not be ascribed to the absence of the disulfide. CD of the de-
`stabilized RGD4 in native buffer indicates secondary structure
`substantially different from the folded wild-type REI VL, but
`not statistical coil. Whatever the structure of this nonnative
`state, it must be substantially reorganized from the native be-
`cause the normally buried tryptophan is accessible to dynamic
`quenching by iodide. Further studies on the solution structures
`of the highly destabilized REI VL mutants are warranted, espe-
`cially because these molecules may hold important clues to the
`
`Ex. 1044 - Page 5 of 9
`
`

`

`2078
`
`L.R. Helms and R. Wetzel
`
`and 90-97. There are only a few long-range H-bonds in each
`case. In CDRl, the serine side-chain O H is involved in several
`H-bonds thqat would be lost if Ser were replaced by another res-
`idue; however, REI-RGD35 retains this serine and is just as de-
`stabilized as the related REI-RGD34, which lacks this residue.
`The other long-range H-bonds, at residues 3-26 and 29-68, may
`be lost if the inserted loop occupies a conformation that brings
`atoms 26:N and 29:N out of position with respect to the wild-
`type structure.
`Table 2 also lists a number of long-range H-bonds that may
`be lost in loop-swap mutants involving CDR3. The side chain
`of Gln 90 plays a very important role in orienting the wild-type
`loop, by lying in the loop plane and making H-bonds with resi-
`dues 93,95, and 97 (Table 2). This is a standard feature of CDR3
`loops (Tramontano et al., 1989). This may be important in the
`destabilization of some of the mutants reported here, because
`even if Gln 90 is retained (as it is in many of the mutants), loops
`of different sizes compared with WT may not be able to adopt
`these H-bonds.
`Although it is thus possible that lost H-bond contributions
`may account for some stability decrease in the mutants, further
`experiments and structural determination will be required to elu-
`cidate the mechanism of destabilization. X-ray crystal struc-
`ture analysis of REI-RGD34, in progress, shows that the basic
`immunoglobulin fold is maintained in the folded state of this
`destabilized mutant (B. Zhao, L.R. Helms, E. Winborne, S.
`Abdel-Meguid, & R. Wetzel, unpubl. results).
`
`~
`
`~
`
`~
`
`
`
`Table 2. Hydrogen bondsa in wild-type REI- VI-
`
`~-~ ~- _" ~- - ~
`~
`
`
`
`
`Donor Acceptor Molecule A
`
`3.34
`
`2.98
`3.14
`2.48
`
`3.36
`-
`
`-
`3.22
`-
`-
`3.02
`3.16
`-
`-
`2.71
`3.12
`3.28
`
`32:O
`
`93:O
`1 :OD2
`95 :O
`I:ODl
`
`90:OE 1
`
`3.02
`3.21
`3.29
`3.32
`2.88
`3.18
`-
`3.10
`2.94
`2.85
`3.16
`-
`____
`a Hydrogen bonds of acceptable bond lengths and geometries were
`identified in the coordinate set 1REI in the Brookhaven database (Bern-
`stein et al., 1977; Abola et al., 1987) using Insight 11"' (Biosym Tech-
`nologies, San Diego, California). The independent data for Molecule
`A and Molecule B reflect the two independent monomers in the single
`dimer of the unit cell (Epp et al., 1974).
`
`partially unfolded states implicated in light chain amyloidosis
`and deposition disease (Hurle et al., 1994).
`
`Loop swaps and destabilization
`A number of mutagenesis studies have been conducted suggest-
`ing that loop sequences have relatively little effect on the thermo-
`dynamic stability of globular proteins (reviewed in El Hawrani
`et al., 1994). Our observation of significant destabilization by
`loop replacements is not unprecedented, however. Two Staph-
`ylococcus nuclease mutants constructed by replacing a five-
`residue 0-turn with a @-turn sequence from concanavalin A were
`found to be destabilized by about 4.5 kcal/mol with a corre-
`sponding reduction in T, of about 20 "C (Eftink et al., 1991).
`A mutant constructed by replacing the loop region of chymo-
`trypsin inhibitor 2 with a nonapeptide sequence found in an
`a-helix in subtilisin Carlsberg is also significantly destabilized
`by about 8 kcal/mol (Osmark et al., 1993). In both of these
`cases, crystallographic analysis suggests that destabilization may
`be attributable to the loss of favorable long-range interactions
`in the WT between loop residues and residues elsewhere in the
`"host" structure (Hynes et al., 1989; Osmark et al., 1993).
`The mechanism(s) by which the loop replacements described
`here destabilize REI V, are not obvious. Most of the installed
`sequences are derived from the long, relatively disordered loops
`of the platelet antagonist disintegrins (Gould et al., 1990) and
`thus might be expected to be both relatively flexible and com-
`patible with extended, solvated loop conformations. The RGD
`(or KGD) sequence itself, which is common to all of the inserted
`loops, would also be expected to be compatible with a solvated
`loop environment; in fact, the RGD sequence has been observed
`in a topologically similar loop to CDR3 in the fibronectin do-
`main of several matrix proteins (Leahy et al., 1992; Main et al.,
`1992; Dickinson et al., 1994) as well as in the loops of disinte-
`grins. The inserted loops vary considerably in length (leading to
`net changes in domain length of between 0 and + 12 residues)
`3:O
`and in the presence or absence of residues that increase (Gly) or
`decrease (Pro) peptide backbone configurational flexibility. AI-
`though many of the destabilizing replacements contain proline
`residues, two CDRl mutants lacking prolines, REI-RGD32 (Ta-
`ble l) and REI-RGD37 (Results), suffer destabilization equal to
`or greater than that of a mutant containing an insert of equal
`length and including prolines (REI-RGD34).
`3.14
`
`
`93:O
`Two different CDRs-one between sheets (CDRI) and one
`
`95:O
`within a sheet (CDR3) - were replaced with very similar effects
`97:OGl
`on stability. In CDR3, different insertion points were explored,
`including points in the loop removed from any possible inter-
`90:OEl
`
`action with the body of the protein. In all cases, however, the
`90:OEI
`
`destabilization was significant. For example, although the en-
`tire 90-97 sequence is within the CDR3 as defined by sequence
`homology (Fig. 2), it is possible that some replacements of res-
`idues 90,91, and 97 may make altered packing interactions with
`
`2:o
`framework residues, resulting in destabilization. However, re-
`
`97:o
`placements that leave residues 90, 91, 96, and 97 unchanged
`are also observed to destabilize the domain by 5 (RGD12) to 7
`(RGD15) kcal/mol (Table 1).
`As in the other examples of destabilizing loop replacements
`described previously (see above), interactions lost from the wild
`type in the process of replacement may account for some of the
`observed destabilization. Table 2 lists the hydrogen bonds found
`in the crystal structure of wild-type REI involving residues 26-29
`
`
`
`
`
`~
`
`~
`
`
`
`
`3.16
`
`~
`
`26:OG
`3:O
`26:O
`
`27:O
`28:OD2
`68:O
`28:OD2
`97:O
`1 :OD2
`
`
`
`- -~ "" ~
`3.24
`3.27
`3.23
`
`3.29
`2.75
`3.31
`
`3:N
`26:N
`26:OG
`26:OG
`29: N
`3.16
`
`29:N 3.02
`29:N
`30:N
`90:N
`90:NE2
`90:NE2
`90:NE2
`90:NE2
`91:N
`91:N
`92:N
`93:OG
`95:N
`97:N
`97:OGl
`97:OGl
`97:OGl
`97:OGl
`
`Ex. 1044 - Page 6 of 9
`
`

`

`Destabilizing loop swaps
`
`2079
`
`An important implication of these results is that there may
`be significant limitations on the allowed sequences and lengths
`of protein loops, including antibody hypervariable loops.
`
`Thermal dependence of binding activity
`
`The temperature dependence of receptor binding by these mu-
`tants confirms the ability of appropriately chosen scaffolds to
`control the conformation of an installed bioactive ligand (Wet-
`zel, 1991). At the same time, the results underscore the impor-
`tance of considering domain stability in the choice of a domain
`for a presentation scaffold. If the destabilizations of 5-7 kcal/
`mol found here and elsewhere for loop swaps prove to be some-
`what general, it is clearly advantageous to choose a domain with
`a free energy of folding of significantly more than 7 kcal/mol
`for phage display and other scaffoId experiments.
`
`Stability of the immunoglobulin fold
`
`VH and V, domains (Pantoliano et al., 1991) and