U.S. patent application number 10/057321 was filed with the patent office on 2003-01-02 for crystalline tnf-alpha-converting enzyme and uses thereof.
Invention is credited to Black, Roy A., Bode, Wolfram, Chen, James Ming, Fernandez-Catalan, Carlos, Levin, Jeremy Ian, Maskos, Klaus, Paxton, Raymond James.
Application Number | 20030004651 10/057321 |
Document ID | / |
Family ID | 33545494 |
Filed Date | 2003-01-02 |
United States Patent
Application |
20030004651 |
Kind Code |
A1 |
Black, Roy A. ; et
al. |
January 2, 2003 |
Crystalline TNF-alpha-converting enzyme and uses thereof
Abstract
A tumor necrosis factor-.alpha. converting enzyme (TACE) is
produced, purified, and crystallized. The three-dimensional
coordinates of the crystal are obtained by X-ray diffraction. The
coordinates can be recorded on a computer readable medium, or are
part of a video memory, where they can be used as part of a system
for studying for studying TACE. The coordinates are also used in
designing, screening, and developing compounds that associate with
TACE.
Inventors: |
Black, Roy A.; (Seattle,
WA) ; Paxton, Raymond James; (Bellevue, WA) ;
Bode, Wolfram; (Gauting, DE) ; Maskos, Klaus;
(Holzkirchen, DE) ; Fernandez-Catalan, Carlos;
(Martinsried-Planegg, DE) ; Chen, James Ming;
(Bedminister, NJ) ; Levin, Jeremy Ian; (New City,
NY) |
Correspondence
Address: |
Stephen A. Bent
FOLEY & LARDNER
Suite 500
3000 K Street, N.W.
Washington
DC
20007-5109
US
|
Family ID: |
33545494 |
Appl. No.: |
10/057321 |
Filed: |
September 24, 2001 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
10057321 |
Sep 24, 2001 |
|
|
|
09611722 |
Jul 6, 2000 |
|
|
|
09611722 |
Jul 6, 2000 |
|
|
|
09244984 |
Feb 4, 1999 |
|
|
|
60117476 |
Jan 27, 1999 |
|
|
|
60135499 |
Mar 30, 1998 |
|
|
|
60073709 |
Feb 4, 1998 |
|
|
|
Current U.S.
Class: |
702/19 ; 435/226;
435/7.1; 702/20 |
Current CPC
Class: |
C12N 9/6489
20130101 |
Class at
Publication: |
702/19 ; 435/226;
435/7.1; 702/20 |
International
Class: |
G01N 033/53; G06F
019/00; G01N 033/48; G01N 033/50; C12N 009/64 |
Claims
What we claim is:
1. A composition comprising a polypeptide in crystalline form,
wherein the polypeptide is a TNF-.alpha.-converting enzyme
polypeptide.
2. A composition according to claim 1, wherein the
TNF-.alpha.-converting enzyme polypeptide comprises the
TNF-.alpha.-converting enzyme catalytic domain.
3. A composition according to claim 1, wherein the
TNF-.alpha.-converting enzyme polypeptide is the expression product
of a polynucleotide encoding the pro and catalytic domains of
TNF-.alpha.-converting enzyme.
4. A composition according to claim 1, wherein the
TNF-.alpha.-converting enzyme polypeptide is the expression product
of a polynucleotide encoding the amino acid residues 1-477 of
TNF-.alpha.-converting enzyme.
5. A composition according to claim 4, wherein the polynucleotide
is substituted such that amino acid residue Ser266 is changed to
Ala and amino acid residue Asn542 is changed to Gln, and wherein a
second polynucleotide encoding the sequence Gly-Ser-(His).sub.6 is
fused to the C-terminus.
6. A composition according to claim 1, further comprising a binding
partner suitable for co-crystallization with the
TNF-.alpha.-converting enzyme polypeptide.
7. A composition according to claim 6, wherein the binding partner
is a hydroxamate-based binding partner.
8. A composition according to claim 6, wherein the binding partner
is
N-{D,L-2-(hydroxyaminocarbonyl)methylmethylpentanoyl}-L-3-amino-2-dimethy-
lbutanoyl-L-alanine,2-(amino)ethyl amide.
9. A composition according to claim 1, wherein the crystal has a
crystal structure diffracting to 2.0 .ANG..
10. A composition according to claim 1, wherein the crystal is
monoclinic.
11. A composition according to claim 1, wherein the unit cell of
the crystal comprises four crystallographically independent
TNF-.alpha.-converting enzyme catalytic domain (TCD) molecules.
12. A composition according to claim 11, wherein the TCD molecules
are in an asymmetric unit.
13. A composition according to claim 1, wherein the crystal is of
monoclinic space group P2.sub.1 and the cell has the constants
a=61.38 A, b=126.27 A, c=81.27 A, and .beta.=107.4.degree..
14. A composition according to claim 1, wherein the polypeptide is
characterized by the structure coordinates according to Table 1, or
a substantial part thereof.
15. A method for crystallizing a TNF-.alpha.-converting enzyme
polypeptide, comprising: (A) mixing a solution comprising a TACE
polypeptide and a binding partner with a crystallization buffer;
and (B) crystallizing the mixture of step (A) by drop vapor
diffusion to form a crystalline precipitate.
16. The method according to claim 15, further comprising: (C)
transferring seeds from the crystalline precipitate formed by the
drop vapor diffusion and a crystallization promotor into a mixture
of a concentrated solution comprising a TACE polypeptide and
binding partner substrate, and a crystallization buffer; and (D)
crystallizing the mixture of step (C) by drop vapor diffusion to
form a crystal.
17. The method of claim 15, wherein said crystallization buffer is
0.1M Na Citrate pH 5.4, 20% w/v PEG 4000, and 20% v/v
Isopropanol.
18. The method of claim 15 or 16, wherein the binding partner is
N-{D,L-2-(hydroxyaminocarbonyl)methyl-4-methylpentanoyl}-L-3-amino-2-dime-
thylbutanoyl-L-alanine, 2-(amino)ethyl amide.
18. The method of claims 15, wherein crystallization is at a
temperature ranging from 4 to 20 degrees Celsius.
19. The method of claim 15, wherein the solution comprising the
TACE polypeptide and the inhibitor is at a concentration of about 5
mg/mL to about 12 mg/mL in a buffer.
20. The method of claim 19, wherein the solution comprising a TACE
polypeptide and the binding partner is mixed with the
crystallization buffer in a 1:1 ratio.
21. A tumor necrosis factor-.alpha. (TNF-.alpha.)-converting enzyme
crystal made by co-crystallizing a TNF-.alpha.-converting enzyme
polypeptide with a co-crystallization substrate.
22. A computer-readable medium having recorded thereon x-ray
crystallographic coordinate data for the catalytic domain of
TNF-.alpha. converting enzyme, or a portion thereof.
23. A computer-readable medium having recorded thereon the x-ray
crystallographic coordinate data set forth in Table 1, or a portion
thereof.
24. A computer-readable medium of claim 22, wherein the medium is
selected from the group consisting of a floppy disc, a hard disc,
computer tape, RAM, ROM, CD, DVD, a magnetic disk, and an optical
disk.
25. A computer-readable medium having recorded thereon
machine-readable data, wherein the computer-readable medium, when
used in conjunction with a machine programmed with instructions for
using the data, is capable of generating image signals for
depicting a graphical, three-dimensional representation of a
TNF-.alpha. converting enzyme polypeptide, or portion thereof.
26. A system for studying a TNF-.alpha. converting enzyme
polypeptide, said system comprising: (a) a memory capable of
storing information representing at least a portion of a
TNF-.alpha. converting enzyme polypeptide, wherein said memory
comprises at least one first-type storage region, including a set
of spatial coordinates specifying a location in a three dimensional
space, and at least one second-type storage region comprising
information representing a characteristic of one of a plurality of
amino acids, said second-type storage regions being logically
associated with said first-type storage regions in said memory to
represent a geometric arrangement of at least one characteristic of
said at least a portion of said TNF-.alpha. converting enzyme
peptide in said three dimensional space; (b) a processor coupled to
said memory to access said first-type storage regions and said
second-type storage regions, wherein the processor generates image
signals for depicting a visual image representing three dimensional
image of said at least one characteristic of said at least a
portion of said TNF-.alpha. converting enzyme polypeptide in said
three dimensional space based on data from said memory; and (c) a
display coupled to said processor to receive said image signals,
wherein the display depicts a visual three dimensional image of
said at least one characteristic of said at least a portion of said
TNF-.alpha. converting enzyme polypeptide in said three dimensional
space based on said image signals.
27. A system as set forth in claim 26, wherein said image signals
include signals for depicting a visual three dimensional image of a
ribbon structure of said at least a portion of said TNF-.alpha.
converting enzyme polypeptide in said three dimensional space.
28. A system as set forth in claim 26, wherein said image signals
include signals for depicting a visual image of a solid model
representation of said at least a portion of said TNF-.alpha.
converting enzyme polypeptide in said three dimensional space.
29. A system as set forth in claim 26, wherein said image signals
include signals for depicting a visual three dimensional image of
electrostatic surface potential of said at least a portion of said
TNF-.alpha. converting enzyme polypeptide in said three dimensional
space.
30. A system as set forth in claim 26, wherein said image signals
include signals for depicting a visual three dimensional stereo
image of said at least a portion of said TNF-.alpha. converting
enzyme polypeptide in said three dimensional space.
31. A system as set forth in claim 26, further comprising: a
storage device capable of storing data representing a geometric
arrangement of a characteristic of a composition other than said
TNF-.alpha. converting enzyme polypeptide; and an operator
interface for receiving instructions from a operator; and wherein
said processor is coupled to said storage device and to said
operator interface and generates additional image signals for
depicting said geometric arrangement of said characteristic of said
composition relative to said visual three dimensional image of said
at least one characteristic of said at least a portion of said TNF
converting enzyme polypeptide on said display based on instructions
from the operator interface.
32. A system as set forth in claim 31, wherein said storage device
is part of said memory.
33. A system as set forth in claim 26, comprising a plurality of
first-type and second-type storage regions.
34. A video memory capable of storing information for generating a
visual display of at least a portion of a TNF-.alpha. converting
enzyme polypeptide, said video memory comprising: (a) at least one
first-type storage region, each of said first-type storage regions
including a set of spatial coordinates specifying a location in a
three dimensional space; and (b) at least one second-type storage
region, each of said second-type storage regions containing
information for visually depicting a characteristic of one of a
plurality of amino acids; wherein said second-type storage regions
are logically associated with said first-type storage regions in
said video memory to represent a geometric arrangement of at least
one characteristic of said at least a portion of said TNF-.alpha.
converting enzyme polypeptide in said three dimensional space.
35. A video memory as set forth in claim 34, wherein said
second-type storage regions are logically associated with said
first-type storage regions in said video memory to represent a
geometric arrangement of at least one characteristic of a catalytic
domain portion of said TNF-.alpha. converting enzyme polypeptide in
said three dimensional space.
36. A video memory as set forth in claim 34, wherein said
first-type storage regions and said second-type storage regions are
regions of a semiconductor memory.
37. A video memory as set forth in claim 34, wherein said
first-type storage regions and said second-type storage regions are
regions of an optical disk.
38. A video memory as set forth in claim 34, wherein said
first-type storage regions and said second-type storage regions are
regions of a magnetic memory.
39. A video memory as set forth in claim 34, comprising a plurality
of first-type and second-type storage regions.
40. A method of identifying a compound that associates with
TNF-.alpha.-converting enzyme, comprising: (A) designing an
associating compound for said polypeptide that forms a bond with
the TNF-.alpha.-converting enzyme catalytic domain based on x-ray
diffraction coordinates of a TNF-.alpha.-converting enzyme
polypeptide crystal; (B) synthesizing said compound; and (C)
determining the associate capability of said compound with said
TNF-.alpha.-converting enzyme.
41. The method according to claim 40, wherein said associating
compound is an inhibitor, mediator, or other compound that
regulates TNF-.alpha.-converting enzyme activity.
42. The method of claim 41, wherein said associating compound is a
competitive inhibitor, uncompetitive inhibitor, or non-competitive
inhibitor.
43. The method according to claim 40, wherein the coordinates are
the coordinates of Table 1, or a substantial part thereof.
44. The method of claim 40, wherein said TNF-.alpha.-converting
enzyme polypeptide crystal comprises the TNF-.alpha.-converting
enzyme catalytic domain.
45. The method of claim 40, wherein said TNF-.alpha.-converting
enzyme polypeptide is the expression product of a polynucleotide
encoding the pro and catalytic domains of TNF-.alpha.-converting
enzyme.
46. The method of claim 40, wherein said TNF-.alpha.-converting
enzyme polypeptide is the expression product of a polynucleotide
encoding the amino acid residues 1-477 of TNF-.alpha.-converting
enzyme.
47. The method of claim 46, wherein the polynucleotide is
substituted such that amino acid residue Ser266 is changed to Ala
and amino acid residue Asn542 is changed to Gln, and wherein a
second polynucleotide encoding the sequence Gly-Ser-(His).sub.6 is
fused to the C-terminus.
48. The method of claim 40, wherein said TNF-.alpha.-converting
enzyme polypeptide crystal is co-crystallized with a binding
partner.
49. The method of claim 48, wherein the binding partner is a
hydroxamate-based binding partner.
50. The method of claim 48, wherein the binding partner is
N-{D,L-2-(hydroxyaminocarbonyl)methyl-4-methylpentanoyl}-L-3-amino-2dimet-
hylbutanoyl-L-alanine, 2-(amino)ethyl amide.
51. The method of claim 40, wherein said TNF-.alpha.-converting
enzyme polypeptide crystal has a crystal structure diffracting to
2.0 .ANG..
52. The method of claim 40, wherein said TNF-.alpha.-converting
enzyme polypeptide crystal is monoclinic.
53. The method of claim 40, wherein said TNF-.alpha.-converting
enzyme polypeptide crystal has a unit cell comprising four
crystallographically independent TNF-.alpha.-converting enzyme
catalytic domain (TCD) molecules.
54. The method of claim 53, wherein the TCD molecules are in an
asymmetric unit.
55. The method of claim 40, wherein said TNF-.alpha.-converting
enzyme polypeptide crystal is of monoclinic space group P2.sub.1
and the cell has the constants a=61.38 .ANG., b=126.27 .ANG.,
c=81.27 .ANG., and .beta.=107.41.degree..
56. The method of claim 40, wherein the associating compound is
designed to associate with the S1' region of TNF-.alpha.-converting
enzyme.
57. The method of claim 40, wherein the associating compound is
designed to associate with the S1'S3' pocket of
TNF-.alpha.-converting enzyme.
58. The method of claim 40, wherein the associating compound is
designed to incorporate a moiety that chelates zinc.
59. The method of claim 40, wherein the associating compound is
designed to form a hydrogen bond with Leu348 or Gly349 of
TNF-.alpha.-converting enzyme.
60. The method of claim 40, wherein the associating compound is
designed to introduce a non-polar group which occupies the S1'
pocket of TNF-.alpha.-converting enzyme.
61. The method of claim 40, wherein the associating compound is
designed to introduce a group which lies within the channel joining
S1'-S3' pockets of TNF-.alpha.-converting enzyme and which makes
appropriate van der Waal contact with the channel.
62. The method of claim 40, wherein the associating compound is
designed to form a hydrogen bond with Leu348 or Gly349 on the
backbone amide groups of TNF-.alpha.-converting enzyme.
Description
INFORMATION ON RELATED APPLICATIONS
[0001] This application claims the priority benefit of U.S.
provisional patent application serial No. 60/073,709, filed Feb. 4,
1998, U.S. patent application Ser. No. 09/050,083, filed Mar. 30,
1998 (which will be converted to a US provisional application
pursuant to a petition filed on Jan. 27, 1999), and US provisional
patent application titled "Crystalline TNF-.alpha.-Converting
Enzyme and Uses Thereof," filed Jan. 27, 1999.
BACKGROUND OF THE INVENTION
[0002] The cytokine tumor necrosis factor-.alpha. (TNF.alpha.)
plays a role in the induction of inflammatory reactions and is
known to be cytotoxic towards tumor cells. TNF.alpha., however,
also may cause severe damage to the human body when produced in
excess by eventually leading to multiple organ failure and death.
See Bemelmans et al., "Tumor Necrosis Factor: Function, Release and
Clearance," Crit. Rev. Immun. 16: 1-11 (1996).
[0003] Tumor necrosis factor-.alpha. is produced by activated
cells, such as mononuclear phagocytes, T-Cells, B-Cells, mast cells
and NK cells. TNF.alpha. exists in two forms: a type II membrane
protein having a relative molecular mass of 26 kD and a soluble 17
kD form generated from the membrane form by proteolytic cleavage.
The TNF.alpha. membrane protein is synthesized as a 223 amino acid
membrane-anchored precursor. The soluble TNF.alpha. is released
from the membrane-bound precursor by a membrane-anchored
proteinase. This proteinase was recently identified as a
multidomain metalloproteinase called TNF-.alpha.-converting enzyme
(TACE). See, Black et al., "A metalloproteinase disintigrin that
releases tumor-necrosis factor-.alpha. from cells," Nature 385:
729-733 (1997), Moss et al., "Cloning of a disintigrin
metalloproteinase that processes precursor tumor-necrosis
factor-.alpha.," Nature 385: 733-736 (1997). TACE has recently been
identified as a zinc endopeptidase consisting of an extracellular
region comprising an N-terminal signal peptide, a pro-domain, a 263
residue catalytic domain (TCD) that is preceded by a furin cleavage
site (residues 211-214), a disintegrin domain, an EGF-like domain,
and a crambin-like domain, an apparent transmembrane helix and the
intracellular C-terminal tail. Tumor necrosis factor-.alpha.
converting enzyme (TACE), including a polynucleotide sequence, is
described in detail in the published PCT application No. WO
96/41624, herein incorporated in the entirety by reference.
[0004] As noted above, the over-production or unregulated
production of TNF.alpha. presents serious physiological dangers. It
has been implicated in various deleterious physiological diseases
such as rheumatoid arthritis, cachexia and endotoxic shock. It also
may eventually lead to organ failure and death. Thus, a way to
control or block release of TNF.alpha. into the circulation is
needed. Because of TACE's role in the conversion of TNF.alpha.,
inhibition, modulation, or regulation of TACE would affect the
release of TNF.alpha. into circulation. Inhibitors of
metalloproteinases and structure based design thereof are described
in Zask et al., "Inhibition of Matrix Metalloproteinases: Structure
Based Design" Current Pharmaceutical Design, 2:624-661 (1996).
Thus, compounds that associate with TACE, such as inhibitors,
receptors or modulators will be useful to protect patients from
adverse effects associated with the over-production or unregulated
production of tumor necrosis factor-.alpha..
SUMMARY OF THE INVENTION
[0005] According to one aspect of the invention, there is provided
a composition comprising a polypeptide in crystalline form, wherein
the polypeptide is a TNF-.alpha.-converting enzyme polypeptide. In
one embodiment, the TNF-.alpha.-converting enzyme polypeptide
comprises the TNF-.alpha.-converting enzyme catalytic domain. In
another embodiment, the TNF-.alpha.-converting enzyme polypeptide
is the expression product of a polynucleotide encoding the pro and
catalytic domains of TNF-.alpha.-converting enzyme. In a further
embodiment, the TNF-.alpha.-converting enzyme polypeptide is the
expression product of a polynucleotide encoding the amino acid
residues 1-477 of TNF-converting enzyme. In yet another embodiment,
the polynucleotide is substituted such that amino acid residue
Ser266 is changed to Ala and amino acid residue Asn542 is changed
to Gin, and wherein a second polynucleotide encoding the sequence
Gly-Ser-(His).sub.6 is fused to the C-terminus.
[0006] According to another aspect of the invention, the
compositions above further comprising a binding partner suitable
for co-crystallization with the TNF-.alpha.-converting enzyme
polypeptide. In one embodiment, the binding partner is a
hydroxamate-based binding partner. In another embodiment, the
binding partner is
N-{D,L-2-(hydroxyaminocarbonyl)methyl-4-methylpentanoyl}-L-3-amino-2-dime-
thylbutanoyl-L-alanine,2-(amino)ethyl amide.
[0007] According to further embodiments, the compositions above
have a crystal structure diffracting to 2.0 .ANG., are monoclinic,
have a unit cell comprising four crystallographically independent
TNF-.alpha.-converting enzyme catalytic domain (TCD) molecules,
have the TCD molecules are in an asymmetric unit, and/or have
monoclimc space group P2.sub.1 and the cell has the constants
a=61.38 A, b=126.27 A, c=81.27 A, and .beta.=107.41.degree..
[0008] In still another embodiment of the invention, the
polypeptides above are characterized by the structure coordinates
according to Table 1, or a substantial part thereof.
[0009] According to a further aspect of the invention, there is
provided a method for crystallizing a TNF-.alpha.-converting enzyme
polypeptide, comprising (A) mixing a solution comprising a TACE
polypeptide and a binding partner with a crystallization buffer;
and (B) crystallizing the mixture of step (A) by drop vapor
diffusion to form a crystalline precipitate. In one embodiment, the
method further comprises (C) transferring seeds from the
crystalline precipitate formed by the drop vapor diffusion and a
crystallization promotor into a mixture of a concentrated solution
comprising a TACE polypeptide and binding partner substrate, and a
crystallization buffer; and (D) crystallizing the mixture of step
(C) by drop vapor diffusion to form a crystal. In another
embodiment, the crystallization buffer is 0.1M Na Citrate pH 5.4,
20% w/v PEG 4000, and 20% v/v Isopropanol. In still another
embodiment, the binding partner is
N-{D,L-2-(hydroxyaminocarbonyl)methyl-4-methylpentanoy-
l}-L-3-amino-2-dinethylbutanoyl-L-alanine, 2-(amino)ethyl amide. In
yet another embodiment, crystalization is at a temperature ranging
from 4 to 20 degrees Celsius. In another embodiment, the solution
comprising the TACE polypeptide and the inhibitor is at a
concentration of about 5 mg/mL to about 12 mg/mL in a buffer. In a
further embodiment, the solution comprising a TACE polypeptide and
the binding partner is mixed with the crystallization buffer in a
1:1 ratio.
[0010] According to still another aspect of the invention, there is
provided a tumor necrosis factor-.alpha. (TNF-.alpha.)-converting
enzyme crystal made by co-crystallizing a TNF-.alpha.-converting
enzyme polypeptide with a co-crystallization substrate.
[0011] According to yet another aspect of the invention, there is
provided a computer-readable medium having recorded thereon x-ray
crystallographic coordinate data for the catalytic domain of
TNF-.alpha. converting enzyme, or a portion thereof. In one
embodiment, the computer-readable medium has recorded thereon the
x-ray crystallographic coordinate data set forth in Table 1, or a
portion thereof. In another embodiment, the medium is selected from
the group consisting of a floppy disc, a hard disc, computer tape,
RAM, ROM, CD, DVD, a magnetic disk, and an optical disk. In still
another embodiment, the computer-readable medium has recorded
thereon machine-readable data, wherein the computer-readable
medium, when used in conjunction with a machine programmed with
instructions for using the data, is capable of generating image
signals for depicting a graphical, three-dimensional representation
of a TNF-.alpha. converting enzyme polypeptide, or portion
thereof.
[0012] According to a further aspect of the invention, there is
provided a system for studying a TNF.alpha. converting enzyme
polypeptide, said system comprising (a) a memory capable of storing
information representing at least a portion of a TNF-.alpha.
converting enzyme polypeptide, wherein said memory comprises at
least one first-type storage region, including a set of spatial
coordinates specifying a location in a three dimensional space, and
at least one second-type storage region comprising information
representing a characteristic of one of a plurality of amino acids,
said second-type storage regions being logically associated with
said first-type storage regions in said memory to represent a
geometric arrangement of at least one characteristic of said at
least a portion of said TNF-.alpha. converting enzyme peptide in
said three dimensional space; (b) a processor coupled to said
memory to access said first-type storage regions and said
second-type storage regions, wherein the processor generates image
signals for depicting a visual image representing three dimensional
image of said at least one characteristic of said at least a
portion of said TNF-.alpha. converting enzyme polypeptide in said
three dimensional space based on data from said memory; and (c) a
display coupled to said processor to receive said image signals,
wherein the display depicts a visual three dimensional image of
said at least one characteristic of said at least a portion of said
TNF.alpha. converting enzyme polypeptide in said three dimensional
space based on said image signals. In one embodiment of the
invention, the image signals include signals for depicting a visual
three dimensional image of a ribbon structure of said at least a
portion of said TNF-.alpha. converting enzyme polypeptide in said
three dimensional space. In another embodiment of the invention,
the image signals include signals for depicting a visual image of a
solid model representation of said at least a portion of said
TNF-.alpha. converting enzyme polypeptide in said three dimensional
space. In still another embodiment of the invention, the image
signals include signals for depicting a visual three dimensional
image of electrostatic surface potential of said at least a portion
of said TNF-.alpha. converting enzyme polypeptide in said three
dimensional space. In yet another embodiment of the invention, the
image signals include signals for depicting a visual three
dimensional stereo image of said at least a portion of said
TNF-.alpha. converting enzyme polypeptide in said three dimensional
space. In a further embodiment of the invention, the system further
comprises a storage device capable of storing data representing a
geometric arrangement of a characteristic of a composition other
than said TNF-.alpha. converting enzyme polypeptide; and an
operator interface for receiving instructions from a operator; and
wherein said processor is coupled to said storage device and to
said operator interface and generates additional image signals for
depicting said geometric arrangement of said characteristic of said
composition relative to said visual three dimensional image of said
at least one characteristic of said at least a portion of said
TNF-.alpha. converting enzyme polypeptide on said display based on
instructions from the operator interface. In one embodiment, the
storage device is part of said memory. In another embodiment, the
system comprises a plurality of first-type and second-type storage
regions.
[0013] According to another aspect of the invention, there is
provided a video memory capable of storing information for
generating a visual display of at least a portion of a TNF-.alpha.
converting enzyme polypeptide, said video memory comprising (a) at
least one first-type storage region, each of said first-type
storage regions including a set of spatial coordinates specifying a
location in a three dimensional space; and (b) at least one
second-type storage region, each of said second-type storage
regions containing information for visually depicting a
characteristic of one of a plurality of amino acids; wherein said
second-type storage regions are logically associated with said
first-type storage regions in said video memory to represent a
geometric arrangement of at least one characteristic of said at
least a portion of said TNF-.alpha. converting enzyme polypeptide
in said three dimensional space. In one embodiment, the second-type
storage regions are logically associated with said first-type
storage regions in said video memory to represent a geometric
arrangement of at least one characteristic of a catalytic domain
portion of said TNF converting enzyme polypeptide in said three
dimensional space. In another embodiment, the first-type storage
regions and said second-type storage regions are regions of a
semiconductor memory. In yet another embodiment, the first-type
storage regions and said second-type storage regions are regions of
an optical disk. In still another embodiment, the first-type
storage regions and said second-type storage regions are regions of
a magnetic memory. In a further embodiment, the video memory
comprises a plurality of first-type and second-type storage
regions.
[0014] In a still further aspect of the invention, there is
provided a method of identifying a compound that associates with
TNF-.alpha.-converting enzyme, comprising (A) designing an
associating compound for said polypeptide that forms a bond with
the TNF-.alpha.-converting enzyme catalytic domain based on x-ray
diffraction coordinates of a TNF.alpha.-converting enzyme
polypeptide crystal; (B) synthesizing said compound; and (C)
determining the associate capability of said compound with said
TNF-.alpha.-converting enzyme. In one embodiment, the associating
compound is an inhibitor, mediator, or other compound that
regulates TNF-.alpha.-converting enzyme activity. In another
embodiment, the associating compound is a competitive inhibitor,
uncompetitive inhibitor, or non-competitive inhibitor. In still
another embodiment, the coordinates are the coordinates of Table 1,
or a substantial part thereof. In a further embodiment, the
TNF-.alpha.-converting enzyme polypeptide crystal comprises the
TNF-.alpha.-converting enzyme catalytic domain. In still another
embodiment, the TNF-.alpha.-converting enzyme polypeptide is the
expression product of a polynucleotide encoding the pro and
catalytic domains of TNF.alpha.-converting enzyme. In yet another
embodiment, the TNF-.alpha.-converting enzyme polypeptide is the
expression product of a polynucleotide encoding the amino acid
residues 1-477 of TNF-.alpha.-converting enzyme. In another
embodiment, the polynucleotide is substituted such that amino acid
residue Ser266 is changed to Ala and amino acid residue Asn542 is
changed to Gln, and wherein a second polynucleotide encoding the
sequence Gly-Ser-(His).sub.6 is fused to the C-terminus. In a
further embodiment, the TNF-.alpha.-converting enzyme polypeptide
crystal is co-crystallized with a binding partner. In still another
embodiment, the binding partner is a hydroxamate-based binding
partner or
N-{D,L-2-(hydroxyaminocarbonyl)methylmethylpentanoyl}-L-3-amin-
o-2Amethylbutanoyl-L-alanine,2-(amino)ethyl amide. In yet other
embodiments, the TNF-.alpha.-converting enzyme polypeptide crystal
has a crystal structure diffracting to 2.0 .ANG., is monoclinic,
has a unit cell comprising four crystallographically independent
TNF-.alpha.-converting enzyme catalytic domain (TCD) molecules, has
the TCD molecules are in an asymmetric unit, and/or is of
monoclinic space group P2.sub.1 and the cell has the constants
a=61.38 .ANG., b=126.27 .ANG., c=81.27 .ANG., and
.beta.=107.41.degree.. In still another embodiment, the invention
the the associating compound is designed to associate with the S1'
region of TNF-.alpha.-converting enzyme. In yet another embodiment,
the associating compound is designed to associate with the S1'S3'
pocket of TNF.alpha.-converting enzyme. In still other embodiments
of the invention, the associating compound is designed to (i)
incorporate a moiety that chelates zinc, (ii) form a hydrogen bond
with Leu348 or Gly349 of TNF-.alpha.-converting enzyme, (iii)
introduce a non-polar group which occupies the S1' pocket of
TNF-.alpha.-converting enzyme, (iv) introduce a group which lies
within the channel joining S1'-S3' pockets of
TNF-.alpha.-converting enzyme and which makes appropriate van der
Waal contact with the channel, and/or (v) form a hydrogen bond with
Leu348 or Gly349 on the backbone amide groups of
TNF-.alpha.-converting enzyme.
[0015] These and other aspects of the invention will become
apparent to the skilled artisan in view of the teachings contained
herein.
BRIEF DESCRIPTION OF THE FIGURES
[0016] FIG. 1: FIG. 1 is a ribbon diagram of the TACE catalytic
domain (TCD). The chain starts on the lower left back side, runs
through the structural elements sI, hAI, hA, sII, hB, hB2, sIII,
IV, IVa, sIVb, sV, hC, Met-turn and hD, and ends in the upper left
back. The three disulfides are shown as connections, with the
sulphurs given as small spheres. The catalytic zinc (central
sphere) is liganded by the three imidazoles of His4O5, His4O9 and
His415, and by the hydroxyl and the carbonyl oxygen atoms of the
inhibitor hydroxamic acid group. The inhibitor mimicking
interaction of primed-site residues of a peptide substrate is shown
in full. FIG. 1 was made using SETOR. See Evans, S. "SETOR:
Hardware Lighted Three-Dimensional Solid Model Representations of
Macromolecules" J. Mol. Graph. 11:134-138 (1993).
[0017] FIGS. 2a and 2b: FIGS. 2a and 2b are solid surface
representations of the catalytic domains of TACE (TCD) (FIG. 2a)
and MMP-3 (FIG. 2b). The electrostatic surface potential is
contoured from -15 (intense red) to 15 (intense blue) k.sub.8T/e.
Both active-site clefts run from left to right, with the catalytic
zinc atoms (spheres) in the centers. In TACE, the bound inhibitor
is shown in full structure, binding with its isobutyl (P1') and its
Ala (P3') sidechains into the deep S1' and the novel S3' pockets.
The orientation is similar to FIG. 1. FIGS. 2a and 2b were made
using GRASP. Nicolls, A., Bharadwaj, R. and Houig, B.,
"Grasp--Graphical representation and analysis of surface
properties," Biophs. 64, A166 (1993).
[0018] FIG. 3: FIG. 3 aligns the catalytic domain sequences of
adamalysin II (ADAM_CROAD), TACE and human ADAM 10 (hADAM1O),
according to their topological equivalence and sequence similarity,
respectively. The residue numbers are due to the generic TACE
numbering. Arrows and braces represent .beta.-strands and
.alpha.-helices in TACE.
[0019] FIG. 4: FIG. 4 is a stereo section of the final 2.0 .ANG.
electron density around the catalytic zinc (large, central sphere)
superimposed with the final TACE model. Visible are the three zinc
liganding imidazole rings of His4O5 (top), His4O9 (left) and His415
(bottom), the "catalytic" Glu406, and the hydroxamic acid moiety of
the inhibitor. The orientation is similar to FIG. 1. FIG. 4 was
made using TURBO-FRODO. See Roussel, A. & Cambilleau, C.,
"Turbo-Frodo in Silicon Graphics Geometry," Partners Directory,
Silicon Graphics, Mountain View, Calif. (1989).
[0020] FIG. 5: FIG. 5 is a superposition of the ribbon plots of the
catalytic domain of TACE (light) and adamalysin (dark). Also shown
is the catalytic zinc of TACE (sphere) and the three (TACE) and two
(adamalysin) disulfide bridges. The orientation is similar to FIG.
1. FIG. 5 was made using GRASP.
[0021] FIG. 6: FIG. 6 illustrates a system for studying a
TNF-.alpha. converting enzyme, including a video memory storing
information for generating a visual display of at least a portion
of a TNF-.alpha. converting enzyme.
DETAILED DESCRIPTION OF THE INVENTION
[0022] The present invention relates to a highly purified tumor
necrosis factor-.alpha. converting enzyme (TACE) polypeptide, a
method of producing and purifying a TACE polypeptide, a method of
crystallizing a TACE polypeptide, and a TACE polypeptide crystal.
The invention further relates to a X-ray diffraction method using a
TACE polypeptide crystal, and to a method of obtaining the X-ray
crystallographic structural coordinates of a TACE polypeptide as
well as the structural coordinates themselves. Still finer, the
present invention relates using the structural coordinates of a
TACE polypeptide to elucidate the three-dimensional structure of a
TACE polypeptide and designing and developing compounds that
associate with TACE. Knowledge of the three-dimensional structure
and structure coordinates provided according to the invention
permit the skilled person to make compounds that will interact with
TACE. Such interacting compounds can be made by a variety of
techniques and design criteria, including those disclosed in
Protein Engineering (Oxender and Fox, eds.) (Alan R. Liss, Inc.
1987).
[0023] As used herein, TACE refers to a group of polypeptides that
are capable of converting the 26 kD cell membrane-bound form of
TNF.alpha. into the soluble 17 kD form that comprises the
C-terminal 156 residues of the TNF.alpha. protein. TACE encompasses
proteins having the amino acid sequence described in PCT
application No. WO 96/41624, herein incorporated in its entirety by
reference, as well as any of those proteins having homology,
preferably no less than 50%, more preferably at least 80% homology,
still more preferably 90% homology to such sequence, at the amino
acid level. Additionally, TACE further refers to the expression
products of nucleotide sequences disclosed in PCT application No.
WO 96/41624. TACE further encompasses the membrane-bound protein
and soluble or truncated proteins comprising the extracellular
portion of the protein and which retain biological activity and are
capable of being secreted. Examples of such proteins are described
in PCT application No. WO 96/41624.
[0024] The TACE amino acid sequence, or any part or residue
thereof, can be found in Black et al., "A Metalloproteinase
disintigrin that releases tumour-necrosis factor-.alpha. from
cells," Nature 385: 729-733 (February 1997), herein incorporated in
the entirety by reference. Variations in the amino acid sequence of
TACE are within the present invention as well. All references to
the TACE ammo acid sequence contained herein refer to the sequence
in Black et al., supra.
[0025] As used herein, the TACE catalytic domain (TCD) refers to
the portion of a TACE polypeptide between residues 215 and 477-and
including the preceding furin cleavage site (residues 211-214), or
any part thereof that is capable of cleaving the peptide
PLAQAVRSSS.
[0026] Expression, Isolation and Purification of TACE
Polypeptides
[0027] Tumor necrosis factor-.alpha. converting enzyme (TACE) is
described in the published PCT application No. WO 96/41624. The
application describes isolated nucleic acids encoding TACE or
portions of TACE, expression vectors comprising a cDNA encoding
TACE or portions thereof, and host cells transformed or transfected
with the expression vectors comprising a cDNA encoding TACE or
portions of TACE. The application further describes processes for
producing TACE and portions thereof, for example by culturing
transfected cells engineered to express TACE, followed by
purification of the recombinantly produced TACE or portion thereof.
Methods of isolating, expressing, and purifying a TACE polypeptide
are described in detail in published PCT application No. WO
96/41624. The entirety of PCT 96/41624 is incorporated herein by
reference.
[0028] According to the invention, cDNA encoding the signal
peptide, pro and catalytic domains of TACE, i.e., amino acid
residues 1-477 is inserted into a suitable expression vector and
expressed in a suitable cell line. The cDNA also may include other
regions that facilitate expression or achieve other objects that
otherwise that do not depart from the essence of the invention,
such as flanking regions.
[0029] The cDNAs encoding the TACE polypeptide, or functional
portions thereof, such as the TCD, may be altered by addition,
substitution, deletion, or insertion. Such alterations may be made,
for example, to prevent glycosolation, prevent formation of
incorrect or undesired disulfide bridges, and/or enhance
expression. Examples of such alterations are described in WO
96/41624 and can be carried out by the methods described therein
and other conventional methods. TACE may also be conjugated. Such
conjugates may comprise peptides added to facilitate purification
and/or identification. Such peptides include, for example, poly-His
peptides. Conjugation is described in U.S. Pat. No. 5,011,912 and
Hopp et al., Bio/Technology 6:1204 (1988).
[0030] In one embodiment of the invention, the cDNA encodes a
TNF-.alpha. converting enzyme polypeptide comprising the signal
peptide, pro and catalytic domains of TACE (TCD), residues 1-477,
with Ser266 changed to Ala and Asn452 changed to Gln. These
substitutions are useful in preventing N-linked glycosolation.
Additionally, the sequence Gly-Ser(His).sub.6 may be added to the
C-terminus. The addition of the sequence Gly-Ser(His).sub.6
facilitates purification of the polypeptide using metal-chelate
affinity resins, such as Ni-NTA resins.
[0031] Recombinant expression vectors containing the nucleotide
sequence encoding TACE, or a portion thereof, may be prepared using
well known methods. Suitable host cells for expression of TACE
polypeptides include prokaryotic, yeast, and higher eukaryotic
cells. Vectors and host cells suitable for use in the present
invention are described in WO 96/41624. Further examples of
suitable expression systems that can be employed to express
recombinant TACE according to the present invention include
mammalian or insect host cell culture expression systems, including
baculovirus systems in insect cells (See Luckow and Summers,
Bio/Technology 6:47 (1988))and mammalian cell lines such as COS-7
cells (Gluzman et al., Cell 23:175 (1981)). Additional examples are
known in the art and include those described in WO 96141624. In one
embodiment of the invention, the TACE polypeptide is expressed in
CHO cells. In this embodiment, the cells secrete a mixture of TACE
polypeptide beginning with Val212 and Arg215.
[0032] In one embodiment, stable expressing cells may be selected
by culturing the cells in a drug that kills those cells that do not
incorporate the vector. Examples of suitable selection methods are
described in, for example, Kaufman, R. J., "Selection and
coamplification of heterologous genes in mammalian cells," Methods
in Enzymology, 185:537-566 (1990).
[0033] Purification of the expressed TACE polypeptide may be
carried out by any suitable means, such as those described in WO
96/41624. According to one aspect of the invention, it is
preferable to obtain a TACE polypeptide that is suitable for
crystallization. In obtaining a TACE polypeptide suitable for
crystallization, it is important that the process for purifying the
TACE polypeptide is sufficient to yield a polypeptide pure enough
to properly crystallize.
[0034] A preferred method of purification starts with a suitable
amount of medium from the culture of TACE-secreting cells. This
medium is generally a supernate of the culture. The medium contains
the TACE polypeptide to be purified. Preferably, the TACE
polypeptide is recombinantly produced using DNA coding for the TACE
polypeptide with the sequence altered to encode a conjugate or
conjugates that facilitate purification. For example, the sequence
encoding Gly-Ser-(His).sub.6 may be added to the C-terminus to
facilitate purification using metal-chelate resins.
[0035] The medium is concentrated, for example, by diafiltration.
Suitable diafiltration units include a Millipore 10K cut-off, 1
ft.sup.2 TFF diafiltration unit. A suitable buffer solution is then
added to the concentrated medium. Any suitable buffer may be used.
One such suitable buffer contains 20 mM Tris (pH 7.5) and 300 mM
NaCl.
[0036] The sample is reconcentrated and diluted numerous times. For
example, the sample may be reconcentrated and diluted a second time
with the buffer, reconcentrated again, diluted a third time with
the buffer, and reconcentrated a final time. The sample retained in
the diafiltration unit is recovered by a suitable method, such as
by a back-flush method. The recovered material may then be filtered
through a suitable membrane. Suitable membranes include, for
example, 0.45 or 0.22 micron pore-size membranes. Azide is then
added. The filtered sample may then be stored overnight at a low
temperature, such as about 2-9.degree. C. After overnight storage,
imidazole from a stock solution in water and ZnCl.sub.2 from a
stock solution in water are added to the filtered sample. The
sample then is pumped over a suitable column. One suitable column,
particularly when the TACE polypeptide is conjugated with the
sequence Gly-Ser-(His).sub.6, is a metal-chelate resin, such as a
Ni-NTA resin.
[0037] The column is washed with a buffer, such as a buffer of 20
mM Tris pH 7.5, 300 mM NaCl, 5 mM imidazole, and 5 uM ZnCl.sub.2.
The TACE polypeptide is then eluted with an increasing gradient of
imidazole. Fractions are collected in tubes containing glycerol in
water Tris pH 8. Preferably, the glycerol solution is prepared the
day of the column run.
[0038] An aliquot from each fraction is spotted on a membrane which
is stained with amido black to determine which fractions contain a
significant amount of protein. Alternatively, a small amount, for
example 5 .mu.l, from each fraction may be used for gel analysis
using Coomassie staining. The fractions with a significant amount
of protein are pooled, and the pool is then concentrated with, for
example, a diafiltration unit.
[0039] In some cases, aggregation of polypeptide may occur. In
order to eliminate aggregates and further facilitate purification,
an inhibitor of TACE, such as a hydroxamate-based inhibitor, may be
added to the concentrated sample from a stock solution in water,
and octylglucoside (commercially available from Boehringer
Mannheim) is added from a stock solution in water. The sample is
then incubated at room temperature for 15-24 hours.
[0040] Following incubation, the sample is applied to a size
exclusion column. The column is first equilibrated with a suitable
buffer, such as a buffer of 10 mM Tris pH 7.5, 100 mM NaCl, 10%
glycerol. Suitable size exclusion columns include, for example, LKB
2135-365, packed with TSK-G3000 SWG or the like such as
Superdex-200. The buffer is then pumped through the column. The
highly purified TACE polypeptide can be detected by absorption at
280 nm.
[0041] A gel analysis of all fractions with significant protein is
carried out to determine which fractions should be pooled. The size
exclusion chromatography pool is concentrated using, for example, a
diafiltration unit.
[0042] A binding partner, such as an inhibitor, may then be added
to the purified sample. The binding partner is particularly usefull
in stabilizing the TACE polypeptide. The binding partner may be any
suitable compound. Suitable binding partners include, for example,
hydroxamate-based inhibitors. One suitable inhibitor is
N-{D,L-2-(hydroxyaminocarbonyl)methyl-4methylpentanoyl}-L-3-amino-2-dimet-
hylbutanoyl-L-alanine, 2-(amino)ethyl amide. This inhibitor, as
well as other inhibitors, are described in U.S. Pat. No. 5,594,106
(Black et al.), herein incorporated in its entirety by
reference.
[0043] The protein complex can be stored at low temperature, for
example, at about 4.degree. C.
[0044] TACE Crystal and Methods of Crystallization of TACE
Polypeptides
[0045] One aspect of the invention relates to a method of
crystallizing a TACE polypeptide. A preferred method comprises
co-crystallizing a TACE polypeptide with a binding partner
described above. Exemplary means for obtaining the TACE
polypeptide, as well as purification of the polypeptide are
described above.
[0046] Crystals may be grown or formed by any suitable method,
including drop vapor diffusion, batch, liquid bridge, and dialysis,
and under any suitable conditions. Crystallization by drop vapor
diffusion is often preferable. In addition, those of skill in the
art will appreciate that the crystallization conditions may be
varied. Various methods of crystallizing polypeptides are generally
known in the art. See, for example, WO 95/35367, WO 97/15588, EP
646 599 A2, GB 2 306 961 A, and WO 97/08300.
[0047] In one embodiment of the invention, a DNA construct
comprising TACE residues 1-477, with Ser266 changed to Ala, Asn452
changed to Gln, and the sequence Gly-Ser-(His).sub.6 added to the
C-terminus, may be expressed in CHO cells. These cells primarily
secrete a processed mixture of TACE, about half beginning with
Val212 and about half with Arg215. The mixture is purified as
described above. The purified TACE polypeptide, with the added
binding partner, is stored in a buffer as described above.
[0048] The TACE polypeptide and binding partner are
co-crystallized. The TACE/binding partner solution, at a
polypeptide concentration of about 5 mg/mL to about 12 mg/mL in a
TACE buffer described above, is mixed with a suitable
crystallization buffer and crystallized using a suitable
crystallization technique, for example drop vapor diffusion.
Suitable crystallization buffers, for example, include: 0.1 M Na
Acetate pH 5.3, 0.2 M CaCl.sub.2, 30% v/v Ethanol; 0.1 M Na Citrate
pH 5.0, 40% v/v Ethanol; 0.1 M Na Citrate pH 8.7, 20% w/v PEG 4000,
20% v/v Isopropanol; and 0.1 M Na Citrate pH 5.4, 20% w/v PEG 4000,
20% v/v Isopropanol. The sample is incubated at a temperature
ranging from about 4 to 20 degrees Celsius. A crystalline
precipitate is formed.
[0049] Seeds from the crystalline precipitate obtained, as whole
crystals or crushed crystal suspensions, are transferred, along
with a suitable crystallization promoter, such as hair of rabbit,
to a solution of concentrated TACE/substrate in a crystallization
buffer. Crystals suitable for X-ray data collection are formed.
[0050] Another aspect of the invention relates to a TACE
polypeptide crystal. One such crystal comprises a TNF.alpha.
converting enzyme catalytic domain (TCD) polypeptide
co-crystallized with an inhibitor. The crystal diffracts to about 2
A and belongs to the monoclinic space group P2.sub.1. The crystal's
unit cell comprises four crystallographically independent TCD
molecules. The TCD molecules are in an asymmetric unit and are not
clustered into separate tetrameres, but are integrated into the
infinite periodic structure. The crystal has the cell constants:
a=61.38 .ANG. (angstrom), b=126.27 .ANG., C=81.27 .ANG. and
.beta.=107.41.degree..
[0051] X-Ray Diffraction
[0052] Another aspect of the invention relates to the structure of
TACE, particularly the structure of the TACE catalytic domain
(TCD). The structure of TACE can be determined utilizing a crystal
comprising a TACE polypeptide as described above. According to the
present invention, the structure of TACE, and particularly the TCD,
is determined using X-ray crystallography. Any suitable X-ray
diffraction method for obtaining three dimensional structural
coordinates of a polypeptide may be used. The three-dimensional
structure coordinates, or any part thereof that characterizes the
part of the TACE polypeptide of interest, such as the TACE
catalytic domain or part thereof that is capable of cleaving the
peptide PLAQAVRSSS, can be used as described herein.
[0053] Methods of Using TACE X-Ray Diffraction Coordinates
[0054] The invention also relates to use of the structure
coordinates obtained from the above described X-ray diffraction
studies of the TACE catalytic domain. The coordinates may be
utilized, by direct analysis, with the aide of computers, or
combinations thereof, to determine the structure, including
secondary and tertiary structure, of the TACE catalytic domain. The
TACE catalytic domain structure coordinates also may be used to
develop, design, and/or screen compounds that associate with TACE.
As used herein, "associate" means that the compound may bind to or
interact with TACE ionically, covalently, by hydrogen bond, van der
Waals interaction, salt bridges, steric interaction, hydophilic
interactions and hydrophobic interaction. Moreover, the term
"associate" encompasses associations with any portion of the TACE
catalytic domain. For example, compounds that associate with TACE
may be compounds that act as competitive inhibitors, un-competitive
inhibitors, and non-competitive inhibitors. Compounds that
associate with TACE also may be compounds that act as mediators or
other regulatory compounds. Compounds that associate with TACE also
may be compounds that isomerize to short-lived reaction
intermediates in the chemical reaction of substrate with TACE. In
particular, compounds designed to associate with TACE may be used
therapeutically as inhibitors, mediators and other regulatory
compounds.
[0055] The use of X-ray coordinates for structure determination,
molecular design and selection and synthesis of compounds that
associate with other polypeptides is known in the art. Published
PCT application WO 95/35367 describes the use of X-ray structure
coordinates to design, evaluate, synthesize and use compounds that
associate with the active site of an enzyme. UK Patent Application
2306961A describes the use of X-ray coordinates in rational drug
design. Published PCT application, WO 97/15588 describes the
structural determination of a polypeptide using x-ray diffraction
patterns as well as use of the coordinates and three-dimensional
structure in finding compounds that associate with the polypeptide
of interest. This invention, however, for the first time allows the
use of X-ray coordinates for a TACE polypeptide for structural
determination, molecular design, and selection and synthesis of
compounds that associate with TACE.
[0056] In one aspect of the invention, the structure coordinates
obtained by the foregoing methods may be displayed as, or converted
to, a graphical representation, including three-dimensional shape
representations. This may be accomplished using commercially
available computer programs capable of generating graphical
representations of molecules, or parts thereof, from a set of
structural coordinates. Examples of computer programs capable of
generating graphical representations of molecules, or parts
thereof, from a set of structural coordinates are described in
published PCT application WO 97/08300, incorporated in the entirety
by reference.
[0057] In another aspect of the invention, the structure
coordinates and structure may be compared to, or superimposed over,
other similar molecules, such as other metalloproteinases. For
example, the TACE structure coordinates and structure may be
compared to or superimposed over the structure coordinates or
structure of snake venom metalloproteinases, such as, for example,
adamalysin II. The TACE structure coordinates and structure also
may be compared to or superimposed over the structure coordinates
or structure of matrix metalloproteinases, such as ADAM 10,
including human ADAM 10. Comparison of TACE and other molecules for
which a graphical structure or three-dimensional structural
coordinates are available may be carried out with the aide of
available software applications, such as the Molecular Similarity
application of QUANTA (Molecular Simulations, Inc., Waltham,
Mass.).
[0058] Compounds that associate with TACE also may be
computationally evaluated and designed by screening and selecting
chemical entities or fragments for their ability to associate with
TACE, and specifically the TACE catalytic domain. Several methods
may be used to accomplish this aspect of the invention. In one
embodiment, one may visually inspect a computer-generated model of
TACE, and specifically the catalytic domain, based on the structure
coordinates described herein. Computer generated models of chemical
entities or specific chemical moieties can then be positioned in or
around the catalytic domain and evaluated based on energy
minimization and molecular dynamics, using, for example, available
programs such as CHARMM or AMBER. Positioning of the chemical
entity or fragment can be accomplished, for example with docking
software such as Quanta and Sybyl. Additionally, known and
commercially available computer programs may be used in selecting
chemical entities or fragments. Once suitable chemical entities or
fragments are selected, they may be assembled into a single
compound, such as an inhibitor, mediator, or other regulatory
compound. Known and commercially available model building software
may assist in assembly.
[0059] In one aspect of the invention, compounds that associate
with TACE and specifically the TACE catalytic domain may be
designed as a whole, rather than by assembly of specific chemical
moieties or chemical entities. This embodiment may be carried out
using computer programs such as LUDI (Biosym Technologies, San
Diego, Calif.), LEGEND (Molecular Simulations, Burlington, Mass.),
and Leap Frog (Tripos Associates, St. Louis, Mo.).
[0060] In one embodiment, a candidate compound is chosen based upon
the desired sites of interaction with TACE and the candidate
compound in light of the sites of interaction identified
previously. Once the specific candidate compound-TACE interactions
are determined, docking studies, using commercially available
docking software, are performed to provide preliminary "modeled"
complexes of selected candidate compound with TACE.
[0061] Constrained conformational analysis is performed using, for
example, molecular dynamics (MD) to check the integrity of the
modeled TACE-inhibitor complex. Once the complex reaches its most
favorable conformational state, the structure as proposed by the MD
study is analyzed visually to ensure that the modeled complex
complies with known experimental SAR/QSAR (structure-activity
relationship/quantitative structure-activity relationship) based on
measured binding affinities.
[0062] Other modeling techniques also may be used in accordance
with the invention. Examples of these techniques are disclosed in
Cohen et al., "Molecular Modeling Software and Methods for
Medicinal Chemistry," J. Med. Chem., 33:883-894 (1990) and Navia et
al., "The Use of Structural Information in Drug Design," Current
Opinions in Structural Biology, 2:202-210 (1992), herein
incorporated by reference in the entirety.
[0063] Compounds developed or designed to associate with TACE may
be optimized or the efficiency of association can be tested using a
number of methods known in the art. For example, the deformation
energy and electrostatic interactions may be determined and
optimized. Known and commercially available software and hardware
systems may be used. Examples of such software are disclosed in WO
95/07619. Structure-based analoging for optimization of the
inhibitor potency, selectivity and physical drug-like properties in
an iterative manner also may be performed by one skilled in the art
of drug design.
[0064] Substitutions also may be made to selected or designed
compounds. These substitutions can be made to improve or modify the
association properties of the compound. Such substitutions may be
made, for example, in side groups or particular atoms of the
compounds. Generally, one should begin with conservative
substitutions that have approximately the same size, shape, charge
and other characteristics of the original group or atom.
Substituted compounds may be further analyzed and optimized as
described above.
[0065] In a further aspect of the invention, the potential
inhibitory, mediatory, regulatory, or other binding effect of a
compound may be analyzed and evaluated, using, for example,
commercially available computer software, prior to actual synthesis
and testing of such compound. In this way, one can evaluate the
probability of synthesizing and testing of inoperative
compounds.
[0066] Procedures for measuring inhibition generally are known in
the art and are disclosed, for example, in PCT 96141624. Such
methods include assays based on reaction with a peptide
substrate.
[0067] TACE Catalytic Domain Structure
[0068] The physical features of the TCD, determined based on the
X-ray diffraction data obtained using the methods described and its
use in creating molecular models of the TCD, are further described,
with reference to the Figures.
[0069] The domain depicted in FIG. 1 has the shape of an oblate
ellipsoid, notched at its flat side to give a relatively small
active-site cleft separating the small "lower" subdomain from the
"upper" main molecular body (FIG. 2a). The TCD polypeptide chain
start on the molecular surface (in the lower back, FIG. 1), with
the chain becoming well defined between Asp217 and Met221 (see FIG.
3). Central to the molecule is the five-stranded .beta.-pleated
sheet, with the .beta.-strands arranged in the order (from back to
front, see FIG. 1) sII, sI, sIII, sV and sIV (see FIG. 3), with
sIV, the "edge" strand, running antiparallel to the others. This
.beta.-sheet is highly twisted flanked by two .alpha.-helices (hB
and hB2) on its convex and two helices (hA and hC) on its concave
side. The .beta.-strands sI and sII are connected by the short
.alpha.-helix hAI and the long .alpha.-helix hA (the obliquely
running helix on the backside, FIG. 1). The .beta.-strands sII and
sIII are linked by the large "multiple-turn loop", the long
"intermediate" .alpha.-helix hB and the adjacent short =60 -helix
hB2, all of them arranged on "top" of the .beta.-sheet thus fully
shielding its central part from bulk water (FIG. 1). The
multiple-turn loop is bulged out at two sites giving rise to a
"spur-like" and a quite acidic protuberance, respectively (visible
in FIG. 2a on top of the molecule). The sIII-sIV linker terminates
in a short "bulge", before it enters the edge strand sIV. The
sIV-sV connecting segment is dissected into two large "ear-like"
surface-located loops, a first one nestling to the main molecular
body (giving rise to the "blue" surface, center left, in FIG. 2a),
and a long .beta.-hairpin loop (sIIa-sIIb) projecting from the
molecular surface (top left in FIGS. 1 and 2). A bulged-out loop
links sV with the "active-site helix" hC, which is located in the
center of the molecule and stops abruptly at the strictly conserved
Gly412, where the chain kinks down to build the lower
subdomain.
[0070] The C-terminal chain comprising the last 61 TCD residues
(FIG. 3) first forms three short straight almost perpendicularly
arranged segments linked by two "narrow" supertwisted loops,
returns via the tight "Met-turn" Tyr433-Val434-Met435-Tyr436 back
to the surface where it kinks at Pro437 to form the
Pro437-Ile438-Ala439 outer "wall" of the S1' crevice, approaches in
a wide loop the C-terminal .alpha.-helix hD and runs through it,
and ends up on the molecular "back" surface close to the
N-terminus, with the last defined residues Arg473-Ser474 fixed via
hydrogen bonds to the main molecular body. Via Cys423-Cys453, the
first of the two "narrow" loops is disulfide-linked with the
N-terminus of helix hD, whose C-terminal end in turn is clamped to
the "ear-like" sIV-sV linker peptide through Cys365-Cys469.
Spatially adjacent, the third disulfide bridge of TCD,
Cys225-Cys333, connects the N-terminal parts of .beta.-strands sI
and sIII. In the intact TACE molecule, four residues downstream of
Ser474 would reside Cys478, which is already integral part of the
compact elongated disintegrin domain (Saudek et al.,
"Three-dimensional structure of echistatin, the smallest active RGD
protein" Biochem. 30, 7369-7372 (1991)). Considering Ser474 and
this Cys478 as pivot points of their respective domains, the three
residue linker would allow relatively unconstrained docking of the
disintegrin domain to the "left" surface side of the catalytic
domain.
[0071] The active-site cleft of TACE (FIG. 2a) is relatively flat
on the left hand (non-primed) side, but becomes notched towards the
right. The catalytic zinc residing in its center is pentaoordinated
by the three imidazole N.epsilon.2 atoms of His4O5, His4O9 and
His415 (provided by the active-site helix and the following
"descending" chain comprising the conserved zinc binding consensus
motif HEXXHXXGXXH), and by the carbonyl and the hydroxyl oxygen of
the hydroxamic acid moiety of the inhibiter (see FIGS. 1, 2a and
4). This zinc-imidazole ensemble is based on the distal
.epsilon.-methyl-sulphur moiety of the strictly conserved Met435,
harbored in the Met-turn characteristic for the metzincin clan
(Bode et al., "Astacins, serralysins, snake venom and matrix
metalloproteinases exhibit identical zinc binding enviromnents
(HEXXHGXXH and Met-turn) and topologies and should be grouped into
a common family, the `metzincins`" FEBS Lett. 331, 134 140 (1993);
Stocker et a., "The metzincins: Topological and sequential
relations between the astacins, adamalysins, serralysins,
andmatrixins (collagenases) define a superfamily of
zinc-peptidases" Protein Sci. 4, 823-840 (1995)). Both carboxylate
oxygens of the "catalytic" Glu4O6 (which acts as a general base
during catalysis (Grams et al., "X-ray structures of human
neutrophil collagenase complexed with peptide hydroxamate and
peptide thiol inhibitors: Implications for substrate binding and
rational drug design" Eur. J. Biochem. 228, 830-841 (1995))
squeezed between the zinc-liganding imidazole of His4O5 and the
edge strand, are hydrogen bonded to the hydroxyl and the N--H group
of the hydroxamic acid (see FIG. 4). To the right of the catalytic
zinc opens the deep S1' pocket, which, besides the S1' wall-forming
segment (bottom, FIGS. 1 and 2a), is bordered by the side chains of
His4O5 and Glu4O6 (left), the sIV main chain and the Leu345 side
chain (top), and the side chains of Val440 (back) and Ala439
(right). To the right of Ala439 opens a second (S3') pocket, which
inside the molecule merges with the S1' pocket, leaving a small
bridge made of the opposing side chains of Ala439 and Leu348 (FIG.
2a).
[0072] The (pseudo)peptidic part of the inhibitor binds in an
extended geometry to the notched right-hand side of the active-site
cleft, mimicking the interaction of the primed residues of a
productively bound peptide substrate (FIG. 2a). It runs
antiparallel to the upper short bulge Gly346-Thr347-Leu348 and
parallel to the S1' wall-forming segment Pro437-Ile438-Ala439,
making two and two inter-main chain hydrogen bonds, respectively.
The dominant intermolecular interactions are made by the P1'
isobutyl (pseudo-leucyl) side chain of the inhibitor and the
essentially hydrophobic S1' pocket, however, is large and
accommodates three partially ordered solvent molecules in addition.
The P2' t-butyl side chain extends away from the enzyme, but
nestles to the hydrophobic canopy above formed by the enzyme's
bulge. The P3' Ala side chain points into the large negatively
charged S3' pocket, but is too short to make favorable contacts.
The C-terminal diaminoethyl group has different conformations in
the four molecules.
[0073] The P1' to P3' segment Val77-Arg78-Ser79 of a bound
pro-TNF.alpha. probably binds in a similar manner, possibly under
better matching with the underlying cleft surface; the preceding P3
to P1 residues Ala74Gln75-Ala76 certainly will align antiparallel
to the edge strand, with their side chains extending into the
(partially charged) S3 pocket and the (negatively charged) shallow
S2 depression, and projecting out of the central cleft,
respectively. The primed subsites and surrounding molecular
surfaces of TACE are dominated by negative charges, while the
non-primed subsites are essentially hydrophobic in nature (FIG.
2a). More distant interactions may be involved in the specificity
of TACE for processing pro-TNF.alpha.. The 12 residue substrate
comprising the pro-TNF.alpha. cleavage site can also be split by
some of the MMPs, although with less specificity and efficacy
(Black et al., "Relaxed specificity of matrix metalloproteinases
(MMPs) and TIMP intensity of tumor necrosis factor-.alpha.
(TNF-.alpha.) production suggest the major TNF-.alpha. converting
enzyme is not an MMP" Biochem. Biophys. Res. Comm. 225, 400-405
(1996)). Thus, the preferential processing of the (probably
trimeric) (Tang et al., "Human pro-tumor necrosis factor is a
homotrimer" Biochem. 35, 8216-8225 (1996a); Tang et al., "Length of
the linking domain of human pro-tumor necrosis factor determines
the cleavage processing" Biochem. 35, 8226-8233 (1996b))
membrane-bound pro-TNF.alpha. in vivo might in part be due to
correct assembling, i.e. suitable presentation of the
pro-TNF.alpha. cleavage segment to the TACE active site in a
distinct distance from the anchoring membrane. Some experimental
evidence (Tang et al., Biochem. 35, 8216-8225 (1996a); Tang et al.,
Biochem. 35, 8226-8233 (1996b)) suggests that the cleavage site
might not be determined by the cleavage sequence alone, but that
also the distance to the base of the compact cone formed by the
associated C-terminal segments of three TNF.alpha. molecules (Jones
et al., "Structure of tumor necrosis factor" Nature 338, 225-228
(1989)) plays a role. In a productive TACE-proTNF.alpha. complex,
the base of this TNF.alpha.-trimer cone (into which the disordered
N-termini run up) may be recognized by the "right" side of the TACE
catalytic domain (FIG. 2a), with the about 10 residues long spacer
favoring the correct placement of the proTNF.alpha. Ala76-Val77
scissile peptide bond in the active site of TACE.
[0074] The polypeptide topology and in particular the surface
presentation of the catalytic zinc prove the catalytic domain of
TACE to be a typical metzincin. (Bode et al., "Astacins,
serralysins, snake venom and matnrix metalloproteinases exhibit
identical zinc binding environments (HEXXHXXGXXH and Met-turn) and
topologies and should be grouped into a common family, the
`metzincins`" FEBS klet. 331, 134-140 (1993); Stocker et al., "The
metzincins: Topological and sequential relations between the
astacins, adamalysins, serralysins, and matrixins (collagenases)
define a superfamily of zinc-peptidases" Protein Sci. 4, 823-840
(1995)) A superposition with the other metzincins shows, however,
that its topology is most similar to that of the catalytic domain
of snake venom metalloproteinases such as adamalysin II (FIG. 5).
(Gomis-Ruth et al., "First structure of a snake venom
metalloproteinase: prototype for matrix
metalloproteinases/collagenases" EMBO J. 12, 4151-4157 (1993);
Zhang et al., "Structural interaction of natural and synthetic
inhibitors with the venom metalloproteinase, atrolysin C (form d)"
Proc. Natl. Acad. Sci. USA 91, 8447-8451 (1994); Kumasaka et al.,
"Crystal structure of H2-proteinase from the venom of Trimeresurus
flavoviridis" J. Biochem. 119, 49-57 (1996)) This close homology is
reflected by the much better simultaneous superposition of the
central sheet and the large helices, but in particularly also by a
couple of structural features, which TACE shares exclusively with
the adamalysins such as: the long helix hB and the preceding
multiple-turn loop arranged on top of the .beta.-sheet; the
typically arranged and shaped C-terminal helix hC; and the extended
C-terminus placed on the backside surface. About 175 of the 263
TACE and 201 adamalysin .alpha.-atoms are topologically equivalent
(with an rms deviation of 1.3 .ANG., 39 of which have identical
side chains (FIG. 3). These numbers are close to those obtained
from a comparison of members within the different metzincin
families. (Stocker et at., supra) In addition, detailed structural
features prove the close relationship of TACE to the adamalysins: a
more conserved core structure; the loosely arranged N-terminus; the
characteristic Asp416 (directly following the zinc binding
consensus motif, FIG. 3) involved in identical intramolecular
hydrogen bond interactions; the adjacent disulfide bridge
Cys423-Cys453 linking the first narrow loop to the C-terminal helix
hD (which TACE does not share with adamalysin II, but with the
H2-proteinase from the snake venom of T. flavoviridis) (Kumasaka et
al., supra); disulfide bridge Cys365-Cys469 connecting the sIV-sV
linker with the C-terminal helix hD; a similarly shaped active-site
cleft, with particularly strong similarities in the S1' pocket and
other primed subsites.
[0075] The catalytic domain of TACE (TCD) also differs from
adamalysin II in several respects: with 263 residues, its chain is
much longer; most of the additional residues of TACE are clustered
giving rise to a more projecting hA-sII turn, to the two surface
protuberances of the multiple-turn loop, to the two "ears" of the
sIV-sV linker, and to a more bulged-out sV-hC connector (see FIGS.
3 and 5); lack of a calcium binding site but presence of a third
disulfide bridge Cys225-Cys333 in TACE, both elements serving,
however, for the same function namely to clamp the N-terminal chain
to strand sill; the quite deep S3' pocket of TACE which merges with
its S1' pocket; an almost inverted charge pattern in and around the
primed subsites, with an absolute predominance of positive charges
in adamalysin.
[0076] According to its sequence, and probably with respect to its
three-dimensional structure, the TACE catalytic domain is thus not
a typical member of the mammalian ADAMs proper (a family of
membrane-anchored cell-surface proteins, with the catalytic domain
quite homologous to adamalysin (Wolfsberg et al., "ADAMs in
Fertilization and Development" Developm. Biol. 180, 389-401
(1996))) TACE presumably shares this "outsider" role with (bovine)
ADAM 10 (FIG. 3), which does also possess some TACE-like activity
(Lumn et al., "Purification of ADAM 10 from bovine spleen as a
TNF.alpha. convertase," FEBS Lett. 400, 333-335 (1997)), and whose
Drosophila version (kuz) has recently been shown to process the
Notch receptor (Rooke et al., Science 273, 1227-1231 (1996)). Also
ADAM 10 probably exhibits an elongated hA-sII loop and the two
"ears" typical for TACE, but might have a multiple-turn
intermediate in size between TACE and adamalysin (see FIG. 3).
Ninety of the ADAM 10 catalytic domain residues are identical to
TACE farther underlining the close homology (see FIG. 3), whereas
the other mammalian ADAMs probably resemble much more adamalysin II
(Gomis-Ruth et al., "Refined 2.0 A crystal structure of snake venom
zinc endopeptidase adamalysin II" J. Mol. Biol. 239, 513-544
(1994)).
[0077] The structural homology of TACE to the MMPs is significantly
lower. The relative arrangement of the common secondary structural
elements differs more (reflected by the significantly larger rms
deviation of 1.6 .ANG. of the about 120 topologically equivalent
C.alpha.-atoms), and the MMPs lack characteristic TACE/adamalysin
structural elements (such as the intermediate helix hB and the
multiple-turn loop, the Asp residue behind the third zinc-binding
histidine), or exhibit typical determinants (such as the structural
zinc and the integrated calcium ions) not seen in TACE.
Notwithstanding the differences in secondary structure, the
active-site cleft of TACE bears some similarity with that of the
MMPs, with the flat nonprimed (left-hand) side, and the narrow
primed side centering in the deep S1' pocket (FIG. 2b). This
subsite similarity to the MMPs explains the observed partial
sensitivity of TNF-.alpha.-convertase activity towards synthetic
hydroxamic acid inhibitors originally designed for inhibition of
various MMPs (DiMartino et al., "Anti-arthritic activity of
hydroxamic acid-based pseudopeptide inhibitors of matrix
metalloproteinases and TNF.alpha. processing" Inflamm. Res. 46,
211-215 (1997)). Model building experiments with TIMP-1 structure
(Gomis-Ruth et al., "Mechanism of inhibition of the human matrix
metalloproteinase stromelysin-1 by TIMP-1" Nature 389, 77-81(1997))
show no obvious obstacles in the active-site region of TACE that
would easily explain its resistance to blockage by the TIMPs.
[0078] This TCD crystal structure thus gives evidence for a
topological similarity of the catalytic domain of TACE with that of
the adamalysins/ADAMs, and for a share of its substrate binding
site to that of the MMPs. TACE exhibits, however, several
structural peculiarities regarding surface contour, charge and
shape, which facilitates the design of potent selective synthetic
inhibitors.
[0079] In designing and developing compounds, such as inhibitors,
mediators and other compounds having activities with biological
significance, that associate with TACE, it is desirable to select
compounds with a view toward the particular surface contour,
charge, shape, and other physical characteristics of the TACE
catalytic domain. Generally, the compounds should be capable of
physically and structurally associating with TACE, as well as be
able to assume a conformation that allows it to associate with
TACE. The features described above will direct the skilled artisan
in this regard. In particular, compounds with a linear
functionality should be particularly suitable. Such compounds will
be particularly suitable in light of the deep pockets of the TACE
catalytic domain.
[0080] The compounds that associate with TACE, for example, may be
designed to associate with the S1' region or the S1'S3' pocket of
TACE. Compounds that associate with TACE also may be designed to
(i) incorporate a moiety that chelates zinc. Further exemplary
compounds include compounds are designed to form a hydrogen bond
with Leu348 or Gly349 of TACE, (ii) introduce a non-polar group
which occupies the S1' pocket of TACE, (iii) introduce a group
which lies within the channel joining S1'-S3' pockets of TACE and
which makes appropriate van der Waal contact with the channel, and
(iv) form a hydrogen bond with Leu348 or Gly349 on the backbone
amide groups of TNF-converting enzyme, or (v) any combination of
the above.
[0081] Computer-Readable Medium
[0082] The present invention also relates to a computer-readable
medium having recorded thereon the x-ray diffraction structure
coordinates of a crystalline TACE polypeptide. The
computer-readable media of the invention are useful for storage,
transfer, and use with software of the TACE structural coordinates.
The computer readable medium may be any suitable data storage
material, including, but not limited to, a floppy disc, a hard
disc, computer-type Random Access Memory, Read-Only Memory flash
memory, CD-ROM, recordable and rewritable CDs, recordable and
rewritable DVDs, magneticoptical disk, ZIP drive, JAZ drive,
Syquist drive, digital tape drive, or the like. Other suitable
media will be known to those of skill in the art.
[0083] In one embodiment, the computer readable medium comprises
the coordinates of Table 1 or a substantial portion thereof. The
computer-readable medium may be used in conjunction with a machine
programmed with instructions for using the data recorded on the
medium, such as a computer loaded with one or more programs
identified throughout the specification, to display a graphical,
three-dimensional representation of a TACE polypeptide, or any part
thereof.
[0084] Computer Based System
[0085] FIG. 6 illustrates a system 1000 for studying a TACE
polypeptide. The system includes a video memory 110 that stores
information representing at least a portion of a TACE polypeptide.
The memory has at least one first-type storage region 112, having
recorded thereon a set of spatial coordinates specifying a location
in a three dimensional space, and at least one second-type storage
region 114, having recording thereon information representing a
characteristic of one of a plurality of amino acids. The
second-type storage regions are logically associated with the
first-type storage regions in the video memory 110 to represent a
geometric arrangement of at least one characteristic of at least a
portion of the TACE polypeptide in the three dimensional space.
Memory, 112 and 114 can comprise, for example, the data shown in
Table 1. The system 1000 also includes a processor, coupled to the
memory to access the first-type storage regions 112 and the
second-type storage regions 114, to generate image signals for
depicting a visual three dimensional image of at least one
characteristic of at least a portion of the TACE polypeptide in the
three dimensional space based on data from the memory 110. The
processor can be any general purpose processor with a CPU,
register, memory and the like. A display 130 coupled to the
processor 120 via lines 125 to receive the image signals, for
depicting a visual three dimensional image of at least one
characteristic of at least a portion of the TACE polypeptide in the
three dimensional space based on the image data on a screen
132.
[0086] In one embodiment of the invention, the image data includes
data for depicting a visual three dimensional image of a ribbon
structure of at least a portion of a TACE polypeptide in three
dimensional space, such as shown in FIG. 1. In another embodiment,
the image data includes data for depicting a visual three
dimensional image of a solid model representation of at least a
portion of said TACE polypeptide in three dimensional space, such
as shown in FIG. 2. In still another embodiment, the image data
includes data for depicting a visual three dimensional image of
electrostatic surface potential of at least a portion of TACE
polypeptide in three dimensional space, such as shown in FIG. 2. In
yet another embodiment, the image data includes data for depicting
a visual three dimensional stereo image of at least a portion of a
TACE polypeptide in three dimensional space, such as shown in FIG.
4.
[0087] The system 1000 of the present invention may further
comprise a storage device 145 that stores data representing a
geometric arrangement of a characteristic of a composition other
than the TACE polypeptide and an operator interface, such as a
mouse 135, for receiving instructions from a Operator. Storage
device 145 can include, for example, the three-dimensional X-ray
coordinate data for other chemical entities. The processor 120 is
coupled to the storage device 145 and to said operator interface
135 and generates additional image data for depicting the geometric
arrangement of the characteristic of the composition relative to
said visual three dimensional image of said at least one
characteristic of said at least a portion of TACE polypeptide on
the screen 132 based on instructions from the operator interface.
In the FIG. 6 embodiment, the storage device 145 is part of the
memory 110.
[0088] The first-type storage regions 112 and said second-type
storage regions 114 are regions of, for example, a semiconductor
memory, regions of an optical disk, or regions of a magnetic
memory.
[0089] In one embodiment, processor 120 and video memory 110 are in
the form of a UNIX or VAX computer, such those available from
Silicon Graphics, Sun Microsystems, and IBM. However, the invention
is not limited to use of this particular hardware and software.
[0090] The invention is described in more detail in the following
illustrative examples. Although the examples may represent only
selected embodiments of the invention, it should be understood that
the following examples are illustrative and not limiting.
EXAMPLE 1
TACE Polypeptide Expression, Isolation, and Purification
[0091] A cDNA encoding the signal peptide, pro and catalytic
domains of TACE, amino acid residues 1-477, as disclosed in Black
et al., "A Metalloproteinase disintigrin that releases
tumour-necrosis factor-.alpha. from cells," Nature 385: 729-733
(February 1997), with Ser266 changed to Ala, Asn452 changed to Gln
and the sequence Gly-Ser-(His).sub.6 added to the C-terminus, was
inserted into an expression vector for CHO cells. The TACE
polypeptide was expressed in CHO cells and a mixture of the TACE
polypeptide beginning either with Val212 or Arg215 was secreted.
The cells were cultured in the drug, methotrexate, which kills
those cells that did not incorporate the vector.
[0092] The expressed TACE polypeptide was then purified.
Purification started with 5 liters of the medium containing the
expressed TACE polypeptide. The medium was concentrated to about
200 mL with a Millipore 10K cut-off, 1 ft.sup.2 TFF diafiltration
unit. The pumping rate was 50-100 mL/min. Two liters of a buffer
solution of 20 mM Tris (pH 7.5) and 300 mM NaCl (Buffer E) was then
added to the sample.
[0093] The sample was reconcentrated as described above and diluted
a second time with 2 liters of Buffer E, reconcentrated again,
diluted a third time with 2 liters of Buffer E, and reconcentrated
to about 100 mL. The sample retained in the diafiltration unit was
recovered by a back-flush. This material was then filtered through
a 0.45 .mu.m and was azide added to 0.05%. The filtered sample was
stored overnight at 4.degree. C.
[0094] After overnight storage, imidazole was added to the filtered
sample to 5 mM from a 200 mM stock in water and ZnCl.sub.2 was
added to 5 uM from a 1 M stock in water. The sample then was pumped
over 2.2 mL of Qiagen Ni-NTA Superflow resin (Cat. #30430) at 3
mL/min (column size 7.5.times.50 mm).
[0095] The column was washed at 5 mL/min with 100 mL of a buffer of
20 mM Tris pH 7.5, 300 mM NaCl, 5 mM imidazole, and S uM ZnCl.sub.2
(Buffer A). The protein was then eluted with an increasing gradient
of imidazole, going up to 200 mM in 1 minute (5 mL total volume),
followed by 35 mL of 200 mM imidazole in Buffer A. Two mL fractions
were collected, TACE generally coming off about 6 mL into the
elution. The fractions were collected in tubes containing 500 ul of
50% glycerol in water and 200 ul of 1 M Tris pH 8. The glycerol in
water was prepared the day of the column run.
[0096] A dot blot, with 3 .mu.l from each fraction, was stained
with amido black to determine which fractions contained a
significant amount of protein. The fractions with a significant
amount of protein were pooled. The pool was then concentrated to
1-2 mL with a 10 K cut-off Amicon Centriprep concentrator.
[0097] The inhibitor
N-{D,L-2(hydroxyaminocarbonyl)methylmethylpentanoyl}--
L-3-amino-2-dimethylbutanoyl-L-alanine, 2-(amino)ethyl amide was
added to the concentrated sample to 1 mM from a 50 mM stock in
water, and octylglucoside was added to 1% from a 10% stock in
water. The sample was then incubated at room temperature for 15-24
hours.
[0098] Following incubation, the sample was applied to a
21.5.times.600 mm size exclusion column, LKB 2135-365, packed with
TSK-G3000 SWG, and equilibrated with 10 mM Tris pH 7.5, 100 mM
NaCl, 10% glycerol. This buffer was then pumped through the column
at 2.5 mL/min for 100 minutes. The TACE polypeptide in the effluent
was detected by absorption at 280 nm. Excluded material generally
eluted at about 38 minutes. The pure TACE generally eluted at about
78 minutes or longer.
[0099] A gel analysis, with 15 .mu.l of all fractions with
significant protein was then carried out to determine which
fractions should be pooled. The size-exclusion chromatography pool
was concentrated to about 1 mL with a 10 K cut-off Amicon
Centriprep concentrator.
[0100] The inhibitor
N-{D,L-2(hydroxyaminocarbonyl)methylmethylpentanoyl}--
L-3-amino-2dimethylbutanoyl-L-alanine, 2-(amino)ethyl amide was
then added to the purified sample to a concentration of 1 mM. The
protein can be stored at 4.degree. C.
EXAMPLE 2
Protein Crystallization
[0101] A DNA construct comprising the prodomain and the catalytic
domain of human TACE (resides 1-477) was fused to the sequence
Gly-Ser-(His).sub.6 to facilitate purification of the protein on a
Ni-NTA affinity column. Chinese Hamster Ovary (CHO) were cells used
for protein expression. The cells secreted a mixture of mature TACE
beginning with either Val212 or Arg215. TACE-containing fractions
from the Ni-NTA column were incubated in a buffer containing
octylglucoside and the binding partner
N-[D,L-[2-(hydroxyaminocarnbonyl)methyl]-4-methylpentanoyl)-L-3-(-
tert-butyl)-glycyl-L-alanine. The final purification step was
performed on a gel filtration column. Purified TACE was stored in a
buffer containing 10 mM Tris/HCL pH 7.5, 100 mM NaCl, 10% glycerol
and 1 mM of inhibitor (TACE buffer).
[0102] Crystallization experiments were set up at a TACE
concentration of approximately 5 mg/mL by mixing TACE (in TACE
buffer) in a 1:1 ratio with the crystallization buffers listed
below and using the sitting drop vapor diffusion technique. The
experiments were performed in duplicate and incubated either at
about 4.degree. C. or at 20.degree. C. Crystalline precipitate was
obtained at 20.degree. C. in the following crystallization
buffers:
[0103] Buffer A) 0.1 M Na Acetate pH 5.3, 0.2 M CaCl.sub.2, 30% v/v
Ethanol
[0104] Buffer B) 0.1 M Na Citrate pH 5.0, 40% v/v Ethanol
[0105] Buffer C) 0.1 M Na Citrate pH 8.7, 20% w/v PEG 4000, 20% v/v
Isopropanol
[0106] Small crystals were obtained upon transferring seeds from
the crystalline precipitate with a hair of a rabbit into a 1:1
mixture of a concentrated sample of TACE (12 mg/mL in TACE buffer)
with either buffer B or C. Further refinement of buffer C resulted
in buffer D, which allowed the production of crystals suitable for
X-ray data collection.
[0107] Buffer D) 0.1 M Na Citrate pH 5.4, 20% w/v PEG 4000, 20% v/v
Isopropanol
[0108] The first data set was measured to a reduction of 2.5 .ANG.
on a MAR300 imaging plate scanner attached to a Rigaku-Denki
totaling Cu-anode generator operated at 5.4 kW providing
graphite-monochromatized CuK.alpha. radiation. The data were
processed with MOSFLM v. 5.23 program and routines of the CCP4
suite. All attempts to solve the structure by molecular replacement
methods using either adamalysin II, an all-alanine model of
adamalysin II and models generated failed to produce useful
starting points for phasing. Thus the locations of four independent
zinc atoms were determined with the help of an anomalous difference
Patterson synthesis. In order to measure MAD data, the crystals
were deep-frozen in liquid nitrogen. Therefore, crystals were
transferred into a cryo buffer (80% v/v buffer D containing 17% v/v
glycerol) with the help of a silk loop of appropriate size, soaked
for about 10 seconds and then immediately deep-frozen at 90 degrees
K.
[0109] The crystals obtained belong to the monoclinic space group
P2.sub.1, have cell constants a=61.38 .ANG. (angstrom), b=126.27
.ANG., c=81.27 .ANG., .beta.=107.41.degree., and contain four
molecules in the asymmetric unit.
EXAMPLE 3
X-Ray Diffraction
[0110] Using the crystals described in Example 2, a first data set
was measured to a resolution of 2.5 .ANG. on a MAR300 imaging plate
scanner attached to a Rigaku-Denki rotating Cu-anode generator
operated at 5.4 kW providing graphite-monochromatized CuK.alpha.
radiation. The data were processed at with the MOSFLM v. 5.23
program and routines of the CCP4 suite.
[0111] All attempts to solve the structure by molecular replacement
methods using either adamalysin II, an all-alanine model of
adamalysin II, and other models failed to produce useful starting
points for phasing.
[0112] Thus, the locations of the four independent zinc atoms were
determined with the help of an anomalous difference Patterson
synthesis. In order to measure MAD data, the crystals were
deep-frozen in a nitrogen gas stream cooled down to the temperature
of liquid nitrogen. The crystals were first transferred into a
cryo-buffer of 80% v/v Buffer D (0.1 M Na Citrate pH 5.4, 20% w/v
PEG 4000, 20% v/v Isopropanol) containing 17% v/v glycerol.
Transfer to the cryo-buffer was performed with the help of a silk
loop of appropriate size. The crystals were soaked in the
cryo-buffer for about 10 seconds and then immediately deep-frozen
at 90 K.
[0113] Anomalous diffraction data to 2.0 .ANG. were collected with
MAR345 imaging plate scanner at 90 K on the BW6 wiggler beamline of
DORIS (DESY, Hamburg, Germany), using monochromatic X-ray radiation
at the wavelengths of maximal f" (1.2769 .ANG.) and minimal f"
(1.2776 .ANG.) at the K absorption edge of zinc and at a remote
wavelength (1.060 .ANG.). The data were scanned and evaluated using
DENZO/SCALEPACK, yielding 77653 independent reflections from
1051836 measurements (96.9% completeness, R-merge 0.031 in
intensities).
[0114] MAD phases were refined and calculated with MLPHARE
including all measured data to 2.0 .ANG. resolution. Their initial
mean-figure-of-merit of 0.53 was increased to 0.76 by solvent
flattening/histogram matching methods applying DM. This density
allowed building of the complete chains of the four independent
TACE catalytic domains and the bound hydroxamic acid substrates on
an SGI system using TURBO-FRODO. This model was
crystallographically refined with XPLOR and with CCP4 routines to a
crystallographic R factor of 18.6% (R.sub.free 27.4%) using 79400
independent reflections from, 12.0 to 2A. resolution.
[0115] Four independent TACE molecules form the periodic
arrangement.
[0116] Molecules 1 and 2, and 3 and 4 are defined from Asp219 and
Met221, respectively, to Ser474.
EXAMPLE 4
X-Ray Diffraction
[0117] Anomalous dispersion diffraction data to 2.0 .ANG. were
collected with a MAR345 imaging plate scanner at 100 K on the
wiggler beamline of DORIS (DESY, Hamburg, Germany), using
monochromatic X-ray radiation of maximal f' (1.2797 .ANG.) and
minimal f' (1.2804 .ANG.) at the K absorption edge of zinc and at a
remote wavelength (1.060 .ANG.). These data were evaluated and
scanned using DENZO/SCALEPACK, yielding 77,653 independent
reflections (96.9% completeness, R-merge 0.031).
[0118] The structure coordinates obtained are reproduced in Table
1.
1TABLE 1 REMARK Created by MOLEMAN V. 961218/7.2.5 at Fri Sep 19
20:05:05 1997 for user carlos REMARK MoleMan PDB file CRYST1 61.387
126.278 81.273 90.00 107.42 90.00 P 21 4 ORIGX1 1.000000 0.000000
0.000000 0.00000 ORIGX2 0.000000 1.000000 0.000000 0.00000 ORIGX3
0.000000 0.000000 1.000000 0.00000 SCALE1 0.016290 0.000000
0.005111 0.00000 SCALE2 0.000000 0.007919 0.000000 0.00000 SCALE3
0.000000 0.000000 0.012896 0.00000 Atom Type Residue ? # X Y Z OCC
B ATOM 1 N ASP A 219 0.865 33.077 15.204 1.00 20.00 ATOM 2 OD2 ASP
A 219 5.154 33.868 14.335 1.00 20.00 ATOM 3 OD1 ASP A 219 4.450
35.924 14.844 1.00 20.00 ATOM 4 CG ASP A 219 4.191 34.718 14.461
1.00 20.00 ATOM 5 CB ASP A 219 2.738 34.303 14.156 1.00 20.00 ATOM
6 CA ASP A 219 2.290 33.026 14.883 1.00 20.00 ATOM 7 C ASP A 219
3.166 32.889 16.123 1.00 20.00 ATOM 8 O ASP A 219 3.439 33.884
16.819 1.00 20.00 ATOM 9 N PRO A 220 3.629 31.679 16.386 1.00 20.00
ATOM 10 CG PRO A 220 4.073 29.436 16.118 1.00 20.00 ATOM 11 CD PRO
A 220 3.224 30.531 15.588 1.00 20.00 ATOM 12 CB PRO A 220 4.893
29.974 17.303 1.00 20.00 ATOM 13 CA PRO A 220 4.523 31.452 17.495
1.00 20.00 ATOM 14 C PRO A 220 5.649 32.530 17.443 1.00 20.00 ATOM
15 O PRO A 220 6.513 32.741 18.173 1.00 20.00 ATOM 16 N MET A 221
5.766 33.341 16.625 1.00 48.83 ATOM 17 CE MET A 221 9.090 36.336
12.584 1.00 53.01 ATOM 18 SD MET A 221 9.248 36.147 14.337 1.00
54.21 ATOM 19 CG MET A 221 8.515 34.606 14.801 1.00 51.15 ATOM 20
CB MET A 221 7.101 34.778 15.298 1.00 48.69 ATOM 21 CA MET A 221
6.875 34.306 16.701 1.00 46.22 ATOM 22 C MET A 221 6.485 35.500
17.614 1.00 42.51 ATOM 23 O MET A 221 7.279 36.002 18.427 1.00
43.93 ATOM 24 N LYS A 222 5.215 35.817 17.508 1.00 36.53 ATOM 25 NZ
LYS A 222 1.844 39.934 13.657 1.00 40.05 ATOM 26 CE LYS A 222 2.513
39.901 14.974 1.00 39.09 ATOM 27 CD LYS A 222 2.353 38.522 15.613
1.00 38.20 ATOM 28 CG LYS A 222 3.646 38.146 16.312 1.00 36.27 ATOM
29 CB LYS A 222 3.345 37.404 17.597 1.00 33.97 ATOM 30 CA LYS A 222
4.567 36.853 18.299 1.00 32.39 ATOM 31 C LYS A 222 4.144 36.220
19.633 1.00 29.13 ATOM 32 O LYS A 222 2.999 35.844 19.866 1.00
26.54 ATOM 33 N ASN A 223 5.157 36.011 20.462 1.00 23.62 ATOM 34 CA
ASN A 223 4.951 35.295 21.704 1.00 22.97 ATOM 35 CB ASN A 223 5.756
33.987 21.611 1.00 25.44 ATOM 36 CG ASN A 223 7.229 34.245 21.372
1.00 26.32 ATOM 37 OD1 ASN A 223 7.973 33.261 21.243 1.00 29.74
ATOM 38 ND2 ASN A 223 7.688 35.482 21.319 1.00 25.96 ATOM 39 C ASN
A 223 5.327 36.123 22.908 1.00 18.46 ATOM 40 O ASN A 223 5.365
35.556 23.983 1.00 18.08 ATOM 41 N THR A 224 5.611 37.408 22.709
1.00 17.03 ATOM 42 CA THR A 224 6.035 38.246 23.824 1.00 16.24 ATOM
43 CB THR A 224 7.507 38.721 23.599 1.00 17.52 ATOM 44 OG1 THR A
224 8.317 37.590 23.318 1.00 16.14 ATOM 45 CG2 THR A 224 8.002
39.440 24.840 1.00 17.72 ATOM 46 C THR A 224 5.152 39.464 24.033
1.00 16.13 ATOM 47 O THR A 224 4.863 40.275 23.152 1.00 15.14 ATOM
48 N CYS A 225 4.708 39.650 25.253 1.00 16.61 ATOM 49 CA CYS A 225
3.915 40.833 25.646 1.00 17.81 ATOM 50 CB CYS A 225 2.895 40.460
26.723 1.00 18.01 ATOM 51 SG CYS A 225 2.120 41.843 27.562 1.00
18.77 ATOM 52 C CYS A 225 4.899 41.914 26.101 1.00 17.46 ATOM 53 O
CYS A 225 5.614 41.703 27.093 1.00 18.52 ATOM 54 N LYS A 226 5.070
42.945 25.285 1.00 17.94 ATOM 55 CA LYS A 226 6.011 44.033 25.573
1.00 18.61 ATOM 56 CB LYS A 226 6.373 44.816 24.311 1.00 21.04 ATOM
57 CG LYS A 226 6.985 43.974 23.202 1.00 22.16 ATOM 58 CD LYS A 226
8.395 43.451 23.514 1.00 24.95 ATOM 59 CE LYS A 226 8.867 42.585
22.365 1.00 28.75 ATOM 60 NZ LYS A 226 10.336 42.445 22.185 1.00
31.31 ATOM 61 C LYS A 226 5.461 44.940 26.658 1.00 17.48 ATOM 62 O
LYS A 226 4.295 45.336 26.642 1.00 16.96 ATOM 63 N LEU A 227 6.281
45.274 27.641 1.00 15.78 ATOM 64 CA LEU A 227 5.848 46.025 28.777
1.00 15.23 ATOM 65 CB LEU A 227 6.182 45.347 30.117 1.00 15.96 ATOM
66 CG LEU A 227 5.848 43.884 30.334 1.00 15.88 ATOM 67 CD1 LEU A
227 6.375 43.381 31.692 1.00 15.72 ATOM 68 CD2 LEU A 227 4.356
43.646 30.314 1.00 13.61 ATOM 69 C LEU A 227 6.462 47.398 28.965
1.00 16.89 ATOM 70 O LEU A 227 7.639 47.635 28.725 1.00 17.35 ATOM
71 N LEU A 228 5.585 48.248 29.488 1.00 16.01 ATOM 72 CA LEU A 228
6.024 49.559 29.935 1.00 15.78 ATOM 73 CB LEU A 228 5.105 50.721
29.644 1.00 15.98 ATOM 74 CG LEU A 228 5.360 52.012 30.426 1.00
17.60 ATOM 75 CD1 LEU A 228 6.596 52.712 29.853 1.00 15.81 ATOM 76
CD2 LEU A 228 4.154 52.945 30.340 1.00 19.40 ATOM 77 C LEU A 228
6.144 49.360 31.455 1.00 16.70 ATOM 78 O LEU A 228 5.124 49.074
32.104 1.00 16.99 ATOM 79 N VAL A 229 7.356 49.488 31.983 1.00
13.83 ATOM 80 CA VAL A 229 7.484 49.343 33.450 1.00 12.75 ATOM 81
CB VAL A 229 8.600 48.320 33.747 1.00 15.41 ATOM 82 CG1 VAL A 229
9.015 48.421 35.199 1.00 15.21 ATOM 83 CG2 VAL A 229 8.062 46.910
33.451 1.00 16.11 ATOM 84 C VAL A 229 7.758 50.710 34.055 1.00
11.45 ATOM 85 O VAL A 229 8.592 51.462 33.529 1.00 10.78 ATOM 86 N
VAL A 230 7.029 51.092 35.090 1.00 11.10 ATOM 87 CA VAL A 230 7.169
52.397 35.688 1.00 13.35 ATOM 88 CB VAL A 230 5.910 53.299 35.518
1.00 13.78 ATOM 89 CG1 VAL A 230 6.096 54.643 36.192 1.00 13.70
ATOM 90 CG2 VAL A 230 5.577 53.586 34.050 1.00 12.14 ATOM 91 C VAL
A 230 7.495 52.252 37.154 1.00 12.98 ATOM 92 O VAL A 230 6.791
51.546 37.877 1.00 14.51 ATOM 93 N ALA A 231 8.582 52.891 37.570
1.00 14.57 ATOM 94 CA ALA A 231 8.915 52.901 39.001 1.00 12.63 ATOM
95 CB ALA A 231 10.382 52.556 39.219 1.00 14.54 ATOM 96 C ALA A 231
8.616 54.305 39.505 1.00 13.54 ATOM 97 O ALA A 231 9.215 55.290
39.004 1.00 12.89 ATOM 98 N ASP A 232 7.739 54.412 40.500 1.00
13.89 ATOM 99 CA ASP A 232 7.440 55.781 40.947 1.00 15.63 ATOM 100
CB ASP A 232 6.061 55.875 41.560 1.00 16.03 ATOM 101 CG ASP A 232
5.856 55.198 42.876 1.00 17.24 ATOM 102 OD1 ASP A 232 4.725 55.249
43.432 1.00 18.20 ATOM 103 OD2 ASP A 232 6.804 54.565 43.403 1.00
15.69 ATOM 104 C ASP A 232 8.559 56.276 41.875 1.00 15.96 ATOM 105
O ASP A 232 9.395 55.459 42.253 1.00 13.21 ATOM 106 N HIS A 233
8.375 57.492 42.392 1.00 16.98 ATOM 107 CA HIS A 233 9.399 58.118
43.244 1.00 16.61 ATOM 108 CB HIS A 233 9.075 59.600 43.522 1.00
17.60 ATOM 109 CG HIS A 233 7.977 59.866 44.494 1.00 17.74 ATOM 110
CD2 HIS A 233 8.012 60.040 45.836 1.00 18.92 ATOM 111 ND1 HIS A 233
6.648 59.946 44.152 1.00 17.72 ATOM 112 CE1 HIS A 233 5.910 60.144
45.222 1.00 18.19 ATOM 113 NE2 HIS A 233 6.730 60.214 46.266 1.00
19.56 ATOM 114 C HIS A 233 9.562 57.364 44.535 1.00 16.98 ATOM 115
O HIS A 233 10.626 57.385 45.170 1.00 14.73 ATOM 116 N ARG A 234
8.457 56.762 45.005 1.00 16.26 ATOM 117 CA ARG A 234 8.476 55.997
46.231 1.00 16.71 ATOM 118 CB ARG A 234 7.083 55.600 46.688 1.00
19.19 ATOM 119 CG ARG A 234 6.078 56.729 46.814 1.00 18.94 ATOM 120
CD ARG A 234 4.726 56.181 47.250 1.00 21.89 ATOM 121 NE ARG A 234
3.696 57.214 47.125 1.00 23.72 ATOM 122 CZ ARG A 234 2.872 57.469
48.130 1.00 26.70 ATOM 123 NH1 ARG A 234 2.923 56.798 49.270 1.00
27.34 ATOM 124 NH2 ARG A 234 1.953 58.411 47.989 1.00 28.96 ATOM
125 C ARG A 234 9.319 54.728 46.062 1.00 17.39 ATOM 126 O ARG A 234
10.072 54.359 46.955 1.00 17.02 ATOM 127 N PHE A 235 9.149 54.039
44.954 1.00 15.45 ATOM 128 CA PHE A 235 9.913 52.829 44.669 1.00
17.15 ATOM 129 CB PHE A 235 9.458 52.167 43.370 1.00 15.92 ATOM 130
CG PHE A 235 10.063 50.804 43.165 1.00 14.51 ATOM 131 CD1 PHE A 235
11.226 50.638 42.442 1.00 14.09 ATOM 132 CD2 PHE A 235 9.429 49.693
43.697 1.00 13.27 ATOM 133 CE1 PHE A 235 11.786 49.394 42.283 1.00
12.49 ATOM 134 CE2 PHE A 235 9.979 48.436 43.514 1.00 14.15 ATOM
135 CZ PHE A 235 11.159 48.280 42.812 1.00 11.73 ATOM 136 C PHE A
235 11.391 53.211 44.502 1.00 15.85 ATOM 137 O PHE A 235 12.309
52.611 45.041 1.00 15.02 ATOM 138 N TYR A 236 11.573 54.282 43.750
1.00 15.65 ATOM 139 CA TYR A 236 12.920 54.781 43.501 1.00 18.77
ATOM 140 CB TYR A 236 12.809 56.087 42.744 1.00 20.15 ATOM 141 CG
TYR A 236 14.079 56.831 42.453 1.00 20.77 ATOM 142 CD1 TYR A 236
15.006 56.379 41.554 1.00 23.25 ATOM 143 CE1 TYR A 236 16.171
57.111 41.294 1.00 25.10 ATOM 144 CD2 TYR A 236 14.303 58.043
43.094 1.00 22.97 ATOM 145 CE2 TYR A 236 15.434 58.789 42.840 1.00
23.38 ATOM 146 CZ TYR A 236 16.355 58.309 41.944 1.00 24.28 ATOM
147 OH TYR A 236 17.490 59.032 41.699 1.00 27.10 ATOM 148 C TYR A
236 13.668 55.005 44.807 1.00 18.41 ATOM 149 O TYR A 236 14.785
54.538 44.979 1.00 17.82 ATOM 150 N ARG A 237 13.029 55.671 45.741
1.00 18.84 ATOM 151 NH2 ARG A 237 16.164 60.578 45.801 1.00 36.66
ATOM 152 NH1 ARG A 237 15.513 58.642 46.822 1.00 35.42 ATOM 153 CZ
ARG A 237 15.208 59.704 46.106 1.00 35.99 ATOM 154 NE ARG A 237
13.940 59.862 45.712 1.00 36.11 ATOM 155 CD ARG A 237 12.867 58.910
45.981 1.00 32.12 ATOM 156 CG ARG A 237 12.655 58.442 47.386 1.00
28.99 ATOM 157 CB ARG A 237 12.578 56.978 47.698 1.00 22.94 ATOM
158 CA ARG A 237 13.566 56.021 47.029 1.00 21.89 ATOM 159 C ARG A
237 13.824 54.847 47.959 1.00 22.82 ATOM 160 O ARG A 237 14.874
54.759 48.586 1.00 22.33 ATOM 161 N TYR A 238 12.826 53.995 48.165
1.00 20.93 ATOM 162 CA TYR A 238 12.807 52.950 49.142 1.00 22.79
ATOM 163 CB TYR A 238 11.438 52.835 49.835 1.00 25.02 ATOM 164 CG
TYR A 238 11.052 54.133 50.502 1.00 28.43 ATOM 165 CD1 TYR A 238
10.191 55.011 49.873 1.00 30.38 ATOM 166 CE1 TYR A 238 9.827 56.214
50.439 1.00 32.45 ATOM 167 CD2 TYR A 238 11.570 54.501 51.729 1.00
30.38 ATOM 168 CE2 TYR A 238 11.228 55.709 52.309 1.00 32.79 ATOM
169 CZ TYR A 238 10.384 56.572 51.647 1.00 33.85 ATOM 170 OH TYR A
238 10.043 57.784 52.208 1.00 34.86 ATOM 171 C TYR A 238 13.222
51.579 48.683 1.00 22.48 ATOM 172 O TYR A 238 13.682 50.772 49.509
1.00 24.69 ATOM 173 N MET A 239 13.171 51.306 47.405 1.00 20.92
ATOM 174 CA MET A 239 13.680 50.048 46.893 1.00 18.57 ATOM 175 CB
MET A 239 12.667 49.374 45.965 1.00 18.57 ATOM 176 CG MET A 239
11.394 48.971 46.729 1.00 18.44 ATOM 177 SD MET A 239 11.677 47.664
47.929 1.00 17.64 ATOM 178 CE MET A 239 12.084 46.309 46.855 1.00
16.38 ATOM 179 C MET A 239 14.975 50.292 46.121 1.00 17.52 ATOM 180
O MET A 239 15.826 49.422 46.133 1.00 16.58 ATOM 181 N GLY A 240
15.067 51.440 45.447 1.00 16.74 ATOM 182 CA GLY A 240 16.198 51.733
44.602 1.00 16.82 ATOM 183 C GLY A 240 17.334 52.497 45.232 1.00
20.60 ATOM 184 O GLY A 240 18.280 52.875 44.516 1.00 20.21 ATOM 185
N ARG A 241 17.242 52.871 46.503 1.00 20.89 ATOM 186 CA ARG A 241
18.300 53.628 47.162 1.00 24.17 ATOM 187 CB ARG A 241 19.609 52.806
47.126 1.00 26.15 ATOM 188 CG ARG A 241 19.504 51.488 47.875 1.00
29.48 ATOM 189 CD ARG A 241 20.771 50.648 47.896 1.00 31.73 ATOM
190 NE ARG A 241 21.417 50.785 49.228 1.00 32.75 ATOM 191 CZ ARG A
241 22.188 51.837 49.428 1.00 33.64 ATOM 192 NH1 ARG A 241 22.361
52.722 48.450 1.00 35.35 ATOM 193 NH2 ARG A 241 22.752 52.014
50.598 1.00 32.64 ATOM 194 C ARG A 241 18.497 55.001 46.543 1.00
25.05 ATOM 195 O ARG A 241 19.574 55.585 46.696 1.00 21.63 ATOM 196
N GLY A 242 17.470 55.529 45.831 1.00 22.70 ATOM 197 CA GLY A 242
17.603 56.784 45.110 1.00 20.94 ATOM 198 C GLY A 242 18.622 56.667
44.003 1.00 20.36 ATOM 199 O GLY A 242 19.255 57.656 43.639 1.00
20.42 ATOM 200 N GLU A 243 18.841 55.486 43.445 1.00 19.57 ATOM 201
CA GLU A 243 19.832 55.277 42.418 1.00 19.51 ATOM 202 CB GLU A 243
20.951 54.324 42.832 1.00 21.04 ATOM 203 CG GLU A 243 21.992 54.818
43.816 1.00 22.54 ATOM 204 CD GLU A 243 22.782 53.747 44.543 1.00
24.67 ATOM 205 OE1 GLU A 243 23.740 54.097 45.292 1.00 24.23 ATOM
206 OE2 GLU A 243 22.512 52.535 44.451 1.00 22.83 ATOM 207 C GLU A
243 19.104 54.717 41.197 1.00 19.62 ATOM 208 O GLU A 243 18.414
53.698 41.291 1.00 18.85 ATOM 209 N GLU A 244 19.324 55.341 40.075
1.00 19.48 ATOM 210 CA GLU A 244 18.699 54.904 38.833 1.00 22.19
ATOM 211 CB GLU A 244 18.928 55.950 37.739 1.00 25.49 ATOM 212 CG
GLU A 244 18.502 55.433 36.371 1.00 29.88 ATOM 213 CD GLU A 244
18.409 56.498 35.310 1.00 32.59 ATOM 214 OE1 GLU A 244 18.205
57.691 35.608 1.00 33.16 ATOM 215 OE2 GLU A 244 18.528 56.090
34.136 1.00 35.02 ATOM 216 C GLU A 244 19.176 53.524 38.414 1.00
21.10 ATOM 217 O GLU A 244 18.360 52.654 38.029 1.00 17.68 ATOM 218
N SER A 245 20.466 53.247 38.644 1.00 18.06 ATOM 219 CA SER A 245
21.001 51.960 38.207 1.00 20.29 ATOM 220 CB SER A 245 22.549 51.953
38.270 1.00 21.06 ATOM 221 OG SER A 245 22.833 52.394 39.607 1.00
21.74 ATOM 222 C SER A 245 20.409 50.844 39.056 1.00 17.46 ATOM 223
O SER A 245 19.970 49.834 38.536 1.00 18.81 ATOM 224 N THR A 246
20.355 51.020 40.343 1.00 15.88 ATOM 225 CA THR A 246 19.821 50.027
41.275 1.00 18.13 ATOM 226 CB THR A 246 20.051 50.546 42.689 1.00
19.42 ATOM 227 OG1 THR A 246 21.459 50.877 42.861 1.00 21.14 ATOM
228 CG2 THR A 246 19.692 49.547 43.761 1.00 19.86 ATOM 229 C THR A
246 18.337 49.757 41.014 1.00 16.92 ATOM 230 O THR A 246 17.915
48.619 41.068 1.00 14.90 ATOM 231 N THR A 247 17.545 50.799 40.800
1.00 17.19 ATOM 232 CA THR A 247 16.108 50.689 40.560 1.00 16.84
ATOM 233 CB THR A 247 15.458 52.076 40.443 1.00 16.21 ATOM 234 OG1
THR A 247 15.860 52.839 41.581 1.00 14.61 ATOM 235 CG2 THR A 247
13.920 52.069 40.449 1.00 14.87 ATOM 236 C THR A 247 15.848 49.892
39.308 1.00 15.69 ATOM 237 O THR A 247 15.088 48.922 39.295 1.00
14.94 ATOM 238 N THR A 248 16.502 50.305 38.232 1.00 16.92 ATOM 239
CA THR A 248 16.382 49.685 36.926 1.00 16.92 ATOM 240 CB THR A 248
17.322 50.415 35.963 1.00 17.93 ATOM 241 OG1 THR A 248 16.875
51.787 35.915 1.00 18.06 ATOM 242 CG2 THR A 248 17.381 49.792
34.586 1.00 19.15 ATOM 243 C THR A 248 16.712 48.202 36.972 1.00
18.22 ATOM 244 O THR A 248 16.073 47.313 36.427 1.00 14.91 ATOM 245
N ASN A 249 17.857 47.907 37.593 1.00 17.46 ATOM 246 CA ASN A 249
18.342 46.545 37.700 1.00 17.81 ATOM 247 CB ASN A 249 19.723 46.472
38.349 1.00 19.11 ATOM 248 CG ASN A 249 20.854 47.016 37.499 1.00
22.53 ATOM 249 OD1 ASN A 249 20.753 47.173 36.283 1.00 22.82 ATOM
250 ND2 ASN A 249 21.989 47.306 38.157 1.00 21.73 ATOM 251 C ASN A
249 17.364 45.662 38.478 1.00 15.16 ATOM 252 O ASN A 249 17.157
44.510 38.066 1.00 13.94 ATOM 253 N TYR A 250 16.850 46.170 39.578
1.00 14.10 ATOM 254 CA TYR A 250 15.888 45.426 40.360 1.00 15.67
ATOM 255 CB TYR A 250 15.393 46.239 41.546 1.00 16.52 ATOM 256 CG
TYR A 250 14.527 45.444 42.501 1.00 19.76 ATOM 257 CD1 TYR A 250
15.109 44.778 43.571 1.00 20.38 ATOM 258 CE1 TYR A 250 14.344
44.031 44.447 1.00 21.56 ATOM 259 CD2 TYR A 250 13.138 45.344
42.332 1.00 18.76 ATOM 260 CE2 TYR A 250 12.377 44.600 43.203 1.00
19.63 ATOM 261 CZ TYR A 250 12.985 43.943 44.248 1.00 22.22 ATOM
262 OH TYR A 250 12.250 43.189 45.134 1.00 24.91 ATOM 263 C TYR A
250 14.683 45.029 39.472 1.00 14.95 ATOM 264 O TYR A 250 14.234
43.893 39.546 1.00 11.83 ATOM 265 N LEU A 251 14.133 46.021 38.758
1.00 13.07 ATOM 266 CA LEU A 251 12.977 45.742 37.899 1.00 12.72
ATOM 267 CB LEU A 251 12.271 47.059 37.519 1.00 12.66 ATOM 268 CG
LEU A 251 11.696 47.800 38.734 1.00 14.96 ATOM 269 CD1 LEU A 251
11.038 49.099 38.363 1.00 18.53 ATOM 270 CD2 LEU A 251 10.717
46.914 39.501 1.00 19.00 ATOM 271 C LEU A 251 13.288 44.885 36.700
1.00 11.35 ATOM 272 O LEU A 251 12.518 43.962 36.349 1.00 10.33
ATOM 273 N ILE A 252 14.429 45.093 36.050 1.00 10.29 ATOM 274 CA
ILE A 252 14.832 44.209 34.944 1.00 12.09 ATOM 275 CB ILE A 252
16.209 44.654 34.372 1.00 13.58 ATOM 276 CG2 ILE A 252 16.848
43.570 33.523 1.00 14.17 ATOM 277 CG1 ILE A 252 16.059 45.964
33.581 1.00 16.57 ATOM 278 CD1 ILE A 252 17.351 46.450 32.902 1.00
18.53 ATOM 279 C ILE A 252 14.938 42.791 35.448 1.00 10.51 ATOM 280
O ILE A 252 14.592 41.830 34.755 1.00 12.14 ATOM 281 N GLU A 253
15.508 42.598 36.641 1.00 9.27 ATOM 282 CA GLU A 253 15.743 41.263
37.159 1.00 10.71 ATOM 283 CB GLU A 253 16.762 41.233 38.306 1.00
12.25 ATOM 284 CG GLU A 253 18.217 41.412 37.830 1.00 12.76 ATOM
285 CD GLU A 253 19.215 41.029 38.924 1.00 15.62 ATOM 286 OE1 GLU A
253 19.091 40.009 39.615 1.00 14.92 ATOM 287 OE2 GLU A 253 20.134
41.798 39.207 1.00 14.69 ATOM 288 C GLU A 253 14.422 40.619 37.605
1.00 9.88 ATOM 289 O GLU A 253 14.195 39.441 37.388 1.00 7.43 ATOM
290 N LEU A 254 13.559 41.420 38.230 1.00 10.47 ATOM 291 CA LEU A
254 12.266 40.858 38.649 1.00 10.04 ATOM 292 CB
LEU A 254 11.537 41.807 39.572 1.00 9.92 ATOM 293 CG LEU A 254
10.103 41.493 40.022 1.00 10.99 ATOM 294 CD1 LEU A 254 9.778 42.322
41.239 1.00 11.71 ATOM 295 CD2 LEU A 254 9.106 41.854 38.918 1.00
11.46 ATOM 296 C LEU A 254 11.481 40.463 37.408 1.00 9.41 ATOM 297
O LEU A 254 10.959 39.343 37.426 1.00 10.16 ATOM 298 N ILE A 255
11.399 41.269 36.369 1.00 11.15 ATOM 299 CA ILE A 255 10.590 40.910
35.180 1.00 11.06 ATOM 300 CB ILE A 255 10.480 42.108 34.194 1.00
10.34 ATOM 301 CG2 ILE A 255 9.910 41.651 32.851 1.00 12.31 ATOM
302 CG1 ILE A 255 9.646 43.216 34.796 1.00 10.06 ATOM 303 CD1 ILE A
255 8.235 42.810 35.286 1.00 10.90 ATOM 304 C ILE A 255 11.113
39.671 34.495 1.00 11.34 ATOM 305 O ILE A 255 10.466 38.704 34.055
1.00 8.95 ATOM 306 N ASP A 256 12.458 39.596 34.420 1.00 10.09 ATOM
307 CA ASP A 256 13.108 38.362 33.953 1.00 11.00 ATOM 308 CB ASP A
256 14.621 38.617 33.890 1.00 12.60 ATOM 309 CG ASP A 256 15.320
37.358 33.434 1.00 16.54 ATOM 310 OD1 ASP A 256 15.159 37.033
32.249 1.00 17.04 ATOM 311 OD2 ASP A 256 15.977 36.731 34.290 1.00
18.15 ATOM 312 C ASP A 256 12.763 37.124 34.779 1.00 10.53 ATOM 313
O ASP A 256 12.485 36.036 34.246 1.00 11.14 ATOM 314 N ARG A 257
12.719 37.177 36.087 1.00 11.51 ATOM 315 CA ARG A 257 12.362 36.025
36.923 1.00 11.03 ATOM 316 CB ARG A 257 12.646 36.291 38.375 1.00
11.04 ATOM 317 CG ARG A 257 14.194 36.302 38.743 1.00 9.97 ATOM 318
CD ARG A 257 14.347 36.333 40.258 1.00 9.94 ATOM 319 NE ARG A 257
13.581 37.392 40.951 1.00 9.56 ATOM 320 CZ ARG A 257 14.058 38.613
41.124 1.00 10.48 ATOM 321 NH1 ARG A 257 15.283 38.961 40.677 1.00
8.64 ATOM 322 NH2 ARG A 257 13.363 39.539 41.718 1.00 8.45 ATOM 323
C ARG A 257 10.868 35.673 36.655 1.00 10.99 ATOM 324 O ARG A 257
10.495 34.509 36.580 1.00 10.38 ATOM 325 N VAL A 258 10.033 36.685
36.635 1.00 7.90 ATOM 326 CA VAL A 258 8.600 36.497 36.295 1.00
9.48 ATOM 327 CB VAL A 258 7.877 37.856 36.335 1.00 8.97 ATOM 328
CG1 VAL A 258 6.398 37.717 35.905 1.00 7.38 ATOM 329 CG2 VAL A 258
7.915 38.447 37.733 1.00 7.76 ATOM 330 C VAL A 258 8.469 35.851
34.931 1.00 9.20 ATOM 331 O VAL A 258 7.769 34.858 34.706 1.00
11.73 ATOM 332 N ASP A 259 9.193 36.330 33.947 1.00 10.44 ATOM 333
CA ASP A 259 9.196 35.817 32.585 1.00 11.92 ATOM 334 CB ASP A 259
10.101 36.614 31.655 1.00 12.64 ATOM 335 CG ASP A 259 10.059 36.157
30.219 1.00 17.49 ATOM 336 OD1 ASP A 259 8.962 36.007 29.644 1.00
16.26 ATOM 337 OD2 ASP A 259 11.145 35.932 29.637 1.00 20.54 ATOM
338 C ASP A 259 9.618 34.346 32.546 1.00 13.21 ATOM 339 O ASP A 259
9.021 33.572 31.783 1.00 11.13 ATOM 340 N ASP A 260 10.546 33.904
33.412 1.00 12.46 ATOM 341 OD2 ASP A 260 14.081 33.042 35.067 1.00
19.42 ATOM 342 OD1 ASP A 260 13.629 33.074 32.936 1.00 15.68 ATOM
343 CG ASP A 260 13.335 32.789 34.098 1.00 16.09 ATOM 344 CB ASP A
260 11.972 32.171 34.451 1.00 13.52 ATOM 345 CA ASP A 260 10.912
32.498 33.403 1.00 11.20 ATOM 346 C ASP A 260 9.647 31.669 33.761
1.00 13.45 ATOM 347 O ASP A 260 9.461 30.565 33.241 1.00 11.34 ATOM
348 N ILE A 261 8.867 32.202 34.720 1.00 10.81 ATOM 349 CA ILE A
261 7.643 31.469 35.052 1.00 11.86 ATOM 350 CB ILE A 261 6.910
32.167 36.198 1.00 13.06 ATOM 351 CG2 ILE A 261 5.538 31.612 36.416
1.00 12.50 ATOM 352 CG1 ILE A 261 7.825 31.986 37.422 1.00 16.02
ATOM 353 CD1 ILE A 261 7.242 32.547 38.701 1.00 17.40 ATOM 354 C
ILE A 261 6.738 31.396 33.815 1.00 10.07 ATOM 355 O ILE A 261 6.351
30.273 33.480 1.00 11.72 ATOM 356 N TYR A 262 6.434 32.522 33.232
1.00 8.70 ATOM 357 CA TYR A 262 5.515 32.515 32.082 1.00 11.55 ATOM
358 CB TYR A 262 5.275 33.937 31.548 1.00 11.91 ATOM 359 CG TYR A
262 4.257 34.650 32.411 1.00 12.15 ATOM 360 CD1 TYR A 262 4.554
35.110 33.670 1.00 12.29 ATOM 361 CE1 TYR A 262 3.598 35.746 34.457
1.00 12.53 ATOM 362 CD2 TYR A 262 2.940 34.799 31.941 1.00 14.52
ATOM 363 CE2 TYR A 262 1.989 35.420 32.716 1.00 14.04 ATOM 364 CZ
TYR A 262 2.315 35.905 33.956 1.00 14.14 ATOM 365 OH TYR A 262
1.368 36.538 34.724 1.00 12.39 ATOM 366 C TYR A 262 5.982 31.682
30.932 1.00 11.66 ATOM 367 O TYR A 262 5.202 30.842 30.446 1.00
11.39 ATOM 368 N ARG A 263 7.231 31.887 30.478 1.00 10.78 ATOM 369
CA ARG A 263 7.771 31.180 29.333 1.00 14.61 ATOM 370 CB ARG A 263
9.236 31.623 29.117 1.00 18.66 ATOM 371 CG ARG A 263 9.634 32.729
28.194 1.00 21.81 ATOM 372 CD ARG A 263 11.103 32.655 27.718 1.00
25.82 ATOM 373 NE ARG A 263 11.932 31.745 28.516 1.00 27.30 ATOM
374 CZ ARG A 263 12.427 32.092 29.708 1.00 28.71 ATOM 375 NH1 ARG A
263 12.225 33.329 30.166 1.00 29.09 ATOM 376 NH2 ARG A 263 13.087
31.212 30.435 1.00 27.05 ATOM 377 C ARG A 263 7.751 29.674 29.425
1.00 15.13 ATOM 378 O ARG A 263 7.596 28.973 28.384 1.00 17.45 ATOM
379 N ASN A 264 8.019 29.078 30.591 1.00 12.82 ATOM 380 CA ASN A
264 8.003 27.677 30.834 1.00 14.56 ATOM 381 CB ASN A 264 8.878
27.204 32.013 1.00 17.95 ATOM 382 CG ASN A 264 10.323 27.555 31.654
1.00 19.62 ATOM 383 OD1 ASN A 264 10.776 27.002 30.654 1.00 23.25
ATOM 384 ND2 ASN A 264 10.957 28.464 32.335 1.00 19.81 ATOM 385 C
ASN A 264 6.582 27.144 31.076 1.00 16.00 ATOM 386 O ASN A 264 6.449
25.944 31.293 1.00 14.82 ATOM 387 N THR A 265 5.566 28.017 31.073
1.00 13.74 ATOM 388 CA THR A 265 4.221 27.425 31.290 1.00 13.37
ATOM 389 CB THR A 265 3.309 28.463 31.935 1.00 14.37 ATOM 390 OG1
THR A 265 3.800 28.787 33.238 1.00 11.71 ATOM 391 CG2 THR A 265
1.866 27.953 32.101 1.00 12.84 ATOM 392 C THR A 265 3.675 26.917
29.965 1.00 12.82 ATOM 393 O THR A 265 3.698 27.628 28.968 1.00
14.37 ATOM 394 N ALA A 266 3.134 25.708 29.970 1.00 13.65 ATOM 395
CA ALA A 266 2.432 25.151 28.820 1.00 14.19 ATOM 396 CB ALA A 266
2.737 23.645 28.812 1.00 14.17 ATOM 397 C ALA A 266 0.937 25.431
28.997 1.00 11.04 ATOM 398 O ALA A 266 0.250 24.682 29.675 1.00
10.22 ATOM 399 N TRP A 267 0.426 26.539 28.524 1.00 11.35 ATOM 400
CA TRP A 267 -0.962 26.949 28.736 1.00 13.04 ATOM 401 CB TRP A 267
-1.183 28.301 28.045 1.00 11.63 ATOM 402 CG TRP A 267 -0.186 29.330
28.523 1.00 10.70 ATOM 403 CD2 TRP A 267 -0.205 29.920 29.819 1.00
8.77 ATOM 404 CE2 TRP A 267 0.904 30.814 29.863 1.00 12.19 ATOM 405
CE3 TRP A 267 -1.007 29.828 30.931 1.00 9.72 ATOM 406 CD1 TRP A 267
0.865 29.846 27.849 1.00 10.38 ATOM 407 NE1 TRP A 267 1.531 30.754
28.652 1.00 11.35 ATOM 408 CZ2 TRP A 267 1.187 31.580 30.989 1.00
10.57 ATOM 409 CZ3 TRP A 267 -0.718 30.580 32.066 1.00 11.82 ATOM
410 CH2 TRP A 267 0.395 31.439 32.071 1.00 9.78 ATOM 411 C TRP A
267 -2.007 25.919 28.298 1.00 14.01 ATOM 412 O TRP A 267 -3.089
25.832 28.915 1.00 14.19 ATOM 413 N ASP A 268 -1.688 25.091 27.326
1.00 11.86 ATOM 414 CA ASP A 268 -2.615 24.026 26.899 1.00 13.82
ATOM 415 CB ASP A 268 -2.765 24.051 25.382 1.00 13.92 ATOM 416 CG
ASP A 268 -1.517 23.691 24.619 1.00 16.74 ATOM 417 OD1 ASP A 268
-0.422 23.699 25.244 1.00 16.49 ATOM 418 OD2 ASP A 268 -1.561
23.395 23.404 1.00 15.38 ATOM 419 C ASP A 268 -2.165 22.661 27.367
1.00 15.26 ATOM 420 O ASP A 268 -2.554 21.608 26.834 1.00 13.18
ATOM 421 N ASN A 269 -1.225 22.609 28.329 1.00 14.90 ATOM 422 CA
ASN A 269 -0.596 21.391 28.792 1.00 15.45 ATOM 423 CB ASN A 269
-1.547 20.454 29.549 1.00 17.28 ATOM 424 CG ASN A 269 -2.190 21.182
30.719 1.00 18.96 ATOM 425 OD1 ASN A 269 -1.464 21.670 31.560 1.00
20.34 ATOM 426 ND2 ASN A 269 -3.500 21.321 30.788 1.00 18.93 ATOM
427 C ASN A 269 0.108 20.556 27.711 1.00 14.79 ATOM 428 O ASN A 269
0.330 19.352 27.859 1.00 14.35 ATOM 429 N ALA A 270 0.523 21.193
26.654 1.00 14.55 ATOM 430 CA ALA A 270 1.203 20.593 25.533 1.00
17.42 ATOM 431 CB ALA A 270 0.189 20.155 24.481 1.00 15.76 ATOM 432
C ALA A 270 2.239 21.573 25.004 1.00 16.67 ATOM 433 O ALA A 270
3.183 21.866 25.757 1.00 19.13 ATOM 434 N GLY A 271 2.143 22.004
23.765 1.00 16.01 ATOM 435 CA GLY A 271 3.174 22.837 23.177 1.00
18.64 ATOM 436 C GLY A 271 2.951 24.321 23.125 1.00 17.56 ATOM 437
O GLY A 271 3.768 25.075 22.589 1.00 18.20 ATOM 438 N PHE A 272
1.844 24.831 23.659 1.00 17.81 ATOM 439 CA PHE A 272 1.538 26.263
23.614 1.00 16.42 ATOM 440 CB PHE A 272 0.047 26.539 23.710 1.00
16.73 ATOM 441 CG PHE A 272 -0.389 27.890 23.205 1.00 17.94 ATOM
442 CD1 PHE A 272 -0.462 28.168 21.858 1.00 19.17 ATOM 443 CD2 PHE
A 272 -0.716 28.894 24.100 1.00 18.80 ATOM 444 CE1 PHE A 272 -0.813
29.421 21.407 1.00 18.87 ATOM 445 CE2 PHE A 272 -1.120 30.139
23.664 1.00 18.81 ATOM 446 CZ PHE A 272 -1.142 30.404 22.310 1.00
19.11 ATOM 447 C PHE A 272 2.305 26.889 24.796 1.00 16.83 ATOM 448
O PHE A 272 1.786 27.138 25.860 1.00 14.19 ATOM 449 N LYS A 273
3.584 27.067 24.561 1.00 17.93 ATOM 450 CA LYS A 273 4.522 27.543
25.546 1.00 21.35 ATOM 451 CB LYS A 273 5.236 26.296 26.080 1.00
24.32 ATOM 452 CG LYS A 273 6.157 25.698 25.023 1.00 27.94 ATOM 453
CD LYS A 273 6.625 24.300 25.389 1.00 31.29 ATOM 454 CE LYS A 273
7.307 24.300 26.732 1.00 32.80 ATOM 455 NZ LYS A 273 6.347 24.136
27.857 1.00 33.38 ATOM 456 C LYS A 273 5.479 28.518 24.906 1.00
19.96 ATOM 457 O LYS A 273 5.444 28.763 23.706 1.00 18.61 ATOM 458
N GLY A 274 6.382 29.094 25.709 1.00 20.59 ATOM 459 CA GLY A 274
7.323 30.058 25.166 1.00 17.10 ATOM 460 C GLY A 274 6.701 31.449
25.153 1.00 17.80 ATOM 461 O GLY A 274 7.272 32.334 24.547 1.00
17.48 ATOM 462 N TYR A 275 5.527 31.675 25.763 1.00 15.67 ATOM 463
CA TYR A 275 4.892 32.965 25.802 1.00 16.24 ATOM 464 CB TYR A 275
3.340 32.874 25.854 1.00 15.81 ATOM 465 CG TYR A 275 2.838 32.446
24.503 1.00 17.73 ATOM 466 CD1 TYR A 275 2.784 31.102 24.175 1.00
17.51 ATOM 467 CE1 TYR A 275 2.384 30.722 22.919 1.00 20.42 ATOM
468 CD2 TYR A 275 2.524 33.418 23.543 1.00 18.48 ATOM 469 CE2 TYR A
275 2.140 33.017 22.279 1.00 20.02 ATOM 470 CZ TYR A 275 2.065
31.690 21.989 1.00 19.42 ATOM 471 OH TYR A 275 1.670 31.304 20.737
1.00 23.70 ATOM 472 C TYR A 275 5.317 33.689 27.087 1.00 15.60 ATOM
473 O TYR A 275 5.270 33.077 28.160 1.00 15.87 ATOM 474 N GLY A 276
5.638 34.959 26.991 1.00 17.04 ATOM 475 CA GLY A 276 5.978 35.692
28.190 1.00 17.48 ATOM 476 C GLY A 276 5.988 37.183 28.020 1.00
19.17 ATOM 477 O GLY A 276 5.226 37.781 27.278 1.00 16.76 ATOM 478
N ILE A 277 6.893 37.805 28.811 1.00 18.66 ATOM 479 CA ILE A 277
6.938 39.245 28.907 1.00 17.89 ATOM 480 CB ILE A 277 6.413 39.745
30.256 1.00 19.63 ATOM 481 CG2 ILE A 277 4.925 39.409 30.421 1.00
21.02 ATOM 482 CG1 ILE A 277 7.212 39.101 31.390 1.00 20.19 ATOM
483 CD1 ILE A 277 6.754 39.439 32.779 1.00 20.94 ATOM 484 C ILE A
277 8.385 39.741 28.763 1.00 18.46 ATOM 485 O ILE A 277 9.318
39.076 29.168 1.00 16.61 ATOM 486 N GLN A 278 8.505 40.919 28.219
1.00 17.96 ATOM 487 CA GLN A 278 9.743 41.629 28.051 1.00 19.65
ATOM 488 CB GLN A 278 10.224 41.684 26.624 1.00 25.01 ATOM 489 CG
GLN A 278 11.692 41.405 26.382 1.00 28.24 ATOM 490 CD GLN A 278
11.580 40.059 25.643 1.00 32.63 ATOM 491 OE1 GLN A 278 11.660
40.060 24.422 1.00 33.32 ATOM 492 NE2 GLN A 278 11.331 39.078
26.502 1.00 34.14 ATOM 493 C GLN A 278 9.499 43.122 28.316 1.00
17.44 ATOM 494 O GLN A 278 8.589 43.689 27.693 1.00 14.22 ATOM 495
N ILE A 279 10.537 43.709 28.900 1.00 13.53 ATOM 496 CA ILE A 279
10.471 45.146 29.090 1.00 15.82 ATOM 497 CB ILE A 279 11.441 45.658
30.183 1.00 15.54 ATOM 498 CG2 ILE A 279 11.402 47.166 30.140 1.00
15.97 ATOM 499 CG1 ILE A 279 11.007 45.083 31.520 1.00 16.27 ATOM
500 CD1 ILE A 279 11.940 45.388 32.661 1.00 17.52 ATOM 501 C ILE A
279 10.745 45.847 27.771 1.00 16.97 ATOM 502 O ILE A 279 11.741
45.578 27.115 1.00 16.40 ATOM 503 N GLU A 280 9.824 46.719 27.370
1.00 17.63 ATOM 504 CA GLU A 280 10.019 47.523 26.185 1.00 20.86
ATOM 505 CB GLU A 280 8.744 47.778 25.395 1.00 23.07 ATOM 506 CG
GLU A 280 8.890 48.784 24.268 1.00 29.11 ATOM 507 CD GLU A 280
9.843 48.327 23.189 1.00 32.61 ATOM 508 OE1 GLU A 280 10.187 47.124
23.154 1.00 36.14 ATOM 509 OE2 GLU A 280 10.293 49.127 22.343 1.00
34.84 ATOM 510 C GLU A 280 10.599 48.892 26.605 1.00 18.21 ATOM 511
O GLU A 280 11.426 49.511 25.941 1.00 16.53 ATOM 512 N GLN A 281
10.088 49.380 27.716 1.00 18.36 ATOM 513 NE2 GLN A 281 11.497
53.112 26.056 1.00 34.24 ATOM 514 OE1 GLN A 281 10.702 55.093
26.755 1.00 36.11 ATOM 515 CD GLN A 281 10.798 53.869 26.901 1.00
33.14 ATOM 516 CG GLN A 281 10.120 53.163 28.056 1.00 29.67 ATOM
517 CB GLN A 281 9.611 51.793 27.640 1.00 24.17 ATOM 518 CA GLN A
281 10.509 50.675 28.207 1.00 21.85 ATOM 519 C GLN A 281 10.385
50.730 29.723 1.00 22.08 ATOM 520 O GLN A 281 9.444 50.170 30.275
1.00 19.43 ATOM 521 N ILE A 282 11.379 51.377 30.337 1.00 19.26
ATOM 522 CA ILE A 282 11.340 51.564 31.756 1.00 20.21 ATOM 523 CB
ILE A 282 12.456 50.838 32.501 1.00 23.17 ATOM 524 CG2 ILE A 282
12.567 51.331 33.942 1.00 22.63 ATOM 525 CG1 ILE A 282 12.183
49.337 32.486 1.00 24.53 ATOM 526 CD1 ILE A 282 13.292 48.530
33.126 1.00 27.43 ATOM 527 C ILE A 282 11.408 53.062 32.033 1.00
20.13 ATOM 528 O ILE A 282 12.183 53.826 31.441 1.00 21.53 ATOM 529
N ARG A 283 10.595 53.478 32.972 1.00 18.99 ATOM 530 CA ARG A 283
10.541 54.878 33.353 1.00 20.71 ATOM 531 CB ARG A 283 9.159 55.499
33.116 1.00 23.74 ATOM 532 CG ARG A 283 8.779 55.516 31.645 1.00
28.61 ATOM 533 CD ARG A 283 9.030 56.887 31.041 1.00 33.16 ATOM 534
NE ARG A 283 9.596 56.759 29.712 1.00 37.40 ATOM 535 CZ ARG A 283
9.199 57.449 28.659 1.00 40.44 ATOM 536 NH1 ARG A 283 8.191 58.311
28.783 1.00 41.71 ATOM 537 NH2 ARG A 283 9.800 57.245 27.492 1.00
41.03 ATOM 538 C ARG A 283 10.826 54.930 34.844 1.00 20.40 ATOM 539
O ARG A 283 10.049 54.339 35.566 1.00 17.37 ATOM 540 N ILE A 284
11.881 55.634 35.220 1.00 18.94 ATOM 541 CA ILE A 284 12.127 55.779
36.655 1.00 19.58 ATOM 542 CB ILE A 284 13.582 55.519 37.053 1.00
20.09 ATOM 543 CG2 ILE A 284 13.684 55.477 38.573 1.00 20.97 ATOM
544 CG1 ILE A 284 14.178 54.254 36.460 1.00 20.86 ATOM 545 CD1 ILE
A 284 13.442 52.976 36.774 1.00 20.01 ATOM 546 C ILE A 284 11.792
57.228 37.002 1.00 19.71 ATOM 547 O ILE A 284 12.399 58.147 36.427
1.00 19.08 ATOM 548 N LEU A 285 10.816 57.413 37.849 1.00 19.33
ATOM 549 CA LEU A 285 10.379 58.729 38.276 1.00 20.90 ATOM 550 CB
LEU A 285 8.881 58.729 38.557 1.00 20.57 ATOM 551 CG LEU A 285
8.020 58.220 37.391 1.00 18.92 ATOM 552 CD1 LEU A 285 6.569 58.488
37.719 1.00 19.77 ATOM 553 CD2 LEU A 285 8.426 58.759 36.039 1.00
18.79 ATOM 554 C LEU A 285 11.153 59.078 39.542 1.00 22.65 ATOM 555
O LEU A 285 10.861 58.536 40.585 1.00 21.09 ATOM 556 N LYS A 286
12.179 59.897 39.370 1.00 23.47 ATOM 557 CA LYS A 286 13.101 60.223
40.443 1.00 25.88 ATOM 558 CB LYS A 286 14.400 60.777 39.830 1.00
24.64 ATOM 559 CG LYS A 286 15.057 59.703 38.969 1.00 26.69 ATOM
560 CD LYS A 286 16.105 60.262 38.032 1.00 28.58 ATOM 561 CE LYS A
286 17.308 59.334 37.986 1.00 28.89 ATOM 562 NZ LYS A 286 18.328
59.807 37.012 1.00 29.92 ATOM 563 C LYS A 286 12.515 61.157 41.473
1.00 26.73 ATOM 564 O LYS A 286 12.944 61.136 42.622 1.00 28.60
ATOM 565 N SER A 287 11.543 61.980 41.133 1.00 25.80 ATOM 566 CA
SER A 287 10.969 62.865 42.158 1.00 26.18 ATOM 567 CB SER A 287
11.627 64.249 42.056 1.00 26.60 ATOM 568 OG SER A 287 11.272 64.767
40.796 1.00 28.18 ATOM 569 C SER A 287 9.471 62.857 41.977 1.00
24.89 ATOM 570 O SER A 287 8.996 62.358 40.947 1.00 25.47 ATOM 571
N PRO A 288 8.739 63.268 42.989 1.00 26.04 ATOM 572 CD PRO A 288
9.222 63.876 44.247 1.00 27.14 ATOM 573 CA PRO A 288 7.294 63.276
42.938 1.00 26.70 ATOM 574 CB PRO A 288 6.831 63.561 44.339 1.00
26.05 ATOM 575 CG PRO A 288 8.010 63.972 45.128 1.00 27.69 ATOM 576
C PRO A 288 6.826 64.313 41.928 1.00 27.00 ATOM 577 O PRO A 288
7.522 65.301 41.666 1.00 27.28 ATOM 578 N GLN A 289 5.681 64.051
41.341 1.00 25.80 ATOM 579 CA GLN A 289 5.123 64.976 40.376 1.00
27.15 ATOM 580 CB GLN A 289 4.021 64.214 39.628 1.00 27.78 ATOM 581
CG GLN A 289 3.246 65.040 38.617 1.00 27.92 ATOM 582 CD GLN A 289
4.207 65.636 37.616 1.00 28.41 ATOM 583 OE1 GLN A 289 4.970 64.944
36.966 1.00 28.03 ATOM 584 NE2 GLN A 289 4.179 66.966 37.505 1.00
32.04 ATOM 585 C GLN A 289 4.553 66.194 41.093 1.00 27.84 ATOM 586
O GLN A 289 3.693 66.017 41.943 1.00 27.12 ATOM 587 N GLU A 290
4.962 67.414 40.766 1.00 30.72 ATOM 588 OE2 GLU A 290 4.477 72.113
38.486 1.00 39.70 ATOM 589 OE1 GLU A 290 6.067 72.564 39.857 1.00
40.48 ATOM 590 CD GLU A 290 5.336 71.724 39.305 1.00 39.53 ATOM 591
CG GLU A 290 5.461 70.253 39.660 1.00 37.94 ATOM 592 CB GLU A 290
5.094 69.910 41.084 1.00
35.71 ATOM 593 CA GLU A 290 4.384 68.595 41.401 1.00 33.44 ATOM 594
C GLU A 290 2.912 68.679 40.978 1.00 32.66 ATOM 595 O GLU A 290
2.615 68.487 39.807 1.00 32.60 ATOM 596 N VAL A 291 2.027 68.929
41.928 1.00 32.52 ATOM 597 CA VAL A 291 0.589 68.960 41.608 1.00
34.53 ATOM 598 CB VAL A 291 0.016 67.667 42.214 1.00 34.68 ATOM 599
CG1 VAL A 291 -0.992 67.828 43.317 1.00 33.99 ATOM 600 CG2 VAL A
291 -0.509 66.742 41.114 1.00 33.97 ATOM 601 C VAL A 291 -0.040
70.265 42.070 1.00 36.93 ATOM 602 O VAL A 291 0.296 70.786 43.141
1.00 35.77 ATOM 603 N LYS A 292 -0.879 70.860 41.231 1.00 38.91
ATOM 604 NZ LYS A 292 0.250 74.239 36.361 1.00 47.09 ATOM 605 CE
LYS A 292 -0.459 74.753 37.563 1.00 45.89 ATOM 606 CD LYS A 292
-1.298 73.665 38.215 1.00 45.49 ATOM 607 CG LYS A 292 -1.226 73.759
39.722 1.00 44.08 ATOM 608 CB LYS A 292 -2.197 72.809 40.397 1.00
42.31 ATOM 609 CA LYS A 292 -1.595 72.078 41.589 1.00 41.28 ATOM
610 C LYS A 292 -2.701 71.697 42.557 1.00 41.79 ATOM 611 O LYS A
292 -3.085 70.535 42.624 1.00 42.72 ATOM 612 N PRO A 293 -3.192
72.668 43.305 1.00 42.21 ATOM 613 CD PRO A 293 -2.802 74.096 43.247
1.00 42.59 ATOM 614 CA PRO A 293 -4.279 72.420 44.237 1.00 41.91
ATOM 615 CB PRO A 293 -4.605 73.784 44.806 1.00 41.79 ATOM 616 CG
PRO A 293 -3.481 74.676 44.450 1.00 41.90 ATOM 617 C PRO A 293
-5.441 71.807 43.474 1.00 41.40 ATOM 618 O PRO A 293 -5.696 72.212
42.339 1.00 41.02 ATOM 619 N GLY A 294 -6.059 70.755 44.003 1.00
40.85 ATOM 620 CA GLY A 294 -7.176 70.123 43.332 1.00 41.30 ATOM
621 C GLY A 294 -6.832 69.033 42.332 1.00 40.98 ATOM 622 O GLY A
294 -7.688 68.185 42.061 1.00 41.17 ATOM 623 N GLU A 295 -5.636
69.051 41.753 1.00 39.55 ATOM 624 OE2 GLU A 295 -2.049 69.989
38.852 1.00 44.47 ATOM 625 OE1 GLU A 295 -3.356 71.223 37.568 1.00
44.87 ATOM 626 CD GLU A 295 -3.176 70.330 38.422 1.00 44.01 ATOM
627 CG GLU A 295 -4.422 69.647 38.946 1.00 42.26 ATOM 628 CB GLU A
295 -4.099 68.521 39.886 1.00 39.91 ATOM 629 CA GLU A 295 -5.214
68.044 40.813 1.00 38.60 ATOM 630 C GLU A 295 -4.641 66.825 41.552
1.00 36.59 ATOM 631 O GLU A 295 -4.281 66.871 42.720 1.00 36.85
ATOM 632 N LYS A 296 -4.418 65.777 40.773 1.00 33.95 ATOM 633 CA
LYS A 296 -3.782 64.575 41.294 1.00 30.68 ATOM 634 CB LYS A 296
-4.755 63.656 41.996 1.00 31.56 ATOM 635 CG LYS A 296 -5.992 63.312
41.185 1.00 32.22 ATOM 636 CD LYS A 296 -7.149 62.924 42.075 1.00
32.13 ATOM 637 CE LYS A 296 -8.161 62.034 41.381 1.00 33.43 ATOM
638 NZ LYS A 296 -8.106 62.091 39.904 1.00 34.90 ATOM 639 C LYS A
296 -3.028 63.883 40.167 1.00 28.79 ATOM 640 O LYS A 296 -3.275
64.112 38.986 1.00 27.50 ATOM 641 N HIS A 297 -1.999 63.135 40.580
1.00 25.20 ATOM 642 CA HIS A 297 -1.216 62.396 39.584 1.00 20.71
ATOM 643 CB HIS A 297 -0.148 63.267 38.981 1.00 20.06 ATOM 644 CG
HIS A 297 0.824 62.612 38.054 1.00 18.56 ATOM 645 CD2 HIS A 297
1.808 61.741 38.309 1.00 15.40 ATOM 646 ND1 HIS A 297 0.863 62.878
36.701 1.00 18.13 ATOM 647 CE1 HIS A 297 1.844 62.180 36.160 1.00
17.97 ATOM 648 NE2 HIS A 297 2.416 61.463 37.105 1.00 19.72 ATOM
649 C HIS A 297 -0.730 61.151 40.292 1.00 19.15 ATOM 650 O HIS A
297 -0.477 61.256 41.493 1.00 18.54 ATOM 651 N TYR A 298 -0.557
60.030 39.580 1.00 18.21 ATOM 652 CA TYR A 298 -0.155 58.824 40.330
1.00 16.80 ATOM 653 CB TYR A 298 -0.193 57.594 39.437 1.00 16.24
ATOM 654 CG TYR A 298 0.940 57.509 38.444 1.00 17.23 ATOM 655 CD1
TYR A 298 2.077 56.755 38.742 1.00 15.56 ATOM 656 CE1 TYR A 298
3.094 56.667 37.821 1.00 15.21 ATOM 657 CD2 TYR A 298 0.881 58.184
37.236 1.00 16.74 ATOM 658 CE2 TYR A 298 1.913 58.098 36.337 1.00
16.91 ATOM 659 CZ TYR A 298 3.022 57.325 36.636 1.00 16.49 ATOM 660
OH TYR A 298 4.046 57.260 35.719 1.00 18.10 ATOM 661 C TYR A 298
1.185 58.981 41.029 1.00 16.63 ATOM 662 O TYR A 298 1.351 58.330
42.073 1.00 16.70 ATOM 663 N ASN A 299 2.112 59.766 40.523 1.00
16.99 ATOM 664 CA ASN A 299 3.423 60.011 41.085 1.00 18.59 ATOM 665
CB ASN A 299 4.463 60.166 39.965 1.00 18.92 ATOM 666 CG ASN A 299
5.890 60.214 40.494 1.00 18.76 ATOM 667 OD1 ASN A 299 6.205 59.456
41.412 1.00 18.23 ATOM 668 ND2 ASN A 299 6.748 61.070 39.950 1.00
19.19 ATOM 669 C ASN A 299 3.513 61.224 42.009 1.00 20.06 ATOM 670
O ASN A 299 4.606 61.723 42.362 1.00 18.80 ATOM 671 N MET A 300
2.395 61.732 42.509 1.00 20.80 ATOM 672 CA MET A 300 2.449 62.886
43.417 1.00 22.77 ATOM 673 CB MET A 300 1.115 63.568 43.608 1.00
23.16 ATOM 674 CG MET A 300 0.029 62.931 44.436 1.00 25.10 ATOM 675
SD MET A 300 -1.590 63.723 44.020 1.00 28.00 ATOM 676 CE MET A 300
-2.640 62.839 45.144 1.00 24.04 ATOM 677 C MET A 300 3.012 62.458
44.767 1.00 21.89 ATOM 678 O MET A 300 2.833 61.314 45.160 1.00
18.53 ATOM 679 N ALA A 301 3.369 63.413 45.606 1.00 22.83 ATOM 680
CA ALA A 301 3.872 63.134 46.926 1.00 23.87 ATOM 681 CB ALA A 301
4.450 64.429 47.484 1.00 23.72 ATOM 682 C ALA A 301 2.841 62.580
47.893 1.00 24.63 ATOM 683 O ALA A 301 3.085 61.615 48.637 1.00
23.46 ATOM 684 N LYS A 302 1.652 63.178 47.913 1.00 26.09 ATOM 685
CA LYS A 302 0.645 62.719 48.859 1.00 26.00 ATOM 686 CB LYS A 302
-0.215 63.890 49.329 1.00 28.38 ATOM 687 CG LYS A 302 0.554 65.188
49.528 1.00 29.29 ATOM 688 CD LYS A 302 1.697 65.125 50.505 1.00
30.98 ATOM 689 CE LYS A 302 2.522 66.407 50.512 1.00 32.65 ATOM 690
NZ LYS A 302 2.617 67.085 49.191 1.00 33.62 ATOM 691 C LYS A 302
-0.198 61.580 48.291 1.00 25.55 ATOM 692 O LYS A 302 -0.295 61.399
47.087 1.00 24.72 ATOM 693 N SER A 303 -0.793 60.835 49.202 1.00
24.45 ATOM 694 CA SER A 303 -1.621 59.696 48.836 1.00 25.50 ATOM
695 CB SER A 303 -1.434 58.619 49.878 1.00 26.64 ATOM 696 OG SER A
303 -1.869 59.030 51.149 1.00 31.03 ATOM 697 C SER A 303 -3.052
60.182 48.666 1.00 25.87 ATOM 698 O SER A 303 -3.396 61.256 49.185
1.00 22.73 ATOM 699 N TYR A 304 -3.814 59.454 47.855 1.00 24.98
ATOM 700 CA TYR A 304 -5.181 59.862 47.550 1.00 24.27 ATOM 701 CB
TYR A 304 -5.140 60.449 46.132 1.00 25.45 ATOM 702 CG TYR A 304
-6.466 61.015 45.686 1.00 28.35 ATOM 703 CD1 TYR A 304 -6.762
62.357 45.929 1.00 29.08 ATOM 704 CE1 TYR A 304 -7.978 62.891
45.552 1.00 29.59 ATOM 705 CD2 TYR A 304 -7.426 60.214 45.085 1.00
27.54 ATOM 706 CE2 TYR A 304 -8.630 60.754 44.702 1.00 29.33 ATOM
707 CZ TYR A 304 -8.893 62.094 44.944 1.00 29.27 ATOM 708 OH TYR A
304 -10.103 62.610 44.578 1.00 32.08 ATOM 709 C TYR A 304 -6.192
58.769 47.721 1.00 25.14 ATOM 710 O TYR A 304 -5.707 57.735 47.336
1..alpha.00 20.08 ATOM 711 N PRO A 305 -7.356 59.000 48.279 1.00
26.60 ATOM 712 CD PRO A 305 -8.386 58.001 48.459 1.00 25.22 ATOM
713 CA PRO A 305 -7.989 60.240 48.637 1.00 27.02 ATOM 714 CB PRO A
305 -9.492 60.083 48.491 1.00 25.75 ATOM 715 CG PRO A 305 -9.707
58.646 48.789 1.00 25.74 ATOM 716 C PRO A 305 -7.625 60.782 50.007
1.00 28.63 ATOM 717 O PRO A 305 -7.922 61.915 50.430 1.00 28.68
ATOM 718 N ASN A 306 -6.988 59.904 50.783 1.00 29.14 ATOM 719 CA
ASN A 306 -6.713 60.273 52.173 1.00 30.26 ATOM 720 CB ASN A 306
-7.350 59.232 53.075 1.00 32.58 ATOM 721 CG ASN A 306 -8.766 58.861
52.676 1.00 35.48 ATOM 722 OD1 ASN A 306 -9.661 59.712 52.612 1.00
36.24 ATOM 723 ND2 ASN A 306 -8.992 57.588 52.368 1.00 36.01 ATOM
724 C ASN A 306 -5.229 60.477 52.375 1.00 29.94 ATOM 725 O ASN A
306 -4.444 59.536 52.439 1.00 28.29 ATOM 726 N GLU A 307 -4.819
61.724 52.507 1.00 29.94 ATOM 727 OE2 GLU A 307 -2.295 66.669
51.130 1.00 41.44 ATOM 728 OE1 GLU A 307 -2.711 66.223 53.238 1.00
41.35 ATOM 729 CD GLU A 307 -2.656 65.896 52.037 1.00 40.02 ATOM
730 CG GLU A 307 -3.034 64.464 51.678 1.00 38.20 ATOM 731 CB GLU A
307 -3.401 63.683 52.914 1.00 33.77 ATOM 732 CA GLU A 307 -3.461
62.152 52.717 1.00 32.32 ATOM 733 C GLU A 307 -2.770 61.594 53.949
1.00 30.24 ATOM 734 O GLU A 307 -1.542 61.522 53.970 1.00 30.44
ATOM 735 N GLU A 308 -3.513 61.273 54.984 1.00 29.39 ATOM 736 OE2
GLU A 308 -6.099 61.320 55.501 1.00 38.47 ATOM 737 OE1 GLU A 308
-7.129 59.531 56.089 1.00 37.95 ATOM 738 CD GLU A 308 -6.176 60.330
56.250 1.00 36.70 ATOM 739 CG GLU A 308 -5.168 60.057 57.322 1.00
35.39 ATOM 740 CB GLU A 308 -3.931 60.913 57.384 1.00 33.10 ATOM
741 CA GLU A 308 -2.950 60.795 56.224 1.00 29.98 ATOM 742 C GLU A
308 -2.487 59.353 56.093 1.00 28.21 ATOM 743 O GLU A 308 -1.607
58.963 56.857 1.00 26.31 ATOM 744 N LYS A 309 -3.004 58.626 55.094
1.00 24.19 ATOM 745 CA LYS A 309 -2.648 57.216 55.015 1.00 24.20
ATOM 746 CB LYS A 309 -3.711 56.413 54.263 1.00 24.89 ATOM 747 CG
LYS A 309 -5.098 56.344 54.884 1.00 27.73 ATOM 748 CD LYS A 309
-5.089 56.571 56.361 1.00 31.22 ATOM 749 CE LYS A 309 -6.317 56.792
57.162 1.00 32.52 ATOM 750 NZ LYS A 309 -6.638 55.694 58.107 1.00
34.53 ATOM 751 C LYS A 309 -1.285 57.012 54.387 1.00 21.98 ATOM 752
O LYS A 309 -0.832 57.851 53.636 1.00 22.27 ATOM 753 N ASP A 310
-0.675 55.855 54.593 1.00 20.75 ATOM 754 CA ASP A 310 0.605 55.530
53.953 1.00 20.70 ATOM 755 CB ASP A 310 1.130 54.173 54.501 1.00
20.19 ATOM 756 CG ASP A 310 2.615 54.085 54.167 1.00 22.21 ATOM 757
OD1 ASP A 310 3.335 54.893 54.831 1.00 24.55 ATOM 758 OD2 ASP A 310
3.040 53.306 53.292 1.00 18.20 ATOM 759 C ASP A 310 0.474 55.419
52.450 1.00 19.43 ATOM 760 O ASP A 310 1.385 55.706 51.651 1.00
20.19 ATOM 761 N ALA A 311 -0.727 55.044 51.990 1.00 18.69 ATOM 762
CA ALA A 311 -0.947 54.834 50.572 1.00 17.44 ATOM 763 CB ALA A 311
-0.991 53.296 50.429 1.00 16.55 ATOM 764 C ALA A 311 -2.302 55.316
50.028 1.00 16.60 ATOM 765 O ALA A 311 -3.221 55.655 50.751 1.00
12.08 ATOM 766 N TRP A 312 -2.339 55.302 48.711 1.00 17.19 ATOM 767
CA TRP A 312 -3.555 55.615 47.963 1.00 15.85 ATOM 768 CB TRP A 312
-3.236 55.569 46.483 1.00 14.37 ATOM 769 CG TRP A 312 -2.713 56.757
45.760 1.00 16.10 ATOM 770 CD2 TRP A 312 -3.413 57.526 44.772 1.00
16.24 ATOM 771 CE2 TRP A 312 -2.554 58.554 44.342 1.00 17.46 ATOM
772 CE3 TRP A 312 -4.685 57.427 44.204 1.00 17.59 ATOM 773 CD1 TRP
A 312 -1.492 57.360 45.884 1.00 15.52 ATOM 774 NE1 TRP A 312 -1.381
58.440 45.050 1.00 16.75 ATOM 775 CZ2 TRP A 312 -2.900 59.475
43.357 1.00 17.80 ATOM 776 CZ3 TRP A 312 -5.022 58.348 43.228 1.00
15.80 ATOM 777 CH2 TRP A 312 -4.154 59.345 42.828 1.00 17.23 ATOM
778 C TRP A 312 -4.562 54.502 48.204 1.00 16.18 ATOM 779 O TRP A
312 -4.185 53.393 48.559 1.00 14.62 ATOM 780 N ASP A 313 -5.841
54.750 47.906 1.00 17.14 ATOM 781 CA ASP A 313 -6.817 53.682 47.741
1.00 17.42 ATOM 782 CB ASP A 313 -8.170 54.298 47.343 1.00 19.70
ATOM 783 CG ASP A 313 -9.106 53.244 46.797 1.00 22.89 ATOM 784 OD1
ASP A 313 -9.073 52.935 45.598 1.00 24.60 ATOM 785 OD2 ASP A 313
-9.841 52.697 47.623 1.00 26.78 ATOM 786 C ASP A 313 -6.273 52.903
46.539 1.00 14.58 ATOM 787 O ASP A 313 -6.033 53.570 45.533 1.00
14.85 ATOM 788 N VAL A 314 -6.101 51.596 46.588 1.00 15.33 ATOM 789
CA VAL A 314 -5.434 50.915 45.475 1.00 14.25 ATOM 790 CB VAL A 314
-5.067 49.478 45.891 1.00 13.94 ATOM 791 CG1 VAL A 314 -6.266
48.578 46.107 1.00 12.58 ATOM 792 CG2 VAL A 314 -4.090 48.891
44.864 1.00 12.94 ATOM 793 C VAL A 314 -6.110 50.966 44.129 1.00
14.22 ATOM 794 O VAL A 314 -5.480 51.152 43.099 1.00 13.01 ATOM 795
N LYS A 315 -7.435 50.819 44.092 1.00 16.35 ATOM 796 NZ LYS A 315
-12.342 48.884 46.668 1.00 27.78 ATOM 797 CE LYS A 315 -11.908
48.279 45.365 1.00 26.99 ATOM 798 CD LYS A 315 -11.499 49.287
44.334 1.00 24.72 ATOM 799 CG LYS A 315 -10.084 49.278 43.775 1.00
21.52 ATOM 800 CB LYS A 315 -9.700 50.621 43.204 1.00 18.30 ATOM
801 CA LYS A 315 -8.223 50.882 42.870 1.00 17.01 ATOM 802 C LYS A
315 -8.111 52.273 42.255 1.00 15.42 ATOM 803 O LYS A 315 -7.967
52.432 41.065 1.00 14.82 ATOM 804 N MET A 316 -8.189 53.317 43.095
1.00 16.67 ATOM 805 CA MET A 316 -8.023 54.673 42.531 1.00 16.96
ATOM 806 CB MET A 316 -8.396 55.717 43.579 1.00 19.66 ATOM 807 CG
MET A 316 -9.874 55.616 43.966 1.00 21.49 ATOM 808 SD MET A 316
-10.303 56.858 45.216 1.00 25.95 ATOM 809 CE MET A 316 -11.921
56.212 45.655 1.00 23.70 ATOM 810 C MET A 316 -6.624 54.927 41.996
1.00 16.13 ATOM 811 O MET A 316 -6.463 55.672 41.038 1.00 16.38
ATOM 812 N LEU A 317 -5.583 54.315 42.552 1.00 15.29 ATOM 813 CA
LEU A 317 -4.218 54.490 42.078 1.00 13.26 ATOM 814 CB LEU A 317
-3.200 53.915 43.097 1.00 11.74 ATOM 815 CG LEU A 317 -1.733 54.036
42.672 1.00 12.04 ATOM 816 CD1 LEU A 317 -1.323 55.505 42.512 1.00
10.99 ATOM 817 CD2 LEU A 317 -0.861 53.326 43.698 1.00 11.54 ATOM
818 C LEU A 317 -4.079 53.785 40.755 1.00 12.38 ATOM 819 O LEU A
317 -3.526 54.334 39.819 1.00 14.48 ATOM 820 N LEU A 318 -4.637
52.580 40.616 1.00 12.87 ATOM 821 CA LEU A 318 -4.561 51.890 39.326
1.00 12.84 ATOM 822 CB LEU A 318 -5.112 50.457 39.453 1.00 12.79
ATOM 823 CG LEU A 318 -4.966 49.629 38.170 1.00 14.82 ATOM 824 CD1
LEU A 318 -3.531 49.542 37.700 1.00 15.86 ATOM 825 CD2 LEU A 318
-5.558 48.241 38.355 1.00 14.73 ATOM 826 C LEU A 318 -5.230 52.673
38.218 1.00 12.07 ATOM 827 O LEU A 318 -4.780 52.829 37.099 1.00
12.74 ATOM 828 N GLU A 319 -6.420 53.193 38.511 1.00 15.24 ATOM 829
OE2 GLU A 319 -11.483 52.210 39.220 1.00 28.26 ATOM 830 OE1 GLU A
319 -10.797 54.093 40.042 1.00 29.42 ATOM 831 CD GLU A 319 -10.674
53.181 39.193 1.00 27.45 ATOM 832 CG GLU A 319 -9.561 53.233 38.177
1.00 24.55 ATOM 833 CB GLU A 319 -8.553 54.375 38.270 1.00 18.18
ATOM 834 CA GLU A 319 -7.199 54.025 37.614 1.00 15.25 ATOM 835 C
GLU A 319 -6.421 55.288 37.289 1.00 14.06 ATOM 836 O GLU A 319
-6.317 55.697 36.141 1.00 13.57 ATOM 837 N GLN A 320 -5.879 55.947
38.303 1.00 14.48 ATOM 838 CA GLN A 320 -5.086 57.164 38.073 1.00
15.95 ATOM 839 CB GLN A 320 -4.631 57.772 39.413 1.00 17.37 ATOM
840 CG GLN A 320 -3.969 59.147 39.262 1.00 20.30 ATOM 841 CD GLN A
320 -4.917 60.189 38.681 1.00 20.82 ATOM 842 OE1 GLN A 320 -6.069
60.198 39.101 1.00 22.55 ATOM 843 NE2 GLN A 320 -4.480 61.016
37.768 1.00 20.50 ATOM 844 C GLN A 320 -3.896 56.849 37.206 1.00
15.14 ATOM 845 O GLN A 320 -3.545 57.572 36.274 1.00 16.35 ATOM 846
N PHE A 321 -3.178 55.751 37.535 1.00 14.19 ATOM 847 CA PHE A 321
-1.997 55.416 36.728 1.00 13.62 ATOM 848 CB PHE A 321 -1.389 54.100
37.277 1.00 14.84 ATOM 849 CG PHE A 321 -0.286 53.531 36.462 1.00
14.25 ATOM 850 CD1 PHE A 321 0.874 54.232 36.256 1.00 16.53 ATOM
851 CD2 PHE A 321 -0.407 52.275 35.889 1.00 15.71 ATOM 852 CE1 PHE
A 321 1.915 53.714 35.499 1.00 16.96 ATOM 853 CE2 PHE A 321 0.594
51.737 35.109 1.00 16.98 ATOM 854 CZ PHE A 321 1.760 52.464 34.920
1.00 19.64 ATOM 855 C PHE A 321 -2.367 55.265 35.284 1.00 15.37
ATOM 856 O PHE A 321 -1.743 55.838 34.391 1.00 17.95 ATOM 857 N SER
A 322 -3.440 54.502 35.011 1.00 15.65 ATOM 858 CA SER A 322 -3.862
54.261 33.650 1.00 16.47 ATOM 859 CB SER A 322 -5.052 53.312 33.545
1.00 17.62 ATOM 860 OG SER A 322 -4.956 52.111 34.281 1.00 21.90
ATOM 861 C SER A 322 -4.219 55.608 32.980 1.00 15.02 ATOM 862 O SER
A 322 -3.836 55.773 31.839 1.00 15.73 ATOM 863 N PHE A 323 -4.855
56.499 33.694 1.00 15.76 ATOM 864 CA PHE A 323 -5.163 57.831 33.153
1.00 17.15 ATOM 865 CB PHE A 323 -5.953 58.697 34.136 1.00 18.22
ATOM 866 CG PHE A 323 -6.374 60.050 33.604 1.00 20.68 ATOM 867 CD1
PHE A 323 -5.660 61.195 33.861 1.00 21.45 ATOM 868 CD2 PHE A 323
-7.524 60.161 32.852 1.00 22.31 ATOM 869 CE1 PHE A 323 -6.032
62.425 33.329 1.00 22.60 ATOM 870 CE2 PHE A 323 -7.945 61.384
32.325 1.00 24.15 ATOM 871 CZ PHE A 323 -7.175 62.502 32.558 1.00
23.26 ATOM 872 C PHE A 323 -3.897 58.570 32.756 1.00 18.08 ATOM 873
O PHE A 323 -3.807 59.053 31.648 1.00 15.95 ATOM 874 N ASP A 324
-2.913 58.698 33.662 1.00 19.20 ATOM 875 CA ASP A 324 -1.748 59.521
33.405 1.00 20.57 ATOM 876 CB ASP A 324 -0.994 59.740 34.742 1.00
23.05 ATOM 877 CG ASP A 324 -1.773 60.624 35.687 1.00 23.08 ATOM
878 OD1 ASP A 324 -1.894 60.419 36.896 1.00 22.69 ATOM 879 OD2 ASP
A 324 -2.335 61.620 35.188 1.00 26.34 ATOM 880 C ASP A 324 -0.746
58.999 32.402 1.00 21.82 ATOM 881 O ASP A 324 -0.058 59.793 31.764
1.00 21.73 ATOM 882 N ILE A 325 -0.672 57.679 32.240 1.00 21.33
ATOM 883 CA ILE A 325 0.264 57.038 31.339 1.00 22.61 ATOM 884 CB
ILE A 325 0.922 55.889 32.161 1.00 25.69 ATOM 885 CG2 ILE A 325
0.154 54.579 32.043 1.00 24.65 ATOM 886 CG1 ILE A 325 2.370 55.730
31.765 1.00 26.90 ATOM 887 CD1 ILE A 325 3.369 56.465 32.631 1.00
27.55 ATOM 888 C ILE A 325 -0.394 56.551 30.073 1.00 22.34 ATOM 889
O ILE A 325 0.238 55.950 29.209 1.00
19.22 ATOM 890 N ALA A 326 -1.663 56.970 29.849 1.00 21.30 ATOM 891
CA ALA A 326 -2.412 56.538 28.677 1.00 21.14 ATOM 892 CB ALA A 326
-3.743 57.301 28.568 1.00 20.96 ATOM 893 C ALA A 326 -1.701 56.616
27.350 1.00 20.02 ATOM 894 O ALA A 326 -1.716 55.660 26.557 1.00
18.13 ATOM 895 N GLU A 327 -1.087 57.750 27.055 1.00 21.17 ATOM 896
OE2 GLU A 327 1.606 59.498 22.827 1.00 38.00 ATOM 897 OE1 GLU A 327
-0.468 59.073 22.310 1.00 37.16 ATOM 898 CD GLU A 327 0.389 59.482
23.114 1.00 35.40 ATOM 899 CG GLU A 327 -0.062 59.993 24.456 1.00
33.57 ATOM 900 CB GLU A 327 0.333 59.223 25.685 1.00 26.72 ATOM 901
CA GLU A 327 -0.310 57.845 25.805 1.00 23.82 ATOM 902 C GLU A 327
0.748 56.760 25.705 1.00 20.97 ATOM 903 O GLU A 327 0.932 56.156
24.626 1.00 20.68 ATOM 904 N GLU A 328 1.536 56.501 26.758 1.00
20.00 ATOM 905 OE2 GLU A 328 6.485 57.857 28.558 1.00 20.00 ATOM
906 OE1 GLU A 328 5.550 56.151 29.650 1.00 20.00 ATOM 907 CD GLU A
328 5.561 56.965 28.651 1.00 20.00 ATOM 908 CG GLU A 328 4.474
56.875 27.579 1.00 20.00 ATOM 909 CB GLU A 328 3.625 55.608 27.694
1.00 20.00 ATOM 910 CA GLU A 328 2.584 55.483 26.582 1.00 20.00
ATOM 911 C GLU A 328 1.966 54.088 26.633 1.00 20.00 ATOM 912 O GLU
A 328 2.420 53.176 25.932 1.00 20.00 ATOM 913 N ALA A 329 0.930
53.900 27.438 1.00 19.57 ATOM 914 CA ALA A 329 0.280 52.586 27.501
1.00 21.09 ATOM 915 CB ALA A 329 -0.828 52.628 28.545 1.00 20.52
ATOM 916 C ALA A 329 -0.292 52.148 26.167 1.00 21.74 ATOM 917 O ALA
A 329 -0.460 50.963 25.860 1.00 22.92 ATOM 918 N SER A 330 -0.593
53.077 25.263 1.00 22.47 ATOM 919 CA SER A 330 -1.185 52.807 23.972
1.00 23.55 ATOM 920 CB SER A 330 -1.669 54.151 23.397 1.00 23.31
ATOM 921 OG SER A 330 -0.614 54.856 22.781 1.00 25.74 ATOM 922 C
SER A 330 -0.209 52.129 23.038 1.00 23.79 ATOM 923 O SER A 330
-0.545 51.533 22.032 1.00 23.93 ATOM 924 N LYS A 331 1.073 52.141
23.385 1.00 20.00 ATOM 925 NZ LYS A 331 4.948 57.241 21.295 1.00
20.00 ATOM 926 CE LYS A 331 3.827 56.290 21.276 1.00 20.00 ATOM 927
CD LYS A 331 4.162 54.958 21.950 1.00 20.00 ATOM 928 CG LYS A 331
2.995 53.972 21.928 1.00 20.00 ATOM 929 CB LYS A 331 3.327 52.639
22.588 1.00 20.00 ATOM 930 CA LYS A 331 2.147 51.613 22.574 1.00
20.00 ATOM 931 C LYS A 331 2.659 50.261 22.982 1.00 20.00 ATOM 932
O LYS A 331 3.484 49.683 22.278 1.00 20.00 ATOM 933 N VAL A 332
2.207 49.748 24.112 1.00 21.24 ATOM 934 CA VAL A 332 2.660 48.458
24.592 1.00 19.62 ATOM 935 CB VAL A 332 3.467 48.639 25.896 1.00
20.18 ATOM 936 CG1 VAL A 332 4.816 49.290 25.660 1.00 19.67 ATOM
937 CG2 VAL A 332 2.651 49.433 26.897 1.00 18.82 ATOM 938 C VAL A
332 1.495 47.533 24.905 1.00 18.19 ATOM 939 O VAL A 332 0.344
47.965 24.964 1.00 18.61 ATOM 940 N CYS A 333 1.816 46.274 25.098
1.00 16.72 ATOM 941 CA CYS A 333 0.869 45.245 25.475 1.00 18.08
ATOM 942 CB CYS A 333 1.522 43.872 25.365 1.00 20.13 ATOM 943 SG
CYS A 333 0.694 42.448 26.048 1.00 23.29 ATOM 944 C CYS A 333 0.346
45.494 26.878 1.00 16.83 ATOM 945 O CYS A 333 -0.855 45.347 27.075
1.00 15.63 ATOM 946 N LEU A 334 1.226 45.820 27.841 1.00 12.87 ATOM
947 CA LEU A 334 0.838 46.094 29.187 1.00 12.88 ATOM 948 CB LEU A
334 0.857 44.925 30.157 1.00 13.67 ATOM 949 CG LEU A 334 0.004
43.686 29.867 1.00 12.92 ATOM 950 CD1 LEU A 334 0.432 42.513 30.699
1.00 9.29 ATOM 951 CD2 LEU A 334 -1.465 44.030 30.158 1.00 10.80
ATOM 952 C LEU A 334 1.788 47.151 29.816 1.00 14.90 ATOM 953 O LEU
A 334 2.929 47.313 29.428 1.00 13.88 ATOM 954 N ALA A 335 1.203
47.856 30.778 1.00 14.03 ATOM 955 CA ALA A 335 1.920 48.854 31.544
1.00 14.38 ATOM 956 CB ALA A 335 1.292 50.235 31.381 1.00 15.95
ATOM 957 C ALA A 335 1.859 48.404 32.991 1.00 13.27 ATOM 958 O ALA
A 335 0.794 48.020 33.435 1.00 14.32 ATOM 959 N HIS A 336 2.936
48.458 33.750 1.00 12.02 ATOM 960 CA HIS A 336 2.886 48.039 35.141
1.00 10.67 ATOM 961 CB HIS A 336 3.543 46.655 35.252 1.00 9.55 ATOM
962 CG HIS A 336 3.104 45.903 36.449 1.00 12.03 ATOM 963 CD2 HIS A
336 2.026 45.099 36.601 1.00 9.22 ATOM 964 ND1 HIS A 336 3.730
45.915 37.684 1.00 11.53 ATOM 965 CE1 HIS A 336 3.048 45.186 38.533
1.00 8.45 ATOM 966 NE2 HIS A 336 1.995 44.711 37.891 1.00 12.17
ATOM 967 C HIS A 336 3.596 49.089 36.001 1.00 13.40 ATOM 968 O HIS
A 336 4.669 49.576 35.641 1.00 13.52 ATOM 969 N LEU A 337 2.987
49.473 37.104 1.00 12.61 ATOM 970 CA LEU A 337 3.514 50.440 38.026
1.00 12.83 ATOM 971 CB LEU A 337 2.357 51.336 38.563 1.00 13.25
ATOM 972 CG LEU A 337 2.688 52.163 39.816 1.00 14.22 ATOM 973 CD1
LEU A 337 3.886 53.054 39.540 1.00 14.34 ATOM 974 CD2 LEU A 337
1.489 53.027 40.232 1.00 12.48 ATOM 975 C LEU A 337 4.137 49.730
39.243 1.00 11.76 ATOM 976 O LEU A 337 3.486 48.887 39.833 1.00
10.29 ATOM 977 N PHE A 338 5.396 50.012 39.548 1.00 11.12 ATOM 978
CA PHE A 338 6.001 49.502 40.775 1.00 9.77 ATOM 979 CB PHE A 338
7.378 48.877 40.584 1.00 10.95 ATOM 980 CG PHE A 338 7.336 47.661
39.688 1.00 10.01 ATOM 981 CD1 PHE A 338 7.341 47.745 38.336 1.00
6.77 ATOM 982 CD2 PHE A 338 7.347 46.392 40.263 1.00 10.50 ATOM 983
CE1 PHE A 338 7.261 46.629 37.530 1.00 8.39 ATOM 984 CE2 PHE A 338
7.260 45.254 39.504 1.00 7.33 ATOM 985 CZ PHE A 338 7.257 45.387
38.135 1.00 9.45 ATOM 986 C PHE A 338 6.093 50.673 41.742 1.00
10.02 ATOM 987 O PHE A 338 6.621 51.737 41.430 1.00 11.99 ATOM 988
N THR A 339 5.501 50.493 42.897 1.00 9.73 ATOM 989 CA THR A 339
5.455 51.468 43.940 1.00 12.72 ATOM 990 CB THR A 339 4.053 52.092
44.003 1.00 15.30 ATOM 991 OG1 THR A 339 4.060 53.246 44.828 1.00
16.97 ATOM 992 CG2 THR A 339 3.048 51.044 44.495 1.00 15.12 ATOM
993 C THR A 339 5.886 50.830 45.261 1.00 12.86 ATOM 994 O THR A 339
6.169 49.638 45.386 1.00 12.05 ATOM 995 N TYR A 340 5.954 51.663
46.268 1.00 11.17 ATOM 996 CA TYR A 340 6.323 51.315 47.616 1.00
11.82 ATOM 997 CB TYR A 340 7.823 51.632 47.844 1.00 10.81 ATOM 998
CG TYR A 340 8.342 51.050 49.121 1.00 11.94 ATOM 999 CD1 TYR A 340
8.766 49.717 49.157 1.00 13.34 ATOM 1000 CE1 TYR A 340 9.228 49.144
50.342 1.00 13.22 ATOM 1001 CD2 TYR A 340 8.375 51.795 50.286 1.00
11.57 ATOM 1002 CE2 TYR A 340 8.862 51.239 51.481 1.00 13.34 ATOM
1003 CZ TYR A 340 9.294 49.929 51.474 1.00 12.53 ATOM 1004 OH TYR A
340 9.708 49.349 52.630 1.00 13.01 ATOM 1005 C TYR A 340 5.422
52.066 48.560 1.00 12.56 ATOM 1006 O TYR A 340 5.784 53.138 49.086
1.00 13.24 ATOM 1007 N GLN A 341 4.248 51.479 48.801 1.00 12.63
ATOM 1008 CA GLN A 341 3.279 52.124 49.697 1.00 12.73 ATOM 1009 CB
GLN A 341 2.457 53.199 48.996 1.00 15.96 ATOM 1010 CG GLN A 341
1.630 52.745 47.790 1.00 12.86 ATOM 1011 CD GLN A 341 1.063 53.915
47.015 1.00 14.37 ATOM 1012 OE1 GLN A 341 0.019 54.455 47.412 1.00
14.26 ATOM 1013 NE2 GLN A 341 1.690 54.336 45.913 1.00 12.87 ATOM
1014 C GLN A 341 2.409 51.026 50.281 1.00 13.05 ATOM 1015 O GLN A
341 2.135 50.001 49.651 1.00 12.36 ATOM 1016 N ASP A 342 2.036
51.214 51.518 1.00 11.62 ATOM 1017 CA ASP A 342 1.313 50.221 52.293
1.00 12.36 ATOM 1018 CB ASP A 342 1.766 50.350 53.748 1.00 11.58
ATOM 1019 CG ASP A 342 1.603 49.084 54.533 1.00 14.83 ATOM 1020 OD1
ASP A 342 1.136 48.053 53.979 1.00 12.25 ATOM 1021 OD2 ASP A 342
1.976 49.034 55.734 1.00 16.16 ATOM 1022 C ASP A 342 -0.202 50.344
52.159 1.00 13.76 ATOM 1023 O ASP A 342 -0.850 50.940 53.016 1.00
10.06 ATOM 1024 N PHE A 343 -0.705 49.804 51.047 1.00 12.77 ATOM
1025 CA PHE A 343 -2.123 49.686 50.828 1.00 13.67 ATOM 1026 CB PHE
A 343 -2.407 48.846 49.572 1.00 11.23 ATOM 1027 CG PHE A 343 -1.813
49.447 48.336 1.00 13.10 ATOM 1028 CD1 PHE A 343 -2.070 50.761
47.978 1.00 11.42 ATOM 1029 CD2 PHE A 343 -0.968 48.703 47.519 1.00
11.91 ATOM 1030 CE1 PHE A 343 -1.494 51.298 46.845 1.00 11.83 ATOM
1031 CE2 PHE A 343 -0.431 49.235 46.395 1.00 13.17 ATOM 1032 CZ PHE
A 343 -0.693 50.547 46.011 1.00 11.97 ATOM 1033 C PHE A 343 -2.827
49.073 52.018 1.00 13.82 ATOM 1034 O PHE A 343 -2.411 48.111 52.663
1.00 11.97 ATOM 1035 N ASP A 344 -3.976 49.646 52.364 1.00 14.70
ATOM 1036 CA ASP A 344 -4.734 49.159 53.500 1.00 16.30 ATOM 1037 CB
ASP A 344 -6.066 49.896 53.688 1.00 19.86 ATOM 1038 CG ASP A 344
-5.919 51.270 54.287 1.00 24.51 ATOM 1039 OD1 ASP A 344 -4.851
51.804 54.586 1.00 24.70 ATOM 1040 OD2 ASP A 344 -6.983 51.916
54.436 1.00 27.18 ATOM 1041 C ASP A 344 -5.093 47.675 53.396 1.00
13.85 ATOM 1042 O ASP A 344 -5.208 47.157 52.310 1.00 9.77 ATOM
1043 N MET A 345 -5.213 47.069 54.564 1.00 13.16 ATOM 1044 CA MET A
345 -5.708 45.753 54.800 1.00 17.11 ATOM 1045 CB MET A 345 -7.235
45.710 54.450 1.00 23.01 ATOM 1046 CG MET A 345 -7.944 46.687
55.406 1.00 28.99 ATOM 1047 SD MET A 345 -9.349 45.913 56.215 1.00
39.05 ATOM 1048 CE MET A 345 -9.993 47.308 57.172 1.00 36.28 ATOM
1049 C MET A 345 -4.992 44.658 54.023 1.00 16.11 ATOM 1050 O MET A
345 -5.586 43.729 53.496 1.00 11.01 ATOM 1051 N GLY A 346 -3.649
44.811 53.976 1.00 13.02 ATOM 1052 CA GLY A 346 -2.848 43.775
53.364 1.00 12.48 ATOM 1053 C GLY A 346 -2.754 43.637 51.897 1.00
10.57 ATOM 1054 O GLY A 346 -2.092 42.654 51.492 1.00 10.20 ATOM
1055 N THR A 347 -3.392 44.516 51.092 1.00 10.34 ATOM 1056 CA THR A
347 -3.236 44.419 49.645 1.00 10.83 ATOM 1057 CB THR A 347 -4.204
45.426 48.958 1.00 11.57 ATOM 1058 OG1 THR A 347 -5.515 45.104
49.474 1.00 11.02 ATOM 1059 CG2 THR A 347 -4.265 45.214 47.479 1.00
9.59 ATOM 1060 C THR A 347 -1.836 44.663 49.119 1.00 11.95 ATOM
1061 O THR A 347 -1.160 45.651 49.506 1.00 10.72 ATOM 1062 N LEU A
348 -1.400 43.853 48.171 1.00 9.29 ATOM 1063 CA LEU A 348 -0.082
43.950 47.585 1.00 11.62 ATOM 1064 CB LEU A 348 0.550 42.518 47.580
1.00 11.68 ATOM 1065 CG LEU A 348 0.940 41.992 48.967 1.00 13.21
ATOM 1066 CD1 LEU A 348 1.053 40.466 48.924 1.00 13.82 ATOM 1067
CD2 LEU A 348 2.282 42.599 49.398 1.00 13.80 ATOM 1068 C LEU A 348
-0.066 44.478 46.170 1.00 10.88 ATOM 1069 O LEU A 348 0.889 45.107
45.724 1.00 10.81 ATOM 1070 N GLY A 349 -1.177 44.229 45.417 1.00
10.78 ATOM 1071 CA GLY A 349 -1.154 44.718 44.035 1.00 9.96 ATOM
1072 C GLY A 349 -2.577 44.686 43.420 1.00 11.05 ATOM 1073 O GLY A
349 -3.461 44.195 44.075 1.00 9.57 ATOM 1074 N LEU A 350 -2.667
45.177 42.201 1.00 12.61 ATOM 1075 CA LEU A 350 -4.033 45.102
41.592 1.00 12.78 ATOM 1076 CB LEU A 350 -4.766 46.348 42.090 1.00
13.89 ATOM 1077 CG LEU A 350 -6.279 46.356 41.850 1.00 16.31 ATOM
1078 CD1 LEU A 350 -6.981 45.416 42.807 1.00 17.00 ATOM 1079 CD2
LEU A 350 -6.786 47.797 41.945 1.00 15.79 ATOM 1080 C LEU A 350
-3.881 45.086 40.119 1.00 10.61 ATOM 1081 O LEU A 350 -2.867 45.625
39.674 1.00 9.12 ATOM 1082 N ALA A 351 -4.815 44.544 39.319 1.00
8.13 ATOM 1083 CA ALA A 351 -4.645 44.560 37.896 1.00 9.68 ATOM
1084 CB ALA A 351 -3.807 43.358 37.429 1.00 8.30 ATOM 1085 C ALA A
351 -6.006 44.393 37.158 1.00 8.00 ATOM 1086 O ALA A 351 -6.781
43.671 37.749 1.00 8.17 ATOM 1087 N TYR A 352 -6.102 44.923 35.997
1.00 10.56 ATOM 1088 CA TYR A 352 -7.378 44.669 35.243 1.00 11.99
ATOM 1089 CB TYR A 352 -7.517 45.727 34.152 1.00 13.49 ATOM 1090 CG
TYR A 352 -7.816 47.102 34.704 1.00 16.49 ATOM 1091 CD1 TYR A 352
-9.085 47.388 35.198 1.00 18.25 ATOM 1092 CE1 TYR A 352 -9.369
48.643 35.704 1.00 18.58 ATOM 1093 CD2 TYR A 352 -6.835 48.096
34.726 1.00 15.80 ATOM 1094 CE2 TYR A 352 -7.121 49.344 35.231 1.00
17.33 ATOM 1095 CZ TYR A 352 -8.385 49.612 35.726 1.00 19.62 ATOM
1096 OH TYR A 352 -8.671 50.865 36.229 1.00 20.81 ATOM 1097 C TYR A
352 -7.306 43.297 34.623 1.00 11.61 ATOM 1098 O TYR A 352 -6.226
42.856 34.133 1.00 9.99 ATOM 1099 N VAL A 353 -8.414 42.569 34.618
1.00 9.19 ATOM 1100 CA VAL A 353 -8.433 41.240 34.027 1.00 9.63
ATOM 1101 CB VAL A 353 -9.532 40.388 34.675 1.00 12.16 ATOM 1102
CG1 VAL A 353 -9.580 38.963 34.168 1.00 11.71 ATOM 1103 CG2 VAL A
353 -9.337 40.459 36.180 1.00 11.77 ATOM 1104 C VAL A 353 -8.597
41.233 32.532 1.00 12.50 ATOM 1105 O VAL A 353 -9.522 41.869 32.012
1.00 12.29 ATOM 1106 N GLY A 354 -7.759 40.476 31.830 1.00 10.94
ATOM 1107 CA GLY A 354 -7.781 40.328 30.400 1.00 14.31 ATOM 1108 C
GLY A 354 -8.927 39.377 29.976 1.00 14.67 ATOM 1109 O GLY A 354
-9.497 38.711 30.814 1.00 13.50 ATOM 1110 N SER A 355 -9.183 39.300
28.685 1.00 15.31 ATOM 1111 CA SER A 355 -10.226 38.340 28.254 1.00
18.47 ATOM 1112 CB SER A 355 -11.596 39.039 28.449 1.00 19.00 ATOM
1113 OG SER A 355 -12.631 38.360 27.760 1.00 21.32 ATOM 1114 C SER
A 355 -9.959 38.049 26.808 1.00 17.63 ATOM 1115 O SER A 355 -9.378
38.937 26.168 1.00 18.38 ATOM 1116 N PRO A 356 -10.461 36.952
26.271 1.00 19.06 ATOM 1117 CD PRO A 356 -11.125 35.862 27.017 1.00
18.95 ATOM 1118 CA PRO A 356 -10.370 36.661 24.859 1.00 21.26 ATOM
1119 CB PRO A 356 -10.771 35.200 24.707 1.00 20.19 ATOM 1120 CG PRO
A 356 -11.451 34.834 25.974 1.00 20.51 ATOM 1121 C PRO A 356
-11.357 37.502 24.051 1.00 24.27 ATOM 1122 O PRO A 356 -11.249
37.622 22.833 1.00 25.16 ATOM 1123 N ARG A 357 -12.369 38.028
24.714 1.00 27.47 ATOM 1124 NH2 ARG A 357 -14.829 39.823 29.846
1.00 45.75 ATOM 1125 NH1 ARG A 357 -15.278 41.187 28.059 1.00 45.46
ATOM 1126 CZ ARG A 357 -15.278 39.980 28.609 1.00 43.96 ATOM 1127
NE ARG A 357 -15.708 38.909 27.965 1.00 41.93 ATOM 1128 CD ARG A
357 -16.239 38.818 26.627 1.00 38.04 ATOM 1129 CG ARG A 357 -15.252
38.219 25.641 1.00 34.46 ATOM 1130 CB ARG A 357 -14.476 39.345
24.961 1.00 32.97 ATOM 1131 CA ARG A 357 -13.386 38.812 24.032 1.00
31.58 ATOM 1132 C ARG A 357 -12.777 39.943 23.231 1.00 33.38 ATOM
1133 O ARG A 357 -11.979 40.784 23.635 1.00 32.52 ATOM 1134 CB ALA
A 358 -13.971 40.872 19.874 1.00 35.32 ATOM 1135 C ALA A 358
-12.904 42.350 21.524 1.00 36.30 ATOM 1136 O ALA A 358 -11.940
43.090 21.425 1.00 37.79 ATOM 1137 N ALA A 358 -13.235 39.948
21.981 1.00 35.10 ATOM 1138 CA ALA A 358 -12.901 40.933 20.979 1.00
36.40 ATOM 1139 N ASN A 359 -14.027 42.745 22.100 1.00 36.18 ATOM
1140 ND2 ASN A 359 -17.589 44.247 20.981 1.00 38.54 ATOM 1141 OD1
ASN A 359 -15.577 44.791 20.178 1.00 40.48 ATOM 1142 CG ASN A 359
-16.290 44.493 21.132 1.00 39.04 ATOM 1143 CB ASN A 359 -15.760
44.367 22.540 1.00 37.70 ATOM 1144 CA ASN A 359 -14.263 44.061
22.641 1.00 36.27 ATOM 1145 C ASN A 359 -13.834 44.198 24.094 1.00
35.65 ATOM 1146 O ASN A 359 -14.157 45.182 24.751 1.00 35.92 ATOM
1147 N SER A 360 -13.119 43.218 24.615 1.00 34.08 ATOM 1148 OG SER
A 360 -10.429 42.540 26.749 1.00 34.50 ATOM 1149 CB SER A 360
-11.683 42.108 26.257 1.00 33.68 ATOM 1150 CA SER A 360 -12.647
43.261 25.982 1.00 32.95 ATOM 1151 C SER A 360 -11.907 44.577
26.260 1.00 30.62 ATOM 1152 O SER A 360 -11.073 45.055 25.505 1.00
29.59 ATOM 1153 N HIS A 361 -12.244 45.128 27.405 1.00 28.43 ATOM
1154 CD2 HIS A 361 -14.316 47.743 26.576 1.00 37.21 ATOM 1155 NE2
HIS A 361 -14.610 48.664 25.596 1.00 37.98 ATOM 1156 CE1 HIS A 361
-13.624 49.531 25.503 1.00 37.94 ATOM 1157 ND1 HIS A 361 -12.729
49.221 26.413 1.00 38.32 ATOM 1158 CG HIS A 361 -13.138 48.108
27.110 1.00 36.46 ATOM 1159 CB HIS A 361 -12.374 47.488 28.234 1.00
33.76 ATOM 1160 CA HIS A 361 -11.550 46.275 27.920 1.00 30.24 ATOM
1161 C HIS A 361 -10.928 45.653 29.193 1.00 29.98 ATOM 1162 O HIS A
361 -11.660 45.080 30.014 1.00 33.02 ATOM 1163 N GLY A 362 -9.625
45.543 29.184 1.00 25.53 ATOM 1164 CA GLY A 362 -8.942 45.064
30.371 1.00 20.70 ATOM 1165 C GLY A 362 -7.829 44.108 29.983 1.00
16.68 ATOM 1166 O GLY A 362 -7.986 43.330 29.056 1.00 12.10 ATOM
1167 N GLY A 363 -6.695 44.301 30.672 1.00 13.81 ATOM 1168 CA GLY A
363 -5.648 43.339 30.406 1.00 14.18 ATOM 1169 C GLY A 363 -4.908
43.513 29.110 1.00 13.66 ATOM 1170 O GLY A 363 -4.808 44.608 28.557
1.00 12.16 ATOM 1171 N VAL A 364 -4.310 42.413 28.673 1.00 14.17
ATOM 1172 CG2 VAL A 364 -4.030 40.030 27.062 1.00 15.53 ATOM 1173
CG1 VAL A 364 -1.902 40.628 28.266 1.00 15.36 ATOM 1174 CB VAL A
364 -2.897 41.047 27.174 1.00 18.71 ATOM 1175 CA VAL A 364 -3.455
42.442 27.502 1.00 17.47 ATOM 1176 C VAL A 364 -4.138 43.070 26.294
1.00 22.13 ATOM 1177 O VAL A 364 -5.313 42.870 26.039 1.00 18.74
ATOM 1178 N CYS A 365 -3.323 43.852 25.611 1.00 26.09 ATOM 1179 SG
CYS A 365 -3.257 42.201 22.935 1.00 38.70 ATOM 1180 CB CYS A 365
-4.184 43.720 23.283 1.00 34.15 ATOM 1181 CA CYS A 365 -3.608
44.572 24.400 1.00 32.02 ATOM 1182 C CYS A 365 -4.433 45.808 24.719
1.00 32.84 ATOM 1183 O CYS A 365 -5.505 45.800 25.307 1.00 33.41
ATOM 1184 N PRO A 366 -3.826 46.943 24.391 1.00 34.94 ATOM 1185 CG
PRO A 366 -2.315 48.596 23.731 1.00
34.57 ATOM 1186 CD PRO A 366 -2.542 47.119 23.698 1.00 34.14 ATOM
1187 CB PRO A 366 -3.403 49.216 24.501 1.00 35.23 ATOM 1188 CA PRO
A 366 -4.515 48.220 24.644 1.00 35.75 ATOM 1189 C PRO A 366 -5.714
48.244 23.715 1.00 36.16 ATOM 1190 O PRO A 366 -5.574 47.917 22.548
1.00 38.73 ATOM 1191 N LYS A 367 -6.905 48.420 24.211 1.00 36.55
ATOM 1192 NZ LYS A 367 -11.060 46.083 19.713 1.00 39.83 ATOM 1193
CE LYS A 367 -10.945 46.830 20.990 1.00 39.58 ATOM 1194 CD LYS A
367 -10.461 45.989 22.141 1.00 39.79 ATOM 1195 CG LYS A 367 -9.071
46.373 22.617 1.00 39.82 ATOM 1196 CB LYS A 367 -9.031 47.227
23.867 1.00 36.78 ATOM 1197 CA LYS A 367 -8.228 48.530 23.676 1.00
37.98 ATOM 1198 C LYS A 367 -8.995 49.679 24.375 1.00 36.54 ATOM
1199 O LYS A 367 -9.725 49.528 25.366 1.00 33.85 ATOM 1200 N ALA A
368 -8.855 50.888 23.839 1.00 35.07 ATOM 1201 CA ALA A 368 -9.376
52.113 24.400 1.00 32.73 ATOM 1202 CB ALA A 368 -9.170 53.209
23.330 1.00 34.00 ATOM 1203 C ALA A 368 -10.770 52.317 24.933 1.00
32.95 ATOM 1204 O ALA A 368 -11.784 51.928 24.364 1.00 32.14 ATOM
1205 N TYR A 369 -10.878 53.117 25.991 1.00 31.03 ATOM 1206 CA TYR
A 369 -12.087 53.643 26.567 1.00 32.06 ATOM 1207 CB TYR A 369
-12.264 53.583 28.057 1.00 34.85 ATOM 1208 CG TYR A 369 -13.026
52.465 28.689 1.00 37.47 ATOM 1209 CD1 TYR A 369 -12.349 51.435
29.321 1.00 38.45 ATOM 1210 CE1 TYR A 369 -13.025 50.390 29.914
1.00 39.73 ATOM 1211 CD2 TYR A 369 -14.410 52.415 28.669 1.00 39.08
ATOM 1212 CE2 TYR A 369 -15.096 51.380 29.271 1.00 40.46 ATOM 1213
CZ TYR A 369 -14.397 50.369 29.891 1.00 41.04 ATOM 1214 OH TYR A
369 -15.067 49.320 30.483 1.00 42.68 ATOM 1215 C TYR A 369 -11.953
55.158 26.287 1.00 32.03 ATOM 1216 O TYR A 369 -10.842 55.634
26.529 1.00 28.95 ATOM 1217 N TYR A 370 -12.996 55.828 25.851 1.00
32.03 ATOM 1218 OH TYR A 370 -13.216 63.382 23.761 1.00 38.96 ATOM
1219 CD2 TYR A 370 -14.752 60.205 24.698 1.00 36.28 ATOM 1220 CE2
TYR A 370 -14.597 61.568 24.490 1.00 36.78 ATOM 1221 CZ TYR A 370
-13.409 62.034 23.975 1.00 37.70 ATOM 1222 CE1 TYR A 370 -12.383
61.168 23.662 1.00 36.85 ATOM 1223 CD1 TYR A 370 -12.563 59.820
23.871 1.00 35.73 ATOM 1224 CG TYR A 370 -13.741 59.312 24.388 1.00
35.35 ATOM 1225 CB TYR A 370 -13.896 57.816 24.613 1.00 34.74 ATOM
1226 CA TYR A 370 -12.850 57.254 25.574 1.00 33.23 ATOM 1227 C TYR
A 370 -12.982 57.999 26.897 1.00 33.56 ATOM 1228 O TYR A 370
-13.819 57.580 27.684 1.00 33.28 ATOM 1229 N SER A 371 -12.169
59.015 27.125 1.00 35.61 ATOM 1230 OG SER A 371 -10.952 60.534
30.263 1.00 37.79 ATOM 1231 CB SER A 371 -10.994 59.623 29.183 1.00
39.02 ATOM 1232 CA SER A 371 -12.297 59.726 28.389 1.00 38.04 ATOM
1233 C SER A 371 -12.666 61.187 28.201 1.00 39.91 ATOM 1234 O SER A
371 -11.843 62.022 27.853 1.00 38.57 ATOM 1235 N PRO A 372 -13.931
61.484 28.420 1.00 43.18 ATOM 1236 CG PRO A 372 -16.243 61.324
28.452 1.00 44.76 ATOM 1237 CD PRO A 372 -15.008 60.574 28.867 1.00
44.03 ATOM 1238 CB PRO A 372 -15.900 62.772 28.425 1.00 45.09 ATOM
1239 CA PRO A 372 -14.403 62.878 28.437 1.00 45.43 ATOM 1240 C PRO
A 372 -13.833 63.366 29.760 1.00 46.59 ATOM 1241 O PRO A 372
-14.102 62.743 30.789 1.00 47.64 ATOM 1242 N VAL A 373 -12.978
64.362 29.767 1.00 48.18 ATOM 1243 CG2 VAL A 373 -13.693 65.070
32.937 1.00 48.74 ATOM 1244 CG1 VAL A 373 -11.354 64.454 33.270
1.00 48.15 ATOM 1245 CB VAL A 373 -12.534 64.295 32.314 1.00 48.02
ATOM 1246 CA VAL A 373 -12.176 64.754 30.916 1.00 47.53 ATOM 1247 C
VAL A 373 -10.843 64.081 30.527 1.00 47.84 ATOM 1248 O VAL A 373
-10.706 62.871 30.685 1.00 48.07 ATOM 1249 N GLY A 374 -9.966
64.870 29.936 1.00 46.71 ATOM 1250 CA GLY A 374 -8.687 64.329
29.474 1.00 45.92 ATOM 1251 C GLY A 374 -8.718 64.327 27.946 1.00
44.34 ATOM 1252 O GLY A 374 -7.703 64.386 27.276 1.00 43.53 ATOM
1253 N LYS A 375 -9.937 64.227 27.433 1.00 20.00 ATOM 1254 NZ LYS A
375 -11.621 70.469 25.481 1.00 20.00 ATOM 1255 CE LYS A 375 -11.822
69.108 25.996 1.00 20.00 ATOM 1256 CD LYS A 375 -10.786 68.111
25.474 1.00 20.00 ATOM 1257 CG LYS A 375 -10.999 66.697 26.013 1.00
20.00 ATOM 1258 CB LYS A 375 -9.976 65.697 25.486 1.00 20.00 ATOM
1259 CA LYS A 375 -10.201 64.262 26.011 1.00 20.00 ATOM 1260 C LYS
A 375 -9.308 63.297 25.260 1.00 20.00 ATOM 1261 O LYS A 375 -8.529
63.725 24.411 1.00 20.00 ATOM 1262 N LYS A 376 -9.386 62.003 25.604
1.00 20.00 ATOM 1263 NZ LYS A 376 -3.285 64.692 24.559 1.00 20.00
ATOM 1264 CE LYS A 376 -4.696 64.381 24.290 1.00 20.00 ATOM 1265 CD
LYS A 376 -5.139 63.045 24.890 1.00 20.00 ATOM 1266 CG LYS A 376
-6.606 62.726 24.608 1.00 20.00 ATOM 1267 CB LYS A 376 -7.050
61.390 25.194 1.00 20.00 ATOM 1268 CA LYS A 376 -8.531 61.050
24.897 1.00 20.00 ATOM 1269 C LYS A 376 -8.810 59.601 25.286 1.00
20.00 ATOM 1270 O LYS A 376 -9.696 59.281 26.077 1.00 20.00 ATOM
1271 N ASN A 377 -8.050 58.726 24.640 1.00 29.74 ATOM 1272 CA ASN A
377 -8.240 57.295 24.862 1.00 29.33 ATOM 1273 CB ASN A 377 -7.825
56.539 23.612 1.00 29.66 ATOM 1274 CG ASN A 377 -8.743 56.803
22.433 1.00 30.76 ATOM 1275 OD1 ASN A 377 -9.964 56.948 22.559 1.00
29.66 ATOM 1276 ND2 ASN A 377 -8.128 56.865 21.249 1.00 31.26 ATOM
1277 C ASN A 377 -7.486 56.848 26.098 1.00 28.35 ATOM 1278 O ASN A
377 -6.411 57.376 26.306 1.00 26.83 ATOM 1279 N ILE A 378 -8.123
56.055 26.952 1.00 26.88 ATOM 1280 CA ILE A 378 -7.431 55.408
28.039 1.00 28.08 ATOM 1281 CB ILE A 378 -7.802 55.780 29.473 1.00
29.37 ATOM 1282 CG2 ILE A 378 -7.624 57.282 29.706 1.00 28.66 ATOM
1283 CG1 ILE A 378 -9.217 55.317 29.789 1.00 31.06 ATOM 1284 CD1
ILE A 378 -9.563 55.301 31.261 1.00 32.92 ATOM 1285 C ILE A 378
-7.567 53.887 27.882 1.00 26.65 ATOM 1286 O ILE A 378 -8.559 53.283
27.479 1.00 25.21 ATOM 1287 N TYR A 379 -6.456 53.246 28.250 1.00
25.14 ATOM 1288 CA TYR A 379 -6.298 51.805 28.141 1.00 23.62 ATOM
1289 CB TYR A 379 -5.057 51.473 27.299 1.00 25.30 ATOM 1290 CG TYR
A 379 -5.076 52.202 25.979 1.00 27.60 ATOM 1291 CD1 TYR A 379
-4.644 53.523 25.919 1.00 28.62 ATOM 1292 CE1 TYR A 379 -4.683
54.224 24.736 1.00 30.42 ATOM 1293 CD2 TYR A 379 -5.571 51.619
24.836 1.00 28.58 ATOM 1294 CE2 TYR A 379 -5.582 52.298 23.634 1.00
29.43 ATOM 1295 CZ TYR A 379 -5.141 53.597 23.603 1.00 30.50 ATOM
1296 OH TYR A 379 -5.162 54.325 22.441 1.00 33.44 ATOM 1297 C TYR A
379 -6.141 51.154 29.498 1.00 22.07 ATOM 1298 O TYR A 379 -5.164
51.460 30.204 1.00 22.84 ATOM 1299 N LEU A 380 -6.941 50.122 29.768
1.00 16.20 ATOM 1300 CA LEU A 380 -6.831 49.396 30.998 1.00 17.68
ATOM 1301 CB LEU A 380 -8.233 49.019 31.490 1.00 19.14 ATOM 1302 CG
LEU A 380 -9.204 50.207 31.624 1.00 21.74 ATOM 1303 CD1 LEU A 380
-10.479 49.690 32.285 1.00 22.06 ATOM 1304 CD2 LEU A 380 -8.614
51.364 32.411 1.00 20.08 ATOM 1305 C LEU A 380 -5.903 48.191 30.845
1.00 13.92 ATOM 1306 O LEU A 380 -6.249 47.139 31.351 1.00 12.26
ATOM 1307 N ASN A 381 -4.762 48.385 30.190 1.00 14.40 ATOM 1308 CA
ASN A 381 -3.798 47.282 30.009 1.00 15.71 ATOM 1309 CB ASN A 381
-3.148 47.364 28.646 1.00 13.12 ATOM 1310 CG ASN A 381 -2.442
48.681 28.373 1.00 13.93 ATOM 1311 OD1 ASN A 381 -2.614 49.693
29.057 1.00 15.98 ATOM 1312 ND2 ASN A 381 -1.584 48.766 27.386 1.00
13.77 ATOM 1313 C ASN A 381 -2.732 47.428 31.111 1.00 15.34 ATOM
1314 O ASN A 381 -1.538 47.496 30.826 1.00 19.25 ATOM 1315 N SER A
382 -3.178 47.587 32.318 1.00 14.28 ATOM 1316 CA SER A 382 -2.363
47.953 33.462 1.00 16.49 ATOM 1317 CB SER A 382 -2.722 49.449
33.712 1.00 19.36 ATOM 1318 OG SER A 382 -4.021 49.622 34.261 1.00
19.22 ATOM 1319 C SER A 382 -2.534 47.168 34.720 1.00 13.50 ATOM
1320 O SER A 382 -3.491 46.443 34.986 1.00 14.79 ATOM 1321 N GLY A
383 -1.543 47.226 35.602 1.00 12.08 ATOM 1322 CA GLY A 383 -1.460
46.552 36.865 1.00 7.68 ATOM 1323 C GLY A 383 -0.407 47.250 37.737
1.00 9.13 ATOM 1324 O GLY A 383 0.356 48.018 37.167 1.00 7.23 ATOM
1325 N LEU A 384 -0.406 46.961 39.018 1.00 9.60 ATOM 1326 CA LEU A
384 0.596 47.604 39.860 1.00 10.98 ATOM 1327 CB LEU A 384 0.108
48.911 40.437 1.00 10.87 ATOM 1328 CG LEU A 384 -1.098 48.840
41.372 1.00 12.55 ATOM 1329 CD1 LEU A 384 -0.751 48.583 42.819 1.00
11.84 ATOM 1330 CD2 LEU A 384 -1.824 50.197 41.305 1.00 14.27 ATOM
1331 C LEU A 384 0.963 46.614 40.982 1.00 11.40 ATOM 1332 O LEU A
384 0.191 45.705 41.269 1.00 12.17 ATOM 1333 N THR A 385 2.174
46.774 41.470 1.00 10.30 ATOM 1334 CA THR A 385 2.773 45.977 42.535
1.00 10.05 ATOM 1335 CB THR A 385 3.958 45.170 41.969 1.00 11.82
ATOM 1336 OG1 THR A 385 3.463 44.161 41.100 1.00 12.53 ATOM 1337
CG2 THR A 385 4.842 44.430 42.991 1.00 7.93 ATOM 1338 C THR A 385
3.429 46.936 43.546 1.00 10.37 ATOM 1339 O THR A 385 4.154 47.869
43.150 1.00 10.04 ATOM 1340 N SER A 386 3.153 46.751 44.806 1.00
10.16 ATOM 1341 CA SER A 386 3.835 47.470 45.853 1.00 12.08 ATOM
1342 CB SER A 386 2.913 47.978 46.925 1.00 12.65 ATOM 1343 OG SER A
386 3.625 48.590 47.976 1.00 12.90 ATOM 1344 C SER A 386 4.773
46.445 46.540 1.00 11.71 ATOM 1345 O SER A 386 4.348 45.314 46.714
1.00 11.70 ATOM 1346 N THR A 387 5.929 46.899 46.978 1.00 13.21
ATOM 1347 CA THR A 387 6.861 46.004 47.677 1.00 11.54 ATOM 1348 CB
THR A 387 8.258 45.976 47.083 1.00 12.46 ATOM 1349 OG1 THR A 387
8.794 47.305 46.902 1.00 12.52 ATOM 1350 CG2 THR A 387 8.193 45.224
45.771 1.00 9.84 ATOM 1351 C THR A 387 6.917 46.429 49.137 1.00
12.59 ATOM 1352 O THR A 387 7.750 45.916 49.857 1.00 9.86 ATOM 1353
N LYS A 388 5.888 47.176 49.559 1.00 10.11 ATOM 1354 CA LYS A 388
5.781 47.451 50.983 1.00 11.18 ATOM 1355 CB LYS A 388 5.844 48.928
51.315 1.00 10.27 ATOM 1356 CG LYS A 388 5.598 49.163 52.809 1.00
11.78 ATOM 1357 CD LYS A 388 5.570 50.665 53.086 1.00 12.59 ATOM
1358 CE LYS A 388 5.610 50.846 54.616 1.00 12.42 ATOM 1359 NZ LYS A
388 5.400 52.289 54.915 1.00 12.86 ATOM 1360 C LYS A 388 4.466
46.836 51.470 1.00 12.79 ATOM 1361 O LYS A 388 3.445 47.068 50.803
1.00 13.47 ATOM 1362 N ASN A 389 4.490 46.117 52.561 1.00 11.20
ATOM 1363 CA ASN A 389 3.235 45.595 53.114 1.00 13.05 ATOM 1364 CB
ASN A 389 2.797 44.247 52.498 1.00 9.98 ATOM 1365 CG ASN A 389
1.328 43.929 52.695 1.00 13.57 ATOM 1366 OD1 ASN A 389 0.562 44.823
53.020 1.00 11.41 ATOM 1367 ND2 ASN A 389 0.828 42.697 52.544 1.00
11.87 ATOM 1368 C ASN A 389 3.401 45.431 54.605 1.00 12.01 ATOM
1369 O ASN A 389 4.389 44.832 55.045 1.00 12.68 ATOM 1370 N TYR A
390 2.463 45.891 55.415 1.00 9.33 ATOM 1371 CA TYR A 390 2.544
45.725 56.834 1.00 11.32 ATOM 1372 CB TYR A 390 2.395 44.317 57.360
1.00 13.01 ATOM 1373 CG TYR A 390 1.034 43.646 57.138 1.00 17.63
ATOM 1374 CD1 TYR A 390 0.888 42.699 56.135 1.00 17.99 ATOM 1375
CE1 TYR A 390 -0.324 42.059 55.906 1.00 18.34 ATOM 1376 CD2 TYR A
390 -0.059 43.950 57.918 1.00 17.56 ATOM 1377 CE2 TYR A 390 -1.295
43.344 57.684 1.00 19.81 ATOM 1378 CZ TYR A 390 -1.408 42.399
56.685 1.00 18.72 ATOM 1379 OH TYR A 390 -2.599 41.770 56.456 1.00
18.16 ATOM 1380 C TYR A 390 3.875 46.390 57.342 1.00 11.32 ATOM
1381 O TYR A 390 4.444 45.867 58.262 1.00 10.06 ATOM 1382 N GLY A
391 4.202 47.568 56.849 1.00 12.38 ATOM 1383 CA GLY A 391 5.264
48.396 57.395 1.00 13.46 ATOM 1384 C GLY A 391 6.671 47.822 57.226
1.00 14.75 ATOM 1385 O GLY A 391 7.615 48.218 57.920 1.00 13.47
ATOM 1386 N LYS A 392 6.864 47.001 56.213 1.00 14.18 ATOM 1387 NZ
LYS A 392 6.391 42.087 59.901 1.00 31.24 ATOM 1388 CE LYS A 392
6.545 43.004 58.726 1.00 30.06 ATOM 1389 CD LYS A 392 7.980 43.237
58.298 1.00 27.75 ATOM 1390 CG LYS A 392 8.231 44.701 57.942 1.00
25.41 ATOM 1391 CB LYS A 392 8.133 44.929 56.460 1.00 20.56 ATOM
1392 CA LYS A 392 8.133 46.376 55.909 1.00 15.94 ATOM 1393 C LYS A
392 8.265 46.221 54.399 1.00 14.65 ATOM 1394 O LYS A 392 7.242
46.070 53.676 1.00 12.21 ATOM 1395 N THR A 393 9.481 45.969 53.965
1.00 9.88 ATOM 1396 CA THR A 393 9.744 45.629 52.574 1.00 10.28
ATOM 1397 CB THR A 393 11.178 45.988 52.155 1.00 10.31 ATOM 1398
OG1 THR A 393 11.412 47.397 52.330 1.00 9.91 ATOM 1399 CG2 THR A
393 11.393 45.631 50.693 1.00 7.83 ATOM 1400 C THR A 393 9.450
44.139 52.450 1.00 11.39 ATOM 1401 O THR A 393 9.872 43.332 53.278
1.00 11.16 ATOM 1402 N ILE A 394 8.715 43.737 51.398 1.00 8.47 ATOM
1403 CA ILE A 394 8.413 42.305 51.297 1.00 8.59 ATOM 1404 CB ILE A
394 7.223 42.076 50.329 1.00 7.43 ATOM 1405 CG2 ILE A 394 6.044
42.929 50.867 1.00 5.89 ATOM 1406 CG1 ILE A 394 7.562 42.401 48.910
1.00 7.72 ATOM 1407 CD1 ILE A 394 6.413 42.268 47.877 1.00 9.17
ATOM 1408 C ILE A 394 9.657 41.573 50.761 1.00 9.33 ATOM 1409 O ILE
A 394 10.482 42.262 50.170 1.00 8.55 ATOM 1410 N LEU A 395 9.689
40.276 50.913 1.00 7.97 ATOM 1411 CA LEU A 395 10.808 39.499 50.347
1.00 11.81 ATOM 1412 CB LEU A 395 10.545 37.998 50.620 1.00 12.21
ATOM 1413 CG LEU A 395 10.370 37.651 52.090 1.00 13.09 ATOM 1414
CD1 LEU A 395 10.246 36.140 52.313 1.00 10.90 ATOM 1415 CD2 LEU A
395 11.569 38.245 52.832 1.00 13.70 ATOM 1416 C LEU A 395 10.920
39.603 48.860 1.00 13.57 ATOM 1417 O LEU A 395 9.864 39.692 48.193
1.00 11.36 ATOM 1418 N THR A 396 12.089 39.426 48.281 1.00 9.31
ATOM 1419 CA THR A 396 12.273 39.345 46.840 1.00 11.95 ATOM 1420 CB
THR A 396 13.746 39.021 46.476 1.00 14.52 ATOM 1421 OG1 THR A 396
14.575 40.034 47.090 1.00 16.31 ATOM 1422 CG2 THR A 396 13.983
39.148 44.996 1.00 16.49 ATOM 1423 C THR A 396 11.376 38.264 46.209
1.00 12.19 ATOM 1424 O THR A 396 10.790 38.544 45.152 1.00 11.09
ATOM 1425 N LYS A 397 11.286 37.104 46.840 1.00 9.73 ATOM 1426 CA
LYS A 397 10.451 36.033 46.298 1.00 12.22 ATOM 1427 CB LYS A 397
10.630 34.680 46.935 1.00 11.19 ATOM 1428 CG LYS A 397 10.365
34.491 48.402 1.00 10.22 ATOM 1429 CD LYS A 397 10.459 32.971
48.678 1.00 12.87 ATOM 1430 CE LYS A 397 9.904 32.634 50.049 1.00
10.59 ATOM 1431 NZ LYS A 397 10.430 31.409 50.690 1.00 13.72 ATOM
1432 C LYS A 397 8.946 36.390 46.379 1.00 11.16 ATOM 1433 O LYS A
397 8.250 35.937 45.488 1.00 10.09 ATOM 1434 N GLU A 398 8.588
37.196 47.346 1.00 10.22 ATOM 1435 CA GLU A 398 7.207 37.628 47.475
1.00 13.12 ATOM 1436 CB GLU A 398 6.931 38.233 48.857 1.00 11.64
ATOM 1437 CG GLU A 398 6.951 37.079 49.876 1.00 15.45 ATOM 1438 CD
GLU A 398 6.952 37.662 51.284 1.00 15.72 ATOM 1439 OE1 GLU A 398
7.139 38.861 51.557 1.00 15.77 ATOM 1440 OE2 GLU A 398 6.722 36.861
52.186 1.00 17.81 ATOM 1441 C GLU A 398 6.875 38.628 46.374 1.00
14.23 ATOM 1442 O GLU A 398 5.819 38.496 45.746 1.00 11.36 ATOM
1443 N ALA A 399 7.791 39.566 46.144 1.00 11.71 ATOM 1444 CA ALA A
399 7.603 40.571 45.121 1.00 10.93 ATOM 1445 CB ALA A 399 8.705
41.611 45.021 1.00 10.46 ATOM 1446 C ALA A 399 7.464 39.901 43.737
1.00 11.66 ATOM 1447 O ALA A 399 6.535 40.310 43.047 1.00 8.52 ATOM
1448 N ASP A 400 8.259 38.892 43.411 1.00 10.58 ATOM 1449 CA ASP A
400 8.092 38.172 42.164 1.00 11.27 ATOM 1450 CB ASP A 400 9.061
37.001 41.994 1.00 13.33 ATOM 1451 CG ASP A 400 10.525 37.443
41.936 1.00 15.25 ATOM 1452 OD1 ASP A 400 10.787 38.685 41.868 1.00
11.52 ATOM 1453 OD2 ASP A 400 11.331 36.459 41.963 1.00 15.55 ATOM
1454 C ASP A 400 6.682 37.543 42.061 1.00 10.74 ATOM 1455 O ASP A
400 6.102 37.681 40.959 1.00 11.54 ATOM 1456 N LEU A 401 6.184
37.034 43.155 1.00 9.91 ATOM 1457 CA LEU A 401 4.860 36.399 43.132
1.00 12.48 ATOM 1458 CB LEU A 401 4.618 35.610 44.411 1.00 13.67
ATOM 1459 CG LEU A 401 5.551 34.464 44.805 1.00 19.11 ATOM 1460 CD1
LEU A 401 5.596 34.451 46.337 1.00 20.83 ATOM 1461 CD2 LEU A 401
5.072 33.125 44.302 1.00 19.44 ATOM 1462 C LEU A 401 3.724 37.434
42.994 1.00 11.96 ATOM 1463 O LEU A 401 2.701 37.092 42.388 1.00
9.62 ATOM 1464 N VAL A 402 3.849 38.626 43.576 1.00 9.23 ATOM 1465
CA VAL A 402 2.856 39.664 43.347 1.00 10.16 ATOM 1466 CB VAL A 402
3.222 40.976 44.086 1.00 11.53 ATOM 1467 CG1 VAL A 402 2.219 42.096
43.805 1.00 10.53 ATOM 1468 CG2 VAL A 402 3.254 40.598 45.565 1.00
8.46 ATOM 1469 C VAL A 402 2.701 40.016 41.882 1.00 12.23 ATOM 1470
O VAL A 402 1.581 40.096 41.336 1.00 9.50 ATOM 1471 N THR A 403
3.821 40.359 41.261 1.00 8.92 ATOM 1472 CA THR A 403 3.845 40.764
39.883 1.00 8.22 ATOM 1473 CB THR A 403 5.220 41.259 39.386 1.00
6.99 ATOM 1474 OG1 THR A 403 5.612 42.325 40.255 1.00 6.97 ATOM
1475 CG2 THR A 403 5.221 41.841 37.998 1.00 6.87 ATOM 1476 C THR A
403 3.381 39.610 38.992 1.00 9.35 ATOM 1477 O THR A 403 2.693
39.954 38.016 1.00 5.99 ATOM 1478 N THR A 404 3.855 38.407 39.301
1.00 6.34 ATOM 1479 CA THR A 404 3.382 37.249 38.521 1.00 7.96
ATOM 1480 CB THR A 404 3.962 35.946 39.104 1.00 8.71 ATOM 1481 OG1
THR A 404 5.404 35.959 38.945 1.00 8.46 ATOM 1482 CG2 THR A 404
3.448 34.743 38.303 1.00 8.11 ATOM 1483 C THR A 404 1.842 37.175
38.572 1.00 6.99 ATOM 1484 O THR A 404 1.211 36.979 37.529 1.00
6.33 ATOM 1485 N HIS A 405 1.298 37.277 39.751 1.00 6.76 ATOM 1486
CA HIS A 405 -0.132 37.232 40.003 1.00 9.62 ATOM 1487 CB HIS A 405
-0.383 37.343 41.495 1.00 7.56 ATOM 1488 CG HIS A 405 -1.828 37.311
41.919 1.00 6.65 ATOM 1489 CD2 HIS A 405 -2.801 38.292 41.862 1.00
4.31 ATOM 1490 ND1 HIS A 405 -2.389 36.204 42.467 1.00 7.16 ATOM
1491 CE1 HIS A 405 -3.664 36.486 42.798 1.00 9.20 ATOM 1492 NE2 HIS
A 405 -3.937 37.722 42.413 1.00 9.52 ATOM 1493 C HIS A 405 -0.867
38.376 39.290 1.00 9.96 ATOM 1494 O HIS A 405 -1.836 38.080 38.597
1.00 9.05 ATOM 1495 N GLU A 406 -0.414 39.612 39.416 1.00 9.09 ATOM
1496 CA GLU A 406 -1.106 40.730 38.755 1.00 11.72 ATOM 1497 CB GLU
A 406 -0.532 42.078 39.207 1.00 12.02 ATOM 1498 CG GLU A 406 -0.582
42.239 40.701 1.00 11.63 ATOM 1499 CD GLU A 406 -1.940 42.013
41.321 1.00 12.61 ATOM 1500 OE1 GLU A 406 -2.985 42.048 40.632 1.00
12.02 ATOM 1501 OE2 GLU A 406 -1.910 41.759 42.535 1.00 12.54 ATOM
1502 C GLU A 406 -1.087 40.659 37.256 1.00 13.24 ATOM 1503 O GLU A
406 -2.110 40.827 36.529 1.00 10.09 ATOM 1504 N LEU A 407 0.115
40.341 36.745 1.00 9.35 ATOM 1505 CA LEU A 407 0.259 40.106 35.318
1.00 11.31 ATOM 1506 CB LEU A 407 1.702 39.802 34.937 1.00 13.34
ATOM 1507 CG LEU A 407 2.671 40.844 34.384 1.00 17.06 ATOM 1508 CD1
LEU A 407 2.232 42.279 34.412 1.00 15.35 ATOM 1509 CD2 LEU A 407
4.112 40.690 34.878 1.00 15.89 ATOM 1510 C LEU A 407 -0.675 38.971
34.886 1.00 8.96 ATOM 1511 O LEU A 407 -1.105 38.869 33.720 1.00
9.09 ATOM 1512 N GLY A 408 -0.837 37.972 35.732 1.00 7.26 ATOM 1513
CA GLY A 408 -1.649 36.774 35.444 1.00 7.95 ATOM 1514 C GLY A 408
-3.107 37.247 35.182 1.00 8.27 ATOM 1515 O GLY A 408 -3.776 36.757
34.277 1.00 7.64 ATOM 1516 N HIS A 409 -3.565 38.171 36.030 1.00
10.19 ATOM 1517 CA HIS A 409 -4.873 38.777 35.752 1.00 11.21 ATOM
1518 CB HIS A 409 -5.162 39.920 36.713 1.00 8.68 ATOM 1519 CG HIS A
409 -5.605 39.404 38.018 1.00 10.30 ATOM 1520 CD2 HIS A 409 -5.193
39.710 39.255 1.00 9.44 ATOM 1521 ND1 HIS A 409 -6.606 38.454
38.114 1.00 10.06 ATOM 1522 CE1 HIS A 409 -6.833 38.237 39.402 1.00
12.63 ATOM 1523 NE2 HIS A 409 -5.961 38.972 40.132 1.00 12.50 ATOM
1524 C HIS A 409 -4.918 39.414 34.381 1.00 12.04 ATOM 1525 O HIS A
409 -5.853 39.223 33.615 1.00 12.02 ATOM 1526 N ASN A 410 -3.910
40.235 34.085 1.00 11.36 ATOM 1527 CA ASN A 410 -3.859 40.925
32.820 1.00 10.24 ATOM 1528 CB ASN A 410 -2.594 41.750 32.616 1.00
10.71 ATOM 1529 CG ASN A 410 -2.499 43.027 33.424 1.00 12.40 ATOM
1530 OD1 ASN A 410 -1.369 43.394 33.748 1.00 9.72 ATOM 1531 ND2 ASN
A 410 -3.598 43.762 33.714 1.00 9.17 ATOM 1532 C ASN A 410 -3.953
39.918 31.675 1.00 12.78 ATOM 1533 O ASN A 410 -4.538 40.276 30.652
1.00 9.40 ATOM 1534 N PHE A 411 -3.292 38.782 31.796 1.00 9.09 ATOM
1535 CA PHE A 411 -3.280 37.693 30.872 1.00 10.91 ATOM 1536 CB PHE
A 411 -2.105 36.703 31.101 1.00 10.93 ATOM 1537 CG PHE A 411 -0.840
37.086 30.375 1.00 13.69 ATOM 1538 CD1 PHE A 411 -0.102 38.216
30.730 1.00 15.12 ATOM 1539 CD2 PHE A 411 -0.309 36.297 29.389 1.00
15.69 ATOM 1540 CE1 PHE A 411 1.076 38.546 30.052 1.00 13.51 ATOM
1541 CE2 PHE A 411 0.862 36.617 28.723 1.00 15.64 ATOM 1542 CZ PHE
A 411 1.556 37.782 29.051 1.00 14.11 ATOM 1543 C PHE A 411 -4.606
36.903 30.849 1.00 10.27 ATOM 1544 O PHE A 411 -4.684 36.100 29.931
1.00 9.02 ATOM 1545 N GLY A 412 -5.578 37.150 31.701 1.00 11.88
ATOM 1546 CA GLY A 412 -6.912 36.517 31.563 1.00 11.02 ATOM 1547 C
GLY A 412 -7.360 35.734 32.748 1.00 12.85 ATOM 1548 O GLY A 412
-8.496 35.256 32.898 1.00 11.80 ATOM 1549 N ALA A 413 -6.418 35.577
33.737 1.00 9.39 ATOM 1550 CA ALA A 413 -6.759 34.738 34.860 1.00
8.60 ATOM 1551 CB ALA A 413 -5.523 34.078 35.504 1.00 6.06 ATOM
1552 C ALA A 413 -7.520 35.451 35.966 1.00 9.94 ATOM 1553 O ALA A
413 -7.216 36.584 36.300 1.00 9.19 ATOM 1554 N GLU A 414 -8.476
34.694 36.530 1.00 9.61 ATOM 1555 CA GLU A 414 -9.172 35.158 37.731
1.00 12.27 ATOM 1556 CB GLU A 414 -10.666 34.812 37.677 1.00 14.74
ATOM 1557 CG GLU A 414 -11.293 35.591 36.517 1.00 16.17 ATOM 1558
CD GLU A 414 -11.696 36.987 36.834 1.00 20.61 ATOM 1559 OE1 GLU A
414 -11.520 37.544 37.933 1.00 21.11 ATOM 1560 OE2 GLU A 414
-12.266 37.646 35.920 1.00 24.06 ATOM 1561 C GLU A 414 -8.509
34.457 38.920 1.00 12.40 ATOM 1562 O GLU A 414 -7.455 33.841 38.757
1.00 12.29 ATOM 1563 N HIS A 415 -9.108 34.478 40.104 1.00 11.96
ATOM 1564 CA HIS A 415 -8.512 33.849 41.259 1.00 13.28 ATOM 1565 CB
HIS A 415 -9.093 34.484 42.514 1.00 12.84 ATOM 1566 CG HIS A 415
-8.513 35.841 42.716 1.00 11.55 ATOM 1567 CD2 HIS A 415 -7.240
36.265 42.433 1.00 11.21 ATOM 1568 ND1 HIS A 415 -9.180 36.899
43.276 1.00 10.77 ATOM 1569 CE1 HIS A 415 -8.380 37.944 43.314 1.00
9.60 ATOM 1570 NE2 HIS A 415 -7.174 37.576 42.791 1.00 11.71 ATOM
1571 C HIS A 415 -8.789 32.362 41.293 1.00 14.34 ATOM 1572 O HIS A
415 -9.851 32.007 40.751 1.00 10.53 ATOM 1573 N ASP A 416 -7.855
31.567 41.826 1.00 10.07 ATOM 1574 CA ASP A 416 -8.239 30.158
41.962 1.00 11.05 ATOM 1575 CB ASP A 416 -6.978 29.338 42.310 1.00
13.78 ATOM 1576 CG ASP A 416 -5.913 29.387 41.229 1.00 14.67 ATOM
1577 OD1 ASP A 416 -6.267 29.373 40.039 1.00 13.81 ATOM 1578 OD2
ASP A 416 -4.717 29.403 41.608 1.00 16.01 ATOM 1579 C ASP A 416
-9.254 29.930 43.052 1.00 10.40 ATOM 1580 O ASP A 416 -9.169 30.468
44.159 1.00 11.17 ATOM 1581 N PRO A 417 -10.296 29.111 42.805 1.00
10.30 ATOM 1582 CD PRO A 417 -10.544 28.473 41.502 1.00 12.83 ATOM
1583 CA PRO A 417 -11.275 28.774 43.805 1.00 12.54 ATOM 1584 CB PRO
A 417 -12.495 28.328 42.975 1.00 13.91 ATOM 1585 CG PRO A 417
-11.940 27.894 41.674 1.00 15.31 ATOM 1586 C PRO A 417 -10.837
27.603 44.689 1.00 13.42 ATOM 1587 O PRO A 417 -10.002 26.839
44.276 1.00 14.77 ATOM 1588 N ASP A 418 -11.346 27.376 45.860 1.00
14.02 ATOM 1589 CA ASP A 418 -11.086 26.217 46.694 1.00 20.43 ATOM
1590 CB ASP A 418 -11.375 26.531 48.172 1.00 22.42 ATOM 1591 CG ASP
A 418 -10.416 27.595 48.658 1.00 25.60 ATOM 1592 OD1 ASP A 418
-9.221 27.416 48.429 1.00 27.33 ATOM 1593 OD2 ASP A 418 -10.724
28.666 49.204 1.00 29.81 ATOM 1594 C ASP A 418 -11.997 25.061
46.257 1.00 23.39 ATOM 1595 O ASP A 418 -13.034 25.354 45.640 1.00
23.78 ATOM 1596 N GLY A 419 -11.533 23.829 46.307 1.00 24.01 ATOM
1597 CA GLY A 419 -12.290 22.650 46.050 1.00 28.51 ATOM 1598 C GLY
A 419 -12.000 21.714 44.916 1.00 31.09 ATOM 1599 O GLY A 419
-11.000 21.726 44.169 1.00 31.19 ATOM 1600 N LEU A 420 -12.976
20.849 44.625 1.00 32.09 ATOM 1601 CD2 LEU A 420 -14.406 17.142
44.111 1.00 38.06 ATOM 1602 CD1 LEU A 420 -16.571 18.444 44.050
1.00 37.27 ATOM 1603 CG LEU A 420 -15.061 18.507 44.238 1.00 36.91
ATOM 1604 CB LEU A 420 -14.424 19.470 43.224 1.00 36.82 ATOM 1605
CA LEU A 420 -12.969 19.843 43.591 1.00 35.26 ATOM 1606 C LEU A 420
-12.193 20.130 42.331 1.00 36.08 ATOM 1607 O LEU A 420 -11.638
19.221 41.701 1.00 37.02 ATOM 1608 N ALA A 421 -12.053 21.380
41.907 1.00 36.27 ATOM 1609 CB ALA A 421 -11.478 23.254 40.472 1.00
37.17 ATOM 1610 CA ALA A 421 -11.309 21.755 40.742 1.00 36.49 ATOM
1611 C ALA A 421 -9.841 21.381 40.794 1.00 35.18 ATOM 1612 O ALA A
421 -9.253 21.402 39.702 1.00 36.10 ATOM 1613 N GLU A 422 -9.183
21.102 41.907 1.00 32.65 ATOM 1614 OE2 GLU A 422 -8.104 20.106
37.709 1.00 41.16 ATOM 1615 OE1 GLU A 422 -6.137 19.284 37.500 1.00
39.97 ATOM 1616 CD GLU A 422 -6.983 19.889 38.198 1.00 38.24 ATOM
1617 CG GLU A 422 -6.628 20.311 39.582 1.00 35.89 ATOM 1618 CB GLU
A 422 -7.426 19.813 40.760 1.00 31.03 ATOM 1619 CA GLU A 422 -7.761
20.769 41.914 1.00 29.38 ATOM 1620 C GLU A 422 -6.901 22.028 41.779
1.00 24.25 ATOM 1621 O GLU A 422 -5.677 21.951 41.645 1.00 21.69
ATOM 1622 N CYS A 423 -7.530 23.185 41.837 1.00 19.59 ATOM 1623 CA
CYS A 423 -6.862 24.457 41.613 1.00 18.31 ATOM 1624 CB CYS A 423
-7.819 25.319 40.797 1.00 21.54 ATOM 1625 SG CYS A 423 -8.112
24.725 39.090 1.00 22.74 ATOM 1626 C CYS A 423 -6.367 25.102 42.893
1.00 16.82 ATOM 1627 O CYS A 423 -5.884 26.233 42.855 1.00 14.86
ATOM 1628 N ALA A 424 -6.516 24.451 44.029 1.00 15.16 ATOM 1629 CA
ALA A 424 -5.974 24.991 45.288 1.00 16.52 ATOM 1630 CB ALA A 424
-6.940 25.972 45.924 1.00 15.07 ATOM 1631 C ALA A 424 -5.603 23.812
46.165 1.00 15.42 ATOM 1632 O ALA A 424 -6.206 23.507 47.182 1.00
16.31 ATOM 1633 N PRO A 425 -4.545 23.118 45.774 1.00 16.63 ATOM
1634 CD PRO A 425 -3.694 23.394 44.593 1.00 14.62 ATOM 1635 CA PRO
A 425 -4.096 21.935 46.484 1.00 18.01 ATOM 1636 CB PRO A 425 -2.965
21.392 45.635 1.00 17.36 ATOM 1637 CG PRO A 425 -3.064 22.047
44.301 1.00 16.07 ATOM 1638 C PRO A 425 -3.655 22.217 47.908 1.00
19.52 ATOM 1639 O PRO A 425 -3.304 23.325 48.294 1.00 16.80 ATOM
1640 N ASN A 426 -3.670 21.143 48.707 1.00 21.90 ATOM 1641 ND2 ASN
A 426 -6.177 19.346 50.674 1.00 32.22 ATOM 1642 OD1 ASN A 426
-5.934 21.437 51.378 1.00 30.53 ATOM 1643 CG ASN A 426 -5.407
20.380 51.018 1.00 30.42 ATOM 1644 CB ASN A 426 -3.898 20.167
50.928 1.00 27.25 ATOM 1645 CA ASN A 426 -3.224 21.249 50.088 1.00
24.17 ATOM 1646 C ASN A 426 -1.695 21.270 50.121 1.00 22.87 ATOM
1647 O ASN A 426 -1.014 21.025 49.127 1.00 19.72 ATOM 1648 N GLU A
427 -1.148 21.695 51.247 1.00 24.03 ATOM 1649 OE2 GLU A 427 3.209
21.859 55.259 1.00 39.86 ATOM 1650 OE1 GLU A 427 2.503 23.818
54.605 1.00 37.70 ATOM 1651 CD GLU A 427 2.640 22.595 54.427 1.00
37.41 ATOM 1652 CG GLU A 427 2.167 21.979 53.132 1.00 36.42 ATOM
1653 CB GLU A 427 0.679 22.110 52.857 1.00 32.16 ATOM 1654 CA GLU A
427 0.305 21.775 51.409 1.00 29.42 ATOM 1655 C GLU A 427 0.998
20.507 50.975 1.00 29.51 ATOM 1656 O GLU A 427 1.971 20.497 50.213
1.00 29.26 ATOM 1657 N ASP A 428 0.488 19.358 51.414 1.00 31.72
ATOM 1658 OD2 ASP A 428 -1.918 16.341 52.011 1.00 41.88 ATOM 1659
OD1 ASP A 428 -1.452 18.328 52.816 1.00 37.31 ATOM 1660 CG ASP A
428 -1.103 17.267 52.267 1.00 38.88 ATOM 1661 CB ASP A 428 0.320
16.958 51.853 1.00 36.50 ATOM 1662 CA ASP A 428 1.042 18.054 51.077
1.00 32.85 ATOM 1663 C ASP A 428 1.081 17.735 49.610 1.00 32.15
ATOM 1664 O ASP A 428 1.827 16.851 49.173 1.00 32.63 ATOM 1665 N
GLN A 429 0.302 18.414 48.772 1.00 29.90 ATOM 1666 NE2 GLN A 429
-0.844 15.151 46.445 1.00 35.05 ATOM 1667 OE1 GLN A 429 -0.895
15.061 48.694 1.00 36.32 ATOM 1668 CD GLN A 429 -1.190 15.629
47.636 1.00 35.53 ATOM 1669 CG GLN A 429 -1.923 16.955 47.621 1.00
32.82 ATOM 1670 CB GLN A 429 -1.138 18.032 46.865 1.00 30.20 ATOM
1671 CA GLN A 429 0.307 18.190 47.347 1.00 26.44 ATOM 1672 C GLN A
429 1.041 19.302 46.621 1.00 22.82 ATOM 1673 O GLN A 429 0.832
19.463 45.426 1.00 21.67 ATOM 1674 N GLY A 430 1.812 20.142 47.320
1.00 18.52 ATOM 1675 CA GLY A 430 2.534 21.185 46.583 1.00 15.09
ATOM 1676 C GLY A 430 1.982 22.581 46.889 1.00 14.53 ATOM 1677 O
GLY A 430 2.582 23.559 46.432 1.00 16.43 ATOM 1678 N GLY A 431
0.898 22.745 47.615 1.00 11.39 ATOM 1679 CA GLY A 431 0.314 24.011
47.958 1.00 12.94 ATOM 1680 C GLY A 431 -0.505 24.702 46.871 1.00
10.35 ATOM 1681 O GLY A 431 -0.764 24.181 45.795 1.00 11.85 ATOM
1682 N LYS A 432 -0.845 25.936 47.147 1.00 11.65 ATOM 1683 CA LYS A
432 -1.637 26.762 46.245 1.00 13.87 ATOM 1684 CB LYS A 432 -2.343
27.862 47.049 1.00 16.54 ATOM 1685 CG LYS A 432 -3.250 27.146
48.043 1.00 20.67 ATOM 1686 CD LYS A 432 -4.058 28.036 48.933 1.00
24.37 ATOM 1687 CE LYS A 432 -5.090 27.228 49.718 1.00 25.86 ATOM
1688 NZ LYS A 432 -4.539 25.962 50.296 1.00 25.58 ATOM 1689 C LYS A
432 -0.810 27.312 45.112 1.00 12.71 ATOM 1690 O LYS A 432 0.414
27.398 45.187 1.00 11.37 ATOM 1691 N TYR A 433 -1.533 27.713 44.072
1.00 8.62 ATOM 1692 CA TYR A 433 -0.997 28.243 42.844 1.00 8.63
ATOM 1693 CB TYR A 433 -1.677 27.746 41.591 1.00 8.94 ATOM 1694 CG
TYR A 433 -1.507 26.270 41.276 1.00 12.81 ATOM 1695 CD1 TYR A 433
-2.460 25.350 41.653 1.00 12.02 ATOM 1696 CE1 TYR A 433 -2.310
23.999 41.361 1.00 14.63 ATOM 1697 CD2 TYR A 433 -0.360 25.813
40.637 1.00 13.00 ATOM 1698 CE2 TYR A 433 -0.202 24.469 40.333 1.00
14.04 ATOM 1699 CZ TYR A 433 -1.180 23.579 40.713 1.00 15.76 ATOM
1700 OH TYR A 433 -1.018 22.241 40.429 1.00 18.68 ATOM 1701 C TYR A
433 -0.990 29.769 42.955 1.00 6.12 ATOM 1702 O TYR A 433 -1.576
30.319 43.866 1.00 7.36 ATOM 1703 N VAL A 434 -0.279 30.385 42.031
1.00 8.06 ATOM 1704 CA VAL A 434 -0.023 31.820 42.080 1.00 9.13
ATOM 1705 CB VAL A 434 1.026 32.189 41.035 1.00 9.24 ATOM 1706 CG1
VAL A 434 0.473 32.255 39.614 1.00 9.78 ATOM 1707 CG2 VAL A 434
1.656 33.531 41.432 1.00 9.67 ATOM 1708 C VAL A 434 -1.247 32.724
41.971 1.00 9.44 ATOM 1709 O VAL A 434 -1.224 33.807 42.521 1.00
8.77 ATOM 1710 N MET A 435 -2.348 32.242 41.373 1.00 9.61 ATOM 1711
CA MET A 435 -3.580 33.044 41.377 1.00 10.35 ATOM 1712 CB MET A 435
-4.313 32.866 40.054 1.00 9.92 ATOM 1713 CG MET A 435 -3.514 33.300
38.848 1.00 10.91 ATOM 1714 SD MET A 435 -2.789 34.945 38.866 1.00
11.06 ATOM 1715 CE MET A 435 -4.229 35.954 39.266 1.00 11.66 ATOM
1716 C MET A 435 -4.479 32.835 42.569 1.00 11.89 ATOM 1717 O MET A
435 -5.625 33.382 42.578 1.00 10.19 ATOM 1718 N TYR A 436 -4.004
32.200 43.636 1.00 9.70 ATOM 1719 CA TYR A 436 -4.776 32.133 44.873
1.00 10.59 ATOM 1720 CB TYR A 436 -4.154 31.274 45.976 1.00 12.19
ATOM 1721 CG TYR A 436 -5.187 30.700 46.929 1.00 12.41 ATOM 1722
CD1 TYR A 436 -6.086 29.710 46.528 1.00 13.54 ATOM 1723 CE1 TYR A
436 -7.026 29.190 47.405 1.00 10.84 ATOM 1724 CD2 TYR A 436 -5.275
31.200 48.218 1.00 12.72 ATOM 1725 CE2 TYR A 436 -6.216 30.710
49.104 1.00 15.95 ATOM 1726 CZ TYR A 436 -7.105 29.725 48.659 1.00
14.43 ATOM 1727 OH TYR A 436 -8.012 29.268 49.572 1.00 17.14 ATOM
1728 C TYR A 436 -4.977 33.571 45.304 1.00 10.80 ATOM 1729 O TYR A
436 -4.106 34.412 45.121 1.00 7.79 ATOM 1730 N PRO A 437 -6.178
33.938 45.778 1.00 11.93 ATOM 1731 CD PRO A 437 -7.340 33.037
45.939 1.00 11.26 ATOM 1732 CA PRO A 437 -6.467 35.326 46.139 1.00
10.62 ATOM 1733 CB PRO A 437 -7.962 35.319 46.495 1.00 12.25 ATOM
1734 CG PRO A 437 -8.320 33.884 46.714 1.00 12.44 ATOM 1735 C PRO A
437 -5.713 35.775 47.385 1.00 10.79 ATOM 1736 O PRO A 437 -5.577
36.995 47.602 1.00 11.50 ATOM 1737 N ILE A 438 -5.411 34.802 48.243
1.00 11.06 ATOM 1738 CA ILE A 438 -4.645 35.046 49.464 1.00 14.21
ATOM 1739 CB ILE A 438 -5.154 34.201 50.625 1.00 16.10 ATOM 1740
CG2 ILE A 438 -4.290 34.449 51.876 1.00 19.74 ATOM 1741 CG1 ILE A
438 -6.619 34.470 50.967 1.00 19.29 ATOM 1742 CD1 ILE A 438 -7.175
33.545 52.055 1.00 20.13 ATOM 1743 C ILE A 438 -3.173 34.778 49.119
1.00 14.66 ATOM 1744 O ILE A 438 -2.796 33.666 48.710 1.00 10.76
ATOM 1745 N ALA A 439 -2.342 35.803 49.202 1.00 12.90 ATOM 1746 CB
ALA A 439 -0.137 36.968 49.129 1.00 13.51 ATOM 1747 CA ALA A 439
-0.946 35.702 48.748 1.00 12.56 ATOM 1748 C ALA A 439 -0.219 34.450
49.179 1.00 11.82 ATOM 1749 O ALA A 439 -0.053 34.221 50.350 1.00
12.72 ATOM 1750 N VAL A 440 0.244 33.624 48.232 1.00 12.54 ATOM
1751 CA VAL A 440 1.038 32.460 48.610 1.00 13.18 ATOM 1752 CB VAL A
440 1.194 31.545 47.405 1.00 13.89 ATOM 1753 CG1 VAL A 440 -0.203
30.964 47.060 1.00 14.71 ATOM 1754 CG2 VAL A 440 1.748 32.335
46.225 1.00 14.29 ATOM 1755 C VAL A 440 2.422 32.915 49.105 1.00
13.24 ATOM 1756 O VAL A 440 2.951 33.937 48.665 1.00 10.97 ATOM
1757 N SER A 441 3.009 32.117 49.958 1.00 13.77 ATOM 1758 CA SER A
441 4.320 32.448 50.541 1.00 15.67 ATOM 1759 CB SER A 441 4.602
31.531 51.741 1.00 14.54 ATOM 1760 OG SER A 441 4.895 30.230 51.251
1.00 12.13 ATOM 1761 C SER A 441 5.443 32.298 49.543 1.00 14.86
ATOM 1762 O SER A 441 6.393 33.042 49.627 1.00 16.15 ATOM 1763 N
GLY A 442 5.365 31.340 48.626 1.00 15.69 ATOM 1764 CA GLY A 442
6.458 31.086 47.680 1.00 15.78 ATOM 1765 C GLY A 442 7.329 29.955
48.292 1.00 17.87 ATOM 1766 O GLY A 442 8.313 29.533 47.700 1.00
17.88 ATOM 1767 N ASP A 443 6.933 29.420 49.433 1.00 16.97 ATOM
1768 CA ASP A 443 7.639 28.297 50.053 1.00 19.48 ATOM 1769 CB ASP A
443 7.315 28.159 51.522 1.00 20.17 ATOM 1770 CG ASP A 443 7.612
29.371 52.357 1.00 23.62 ATOM 1771 OD1 ASP A 443 8.296 30.306
51.896 1.00 23.81 ATOM 1772 OD2 ASP A 443 7.119 29.399 53.505 1.00
25.54 ATOM 1773 C ASP A 443
7.308 26.945 49.435 1.00 21.78 ATOM 1774 O ASP A 443 8.052 25.962
49.619 1.00 20.22 ATOM 1775 N HIS A 444 6.163 26.878 48.737 1.00
18.42 ATOM 1776 CA HIS A 444 5.734 25.593 48.195 1.00 18.25 ATOM
1777 CB HIS A 444 4.263 25.329 48.628 1.00 19.33 ATOM 1778 CG HIS A
444 4.010 25.631 50.066 1.00 21.40 ATOM 1779 CD2 HIS A 444 3.475
26.703 50.697 1.00 21.56 ATOM 1780 ND1 HIS A 444 4.407 24.766
51.066 1.00 22.82 ATOM 1781 CE1 HIS A 444 4.098 25.274 52.242 1.00
22.65 ATOM 1782 NE2 HIS A 444 3.540 26.457 52.047 1.00 23.53 ATOM
1783 C HIS A 444 5.869 25.478 46.699 1.00 18.20 ATOM 1784 O HIS A
444 5.869 26.430 45.924 1.00 16.89 ATOM 1785 N GLU A 445 5.967
24.246 46.228 1.00 18.90 ATOM 1786 OE2 GLU A 445 7.287 19.466
44.909 1.00 37.32 ATOM 1787 OE1 GLU A 445 5.382 19.436 43.805 1.00
36.75 ATOM 1788 CD GLU A 445 6.406 20.025 44.213 1.00 34.73 ATOM
1789 CG GLU A 445 6.551 21.509 43.912 1.00 31.48 ATOM 1790 CB GLU A
445 5.641 22.370 44.753 1.00 24.36 ATOM 1791 CA GLU A 445 6.094
23.856 44.844 1.00 21.30 ATOM 1792 C GLU A 445 5.192 24.573 43.856
1.00 19.12 ATOM 1793 O GLU A 445 5.654 25.106 42.852 1.00 18.43
ATOM 1794 N ASN A 446 3.886 24.568 44.155 1.00 14.02 ATOM 1795 CA
ASN A 446 2.931 25.161 43.217 1.00 12.89 ATOM 1796 CB ASN A 446
1.558 24.575 43.616 1.00 13.52 ATOM 1797 CG ASN A 446 1.492 23.081
43.416 1.00 11.57 ATOM 1798 OD1 ASN A 446 2.288 22.530 42.688 1.00
14.69 ATOM 1799 ND2 ASN A 446 0.580 22.359 44.016 1.00 14.47 ATOM
1800 C ASN A 446 2.880 26.669 43.236 1.00 12.30 ATOM 1801 O ASN A
446 2.299 27.295 42.349 1.00 11.18 ATOM 1802 N ASN A 447 3.432
27.299 44.274 1.00 12.22 ATOM 1803 CA ASN A 447 3.261 28.692 44.560
1.00 12.81 ATOM 1804 CB ASN A 447 3.943 29.244 45.794 1.00 13.02
ATOM 1805 CG ASN A 447 3.432 28.721 47.094 1.00 14.61 ATOM 1806 OD1
ASN A 447 4.151 28.922 48.081 1.00 15.02 ATOM 1807 ND2 ASN A 447
2.276 28.070 47.103 1.00 13.05 ATOM 1808 C ASN A 447 3.662 29.596
43.403 1.00 14.50 ATOM 1809 O ASN A 447 2.990 30.597 43.173 1.00
14.90 ATOM 1810 N LYS A 448 4.698 29.182 42.678 1.00 15.41 ATOM
1811 CA LYS A 448 5.123 30.032 41.582 1.00 19.44 ATOM 1812 CB LYS A
448 6.678 30.007 41.496 1.00 20.29 ATOM 1813 CG LYS A 448 7.319
30.599 42.725 1.00 22.70 ATOM 1814 CD LYS A 448 8.503 31.527 42.408
1.00 25.96 ATOM 1815 CE LYS A 448 8.285 32.968 42.835 1.00 24.34
ATOM 1816 NZ LYS A 448 8.963 33.438 44.090 1.00 16.70 ATOM 1817 C
LYS A 448 4.568 29.566 40.271 1.00 19.62 ATOM 1818 O LYS A 448
5.136 29.990 39.263 1.00 25.92 ATOM 1819 N MET A 449 3.632 28.647
40.230 1.00 13.54 ATOM 1820 CA MET A 449 3.122 28.149 38.956 1.00
14.78 ATOM 1821 CB MET A 449 3.181 26.610 39.000 1.00 17.23 ATOM
1822 CG MET A 449 4.638 26.182 39.231 1.00 23.71 ATOM 1823 SD MET A
449 5.557 26.266 37.701 1.00 35.18 ATOM 1824 CE MET A 449 6.512
27.753 37.876 1.00 29.45 ATOM 1825 C MET A 449 1.660 28.558 38.779
1.00 10.74 ATOM 1826 O MET A 449 1.051 28.898 39.776 1.00 7.03 ATOM
1827 N PHE A 450 1.104 28.330 37.629 1.00 9.72 ATOM 1828 CA PHE A
450 -0.319 28.583 37.393 1.00 11.38 ATOM 1829 CB PHE A 450 -0.503
29.170 35.975 1.00 11.13 ATOM 1830 CG PHE A 450 -0.071 30.631
35.959 1.00 12.52 ATOM 1831 CD1 PHE A 450 -0.930 31.655 36.320 1.00
12.19 ATOM 1832 CD2 PHE A 450 1.207 30.955 35.549 1.00 10.84 ATOM
1833 CE1 PHE A 450 -0.508 32.973 36.265 1.00 11.08 ATOM 1834 CE2
PHE A 450 1.624 32.264 35.490 1.00 10.58 ATOM 1835 CZ PHE A 450
0.770 33.283 35.856 1.00 10.00 ATOM 1836 C PHE A 450 -1.150 27.312
37.490 1.00 11.07 ATOM 1837 O PHE A 450 -0.799 26.260 36.954 1.00
9.07 ATOM 1838 N SER A 451 -2.323 27.410 38.138 1.00 11.36 ATOM
1839 CA SER A 451 -3.268 26.315 38.270 1.00 10.14 ATOM 1840 CB SER
A 451 -4.487 26.735 39.155 1.00 10.95 ATOM 1841 OG SER A 451 -5.179
27.782 38.420 1.00 9.02 ATOM 1842 C SER A 451 -3.844 25.958 36.901
1.00 9.35 ATOM 1843 O SER A 451 -3.770 26.735 35.963 1.00 9.40 ATOM
1844 N GLN A 452 -4.577 24.855 36.824 1.00 12.16 ATOM 1845 CA GLN A
452 -5.351 24.521 35.618 1.00 13.47 ATOM 1846 CB GLN A 452 -5.964
23.125 35.803 1.00 16.94 ATOM 1847 CG GLN A 452 -6.700 22.595
34.601 1.00 21.50 ATOM 1848 CD GLN A 452 -5.827 22.669 33.361 1.00
23.04 ATOM 1849 OE1 GLN A 452 -4.673 22.244 33.400 1.00 21.47 ATOM
1850 NE2 GLN A 452 -6.377 23.285 32.307 1.00 23.25 ATOM 1851 C GLN
A 452 -6.430 25.554 35.338 1.00 13.65 ATOM 1852 O GLN A 452 -6.638
25.907 34.165 1.00 14.17 ATOM 1853 N CYS A 453 -7.045 26.104 36.363
1.00 14.02 ATOM 1854 CA CYS A 453 -8.092 27.127 36.197 1.00 17.31
ATOM 1855 CB CYS A 453 -8.719 27.417 37.560 1.00 21.49 ATOM 1856 SG
CYS A 453 -9.533 26.093 38.447 1.00 29.58 ATOM 1857 C CYS A 453
-7.500 28.361 35.535 1.00 15.91 ATOM 1858 O CYS A 453 -7.993 28.886
34.524 1.00 12.98 ATOM 1859 N SER A 454 -6.301 28.786 36.002 1.00
10.78 ATOM 1860 CA SER A 454 -5.584 29.858 35.390 1.00 9.42 ATOM
1861 CB SER A 454 -4.341 30.410 36.146 1.00 8.01 ATOM 1862 OG SER A
454 -4.692 30.709 37.471 1.00 8.77 ATOM 1863 C SER A 454 -5.186
29.538 33.979 1.00 9.67 ATOM 1864 O SER A 454 -5.237 30.413 33.113
1.00 9.80 ATOM 1865 N LYS A 455 -4.691 28.304 33.763 1.00 11.07
ATOM 1866 CA LYS A 455 -4.240 27.987 32.421 1.00 12.98 ATOM 1867 CB
LYS A 455 -3.599 26.587 32.376 1.00 13.14 ATOM 1868 CG LYS A 455
-2.146 26.611 32.855 1.00 17.01 ATOM 1869 CD LYS A 455 -1.656
25.166 32.882 1.00 19.15 ATOM 1870 CE LYS A 455 -0.242 25.104
33.451 1.00 20.39 ATOM 1871 NZ LYS A 455 0.195 23.673 33.295 1.00
23.90 ATOM 1872 C LYS A 455 -5.367 28.049 31.401 1.00 12.31 ATOM
1873 O LYS A 455 -5.180 28.539 30.315 1.00 12.74 ATOM 1874 N GLN A
456 -6.529 27.527 31.755 1.00 14.47 ATOM 1875 NE2 GLN A 456 -10.676
25.407 33.099 1.00 30.34 ATOM 1876 OE1 GLN A 456 -9.503 23.500
32.899 1.00 29.79 ATOM 1877 CD GLN A 456 -9.664 24.686 32.606 1.00
28.31 ATOM 1878 CG GLN A 456 -8.670 25.342 31.686 1.00 25.00 ATOM
1879 CB GLN A 456 -8.814 26.832 31.511 1.00 19.95 ATOM 1880 CA GLN
A 456 -7.661 27.516 30.804 1.00 16.41 ATOM 1881 C GLN A 456 -7.967
28.942 30.401 1.00 15.24 ATOM 1882 O GLN A 456 -7.936 29.330 29.244
1.00 15.99 ATOM 1883 N SER A 457 -8.038 29.821 31.393 1.00 14.14
ATOM 1884 CA SER A 457 -8.377 31.231 31.194 1.00 14.81 ATOM 1885 CB
SER A 457 -8.518 31.976 32.530 1.00 15.12 ATOM 1886 OG SER A 457
-9.565 31.464 33.315 1.00 16.45 ATOM 1887 C SER A 457 -7.354 31.967
30.377 1.00 12.94 ATOM 1888 O SER A 457 -7.697 32.696 29.467 1.00
12.19 ATOM 1889 N ILE A 458 -6.051 31.825 30.722 1.00 12.21 ATOM
1890 CA ILE A 458 -5.038 32.567 29.982 1.00 9.86 ATOM 1891 CB ILE A
458 -3.699 32.500 30.795 1.00 11.36 ATOM 1892 CG2 ILE A 458 -2.555
32.898 29.890 1.00 6.36 ATOM 1893 CG1 ILE A 458 -3.853 33.384
32.036 1.00 9.88 ATOM 1894 CD1 ILE A 458 -2.730 33.281 33.059 1.00
11.89 ATOM 1895 C ILE A 458 -4.854 32.048 28.573 1.00 9.53 ATOM
1896 O ILE A 458 -4.640 32.816 27.634 1.00 9.61 ATOM 1897 N TYR A
459 -4.939 30.731 28.423 1.00 9.48 ATOM 1898 CA TYR A 459 -4.767
30.100 27.112 1.00 11.60 ATOM 1899 CB TYR A 459 -5.038 28.600
27.250 1.00 11.22 ATOM 1900 CG TYR A 459 -5.045 27.843 25.943 1.00
16.31 ATOM 1901 CD1 TYR A 459 -4.013 27.976 25.041 1.00 16.99 ATOM
1902 CE1 TYR A 459 -4.028 27.303 23.827 1.00 19.40 ATOM 1903 CD2
TYR A 459 -6.103 26.999 25.626 1.00 18.93 ATOM 1904 CE2 TYR A 459
-6.108 26.299 24.422 1.00 21.51 ATOM 1905 CZ TYR A 459 -5.075
26.458 23.532 1.00 20.71 ATOM 1906 OH TYR A 459 -5.095 25.772
22.341 1.00 21.74 ATOM 1907 C TYR A 459 -5.720 30.750 26.110 1.00
12.72 ATOM 1908 O TYR A 459 -5.334 31.252 25.073 1.00 14.82 ATOM
1909 N LYS A 460 -6.974 30.779 26.528 1.00 14.14 ATOM 1910 NZ LYS A
460 -10.468 26.961 25.743 1.00 30.41 ATOM 1911 CE LYS A 460 -9.733
27.718 24.702 1.00 27.82 ATOM 1912 CD LYS A 460 -9.809 29.219
24.921 1.00 26.35 ATOM 1913 CG LYS A 460 -9.794 29.689 26.345 1.00
23.26 ATOM 1914 CB LYS A 460 -9.375 31.156 26.477 1.00 19.80 ATOM
1915 CA LYS A 460 -8.043 31.371 25.721 1.00 18.07 ATOM 1916 C LYS A
460 -7.739 32.806 25.384 1.00 16.65 ATOM 1917 O LYS A 460 -7.735
33.206 24.228 1.00 16.84 ATOM 1918 N THR A 461 -7.225 33.546 26.348
1.00 14.44 ATOM 1919 CA THR A 461 -6.847 34.932 26.129 1.00 14.79
ATOM 1920 CB THR A 461 -6.597 35.590 27.509 1.00 13.79 ATOM 1921
OG1 THR A 461 -7.813 35.528 28.254 1.00 13.87 ATOM 1922 CG2 THR A
461 -6.192 37.054 27.345 1.00 16.01 ATOM 1923 C THR A 461 -5.709
35.061 25.159 1.00 14.88 ATOM 1924 O THR A 461 -5.830 35.765 24.142
1.00 14.86 ATOM 1925 N ILE A 462 -4.569 34.417 25.419 1.00 14.52
ATOM 1926 CA ILE A 462 -3.422 34.528 24.542 1.00 17.34 ATOM 1927 CB
ILE A 462 -2.223 33.642 24.975 1.00 18.43 ATOM 1928 CG2 ILE A 462
-1.130 33.699 23.904 1.00 20.28 ATOM 1929 CG1 ILE A 462 -1.643
34.018 26.316 1.00 18.70 ATOM 1930 CD1 ILE A 462 -0.726 32.967
26.946 1.00 19.10 ATOM 1931 C ILE A 462 -3.757 34.102 23.124 1.00
18.59 ATOM 1932 O ILE A 462 -3.420 34.813 22.189 1.00 17.11 ATOM
1933 N GLU A 463 -4.473 32.979 22.968 1.00 22.82 ATOM 1934 OE2 GLU
A 463 -8.026 29.422 22.532 1.00 41.03 ATOM 1935 OE1 GLU A 463
-7.006 28.787 20.674 1.00 39.73 ATOM 1936 CD GLU A 463 -7.293
29.635 21.536 1.00 39.22 ATOM 1937 CG GLU A 463 -6.723 31.032
21.405 1.00 36.15 ATOM 1938 CB GLU A 463 -5.228 31.095 21.624 1.00
30.56 ATOM 1939 CA GLU A 463 -4.692 32.520 21.587 1.00 26.30 ATOM
1940 C GLU A 463 -5.492 33.530 20.800 1.00 27.72 ATOM 1941 O GLU A
463 -5.233 33.704 19.607 1.00 29.03 ATOM 1942 N SER A 464 -6.399
34.283 21.418 1.00 26.79 ATOM 1943 CA SER A 464 -7.190 35.312
20.764 1.00 26.80 ATOM 1944 CB SER A 464 -8.395 35.611 21.666 1.00
25.15 ATOM 1945 OG SER A 464 -9.060 36.806 21.404 1.00 25.60 ATOM
1946 C SER A 464 -6.434 36.623 20.582 1.00 27.79 ATOM 1947 O SER A
464 -6.686 37.340 19.612 1.00 27.52 ATOM 1948 N LYS A 465 -5.598
36.959 21.560 1.00 25.83 ATOM 1949 CA LYS A 465 -4.956 38.267
21.585 1.00 28.04 ATOM 1950 CB LYS A 465 -5.065 38.846 22.993 1.00
28.65 ATOM 1951 CG LYS A 465 -6.217 39.766 23.316 1.00 30.43 ATOM
1952 CD LYS A 465 -7.570 39.136 23.099 1.00 31.72 ATOM 1953 CE LYS
A 465 -8.699 40.165 23.068 1.00 32.17 ATOM 1954 NZ LYS A 465 -8.593
41.165 24.149 1.00 32.20 ATOM 1955 C LYS A 465 -3.505 38.293 21.145
1.00 27.28 ATOM 1956 O LYS A 465 -3.053 39.353 20.704 1.00 27.45
ATOM 1957 N ALA A 466 -2.767 37.193 21.184 1.00 28.49 ATOM 1958 CA
ALA A 466 -1.373 37.191 20.753 1.00 28.95 ATOM 1959 CB ALA A 466
-0.788 35.788 20.750 1.00 27.44 ATOM 1960 C ALA A 466 -1.152 37.821
19.385 1.00 30.12 ATOM 1961 O ALA A 466 -0.295 38.708 19.256 1.00
28.37 ATOM 1962 N GLN A 467 -1.922 37.440 18.373 1.00 30.63 ATOM
1963 NE2 GLN A 467 0.360 36.582 14.061 1.00 41.77 ATOM 1964 OE1 GLN
A 467 -0.196 35.394 15.846 1.00 42.22 ATOM 1965 CD GLN A 467 -0.524
36.191 14.967 1.00 40.84 ATOM 1966 CG GLN A 467 -1.903 36.768
14.813 1.00 40.14 ATOM 1967 CB GLN A 467 -2.609 37.306 16.027 1.00
36.55 ATOM 1968 CA GLN A 467 -1.712 38.007 17.038 1.00 33.94 ATOM
1969 C GLN A 467 -1.891 39.519 16.996 1.00 33.19 ATOM 1970 O GLN A
467 -1.208 40.231 16.271 1.00 33.32 ATOM 1971 N GLU A 468 -2.797
40.063 17.768 1.00 33.44 ATOM 1972 OE2 GLU A 468 -7.657 40.802
18.969 1.00 44.61 ATOM 1973 OE1 GLU A 468 -6.465 41.900 20.497 1.00
44.24 ATOM 1974 CD GLU A 468 -6.664 41.480 19.332 1.00 44.10 ATOM
1975 CG GLU A 468 -5.602 41.832 18.305 1.00 41.64 ATOM 1976 CB GLU
A 468 -4.238 41.582 18.934 1.00 38.28 ATOM 1977 CA GLU A 468 -3.092
41.457 17.917 1.00 35.15 ATOM 1978 C GLU A 468 -1.956 42.299 18.485
1.00 33.88 ATOM 1979 O GLU A 468 -1.706 43.397 17.980 1.00 32.88
ATOM 1980 N CYS A 469 -1.382 41.873 19.618 1.00 32.54 ATOM 1981 CA
CYS A 469 -0.374 42.738 20.230 1.00 32.43 ATOM 1982 CB CYS A 469
-1.019 43.591 21.320 1.00 34.38 ATOM 1983 SG CYS A 469 -1.294
42.874 22.918 1.00 36.46 ATOM 1984 C CYS A 469 0.888 42.055 20.713
1.00 29.85 ATOM 1985 O CYS A 469 1.786 42.811 21.105 1.00 30.62
ATOM 1986 N PHE A 470 1.027 40.739 20.624 1.00 26.22 ATOM 1987 CA
PHE A 470 2.314 40.147 21.073 1.00 25.07 ATOM 1988 CB PHE A 470
2.163 38.694 21.471 1.00 21.88 ATOM 1989 CG PHE A 470 1.492 38.480
22.808 1.00 21.60 ATOM 1990 CD1 PHE A 470 0.810 39.497 23.450 1.00
19.78 ATOM 1991 CD2 PHE A 470 1.556 37.251 23.414 1.00 19.41 ATOM
1992 CE1 PHE A 470 0.182 39.309 24.658 1.00 20.26 ATOM 1993 CE2 PHE
A 470 0.932 37.046 24.634 1.00 20.13 ATOM 1994 CZ PHE A 470 0.257
38.065 25.257 1.00 20.62 ATOM 1995 C PHE A 470 3.351 40.383 19.999
1.00 24.85 ATOM 1996 O PHE A 470 2.994 40.592 18.836 1.00 23.37
ATOM 1997 N GLN A 471 4.610 40.404 20.389 1.00 23.76 ATOM 1998 CA
GLN A 471 5.714 40.672 19.491 1.00 25.96 ATOM 1999 CB GLN A 471
6.350 42.028 19.794 1.00 27.26 ATOM 2000 CG GLN A 471 5.453 43.201
19.442 1.00 31.81 ATOM 2001 CD GLN A 471 6.014 44.575 19.678 1.00
33.37 ATOM 2002 OE1 GLN A 471 5.258 45.547 19.668 1.00 35.13 ATOM
2003 NE2 GLN A 471 7.323 44.723 19.885 1.00 35.41 ATOM 2004 C GLN A
471 6.772 39.578 19.615 1.00 25.85 ATOM 2005 O GLN A 471 6.668
38.670 20.433 1.00 23.32 ATOM 2006 N GLU A 472 7.771 39.683 18.760
1.00 27.21 ATOM 2007 OE2 GLU A 472 8.898 41.817 17.307 1.00 43.27
ATOM 2008 OE1 GLU A 472 9.685 41.624 15.282 1.00 44.72 ATOM 2009 CD
GLU A 472 9.267 41.115 16.344 1.00 42.17 ATOM 2010 CG GLU A 472
9.207 39.605 16.387 1.00 39.41 ATOM 2011 CB GLU A 472 9.836 38.975
17.600 1.00 34.14 ATOM 2012 CA GLU A 472 8.867 38.730 18.759 1.00
30.87 ATOM 2013 C GLU A 472 9.656 38.993 20.037 1.00 30.26 ATOM
2014 O GLU A 472 9.773 40.161 20.396 1.00 29.23 ATOM 2015 N ARG A
473 10.256 37.970 20.584 1.00 31.97 ATOM 2016 CA ARG A 473 11.073
38.106 21.778 1.00 34.21 ATOM 2017 CB ARG A 473 11.230 36.691
22.340 1.00 34.97 ATOM 2018 CG ARG A 473 12.230 36.454 23.443 1.00
37.52 ATOM 2019 CD ARG A 473 11.661 35.449 24.436 1.00 39.68 ATOM
2020 NE ARG A 473 12.564 35.155 25.532 1.00 41.06 ATOM 2021 CZ ARG
A 473 12.469 35.726 26.729 1.00 42.18 ATOM 2022 NH1 ARG A 473
13.327 35.423 27.693 1.00 42.83 ATOM 2023 NH2 ARG A 473 11.559
36.660 26.960 1.00 42.33 ATOM 2024 C ARG A 473 12.451 38.665 21.456
1.00 36.39 ATOM 2025 O ARG A 473 13.092 38.294 20.489 1.00 36.44
ATOM 2026 N SER A 474 12.918 39.567 22.276 1.00 37.92 ATOM 2027 OG
SER A 474 15.308 38.413 23.524 1.00 36.38 ATOM 2028 CB SER A 474
15.293 39.122 22.290 1.00 37.61 ATOM 2029 CA SER A 474 14.200
40.202 22.359 1.00 37.96 ATOM 2030 C SER A 474 14.436 41.341 21.403
1.00 39.67 ATOM 2031 O SER A 474 13.589 42.259 21.336 1.00 40.16
END
EXAMPLE 5
TACE Inhubitor Design
[0119] The TACE x-ray diffraction coordinates were read into a
Sybyl v.6.3 (Tripos Associates) software package and the x-ray
structure analyzed graphically. The regions within the original
x-ray coordinates were corrected for chirality and atom type. The
modified x-ray model of TACE was energy minimized until all the
TACE structural parameters were at their equilibrium or optimal
values. The energy minimized structure was then compared to the
original structure to confirm the absence of anomalies.
[0120] Sites of specific interaction(s) between TACE and the
co-crystallized inhibitor were identified. The inhibitor was then
removed from the X-ray complex model, leaving only the TACE
structural model.
[0121] Candidate inhibitors were chosen based upon the sites of
interaction with TACE and the candidate inhibitor in light of the
sites of interaction identified previously for the co-crystallized
inhibitor. Once specific candidate inhibitor-TACE interactions were
determined, docking studies were performed to provide preliminary
"modeled" complexes of selected candidate inhibitors with TACE.
[0122] Constrained conformational analysis was performed using
molecular dynamics (MD) to check the integrity of the modeled
TACE-inhibitor complex. Once the complex reached its most favorable
conformational state, the structure as proposed by the MD study was
analyzed visually to insure that the modeled complex complied with
known experimental SAR/QSAR based on measured binding
affinities.
[0123] The modeled candidate inhibitor-TACE complex was analyzed.
The region of the complex associated with the S1' regions of TACE
containing a small solvent exposed channel was chosen as a target
region for modification. A single modification, a benzyl group
which becomes embedded within the target region, was selected based
upon computational and synthetic chemical principles. The benzyl
group was oriented on an appropriate zinc chelator core so as to be
projected into the S1' S3' pocket. This modification converts an
inhibitor which was generally MMP selective to one which is TACE
selective. IC50 data for the inhibitor with a benzyl modification
confirm this selectivity.
[0124] Structure-based analoging for optimization of inhibitor
potency, selectivity and physical drug-like properties was
performed in an iterative manner.
EXAMPLE 6
Measuring TACE Inhibition
[0125] 250 .mu.M peptide substrate (Ac-SPLAQAVRSSSR-NH.sub.2) was
incubate with 3.7 U/.mu.L TACE in a buffer containing 10 mM TRIS
HCl, pH 7.4, 10% glycerol at 25 degrees C. The reaction was
quenched with 1% TFA (final concentration) after two hours. The
reaction mixture was separated by HPLC on a Hewlett-Packard 1150.
The product formation was monitored by absorbance at 220 nm.
[0126] The linearity of the reaction was confirmed
(r.sup.2>0.85). The mean (x.+-.sem) of the control rate was
calculated and compared for statistical significance (p<0.05)
with drug-tested rates using Dunnett's multiple comparison test.
Dose-response relationships were generated using multiple doses of
drug and IC.sub.50 values with 95% CI were estimated using linear
regression.
[0127] From the foregoing description and examples, one skilled in
the art can ascertain the essential characteristics of the
invention and, without departing from the spirit and scope of the
invention, can make changes, modifications, and variations of the
invention to adapt it to various uses and conditions. Additionally,
the disclosure of all publications and patent applications cited
above, including U.S. provisional patent application serial No.
60/073,709; U.S. patent application Ser. No. 09/050,083; and US
provisional patent application titled "Crystalline
TNF-.alpha.-Converdng Enzyme and Uses Thereof," filed Jan. 27,
1999, are expressly incorporated herein by reference in their
entireties to the same extent as if each were incorporated by
reference individually.
* * * * *