`Printed in Great Britain,
`
`0161-5890/91 $3.00 + 000
`Pergamon Press plc
`
`A POSSIBLE PROCEDURE FOR
`
`REDUCING THE IMMUNOGENICITY OF ANTIBODY
`
`VARIABLE DOMAINS WHILE PRESERVING THEIR
`
`LIGAND-BINDING PROPERTIES
`
`Laboratory of Molecular Biology, National Institute of Diabetes and Digestive and Kidney Diseases,
`National Institutes of Health, Bethesda, MD 20892, U.S.A.
`
`EDUARDO A. PADLAN
`
`(First received 24 May 1990; accepted in revised form 17 September 1990)
`
`Abstract—It is proposed to reduce the immunogenicity of allogeneic antibody variable domains, while
`preserving ligand-binding properties, by reducing their antigenicity through replacement of the exposed
`residues in the framework regions which differ from those usually found in host antibodies. The results
`of a comparison of representative murine antibody sequences with those of human origin suggest that the
`number of residues that need to be replaced to “humanize” those antibodies could be small.
`
`489
`
`regardless of species of origin, amino acid sequence,
`or ligand—binding specificity. This is generally taken
`as evidence that the ligand-binding characteristics of
`an antibody combining site are determined primarily
`by the structure and relative disposition of the CDRs,
`although some neighboring framework residues also
`have been found to be involved in antigen binding
`(Amit et al., 1986; Colman et al., 1987; Sheriff et (11.,
`1987; Padlan et al., 1989). Thus, if the fine specificity
`of an antibody is to be preserved, its CDR structures
`(and probably also some of the neighboring residues),
`their interaction with each other, and their interaction
`with the rest of the variable domains must be strictly
`maintained. This may require the retention of most,
`if not all, of the many interior and interdomain
`contact residues; the structural effects of replacing
`only a few of them cannot be predicted.
`It may be possible to simultaneously reduce
`immunogenicity and strictly preserve ligand-binding
`properties. Since the antigenicity of a protein is de-
`pendent on the nature of its surface (Benjamin er £21.,
`1984), replacing the exposed residues of an allogeneic
`antibody with those usually found in host antibodies
`would decrease the probability of its recognition and
`internalization by antigen-processing cells, specifically
`by antigen-specific B-cells which play a critical role in
`priming T-cells (Ron et all, 1983; Kurtdones et at,
`1988; see also the review by Vitetta er al., 1989). Such
`
`variable domains and a proper interdomain contact,
`were kept in the “humanized” molecule in an attempt
`to preserve ligand-binding properties. The “human-
`ized” molecules did exhibit the same ligand—binding
`specificity as the original antibodies, but with only
`about one-third of the affinity.
`X-ray crystallographic studies have repeatedly
`demonstrated that the framework structures of the Fvs
`of dilTerent antibodies assume a “canonica ” structure
`
`INTRODUCTION
`
`Antibodies of predetermined specificity have many
`potential uses in therapy and diagnosis and hybrid-
`oma technology (Koehler and Milstein, 1975) has
`made possible the generation of virtually limitless
`amounts of such antibodies. Unfortunately, hybrid-
`oma proteins are more easily obtained from rodent
`sources and the use of those antibodies in human
`
`subjects will be hindered by the patients’ immune
`system. Various attempts have been made to make
`such allogeneic antibodies less immunogenic while
`preserving their antigenvbinding properties.
`The immunogenicity of rodent-derived antibodies
`had been reduced by generating chimeric antibodies
`in which the variable domains of the rodent proteins
`were grafted onto constant regions of human mol-
`ecules (reviewed by Morrison and Oi, 1988). Such
`chimeric molecules would be expected to possess all
`the antigen-binding characteristics of the original
`antibodies, but would retain the immunogenicity of
`the rodent variable domains.
`
`reduction in the immunogenicity of
`Further
`chimeric antibodies had been achieved by Winter
`and coworkers (Jones et al., 1986; Verhoeyen et (21.,
`1988; Riechmann et al., 1988) by grafting the comple-
`mentarity-determining regions, or CDRs, of the allo-
`geneic proteins onto human framework regions. In
`two published applications of this method (Riechmann
`et al., 1988; Queen at at, 1989), some of the frame.
`work residues of the allogeneic antibody, which were
`thought would ensure a pr0per structure for the
`
`
`Abbreviations: CDR, complementarity-determining region;
`Fab, the antigen-binding fragment of an antibody; Fv,
`the module containing VH and VL, the variable domains
`of the heavy and light chains, respectively.
`
`1 of 10
`
`Celltrion, Inc., Exhibit 1118
`
`
`
`490
`
`E. A. PADLAN
`
`a procedure, in the end, then serves to reduce immuno-
`genicity. The judicious replacement of exterior residues
`should have little, or no, effect on the interior of
`the domains, or on the interdomain contacts. If the
`envisioned replacements are confined to the frame-
`work regions, especially to only the exposed residues,
`then the ligand-binding properties of the antibody
`could be expected to remain unaltered.
`This possibility is explored in this paper using as
`an example the “humanization” of mouse variable
`domains. First, the available sequence data for the
`various human variable domain subgroups are
`analysed in terms of the residues which usually occur
`at each framework position. Then the differences
`between human and representative mouse sequences
`are assessed and correlated with solvent accessibility
`as deduced from the known three-dimensional struc-
`ture of human and mouse Fabs. It is found that the
`
`number of exposed residues, which need to be replaced
`to “humanize” mouse variable domains, could be
`small.
`
`MATERIALS AND METHODS
`
`obtained from the compilation of Kabat et a1. (1987).
`Only those sequences that are completely determined
`in the framework regions, 1—23, 35—49, and 57~88 in
`the light chains, and in the framework regions, 1—30,
`36—49, and 66—94, in the heavy chains (numbering of
`Kabat et al., 1987), were included in the survey. For
`the fourth framework region, 98—107 in the light chain
`and 103—1 13 in the heavy chain, only the residues that
`could be derived from the known J-minigene segments
`(Kabat et al., 1987) were considered. Here, the num-
`bering scheme of Kabat et al. (1987) is used through-
`out and pyrollidone carboxylic acids are counted as
`glutamines.
`The human kappa-chain sequences included in the
`survey were: ROY, AU, REI, HAU, HKlOl’CL,
`SCW, WEA, HK137’CL, HK134’CL, DAUDI’CL,
`WALKER’CL, GAL(1), LAY, WES, Vb’CL,
`HKIOZ’CL, EU, DEN, AMYLOID BAN, MEV,
`Vd’CL, Va’CL, KUE, Ve’CL, V13’CL, V18A’CL,
`Vl9A’CL, Vl9B’CL, and V18C’CL in V-kappa
`subgroup I; NIM, CUM, GM603’CL, FR, and
`RP Ml-6410’CL in V-kappa subgroup II; TI, WOL,
`SIE, NG9’CL, NEU, GOT, PAY, SON, GAR’, PIE,
`FLO, GLO, CUR, IARC/BL41’CL, POM, REE, and
`K-EVlS’CL in V-kappa subgroup III; and VJI’CL,
`VKAPPAIV GERMLINE’CL, PB17IV’CL,
`and
`LEN in V-kappa subgroup IV.
`The human lambda-chain sequences included in the
`survey were: NEWM, HA, NIG-64, NEW, BLZ’CL,
`WAH, NIG-77, VOR, RHE, LOC, OKA, COX, and
`NIG-Sl
`in V-lambda subgroup I; NIG-84, MES,
`WH, NEI, WEIR, TOG, TRO, BOH, NIG-58, VIL,
`WIN, and 4A’CL in V-lambda subgroup II; HIL,
`LAP, GAR, and MOT in V-lambda subgroup III; SH
`in V—lambda subgroup IV (included for completeness
`although missing the residue at position 1); B0 and
`MCG in V-lambda subgroup V; AMYLOID-AR,
`SUT, THO, LBV’CL, and NIG-48 in V—lambda
`subgroup VI.
`The human heavy—chain sequences included in
`the survey were: EU, SIE, HG3’CL, WOL, ND’CL,
`and MOT in VH subgroup I; COR, DAW, OU,
`MCE’, CE—l’CL, HE, SUP-Tl VH-JA’CL, WAH,
`and HIGl’CL in VH subgroup II; TUR, LAMBDA-
`VH26/CL, POM, WAS, Hll’CL, TEI, BRO’IGM,
`LAY, GRA’, ZAP, JON, DOB, NIE, 333’CL,
`lHl’CL, lBll’CL, 126’CL, 112’CL, 115’CL, TRO,
`and KOL in VH subgroup III.
`
`in view of the possibility
`
`(a) Structural data
`
`The structures that were chosen for the analysis
`were those of the variable domains of the antibodies
`KOL and J539. The Fabs of KOL and J539 are those
`
`which have been elucidated to the highest resolution
`and have been subjected to the most extensive refine—
`ments so far. The crystal structure of the human KOL
`Fab had been determined at 1.9 Angstrom resolution
`(Marquart et al., 1980) and had been refined to an
`R-value of 1.89; the structure of the mouse 1539 Fab
`had been elucidated at 1.95 Angstrom resolution and
`had been refined to an R-value of 1.95 (T. N. Bhat,
`E. A. Padlan and D. R. Davies, in preparation). The
`atomic coordinates for the KOL and J539 structures
`are available from the Protein Data Bank (Bernstein
`et al., 1977) (File 2FB4 for KOL, File 2FBJ for J539).
`The solvent accessibility of the individual residues
`in the KOL and J539 VL and VH was assessed using
`program MS of Connolly (1983) and programs devel-
`oped by Sherifi" (Sheriff et al., 1985). A radius of 1.7
`Angstroms was assumed for the solvent probe; stan-
`dard van der Waals’ radii (Case and Karplus, 1979)
`were used. The fractional accessibility values for the
`sidechains (Sheriff et al., 1985; Shrake and Rupley,
`1973) were computed as described previously (Padlan,
`1990). Exposures were computed in the context of
`isolated Fvs.
`Residues in the interface between VL and VH in
`
`the KOL and J 539 Fabs were designated as being in
`contact if they have atoms that come within half an
`Angstrom of the sum of their van der Waals’ radii.
`
`(b) Sequence data
`
`A survey was made of the residues occurring at each
`position in the framework regions of human light and
`heavy chain variable domains. The sequences were
`
`(6) Comparison of representative mouse and human
`sequences
`
`The degree to which mouse variable domains
`would have to be mutated in order to “humanize”
`
`them, was determined by comparing the framework
`regions of representative sequences from the various
`mouse subgroups with the known human sequences
`(Kabat et al., 1987). The human sequence that should
`serve best for comparison purposes is that which is the
`most similar to the particular mouse sequence under
`consideration. However,
`
`2 of 10
`
`Celltrion, Inc., Exhibit 1118
`
`
`
`Reduction of the immunogenicity of antibody variable domains
`
`491
`
`thus, the slightest difference in structure could result
`in different exposure designations for homologous
`glycine residues. The exposure patterns of their VL
`domains (Tables 2 and 3) likewise are very similar
`with large differences only at positions 2, 13, 66, 99,
`and 101. The conformation of the amino-terminal
`
`that individual sequences may have errors, the most
`frequently occurring residue at each position in the
`various human variable domain subgroups was also
`used in the comparisons. In many instances, more
`than one residue is found at high frequency at a given
`position; in those cases, any amino acid that occurs
`at least one-fourth as frequently as the most com—
`monly occurring residue was considered a possible
`alternative in the comparisons. The human subgroup
`to which each mouse sequence most closely corre-
`sponds was determined in this manner; the human
`sequence to which the mouse sequence is most similar
`was also determined.
`
`The sequences that were chosen as representative
`of the various mouse variable domain subgroups were:
`8107 (Subgroup I), TEPC(CAL20)105 (Subgroup II),
`TEPCl ll (Subgroup III), SlO7B’CL (Subgroup IV),
`R167 CRI+ (Subgroup V), and TEPC191 (Sub-
`group VI) of the kappa chains; MOPC104E of the
`lambda chains; and 36-60 CRI- (Subgroup I(A)),
`MCIOl/CL (Subgroup I(B)), MOPC104E (Subgroup
`II(A)), Bl-8’CL (Subgroup II(B)), GS BB 2.2(AB1)
`Subgroup
`11(C)), TEPC15
`(Subgroup
`III(A)),
`XRPC44 (Subgroup III(B)), HPC76’CL (Subgroup
`III(C)), and 93G7 CRI +’CL (Subgroup V(A)) of the
`heavy chains (Kabat et al., 1987). These sequences are
`the first listed of the sequences in each subgroup in
`Kabat et al. (1987) that are completely determined in
`all four framework regions. None of the sequences in
`the VH subgroups V(A) and V(B) in the compilation
`of Kabat et a1.
`(1987) satisfies this criterion.
`In
`addition, MOPC21’CL was chosen to represent the
`miscellaneous mouse heavy chains which have not
`been assigned to any particular subgroup.
`
`RESULTS
`
`domains of KOL or .1539 (those with pB, mE, or Ex
`
`segments of the KOL and J 539 light chains are
`slightly different, resulting in the difference observed
`at position 2; at the other positions, one or both
`proteins again have glycine.
`The very close similarity of the exposure patterns
`for the variable domains of KOL and J539 points to
`the close correspondence of the tertiary structures
`and of the dispositions of individual residues in these
`homologous domains. This is particularly remarkable
`because (a) these antibodies are from different species,
`(b) their light chains are of different types (.1539 has
`a kappa light chain, while KOL has a lambda light
`chain), (c) half of their CDRs, specifically the first
`and third CDRs in VL and the third in VH, have very
`different lengths and backbone conformations, and
`(d) KOL and J 539 have only 44 identical residues out
`of the 79 corresponding positions in the VL frame-
`work and 60 out of 87 in VH. An even closer similarity
`in overall structure and in the exposure patterns might
`be expected for two molecules that are more similar in
`sequence than this pair.
`By and large, the determination of which human
`variable domain subgroup an allogeneic domain most
`closely corresponds to is straightforward. However,
`in some of the human subgroups presented in Tables 1
`to 3, many positions exhibit several possible alterna-
`tives. Of the 76 positions in the first three framework
`regions in VH, 18 in Subgroup I presented more than
`one possible alternative for the amino acid compari-
`sons, 30 in Subgroup II, and 21 in Subgroup 111. Of
`the 70 positions in the first three framework regions
`in the kappa light-chain variable domains, three in
`Subgroup I presented more than one possible alterna-
`tive, two in Subgroup 11, none in Subgroup III, and
`one in Subgroup IV; of the corresponding 69 positions
`in the lambda chains, 11 in Subgroup I presented more
`than one possible alternative, eight in Subgroup II,
`28 in Subgroup III, none in Subgroup IV since there
`is only one available sequence, eight in Subgroup V,
`and five in Subgroup VI. The existence of several
`possible alternatives in some of these subgroups,
`especially in the heavy chains, emphasizes the need for
`further subdivision; finer subgrouping would allow
`greater precision in the sequence correlations and
`greater ease in the identification of the residues which
`differ between human and non-human antibodies.
`
`The solvent accessibility of the framework residues
`in the variable domains of KOL and .1539 and the
`
`amino acids most frequently occurring at each position
`in the various subgroups are given in Table 1 for heavy
`chains, in Table 2 for lambda chains, and in Table 3
`for kappa light chains. The residues that contact the
`opposite domain in the VL—VH interface are indi—
`cated in these tables. The results of the comparison
`of the mouse and human sequences are presented in
`Table 4.
`
`DISCUSSION
`
`A close examination of the fractional accessibility
`values presented in Table 1
`reveals a very close
`similarity in the exposure patterns of the VH domains
`of KOL and J539. Only at positions 88 and 104 are
`the two patterns drastically different and at these
`positions one or both proteins have glycine. According
`to the convention used here (see footnote to Table l),
`glycine is designated as being completely exposed if
`its alpha-carbon is accessible to the solvent probe,
`otherwise it is designated as being completely buried;
`
`A procedure is proposed here for reducing the
`antigenicity of an allogeneic variable domain while
`preserving its ligand-binding properties. It is based on
`the replacement of the residues which differ from
`those of the host with the corresponding residues in
`the most similar host sequence. Only those residues
`which are at least partly exposed in the corresponding
`
`3 of 10
`
`Celltrion, Inc., Exhibit 1118
`
`
`
`492
`
`E.A.PADLAN
`
`Table l. Solvent exposure of sidechains of framework residues in KOL and .1539 Fvs and the residues which occur most frequently at these
`positions in the various human VH subgroups
`
`N00
`N\O
`
`Oemqombwu—
`NNNNNNNN—_.__.._._._._._.
`\iomawN—oxoooqoxmamm—
`
`esgateageesgs
`moooooouuuqxiqxluxiuoooc~kEJN—OKDOONONMAWNHOOOONO‘O
`
`43
`
`W>O<~<~1>~10 :1:
`l“V12ZO1“'-<l"“1259‘”UUm"""--l’1171'10<2?“l“C17<C)'UB>O><1<2‘"!'1'l'-l"1'IO‘/>?I>31>('3"’PWP"111600970<l"OC)¢23"Jl"'!<l“0<l'1'l"11>22m
`7:1>O<<F>HUmmW<7§mZOF<FmZW>ZUWW‘““KO—€171[‘0NC}'U>07°<2‘”'11Um0m>>0wr7<rm00170<rOQQMWFFW<W
`
`Fractional accessibility
`Residues in subgroup
`J539
`KOL
`-—
`11
`Position
`Residue
`Exposure
`Residue
`Exposure
`100 EX
`1.00 Ex
`0.23 11113
`0.37 mB
`0.82 Ex
`0.82 Ex
`0.10 Bu
`0.00 Bu
`0.87 Ex
`1.00 Ex
`0.09 Bu
`0.00 Bu
`0.94 Ex
`0.94 Ex
`1.00 Ex
`1.00 EX
`0.00 Bu
`0.00 Bu
`1.00 Ex
`1.00 Ex
`0.81 Ex
`0.90 Ex
`0.25 11113
`025 m3
`0.87 Ex
`0.71 mE
`0.64 mE
`0.59 p13
`1.00 Ex
`1.00 Ex
`1.00 Ex
`0.73 mE
`0.75 mE
`0.66 mE
`0.26 mB
`0.28 ml!
`0.66 ml?
`0.75 mE
`0.00 Bu
`0.00 Bu
`0.82 Ex
`071 m5
`0.00 Bu
`0.00 Bu
`100 EX
`1.00 Ex
`0.00 Bu
`0.00 Bu
`1.00 Ex
`0.87 Ex
`1.00 Ex
`1.00 Ex
`0.10 Bu
`0.10 Bu
`0.72 mE
`0.85 Ex
`0.00 Bu
`0.00 Bu
`0.83 Ex
`0.74 mE
`0.00 Bu
`0.00 Bu
`0.00 Bu
`0.00 Bu
`0.31 mB
`0.10 Bu
`0.28 mB
`0.15 Bu
`0.75 mE
`0.95 Ex
`0.73 mE
`0.90 Ex
`1.00 Ex
`1.00 Ex
`0.86 Ex
`0.86 Ex
`1.00 Ex
`1.00 Ex
`0.00 Bu
`0.00 Bu
`0.73 mE
`0.75 mE
`0.04 Bu
`0.10 Bu
`0.00 Bu
`0.00 Bu
`0.00 Bu
`0.00 Bu
`0.36 mB
`0.51 pH
`0.00 Bu
`0.00 Bu
`0.88 Ex
`0.87 Ex
`0.00 Bu
`0.00 Bu
`0.79 mE
`0.78 mE
`0.1 l Bu
`0.00 Bu
`0.61 mE
`055pB
`0.43 pH
`044pB
`0.97 Ex
`0.85 Ex
`0.77 mE
`088 EX
`0.69 mE
`0.68 mE
`0.33 m8
`0.41 pH
`0.00 Bu
`000 Bu
`0.35 mB
`0.45 pB
`0.00 Bu
`0.00 Bu
`0.69 mE
`0.53 pH
`0.00 Bu
`0.00 Bu
`0.73 mE
`058pB
`0.98 Ex
`0.96 Ex
`0.00 Bu
`0.00 Bu
`0.83 Ex
`0.73 mE
`0.90 Ex
`075 Me
`0.90 Ex
`082 EX
`0.] 1 Eu
`0.00 Bu
`047 p8
`0.54 pB
`0.00 Bu
`1.00 Ex
`0.63 mE
`0.58 pH
`0.00 Bu
`0.00 Bu
`0.08 Bu
`0.00 Bu
`0.00 Bu
`0.00 Bu
`0.00 Bu
`0.00 Bu
`0.15 Bu
`0.17 Bu
`
`Z
`
`l"
`
`W71>
`
`z
`
`(Am
`
`>
`
`TQ
`
`Hramavx<r
`1“U)gm 0
`rmmowmfiOHr <
`rgmr>7<crew0x~gmHam
`
`> O
`
`<
`
`>
`
`wmaoow>wnm<w<wwovwx<m>Qmo<fio<0
`7:Z<7:< E
`ow4<a<woggmrooov>oz<gWP“:>_.
`
`W '11
`
`W>0~<~<<>~1Umm
`
`EV QLV QSGGGVV QPGRSLRLSCSSSGFIFSWVR QAPG
`
`KGLEWVARFTISRNDSKNTLFL QMDSLRPEDTGVYFCAR
`
`4 of 10
`
`Celltrion, Inc., Exhibit 1118
`
`
`
`Reduction of the immunogenicity of antibody variable domains
`
`Table l~continaed
`
`
`
`
`
`103
`104
`105
`100
`l0?
`108
`109
`HO
`lll
`l 12
`113
`
`0.09 Bu
`0.00 Bu
`0.93 Ex
`0.00 Bu
`0.22 ml?
`099 BX
`000 Bu
`0.76 mE
`0.00 Bu
`0.98 Ex
`0.94 Ex
`
`ww<~i<~6~iooog
`
`0.07 Bu
`1.00 Ex
`099 EX
`0.00 Bu
`0.26 ml!
`0.67 ml?
`0.00 Bu
`0.69 mE
`0.00 Bu
`0.74 mE
`0.84 Ex
`
`>U1<1~l<r‘-JQOD€
`
`w<4<ra00025
`
`S
`
`m<4<haongE
`
`U3
`
`m<~i<zrd000€$
`
`S
`
`ww<~l<PHQOQ€I
`
`mm<4<PHOOQ€I
`
`mw<q<aaooogm
`
`Position
`
`Residue
`
`QsvLT QPPs AsGTPG QR\
`
`ITIsCwv 9
`
`bowwwowwwowem<~rrxw>gmm
`
`Table 2. Solvent exposure of sidechains of framework residues in KOL VI. and the residues which occur most frequently at these positions
`“WWW
`in the various human V-lambda subgroups
`Residues in subgroup
`III
`IV
`
`Exposure
`1.00 Ex
`1.00 Ex
`0.77 mE
`0.00 Bu
`0.92 Ex
`0.00 Bu
`0.62 mE
`1.00 Ex
`LOO Ex
`
`0.34 m8
`071 mE
`1.00 Ex
`0.73 mE
`0.75 mE
`LOO Ex
`0.69 ml?
`0.79 mE
`021 1118
`0.62 mE
`0.00 Bu
`0.92 Ex
`0.00 Bu
`0.00 Bu
`0.00 Bu
`0.46 pB
`0.00 Bu
`0.75 mE
`0.9l Ex
`1.00 Ex
`0.74 mE
`0.62 ml?
`0.00 Bu
`0.95 Ex
`0.23 m3
`015 Bu
`0.00 Bu
`0.39 ml?!
`1.00 Ex
`0J4 Bu
`0.70 mE
`0.95 Ex
`0.3) ml;
`0.12 Bu
`0.85 Ex
`0.00 Bu
`1.00 Ex
`0.41 pH
`1.00 Ex
`1.00 Ex
`0.71 mE
`
`.—ti—
`
`SF
`
`”
`
`0
`
`<b—d
`
`nH3Q~g<lz<m<lwweoArm~<>>r<W
`
`g<w>oowxoo<gna~x>
`
`w“mm
`
`3
`
`r"
`
`'-<"‘<<I“
`
`1:
`
`gnm~aemoovemw>lwvwoar<woT>><
`qgmxmowmmgm<m<~rrx1>~lmnfit~oo<
`
`<:
`
`>
`
`waom<Im>wofir>mo
`
`~w4O
`
`<gow~a ’11
`
`0%“ZF‘W’U>7<O’°:IIOOrn_
`zowawowmx
`
`H
`
`0m~q<moowmom>m~a~eg~lr~>wo<12|
`
`zgmgmgwmwuw<Q3~<rxv>wmwmoo<
`
`71>
`
`"‘<
`
`C:
`
`~H<fl7qmmmmm<Imm-eo-arg-nz
`
`Qmmxgw<o<~<a4v>wowxoo<gnw
`
`’21 Z
`
`W C)
`
`1
`
`min—‘0'; 'fl
`
`2
`
`Fractional accessibility
`Residues in subgroup
`1539
`KOL
`ExposureResidueExposureResiduePosition ll
`La)
`l—It
`_
`&
`KJI
`E—s
`ON
`H
`
`
`
`
`
`
`
`The one-letter amino acid code is used here. The fractional solvent accessibility values for the individual residues were computed as described
`by Padlan (1990); residues whose sidechains have fractional accessibility values between 0.00 and 0.20 are designated as being completely
`buried (Bu), between 0.20 and 0.40 as mostly buried (1113), between 0.40 and 0.60 as partly buried/partly exposed (p13), between 0.60
`and 0.80 as mostly exposed (mE), and at least 0.80 as completely exposed (Ex). In the special case of glycine, the residue is considered
`completely exposed if its alpha-carbon atom is accessible to solvent, otherwise it is considered completely buried. Residues that are
`involved in the VL—VH contact are shown in bold letters.
`
`(continued averted)” )
`mommwowmwov~m<~<rrw>oomwoo<goa~m<ooar><m<l>wm04rmml
`
`5 of 10
`
`Celltrion, Inc., Exhibit 1118
`
`
`
`Residues in subgroup
`5
`III
`IV
`
`Position
`Residue
`Exposure
`TKS
`70
`(/1
`1.00 Ex
`AV
`71
`0.00 Bu
`A
`TI
`72
`1.00 Ex
`73
`0.00 Bu
`74
`0.74 mE
`75
`0.00 Bu
`76
`1.00 Ex
`77
`1.00 Ex
`78
`0.00 Bu
`79
`0.76 mE
`80
`1.00 Ex
`81
`0.78 mE
`82
`0.09 Bu
`83
`0.64 mE
`84
`0.34 mB
`85
`0.30 ml]
`86
`0.00 Bu
`87
`0.16 Bu
`88
`0.00 Bu
`
`E.A.PADLAN
`
`Table 2—c0ntinued
`
`
`
`Qt“<'-1l"7§-IQOO'T1F:Or<h<U>mUm>OFQw“'-ll"m>—l:
`
`
`
`mt"<-H"O"lOmO'TIF0<~<U>m0m>0>Q—l"—H‘VJ>—i
`
`O<<U>momeFQW~%rm>w O a
`
`C}l"<'-ll";O'-1C)WQ’I1'—‘O<<U>mUm>WFQVJ<HFm>~i<
`
`T LII
`
`«1;
`
`-[‘1
`
`
`
`OF<-ll"7<"lQQQ’UFO<<U>MU
`
`QF<H<WHQHQTIFO<<U>m0mmOFQMH>FVJ>UIH
`
`EVW'Um-IO‘AWOO'<€0—lH—1<KOOFW>>HH>’UmO-H‘<Hm
`
`2
`
`
`
`FFWV>7<Qw7<OO<fiafi—‘*—l<WUQ<m>mt‘mmwmoqzo—U—
`
`""0“WOO“WOF<sow—“>vmov4<vrmrwmoqg<~c=
`
`m wrrw1>oov7<oo<gnwra>wmowmrmrflmwmoHP<—m
`
`Frfifivoovxoo<gnz~a>wmorm<>rmgmmoqz<—U
`
`0<<H>>Um>m:HZ“HFM<WHOMOWOW“W>1<Q<H
`
`n<<4>momworwm~aramcaomomommxwv<o<~H
`
`[Ti 0n<<<o<om>m<wm~xrHmcaomomowmwow<o<H
`
`O
`
`O
`
`n<<<>1cmvmer~Heruaowomommwow~o<“
`
`O-<-<<><Dm>0rmm“HF-I'TlU—lQV’OmQVJ'HWU'V<O'<H
`
`._g
`
`._i
`
`max
`
`
`
`n<<oamumwor00">rm
`or<4<weoaom
`
`0.04 Bu
`98
`0.00 Bu
`99
`100
`0.59 pH
`1.00 Ex
`101
`0.00 Bu
`102
`0.82 Ex
`103
`0.00 Bu
`104
`105
`086 EX
`106
`0.19 Bu
`0.70 mE
`106a
`1.00 Ex
`107
`*Additional residues after position 66:
`66a D
`6613 S R D.
`(see footnote to Table l).
`
`(continued next page)
`
`Table 3. Solvent exposure of sidechains of framework residues in J539 VL and the residues which occur most frequently at these positions
`in the various human V-kappa subgroups
`Residues in subgroup
`Residues in subgroup
`2
`
`Position Residue
`Exposure
`III
`II
`III
`IV
`Position Residue
`Exposure
`48
`0008u
`0.99 Ex
`0.16 Bu
`49
`042pB
`57
`LOOEx
`0.87 Ex
`58
`0i3Bu
`0.00 Bu
`0.80 mE
`w
`0MmE
`60
`LOOEx
`0.00 Bu
`61
`036mB
`0.89 Ex
`62
`0008u
`0.67 mE
`63
`094Ex
`1.00 Ex
`64
`QOOBu
`094 EX
`65
`LOOEx
`0.30 mB
`66
`LOOEx
`0.59 pH
`67
`LOOEx
`0.00 Bu
`68
`LOOEx
`0.78 mE
`69
`0.75 mE
`0.79 mE
`1.00 Ex
`70
`098Ex
`7|
`009Bu
`0.64 mE
`72
`0.70 mE
`0.74 mE
`022 m8
`73
`000Bu
`0.65 mE
`74
`043pB
`0.00 Bu
`75
`0OOBu
`0.69 mE
`76
`083Ex
`77
`083Ex
`0.00 Bu
`78
`0.00 Bu
`0.00 Bu
`79
`0.63 mE
`0.00 Bu
`0.14 Bu
`so
`096Ex
`81
`0mm
`024 m8
`0.69 mE
`82
`013Bu
`1.00 Ex
`83
`055pB
`s4
`0003u
`1.00 Ex
`0.90 Ex
`85
`058pB
`86
`0003u
`0.30 mB
`m
`0mm
`0.00 Bu
`090 EX
`88
`ODOBu
`0.43 pH
`0.16 Bu
`
`6 of 10
`
`Celltrion, Inc., Exhibit 1118
`
`
`
`Reduction of the immunogenicity of antibody variable domains
`
`495
`
`Table 3—continued
`
`Residues in subgroup
`Position Residue
`I
`II
`III
`Exposure
`IV
`u
`a
`N
`H
`JK-4 JK-S
`-3
`F
`
`~m<xaooomr
`"U<WHQVOmR
`"mrwaoommfi
`rmrwao>ofi
`~mrxaooofi
`"m<W*-IOOQ
`
`K
`
`W
`
`7‘1
`
`7<
`
`7C
`
`7‘}
`
`0.00 Bu
`1.00 Ex
`100 EX
`0.00 Bu
`0.00 Bu
`0.79 mE
`0.00 Bu
`0.89 Ex
`0.44 pB
`0.77 mE
`
`(data not shown, but see examples below), the amino
`acid replacements by site-directed mutagenesis would
`not be too difficult.
`
`In some of the cases presented in Table 4, many
`buried and contacting residues are found to be differ-
`ent in the mouse and human domains. In those cases,
`unless the interior and contacting residues are also
`replaced, simple grafting of the CDRs to a human
`framework could result in alterations in ligand-binding
`properties.
`One position at which human and mouse sequences
`frequently differ is the N-terminus in both chains.
`The N-termini are contiguous with the CDR surface
`and are in position to be involved in ligand binding.
`It might therefore be wise to keep the N-terminal
`residues of the original antibodies in the “humanized”
`ones.
`
`The replacement of some amino acid types could
`have a significant effect on the tertiary structure of
`the domains. For example, the replacement of proline
`by another amino acid and vice versa, could cause
`a restructuring of the local region;
`this could be
`particularly serious if the proline is
`in the cis
`configuration. A change in the local structure could
`also occur following the replacement of a glycine by
`another amino acid type, if the backbone dihedral
`angles of the glycine are energetically unfavorable for
`other types of residues.
`A replacement that could have a significant effect
`on ligand binding, even if it
`is distant from the
`combining site region, is one which involves a change
`in electrical charge. If electrostatic interactions have
`been determined to be important in the binding of the
`antigen, replacements should be chosen judiciously
`
`,_.
`
`98
`99
`100
`101
`102
`103
`104
`105
`106
`106a
`107
`
`(see footnote to Table 1).
`
`designations in Tables 1—3) are replaced. One would
`retain (a) the CDRs, (b) the residues in the immediate
`neighborhood of the CDRs (the last
`framework
`residue before and the first after each CDR might be
`enough),
`(c)
`the residues corresponding to those
`which in KOL and J539 are completely or mostly
`buried (mB and Bu designations in Tables 1-3), and
`(d) the residues corresponding to those which in KOL
`or J 539 are involved in the interdomain contact.
`
`the number of
`On the basis of this proposal,
`residues in a mouse framework that would need to be
`replaced to “humanize” it can be determined for each
`representative mouse variable domain. This number
`is found to range from 7 to 16, when the mouse
`sequences are correlated with the most frequently
`occurring residues in the human variable domain
`subgroups, and from 6 to 15 when they are compared
`with the human sequences to which they are most
`similar (Table 4). These numbers are not large and,
`since in many instances the differences are bunched
`
`Table 4. Comparison of the frameworks of representative mouse and human variable domains
`Difierences’r
`vs
`
`Most similar human:
`Subgroup
`Sequence
`
`(most similar subgroup)
`Need
`to
`Con-
`replace
`tacting
`
`Total Buried
`
`(most similar sequence)
`Need
`to
`Con-
`replace
`tacting
`
`Total Buried
`
`VKIV, JKZ
`VKII, JK4
`VKIV, JK4
`VKIV, JKZ
`VKI, JKl
`VKIV, 1K2
`
`RPM1-6410’CL
`
`V19A‘CL
`
`VLIII, JL2
`
`4A’CL
`
`VHII, JH4
`VHII, JH4
`VHI, 1H6
`VH1, JH6
`VH1, 1H4
`VHIlI, JH6
`VHIII, JH4
`VHIII, JH6
`VH1, 1114
`VHIII, JH6
`
`SUP-TI-VH-JA’CL
`SUP-T l -VH—JA’CL
`HG3’CL
`HG3’CL
`HGS’CL
`LAMBDA-VH 26’CL
`LAM BDA-VHZé’CL
`LAMBDA-VHZG’CL
`HGS’CL
`H1 l’CL
`
`20
`10
`15
`26
`15
`25
`
`29
`
`231:
`181
`231
`20:
`211
`14
`1 1
`13
`23:
`s
`
`_. 0
`
`omwaoxoxooo
`
`2“
`1"
`
`2+1“l
`3:
`l+l*
`
`3+4.
`
`1+2‘
`1‘
`2
`
`15
`7
`11
`15
`10
`16
`
`16
`
`ll
`ll
`11
`10
`11
`a
`s
`8
`12
`7
`
`23:
`
`271
`281
`251
`211
`25:
`16
`15
`17
`261
`12
`
`10
`13
`10
`9
`13
`6
`5
`7
`11
`3
`
`1o
`
`15
`14
`12
`11
`11
`s
`1o
`10
`14
`9
`
`Mouse
`sequence
`VL (kappa):
`$107
`TEPC(CAL20)105
`TEPCl 1 1
`SlO7B’CL
`R16.7 CRI +
`TEPC191
`
`VL (lambda):
`MOPC104E
`VH1
`36—60 CRI-
`MClOl’CL
`MOPC104E
`Bl-8’CL
`GS BB 2.2(ABl)
`TEPC15
`XRPC44
`HPC76’CL
`9367 CR1 + ’CL
`MOPCZl’CL
`
`When no human sequence is given, it means that none is more similar to the mouse sequence than the consensus subgroup sequence.
`*Those contacting residues are buried.
`'l'The differences in the J-regions have been included.
`iThese include a residue that borders a CDR.
`
`7 of 10
`
`Celltrion, Inc., Exhibit 1118
`
`
`
`E. A. PADLAN
`
`Fig. 1. Stereodrawing of the alpha-carbon backbone trace of J539 Fv. The N - and C -terminal residues
`are labeled, as are every 20th residue in the sequence (numbering of Kabat et al., 1987). The VH domain
`is on the left and VL is on the right. The residues, which need to be replaced in the variable domains
`of the mouse antibody to “humanize” them, are indicated by the large open circles for $107 (a) and for
`MOPC21 (b).
`
`most similar in its framework to the members of the
`
`l7,
`l6,
`replaced are those at positions 9, 10, 14, 15,
`18, 22, 41, 63, 80, 83, 85, 100, and 106. $107 VH
`(Kabat et al., 1987) is most similar in its framework
`to the members of the human subgroup VHIII and
`JH6, from which it diflers at positions 3, 24, 40, 44,
`48, 68, 69, 73, 75, 76, 77, 82b, and 105. Of these,
`positions 24, 48, 69, and 77 are occupied by residues
`that in J539 or KOL are buried (Table 1); in addition,
`the residues at positions 44 and 105 are involved in
`interdomain contacts in 1539 or KOL. Thus the
`
`to preserve as much as possible the density and
`distribution of electrical charges on the molecule.
`The application of the procedure is demonstrated
`by the following examples. In these examples, only
`the use of the most frequently occurring residues in
`the various human variable domain subgroups and
`J-minigenes is explored.
`
`(a) A “worst-case” example
`
`Among the mouse kappa VLs listed in Table 4, for
`which the autologous VH had also been sequenced,
`SlO7 VL has the most residues that need to be replaced
`to “humanize” it. $107 VL is most similar in its
`
`framework to the members of the human subgroup
`VKIV and JK2, from which it difl'ers at positions 9,
`10, 14, 15, 16, 18, 19, 22, 38, 41, 43, 63, 78, 80, 83,
`84, 85, 100, and 106, Of these, positions 19, 38, 43,
`78, and 84 are occupied by residues that in J 539 are
`buried (Table 3); moreover, the residues at positions
`38 and 43 are involved in interdomain contact in
`1539. Thus the residues in $107 VL that need to be
`
`residues in S107 VH that need to be replaced are those
`at positions 3, 40, 68, 73, 75, 76, 82b, and 89. The 23
`residues that need to be replaced to “humanize” the
`variable domains of S107 are indicated in Fig. 1(a).
`
`(b) A “best-case” example
`
`Among the mouse VH included in Table 4, for
`which the autologous VL had also been sequenced,
`MOPC21 has the least number of residues that need
`to be replaced to “humanize” it. MOPC21 VH is
`
`8 of 10
`
`Celltrion, Inc., Exhibit 1118
`
`
`
`Reduction of the immunogenicity of antibody variable domains
`
`497
`
`nevertheless conceivable that antibodies modified by
`the present procedure may be sufficiently “human-
`ized” so that they are no longer immunogenic; this
`can only be proven by actual trials.
`Very recently. a procedure has been described for
`the rapid expression in Escherichia coli of monoclonal
`antigen-binding fragments (Huse et al., 1989). The
`possibility of expressing human antibody fragments
`by this procedure may obviate the need for the
`generation of designer antibodies in a non-human
`system and subsequent “humanization”. Neverthe-
`less,
`there are a number of rodent-derived mono—
`clonal antibodies that are currently being used in
`human therapy and diagnosis, and there will be
`others that for one reason or another are best gener—
`ated using a non-human system for which “human-
`ization” procedures may be needed. For those cases,
`the procedure proposed here will be useful.
`
`REFERENCES
`
`Crystallographic refinement and atomic models of the
`
`Amit A. G., Mariuzza R. A., Phillips S. E. V. and Poljak
`R. J. (1986) Three-dimensional structure of an antigen—
`antibody complex at 2.8 A resolution. Science 233,
`7477753.
`Benjamin D. C., Berzofsky J. A., East I. J., Gurd F. R. N.,
`Hannum C., Leach S. J., Margoliash 13., Michael J. G.,
`Miller A, Prager E. M., Reichlin M., Sercarz E. 13.,
`Smith-Gill S. J ., Todd P. E. and Wilson A. C. (1984) The
`antigenic structure of proteins: A reappraisal. Ann. Rev.
`Immunol. 2, 67—101.
`Bernstein F. C., Koetzle T. F., Williams G. J. 3., Meyer
`E. F., Jr., Brice M. D., Rodgers J. R., Kennard 0.,
`Shimanouchi T. and Tasumi M. (1977) The Protein Data
`Bank. A computer-based archival file for macromolecular
`structures. J. Moles. Biol. 112, 5357542.
`Case D. A. and Karplus M. (1979) Dynamics of ligand
`binding to heme proteins. J. Melee. Biol. 132, 3437368.
`Colman P. M., Laver W. G.. Varghese J. N., Baker A. T.,
`Tulloch P. A., Air G. M. and Webster R. G. (1987)
`Three-dimensional structure of a complex of antibody
`with influenza virus neuraminidase. Nature (London) 326,
`358—363.
`Connolly M. L. (1983) Analytical molecular surface calcu-
`lation. J. Appl. Crystallogr. 16, 548—558.
`Davies D. R., Padlan E. A. and Sheriff S. (1990) Antibody—
`antigen complexes. Arm, Rev, Biochem. 59, 439—473.
`Huse W. D., Sastry L.,
`lverson S. A., Kang A. S.,
`Alting—Mees M., Burton D. R., Benkovic S. J. and Lerner
`R. A. (1989) Generation of a large combinatorial library
`of the immunoglobulin repertoire in phage lambda.
`Science 246, l275~1281.
`Jones P. T., Dear P. 11., Foote J., Neuberger M. S.
`and Winter G. (1986) Replacing the complementarity-
`determining regions in a human antibody with thOSe from
`a mouse. Nature (London) 321, 522—535.
`Kabat E. A., Wu T.-T., Reid-Miller M., Perry H.