BB

ELSEVIER

Biochimica et Biophysica Acta 1247 (1995) 293-297

etBiochi~ic~a
BiophysicaA~ta

Rapid Report

The Temkin isotherm describes heterogeneous protein adsorption

Robert D. Johnson, Frances H. Arnold *
Division of Chemistry and ChemicalEngineering 210-41, CaliforniaInstitute of Technology, Pasadena, CA 91125, USA

Received 6 December 1994; accepted 29 December 1994

Abstract

Here we examine how heterogeneous protein adsorption arises from multivalent interactions with a seemingly homogeneous functional
surface. During adsorption, some arrangement of functional groups on the protein (e.g., charged or hydrophobic amino-acid residues or
specific ligand binding sites) interacts with complementary sites distributed on the adsorbent surface. The protein will show the highest
affinity for the surface arrangements which best match its own distribution of functional sites, resulting in a distribution of binding
energies. To support this interpretation, we show that changing the density of affinity ligands on a surface (immobilized metal ions) is
equivalent to changing the number of target groups on a protein (surface histidines). We also report that reversible protein adsorption
obeys the Temkin isotherm and propose that model as a practical framework for describing the behavior of proteins adsorbing via
multivalent interactions onto surfaces densely derivatized with a random distribution of binding functionalities. This result has important
implications for the design of separations materials and the interpretation of biological recognition phenomena.
Keywords: Biological recognition; Heterogeneous binding; Histidine; Metal affinity chromatography; Copper iminodiacetic acid

The adsorption of proteins to functional surfaces is

often described by the Langmuir isotherm [1,2], originally
derived for gas adsorption to a homogeneous surface.
According to this model, adsorption is characterized by a
single binding energy and a maximum adsorption capacity
corresponding to monolayer surface coverage. Several researchers have noted that the Langmuir model does not
adequately describe the he,terogeneity of protein adsorption
to materials commonly used in ion-exchange and affinity
chromatographic separations [3-7]. We have observed heterogeneous binding in the adsorption of variants of cytochrome c to a metal-affinity chromatography support;
these variants differed only in the number and accessibility
of surface histidines [8]. In the course of these studies, we
noted that the extrapolated maximum capacity appeared to
be correlated with the binding affinities of the cytochrome
c variants. In addition, a correlation between apparent
limiting capacity and binding strength has been observed
for adsorption of the same protein to ion exchange [4] and
affinity [5] chromatography supports when protein adsorption is inhibited by soluble competitors.

Corresponding author. Fax: -~-1 (818) 5688743.

0167-4838/95/$09.50 1995 Elsevier Science B.V. All rights reserved

SSDI 0167-4838(95)00006-2

The Langmuir model predicts a single maximum binding capacity for a given protein in the absence of significant conformational changes. This model therefore cannot
explain the observed increase in limiting capacity with
increasing surface histidine content for what are otherwise
identical proteins. The capacity of these materials to bind
protein is ultimately dictated by the degree of functionalization (the density of ligands) and steric factors (e.g., the
size of the protein) and should not depend on binding
affinity or competitor concentration. The current investigation was motivated by the presumption that a given protein, or a series of its variants, should reach the same
limiting coverage at sufficiently high concentrations during
reversible adsorption to a functional surface 1
Reversible protein adsorption is better described by
another model also widely used in studies of gas adsorption, the Temkin isotherm. This model assumes that adsorption is characterized by a uniform distribution of binding energies, up to some maximum binding energy

1 This argument presumes that the protein's adsorption cross section

(the space it occupies on the surface) does not change. We could envision
situations in which it would change, for exam~ale when the protein
becomes oriented in a certain direction that depends on the specific
interactions with the surface.

294

R.D. Johnson, F.H. Arnold / Biochimica et Biophysica Acta 1247 (1995) 293-297

(Aamax), which results in the following isotherm equation

[9],

2.5

Q = Qrln(1 + K r c )
where K r (M-1) is the equilibrium binding constant corresponding to the maximum binding energy ( K T =
exp(-AGmax/RT)), c (M) is the concentration of protein
in solution at equilibrium, Q (mol protein/ml support) is
the amount of protein adsorbed to the surface, and Qr
(mol protein/ml support) is the differential surface capacity for protein adsorption per unit binding energy.
We have reported equilibrium binding isotherms for ten
yeast cytochrome c surface histidine variants on a copperchelating support (Cu 2 IDA-TSK) used for metal-affinity
chromatography [8]. Increasing the protein's surface histidine content resulted in a dramatic increase in the apparent
binding affinity, presumably a result of multiple histidinecopper coordinate-covalent bonds between the protein and
the surface. Heterogeneity of binding was apparent from
Scatchard plots, and the data were first analyzed using a
bi-Langmuir isotherm with adsorption to two populations
of surface 'binding sites'. Even though the individual
experimental isotherms could all be fit to a bi-Langmuir
isotherm, changes in maximum binding capacity with surface histidine content for otherwise identical proteins could
not be explained. As shown in Fig. 1, semi-logarithmic
plots of the data demonstrate that the Temkin model fits
the individual isotherms as well as does the bi-Langmuir
model. In contrast to the varying limiting capacity that
results from use of the Langmuir model, however, the
.-.

2.5
o

2.0

00

1.5-

~ 1.o
~

0.5
0.0

10 .7

........

........

10 .6

10 5

, .......

10 .4

........

10 .3

free protein c (M)

Fig. 1. E q u i l i b r i u m adsorption o f c y t o c h r o m e c to m e t a l - a f f i n i t y chrom a t o g r a p h y support. E q u i l i b r i u m b i n d i n g data f r o m T o d d et al. [8] f o r

yeast cytochrome c variants differing only in surface histidine content to

Cu 2+ IDA-TSK prepared at maximum copper loading. Protein adsorbed
to support (Q) is plotted against equilibrium protein solution concentration (c) for cytochrome c histidine variants (t2)Hs, (11) H4, (zx)
H26Hs, ( & ) H26H4, ( O ) H26H33Hs, ( O ) H26H33H 4. Lines represent
fit to logarithmic (Temkin) model.

"~ 1.5

i 10

0.5
0.13 . . . . . . . . . .
10 "2

, ........, ........, ........, ........, .......

10

10 2

10 4

K,c
Fig. 2. Equilibrium adsorption of cytochrome c to metal-affinity chromatography support. Equilibrium binding data for protein bound to
support (Q) at increasing solution concentration (c) are from Todd et al.
[8], ( O ) binding data for ten yeast cytochrome c variants differing only
in surface histidine content (see Fig. 3A) to Cu 2+ IDA-TSK at maximum
copper loading, ( O ) binding data for horse cytochrome c to Cu 2+ IDATSK prepared at different immobilized copper concentrations. Protein
concentrations are scaled by the maximum binding constant (Kr, see Fig.

3) determinedby nonlinearregressionto the Temkinisotherm(Qr = 0.21

p~mol/mlTSK).
Temkin parameter Qr (slope of this plot) is approximately
constant ( ~ 0.21 /zmol/ml TSK) for all ten of the cytochrome c variants, despite the fact that the predicted
maximum binding affinity K r varies over five orders of
magnitude.
We have also measured the binding of a single protein,
horse cytochrome c, to the support prepared with different
densities of the copper affinity ligand [8]. These adsorption
isotherms are also described well by the Temkin model,
with the same value of the parameter QT. When each
protein solution concentration (c) is scaled by its maximum binding affinity (Kr), the isotherms of the ten yeast
cytochrome c histidine variants, tuna cytochrome c, and
horse heart cytochrome c at nine different copper loadings
all collapse onto a single curve, as shown in Fig. 2. This
result indicates a fundamental relationship between the
number of target groups of the protein (surface histidines)
and the density of binding sites on the surface (metal ions).
The molecular basis of this result becomes more apparent as we compare the effect of adding histidines to the
surface of the protein (Fig. 3A) to that of adding copper
ions to the surface of the metal-affinity chromatography
support (Fig. 3B). In both cases, adsorption involves more
interactions between protein and surface, with increasing
apparent binding affinities and maximum capacities. According to the Temkin model, adding a single histidine
increases the maximum binding energy of yeast cytochrome c by --4 kcal/mol (roughly a factor of 60

R.D. Johnson, F.H. Arnold / Biochimica et Biophysica Acta 1247 (1995) 293-297

~"

10~ i A

10z

t-

106

II
~

II

10 4

" 1
~

10 3 ~
0

number of histidines
A

B
10 s

0
e-

o
O)

.c_
"o

10 4

e-

0 0

E
-,i
E
x

10 3

E
I

5
10
15
20
accessible copper (~mol/ml TSK)

Fig. 3. Effect of protein histidine content and surface copper loading on

maximum binding constants. (A) Maximum binding constants ( K r ) of
ten yeast cytochrome c variants (labeled by surface histidines) and tuna
cytochrome c are presented frora lower left to upper right : H(-), H26,
tuna, Hs, H4, H26H33 , H26H8, H26H4, H26H33Hs, H26H33H39 ,
H 26H 33H4. (B) Maximum binding constants ( K r ) of horse cytochrome c
to IDA-TSK support prepared at different copper loading. Accessible
copper (Cuac c) is measured by the support's binding capacity for Nacetylhistidine, as determined by regression of equilibrium binding data
to the Langmuir isotherm [13]. ]Maximum binding constants determined
by nonlinear regression to the Temkin isotherm (Qr = 0.21 /,Lmol/ml
TSK). Equilibrium binding experiments from Todd et al. [8].

increase in the maximum binding constant). The same

increase in binding can al,;o be achieved by increasing the
concentration of surface copper ions, from 10 to 18
/zmol/ml TSK (for horse heart cytochrome c) 2. In both
cases, adsorption involves more copper-histidine coordina-

2 This energy gain is consistent with the selectivity displayed ( = 100fold) by rationally designed bis.-mercury receptors for a target bis-imidazole (a two-histidine 'protein analog') over the single imidazole
control [10].

295

tion bonds between protein and surface, increasing apparent binding affinity. This functional equivalence between
histidines on the protein and metal ions on the adsorbent is
to be expected when the binding energy reflects the number of metal-ligand interactions formed between the protein and surface 'binding sites' of multiple metal ions
[3-5,11].
Strictly speaking, the Temkin isotherm stated above
does not reach a finite limiting capacity for protein adsorption at infinite protein concentration. An analogous threeparameter Temkin isotherm includes a minimum binding
energy to satisfy this limit [12]. If, however, cytochrome c
variants are presumed to share the s a m e limiting capacity
and minimum binding energy, then consensus values of
these two parameters can be obtained [13]. The resultant
limiting capacity for protein adsorption (3.5 /~mol/ml
TSK) is consistent with our estimate of the cytochrome c
monolayer coverage (3-4 /zmol/ml TSK) based on the
protein's dimensions in the crystal structure 3
The Temkin model predicts that the binding energy
decreases linearly with increasing amounts of protein bound
to the surface. Why should this be? A distribution of
binding energies can be explained if we consider protein
adsorption to surface sites involving multivalent contacts
in terms of two opposing contributions [15]: a favorable
energy from the specific metal-to-protein contacts ('intrinsic binding energy') and an unfavorable energy required to
match each binding site and protein to make these contacts
('rearrangement energy'). On a densely derivatized surface, a random distribution of functional sites or affinity
ligands can be viewed as a set of individual 'binding sites',
each supporting specific protein-ligand interactions to a
different degree. As illustrated in Fig. 4A, the protein will
adsorb with the highest binding energy to those arrangements of metal ions which most closely match its pattern
of histidines. Less optimal patterns of metal ions would
require some rearrangement to obtain the same number of
interactions, resulting in a lower net binding energy. The
Temkin model predicts a uniform distribution of binding
energies over the population of surface binding sites. Theoretically, such a uniform distribution of binding energies
would arise from a truly random arrangement of surface
binding sites (Wang, Z.-G., unpublished results). The range
and distribution of binding energies should depend strongly
on the density and distribution of functional groups, both
on the protein and on the surface, as demonstrated in Fig.
3.
Binding heterogeneity is a recurrent theme among chromatographic systems in which adsorption involves multi-

3 We estimate a TSK surface area of 16 m2/ml. The average radius of

yeast cytochrome c is 15.2 A, that of the asymmetric unit (protein plus
water) is 17.6 ,~ [14]. Distributing circular shadows of these radii over the
TSK surface area results in a monolayer coverage of 2.7-3.7/.~mol/ml
TSK.

296

R.D. Johnson, F.H. Arnold / Biochimica et Biophysica Acta 1247 (1995) 293-297

1
suboptimal surface sites

optimal surface sites

$
E
t-

net binding energy

B
Langmuir

1
w

E
t-

I/

~ r n Hill

net binding energy

kin

mined parameters and can be expressed in closed form, it

is rigorously grounded in statistical mechanics [18], and it
satisfies Henry's law in the limit of zero protein concentration (a necessary criterion to apply equilibrium adsorption
experiments to chromatographic elution experiments [19]).
Although the original motivation was to describe heterogeneous protein adsorption in chromatographic systems,
we have also examined binding heterogeneity commonly
observed in biological binding systems (e.g., receptor
repertoires, antibodies). We have found that the bell-shaped
energy distribution of the Hill or Freundlich isotherms
(Fig. 4B) describes the relevant features of biological
binding better than the uniform distribution of the Temkin
isotherm [13]. It is interesting to speculate on the origin of
this fundamental difference between the binding energy
distributions for synthetic surfaces (chromatography supports) and biological systems. At the molecular level, the
bias exhibited by biological systems towards particular
values of binding energy reflects a non-random arrangement of functional groups throughout the repertoire of
binding molecules. This bias apparently better prepares the
receptor repertoire to recognize (and discriminate among)
relevant compounds over the range of concentrations it
will encounter in the environment.

Acknowledgements
m

Fig. 4. Heterogeneous protein described as a distribution of binding

energies over the population of surface sites. (A) Distribution in binding
energies results from multiple contacts between protein and random
arrangement of surface ligands (connected to the surface via a flexible
spacer arm). Protein binds with highest affinity to surface sites in which
ligand pattern complements arrangement of protein functional groups.
Less optimal surface sites must incur additional energy penalties to
support the same degree of protein binding, resulting in a range of net
binding energies for protein adsorption. (B) Form of adsorption isotherm
expression depends on distribution on in binding energies. Langmuir
isotherm is described by a spike at a particular binding energy. Hill
isotherm (exponent less than one) is described by a bell-shaped distribution in binding energies. Temkin isotherm is represented by a uniform
distribution, up to some maximum binding energy.

pie interactions between the protein and surface. We find

that this model can also describe adsorption to other
materials (e.g., ion exchange) where random arrangements
of surface ligands give rise to heterogeneous proteinsurface interactions [13]. The Temkin isotherm represents
the binding heterogeneity with a simple expression that
has predictive power over a wide range of concentrations.
In this respect, it has important practical advantages over
other isotherms considered 4: it contains only two undeter-

4 None of the other isotherms considered (Hill [16], Freundlich, Dubinin-Raduskevich, and Toth [17]) resulted in a reasonable fit for a
majority of the binding isotherms.

This research is supported by the U.S. Office of Naval

Research and the National Science Foundation. F.H.A.
acknowledges an NSF PYI award and a fellowship from
the David and Lucile Packard Foundation. R.D.J. is supported by a predoctoral training fellowship from the U.S.
National Institute of General Medical Sciences, Pharmacology Sciences Program.

References
[1] Horstmann, B.J. and Chase, H.A. (1989) Chem. Eng. Res. 67,
243-254.
[2] Belew, M.B., Yip, T.T., Andersson, L. and Porath J. (1987) J.
Chromatogr. 403, 197-206.
[3] Kopaciewicz, W., Rounds, M.A., Fausnaugh, J. and Regnier, F.
(1983) J. Chromatogr. 266, 3-21.
[4] Whitley, R.D., Brown, J.M., Karajgikar, N.D. and Wang, N.H.L.
(1989) J. Chromatogr. 483, 137-155.
[5] Yon, R. (1988) J. Chromatogr. 457, 13-23.
[6] Livingston, A.G. and Chase, H.A. (1989) J. Chromatogr. 481,
159-174.
[7] Gill, D.S., Roush, D.J. and Willson, R.C. (1994) J. Colloid Interface
Sci. 167, 1-7.
[8] Todd, R.J., Johnson, R.D. and Arnold, F.H. (1994) J. Chromatogr. A
662, 13-26.
[9] Sips, R. (1950) J. Chem. Phys. 18, 1024-1026.
[10] Mallik, S. Johnson, R.D. and Arnold, F.H. (1994) J. Am. Chem.
Soc. 116, 8902-8911.
[11] Lancet, D., Sadovsky, E. and Seidelmann, E. (1993) Proc. Natl.
Acad. Sci. USA 90, 3715-3719.

R.D. Johnson, F.H. Arnold/Biochimica et Biophysica Acta 1247 (1995) 293-297

[12] Garrone, E., Bolis, V., Fubini, B. and Morterra, C. (1989) Langmuir
5, 892-899.
[13] Johnson, R.D. (1995) Ph.D. Thesis, California Institute of Technology, Pasadena, CA.
[14] Louie, G.V. and Brayer, G.D. (1990) J. Mol. Biol. 214, 527-555.
[15] Jencks, W.P. (1981) Proc. Natl. Acad. Sci. USA 78, 4046-4050.
[16] Levitzki, A. (1984) Receptors: A Quantitative Approach, pp. 19-36,
Benjamin-Cummings, Menilo Park, CA.

297

[17] Jaroniec, M., Dabrowski, A. and Toth, J. (1984) Chem. Eng. Sci. 39,
65-70.
[18] Yang, C. (1993) J. Phys. Chem. 97, 7097-7101.
[19] Arnold, F.H., Schofield, S.A. and Blanch, H.W. (1986) J. Chromatogr. 335, 1-12.