Вы находитесь на странице: 1из 36

Binding Constants and Their Measurement

Pall Thordarson
The University of New South Wales, Sydney, Australia

1 Introduction
2 Equations, Equilibria, and Experimental
Techniques
3 Determining Stoichiometry
4 Data Analysis and Software
5 Conclusions
Acknowledgments
References

1
2
26
28
34
35
35

INTRODUCTION

In supramolecular chemistry, usually the first question that a


researcher might want to ask is: how strong is this complex
or interaction? In other words, what is the free energy
(G) associated with this interactions. In supramolecular
chemistry, the most straightforward way of measuring this
free energy difference is through the equilibrium constant
(K) for the system of interest since:
G = RT ln K
Binding constants are a special case of equilibrium constants such as acidbase equilibrium constants (best known
as pKa and pKb ) and solubility equilibrium constants (Ks ).
Binding constants measure the bonding affinity between
two or more molecules at equilibrium. In supramolecular
chemistry, binding constants for hostguest complexation
or hosthost aggregation (e.g., dimerization) are usually

the subject of interest. Binding constants are also used in


biochemistry to describe the affinity of an enzyme to a
substrate and in pharmacology to describe the affinity of a
ligand (drug) to a receptor. Researchers from all these disciplines therefore need to deal with the same challenges but
using different terminology and approachesa fact that
can be the source of endless frustration and mistakes to
newcomers. There are many excellent textbooks, reviews,
and software packages available for use in biochemistry
and pharmacology, showing how binding constants can be
obtained. These are extremely useful sources of information but need to be treated carefully in the supramolecular
chemistry context. In biochemistry, it is frequently possible
to determine directly the concentration of the complex of
interest, for example, by electrophoreses or filter-binding
assays,1 simplifying the data treatment considerably. This
is usually not possible in supramolecular chemistry where
methods such as NMR or UVvis are used to indirectly
detect complexation. Another problem in this field is that
the fundamental work was done at a time when computational power was scarce or nonexisting, which meant that
researchers often had to make shortcuts or rely on a statistically questionable approach such as linear regression (e.g.,
LineweaverBurk) for what are really nonlinear problems.
With the power of modern day computers and software
packages, there is absolutely no reason for anyone to use
these outdated methods,2 yet they propagate in the literature adding another layer of confusion to newcomers to the
field.
This chapter covers the fundamentals of determining
binding constants in supramolecular chemistry. Rather than
trying to cover all the different methods that have been
applied to determine binding constants in supramolecular
chemistry, the emphasis is on the basic equations used.
Illustrative examples focus on the most important methods used in determining binding constants including NMR,

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

Techniques

UVvis, fluorescence, and calorimetric titrations. It is probably safe to say that over 90% of all experimentally determined binding constants in supramolecular chemistry are
now determined using one of these four techniques. Other
older or more specialized techniques such as solubility
methods,3, 4 potentiometric,35 and mass spectrometry5 have
been discussed in earlier reviews and are not covered here.
It should, however, be noted that the basic equations shown
here can usually also be readily adapted to fit these older
or more specialized techniques.

2.1

The choice of host and guest is arbitrary but in


supramolecular chemistry the host is usually the larger
molecule (e.g., crown ether) and the guest the smaller
molecule or ion (e.g., potassium). Receptor (usually host),
substrate (usually guest), and ligand (guest) are other definitions that are frequently found in the supramolecular
chemistry literature, some having been borrowed for biochemistry or pharmacology as discussed above.
When m = 1, for each step of the equilibrium process
shown in Figure 1, we can define stepwise equilibrium
binding constants Km according to (1).

If we consider the formation of a complex in solution


between two species, we can define:
Host = H with the concentration of the free host = [H]
Guest = G with the concentration of the free guest = [G]
Complex = Hm Gn with the concentration of the complex
= [Hm Gn ] and n, m = 1, 2, 3, . . .

EQUATIONS, EQUILIBRIA, AND


EXPERIMENTAL TECHNIQUES

Irrespective of the method used, the key step in determining


binding constants involves defining some sort of a model
that relates to the underlying equilibria. The model(s) is
then compared to the data obtained. If the data and model
are in reasonably good agreement, data analysis (fitting)
can be used to extract valuable information, including the
association constant (Ka ). Depending on the situation, this
process can also be used to verify if the model is truly
reasonable for the data obtained from the system under
investigation. Note, however, that just because there is a
good fit between data and a model, it does not automatically
imply that this particular model is the best one for that
system.2 Conversely, one cannot use a poor agreement
between data and a model as the sole criteria to reject
a model, especially if the raw data is of poor quality.
The following section describes some of the most common
binding models and data analysis strategies encountered in
supramolecular chemistry.
The general m:n system
mH + nG

bmn

HmGn

The simple 1 : 1 system


H+G

Ka

HG + G
(a)

(b)

Kn =

Kn =

K1

K2

The 2 : 1 system
H+G

HG

HG + H

HG2

(c)

A+A

A2 + A

K3

A3 + A

K4

A4 + A

K5

A5 + A

K1

K2

HG

H2G

(d)

The general aggregation system


K2

[HGn ]
[HGn1 ][G]

(1)

The stepwise equilibrium constants are also related to


the kn = forward and kn = the backward rate of kinetics
of complexation according to (2).

The 1 : 2 system
H+G

HG

Basic definitions

K6

Ki
...

Ai

kn
kn

(2)

The 2 : 2 system
2H + 2G

H2G2 + 2G

KF

KB

H2G2

2HG2

(e)

Dimerization model:

KD = K2, K3, K4, K5...Ki = 0

EK-aggregation model:

KE = K2 = K3 = K4 = K5... = Ki

coEK-aggregation model: KE = K2/ = K3 = K4 = K5... = Ki


with r = K2/KE
(f)

Figure 1 The different equilibria discussed in this chapter. (a) The general m/n binding model. (b) The 1 : 1 equilibria. (c) The 1 : 2
equilibria. (d) The 2 : 1 equilibria. (e) The 2 : 2 equilibria. (f) The general aggregation system, including dimerization, equal-K linear
aggregation (EK model), and cooperative equal-K linear aggregation (coEK model).
Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

Binding constants and their measurement


We can also define the overall stability constant = mn
according to (3).

mn

[Hm Gn ]
=
[H]m [G]n

(3)

As discussed below, these quantities can be measured


using a range of techniques including spectroscopic and
thermodynamic ones. Binding constants, similar to most
thermodynamic and kinetic properties, are highly dependent
on the solvent and temperature conditions. Even a change
from nondeuterated to deuterated solvent may give slightly
different outcomes. Given the sensitivity of thermodynamic
quantities to temperatures, it is essential to control and
record the temperature of any experiments used to determine binding constants, otherwise the results will be of
little value.
It should be noted that, strictly speaking, one should
also include activity coefficients in (1) and (3) to account
for deviations from ideal behavior in a mixture. This is
usually not done in practicethe usual assumption used
to justify this is that activity coefficients have a negligible
effect in the dilute regime used in typical supramolecular
chemistry experiments. Notable exceptions are experiments
conducted in aqueous buffer or salt solutions where activity
coefficients may have a significant effect on the final
outcome.6
It should also be noted that comparing binding constants that have different units (e.g., M1 vs M2 ) can be
fraught with complications as one then needs to consider
the reference concentrations of the reactants.7 The reason
behind this can be traced back to the fact that the fundamental relation between free energy (G) and the equilibrium constant (K) : G =
RT ln K should be written
as G = RT ln K + RT
ln c, were ln c represents
the reference concentrations of the reactants and products.8
Usually, the reference concentration is taken
as 1 for the
unit used (e.g., M), rendering the term +RT
ln c zero
but using such an arbitrary assigned value (e.g., 1 M) as
the reference state complicates the analysis of many binding phenomena, including the chelate effect and effective molarities. This has resulted in a lively debate to
date,711 on what the chelate effect and effective molarities mean, but this topic is outside the scope of this chapter
(The Thermodynamics of Molecular Recognition, Concepts).
The most important specific cases of the equilibria shown
in Figure 1(a) are discussed in detail below; n = m = 1
(Figure 1b, 1 : 1 equilibria), n = 1, m = 2 (Figure 1c, 1 : 2
equilibria), n = 2, m = 1 (Figure 1d, 2 : 1 equilibria), and
n = m = 2 (Figure 1e, 2 : 2 equilibria) as well as three
selected examples of self-aggregation (Figure 1f).

2.2

The 1 : 1 equilibria

The 1 : 1 equilibria is perhaps the most commonly observed


and studied equilibria in supramolecular chemistry. The
binding constant or association constant = Ka for these
equilibria is defined by (4) which is just a special case
of (1).
Ka =

[HG]
[H][G]

(4)

Here Ka = K1 in (1) but the former notion is chosen


to distinguish the 1 : 1 binding constant from the stepwise
binding constant K1 in the 1 : 2 and more complex equilibria
discussed below.
Measuring Ka according to (4) requires knowledge of at
least one of quantities [HG], [H], and [G]. This may sound
simple but it needs to be remembered that these are free
concentrations of these species in solution and generally
their concentration cannot be measured directly. The total
concentration of the host = [H]0 and the guest = [G]0 ,
however, are usually known and relate to [H], [G], and
[HG] with the mass balance (5) and (6).
[H]0 = [H] + [HG]

(5)

[G]0 = [G] + [HG]

(6)

Taking (5) and (6) and isolating for [H] and [G] and
inserting into (4), then gives us (7).
[HG]
([H]0 [HG])([G]0 [HG])
[HG]
=
[H]0 [G]0 [HG]([H]0 + [G]0 ) + [HG]2

Ka =

(7)

Isolating for [HG] and rearranging (7) gives us the


quadratic (8).


1
[HG] [HG] [G]0 + [H]0 +
Ka

+ [H]0 [G]0 = 0 (8)

The real solution of (8) then gives us [HG] according


to (9).
1
[HG] =
2




1
[G]0 + [H]0 +
Ka

1
[G]0 + [H]0 +
Ka

2
4[H]0 [G]0

(9)

The power of (9) cannot be overstated as the only


unknown quantity here is the binding constant Ka the

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

Techniques

key quantity of interest! However, the problem with measuring [HG] directly still remains. To address this issue,
approaches to monitor changes in [HG] indirectly have
been developed. The most commonly used method in
supramolecular chemistry is the titration method.2 Here
changes in a physical property = Y of the system are monitored on addition of the guest [G]0 , while the concentration of the host [H]0 is kept fixed. The physical property
of interest can be anything from an NMR resonance =
to absorbance in UVvis spectroscopy = A or heat
absorbed/released in calorimetry = Q. The physical change
of interest can usually be described as the aggregate of
the physical property for the host = YH , guest = YG , and
complex Y = YHG and their abundance in solution. If it is
a function of the concentration of [H], [G], and [HG] in
solution, such as in the case of UVvis spectroscopy, the
observed physical property can be described by (10).
Y = YH [H] + YG [G] + YHG [HG]

(11)

The mole fraction notion is of special interest as (4) can


be used to define the mole fraction of the complex according
to (12).
fHG =

Ka [G]
[HG]
=
[H] + [HG]
1 + Ka [G]

[H]0 Ka [G]
1 + Ka [G]

Expanding on (6) with (13) then yields (14).


[G]0 = [G] +

[H]0 Ka [G]
1 + Ka [G]

(14)

[G]0 [H]0

1
Ka


[G]0 [H]0 +

1
Ka

2

[G]0
+4
(15)
Ka

We now first consider the situation where the guest is


silent, that is, YG = 0for instance in UVvis titrations
with a simple noncolored monoatomic ion such as chloride
(Cl ). Further, if we recognize that Y0 = [H]0 YH , we can
simplify (10) considerably using also [H] = [H]0 [HG]
from (5) to obtain (16).
Y = YH ([H]0 [HG]) + YHG ([HG])
= Y0 + (YHG YH )[HG]

(16)

Moving Y0 to the left and defining Y = Y YH and


YHG = YHG YH , we finally obtain (17).
Y = YHG ([HG])

(17)

Equations (9) and (17) can now be combined to give (18),


which describes the experimentally observed change in a
physical property (Y ) as a function of two known ([H]0
and [G]0 ) variables and two unknowns (Ka and YHG ).

Y = YHG

1
2


[G]0 + [H]0 +




(13)




(12)

Equation (12) is known as the general binding isotherm


for a simple 1 : 1 equilibria. It shows that the mole fraction
of the complex follows a hyperbolic relation3 with the free
guest [G] concentration. Equation (12) can also be used as
a starting point to obtain an expression for [G] that only
depends on [H]0 , [G]0, and Ka . Starting from the fact that
we can also write fHG = [HG]/[H]0 or [HG] = fHG [H]0
and combine with (12) to obtain (13).
[HG] =

1
[G] =
2

(10)

In other cases, such as for NMR, the observed physical


change is a function of the mole fractions = f of the host =
fH , guest = fG , and complex = fHG according to (11).
Y = YH fH + YG fG + YHG fHG

Rearrangement of (14) gives a quadratic equation that


has one real relevant solution in the form of (15).

[G]0 + [H]0 +

1
Ka

2

1
Ka

(18)
+ 4[H]0 [G]0

To obtain the two unknown variables, Ka and YHG ,


a titration of some sort is usually carried out, keeping
[H]0 constant, while increasing [G]0 . Using a variety of
software packages or programs, it is then possible to fit the
resulting binding isotherm to (18) by nonlinear regression
methods to obtain the best estimates for Ka and YHG ,
as discussed further below. Equation (18) can be modified
for any method where the physical property of interest is
depended on concentration. In some situation, for example,
NMR, the physical property is not concentration, but rather
mole fraction, dependent with the mole fraction fx depends
on the total concentration of species X = [X]0 according
to (19).
fX =

[X]
[X]0

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

(19)

Binding constants and their measurement


Starting from (11) and (19) and assuming YG is silent,
we use [H] = [H]0 [HG] again from (5) to obtain (20).




[H]0 [HG]
[HG]
Y = YH
+ YHG
[H]0
[H]0


[HG]
(20)
= YH + (YHG YH )
[H]0
As before, we rearrange (20) and define Y = Y YH
and YHG = YHG YH to obtain (21).


[HG]
(21)
Y = YHG
[H]0
Combining this with (9) then gives us (22).


1
YHG 1
Y =
[G]0 + [H]0 +
[H]0 2
Ka





1 2
+ 4[H]0 [G]0 (22)

[G]0 + [H]0 +

Ka
Equation (22) is very similar to (18) and the two
unknown variables, Ka and YHG , can be obtained in a similar manner by titration and fitting the results data to (22)
using a nonlinear regression program.
What happens if the guest is not silent (YG = 0)?
As this is only going to be likely in cases where the
physical property is concentration dependent (e.g., UVvis
spectroscopy), we start again from (10) using the same
approach as with (16) to obtain (23).
Y = YH ([H]0 [HG]) + YG ([G]0 [HG]) + YHG ([HG])
= Y0 + YG [G]0 + (YHG YH YG )[HG]

(23)

Here we again used Y0 = [H]0 YH but now we define


YHG = YHG YH YG and as before Y = Y YH to
obtain (24).
Y = YG [G]0 + YHG ([HG])

(24)

This equation then yields (25) after substitution of [HG]


with the right-hand side of (9).


1
1
Y = YG [G]0 + YHG
[G]0 + [H]0 +
2
Ka





1 2
+ 4[H]0 [G]0 (25)

[G]0 +[H]0 +

Ka
Equation (25) can then be treated exactly as (18) except
we now have one more unknown variable, YG , to deal with

but this does not complicate the fitting process significantly.


Alternatively, YG can be determined independently, for
example, as the molar absorptivity of the free guest = G
and the resulting value used in (25) to simplify the fitting
process.
Equations (18), (22), and (25) can be modified for almost
any form of a titration or a similar study where the
observed physical changes are proportional to the change
in concentration or mole fraction of the 1 : 1 complex. To
illustrate this, the application of these equations in NMR,
UVvis, fluorescence, and calorimetry is detailed below.

2.2.1 NMR measurements of 1 : 1 equilibria


The most important technique for determining binding constants in modern supramolecular chemistry is undoubtedly
NMR, particularly 1 H NMR. Several factors contribute to
the popularity of NMR; the quality of information regarding
the hostguest interactions (shifts in spectra, nOe-contacts,
etc.) and availability being the most important ones (see
also NMR Spectroscopy in Solution, Techniques).
In essence, there are two ways in which one can obtain
binding constants from NMR experiments and these depend
on whether the system can be described as being in slow or
fast exchange on the NMR timescale. The most commonly
used approach assumes that the species of interest are in
fast exchange, where, in the case of hostguest equilibria,
the observed NMR resonance () is the weighted average
of the bound (HG) and unbound host (H) species.5 This
implies that we can immediately apply equations such as
those discussed in Section 2.2 that are dependent on the
relative mole fractions of species of interest. If we define the
NMR resonance for the host as H , the guest as G , and the
hostguest complex as HG , we can also define the change
in resonance for the hostguest complexation as HG =
HG H . If we then define 0 = NMR resonance of the
host before the guest is added (before the start of titration),
we can define the change in physical propertythe change
in resonance as  = 0 . We can now write the NMR
version of our simple 1 : 1 equilibria according to (26).


HG 1
1
 =
[G]0 + [H]0 +
[H]0 2
Ka



2

1
+ 4[H]0 [G]0 (26)

[G]0 + [H]0 +

Ka
Recent work by Leigh, McNab, and coworkers on DDDAAA hydrogen-bonded arrays formed among compounds
2, 5, 6, 7, and 8 in Figure 2(a) illustrates the application of
(26) to determine the relevant association constants from
the NMR-binding isotherms shown in Figure 2(b).12

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

Techniques
NO2

NO2
OEt
H
DDD

AA(A)

O
H

CH3
OEt
H

OEt
O

N
H

N
H
N

N H
H

O
H

N
H

0.6

OEt

N
H

O
N H 2
H

52
O
H
N

H
N

67

0.2

82

0.4

0.0

0.0
(a)

Ka = 8 104 M1

Ka = 6 104 M1

Ka = 2.4 104 M1

(CDCl3, 1H NMR)

(CDCl3, 1H NMR)

(CDCl3, 1H NMR)

(b)

0.4

0.8

1.2
[A]/[D]

1.6

2.0

2.4

Figure 2 (a) Structures, association constants, and (b) binding isotherms of receptor pairs 52, 82, and 67. 1 H NMR titration analyses
performed in CDCl3 using the change in chemical shift () of the amino NH2 groups of 2 (103 M) upon addition of 5 or 8 and
the hydroxyl groups of 7 (103 M) upon addition of 6. The lines indicate best-fitting Ka s for 52 (red), 82 (blue), and 67 (green).12
(Reproduced with permission from Ref. 12. American Chemical Society, 2009.)

For systems in true slow exchange, it is possible to


observe and measure the relative ratios of the free [H]
and bound hostguest species [HG] by integration directly
and then use the mass balance (5) and (6) and then (4) to
determine the binding constant Ka . As an example, Wayland and coworkers recently used this approach to determine the binding constant for methanol to a tetramesityl
rhodium(II) porphyrin, with Ka = 3.3 M1 .13 This situation is, however, relatively rare as complications arising in the slow-to-intermediate exchange region complicate the analysis14 along with the inherent difficulties
in obtaining accurate quantitative integration from NMR
spectra.
For the purpose of determining binding constants, NMR
works well provided (i) the concentration of the host solution is within one or two orders of magnitude from the
millimolar (mM) region and (ii) the association constants of
interest does not greatly exceed Ka = 105 M1 . The former
restriction arises, on the one hand, from the sensitivity of
modern NMR spectrometers but measurements can now
routinely be performed on samples with [H]0 < 104 M
and, on the other, from practical NMR issues with highly
concentrated solutions (>0.1 M). Additionally, it has been
noted that ideally the host concentration in supramolecular
experiments should be as close as possible to the dissociation constant Kd = 1/Ka , which is the inverse of the association constant.2, 15 If the dissociation constant is fairly large
(in other words, the association constant is very small),
large excesses of guest may be required to obtain useful binding data. More importantly, simulations have also
shown that once the dissociation constant is more than two
orders of magnitudes smaller than the host concentration,

the uncertainty from fitting the data becomes too high to


yield any valuable information. For instance, with a [H]0 =
104 M (dilute NMR) and the real Ka = 106 M1 (Kd =
106 M), one could expect up to 3575% uncertainty in the
Ka value obtained.2 This is one reason why Ka  105 M1
obtained by NMR is unreliable; however, equally importantly, assumptions regarding fast exchange may break
down once Ka > 105 M1 .14 Competition experiments16
can circumvent this limitation but this approach is outside
the scope of this chapter (Competition Experiments, Concepts).

2.2.2 UVvis measurements of 1 : 1 equilibria


After NMR, UVvis spectroscopy is probably the second
most important method for determining binding constants.
It has the advantage of being cheap, simple, and relatively
insensitive to minor impurities, provided they are nonabsorbing. With chromophores, such as porphyrins, it is possible to use concentrations in the submicromolar region.
Given what was said above, this implies that binding constants up to Ka = 108 109 M1 can be accurately determined by UVvis spectroscopy.2 The main limitations of
UVvis spectroscopy include the need for a chromophore
and that UVvis titrations need to be carried out within
a concentration range where the absorbance follows the
BeerLambert law (A < 1).
In contrast to NMR, equilibria monitored by UVvis is
always in the slow exchange region as the timescale for
UVvis transitions (<1012 s) is much smaller than the
lifetime of the complex(es) of interest. This implies that

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

lmax ~ 480 nm

S
S

Ka = 6500 M1
(as the TBA salt in CH2Cl2)

S
H2PO4

The molar absorptivity of the guest (G ) can either


determine independently in a separate experiments or
it can be added to the list of unknown parameters
(Ka and HG ) that will be obtained from the fitting
process.

lmax ~ 470 nm

NH

+
NH
N

1
Ka

[G]0 + [H]0 +

+ 4[H]0 [G]0 (28)

2






1
1
A = G [G]0 + HG
[G]0 + [H]0 +
2
Ka

There is a myriad of examples in the literature where


UVvis titrations are used to determine binding constants
in 1 : 1 equilibria systems. For a recent example, see for
instance, the work of Sessler and coworkers on a tetrathiafulvalene diindolylquinoxaline receptor that binds to anions
such as dihydrogen phosphate (Figure 3).17
If the guest is also absorbed at the wavelength of interest
to us, we get (28).

Figure 3 The binding strength of tetrathiafulvalene diindolylquinoxaline to phosphate was determined by a UVvis titration by monitoring the gradual increase in
absorbance at around 480 nm.17

NH
N

NH
N
S
S
Ka



1
1
[G]0 + [H]0 +
A = HG
2
Ka


2


+ 4[H]0 [G]0 (27)


[G]0 + [H]0 +

Ka

the observed absorption spectra are a sum of the spectra of the species in solution. At a given wavelength,
the observed absorbance (A) is therefore the sum of the
absorbance of the host, the guest, and the complex. Each
of these in turn depends on the relative molar absorptivity
() of these species as well as the concentration (c) and
cell path length (b) according to the BeerLamberts law
(A = bc). If we titrate a solution of a host with a guest
solution, the physical change we observe is the change in
absorbance (A), defined as the difference between the
observed (Aobs ) and initial (AH0 ) absorbance of the host or
A = Aobs AH0 . We now recognize that the change in
absorbance depends on two factors; the concentration and
the molar absorptivity of species in solution. Depending on
whether the guest is nonabsorbing (silent) or not, we see
that (18) and (24) could be applied to describe the change in
absorbance if we define the change in molar absorptivity as
HG = HG H , where HG and H are the molar absorptivities of the complex and the host, respectively. When the
guest is nonabsorbing, we obtain (27).

H2PO4

Binding constants and their measurement

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

Techniques

2.2.3 Fluorescence measurements of 1 : 1 equilibria


Fluorescence spectroscopy (see also Luminescent Spectroscopy in Supramolecular Chemistry, Techniques) is
perhaps the most sensitive of all the methods used to
determine binding constants. A range of fluorescent hosts
allow the detection of binding in the submicromolar concentration range, allowing binding constants in excess of
109 M1 to be determined with good accuracy. In fact, it
is essential that the host concentration is kept low in fluorescence spectroscopy titrations as fluorescence response is
only linear to the concentration provided that the absorption of the species at the wavelength used for excitation is
low (A < 0.05).2, 18 In this region, the fluorescence follows
a BeerLamberts-type relation with the observed fluorescence F = kx [X] with kx = the proportionality constant for
species X but kx in turn depends on the incoming light,
molar absorptivity, path length, and quantum yield of the
species of interest.18 The application of the general binding
equations above is complicated by the fact that one needs
to consider whether the fluorescence is quenched or not on
addition of a guest. If it is quenched, two possible types
of quenching are possible, dynamic (collisional) and static
quenching, where only the latter has any real significance
in supramolecular chemistry. We shall first consider the situation when there is no dynamic quenching and where the
change in fluorescence (F ) on addition of a guest can
be described as the difference between the observed (Fobs )
and initial (F0 ) fluorescence (F = Fobs F0 ). Assuming
first that the quest is nonfluorescent (silent) and using the
same approach as with UVvis spectroscopy, we define
the change in proportionality constant (kHG ) as the difference between the proportionality constants of the complex (kHG ) and host (kH ) or kHG = kHG kH . We now
obtain (29).2


1
1
[G]0 + [H]0 +
F = kHG
2
Ka


1
[G]0 + [H]0 +
Ka

2

+ 4[H]0 [G]0

(29)
When the guest is also fluorescent (and there is no
dynamic quenching), we obtain (30).


1
1
[G]0 + [H]0 +
F = kG [G]0 + kHG
2
Ka



2

1
+ 4[H]0 [G]0 (30)

[G]0 + [H]0 +

Ka

If, on the other hand, both the guest and the host are
also nonfluorescent and only the complex is fluorescentin
other words, fluorescence is turned on by addition of the
guest to form a fluorescent complex, we could simplify (29)
to give us (31).


1
1
Fobs = kHG
[G]0 + [H]0 +
2
Ka





1 2
+ 4[H]0 [G]0 (31)

[G]0 + [H]0 +

Ka
Likewise, if the guest completely quenches the host
fluorescence and dynamic quenching does not play a role,
we would get (32).


1
1
[G]0 + [H]0 +
Fobs = kH
2
Ka





1 2

+ 4[H]0 [G]0 (32)


[G]0 + [H]0 +

Ka
However, if dynamic quenching is suspected to play a
role, we need to modify this analysis. We start by noting that pure dynamic quenching is usually described by
the SternVolmer relation, which states that the fluorescence intensity ratio F0 /Fobs = 1 + KSV [Q] where KSV is
the SternVolmer constant and as always, [Q] = free concentration of the quencher. If we wanted to describe static
quenching or binding in the same way, that is, as a function
of the guest/quencher concentration and the fluorescence
intensity ratio F0 /Fobs , we would get an almost identical
expression in the form (34).
F0
= 1 + Ka [G]
Fobs

(33)

Inserting [G] from (20) into the static quenching expression above would then give us (34).4


1
F0
1
[G]0 [H]0
= 1 + Ka
Fobs
2
Ka




1 2
[G]0
+4
+
[G]0 [H]0 +
(34)
Ka
Ka
When the host is fluorescent and the addition of a guest
results in both dynamic and static quenching, we can write a
SternVolmer-like description of the observed fluorescence
intensity ratio Fobs /F0 according to (35).4
kH /kH0 + (kHG /kH0 )Ka [G]
Fobs
=
F0
1 + Ka [G]

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

(35)

Binding constants and their measurement


Note here the two different proportionality constants for
the host, kH and kH0 , with the latter referring to the initial
proportionality constant for the host, that is, in the absence
of any guest, while the former refers to the proportionality
constant of the host after a guest has been added. Also,
please note that we now use the Fobs /F0 ratio in (36) instead
of the F0 /Fobs ratio used in the SternVolmer equations for
dynamic and static quenching (29). Owing to complication
arising from dynamic quenching, the two constants kH and
kH0 may not be equal. However, if it is, we can assume
kH = kH0 and we get (36).12
Fobs
1 + (kHG /kH0 )Ka [G]
=
F0
1 + Ka [G]

(36)

The conditions for this equation are exactly the same


as those for (29) and fitting data to (29 and 36) should
give the same resultswhich equation should be used is
just a matter of preference or fitting program. We also see
that when the hostguest complex fluorescence is fully
quenched (kHG = 0), we obtain the SternVolmer (33) for
static quenching back. Equation (36) can also be modified
to include the situation when the guest is fluorescent to give
(37).12
1 + (kHG /kH0 )Ka [G] kG
Fobs
+ 0 [G]
=
F0
1 + Ka [G]
kH

(37)

This equation should give the same results as (30).


The above-mentioned paper by Leigh, McNab, and
coworkers also illustrates the power of fluorescence titration
as they were able to use (36) to measure Ka > 1010 in their
cationic DDD-AAA array (Figure 4).12

2.2.4 Calorimetric measurements of 1 : 1 equilibria


In supramolecular calorimetric titrations, the enthalpic
(H ) changes on adding a guest to a host are measured and
14

N
H

N+
H

N
H

Ka = 3 1010 M1
(CH2Cl2, 298 K)

the resulting binding isotherm then fitted to obtain the binding constant (Ka ) and hence the free energy change (G)
in the system. Calorimetry can therefore yield all the key
thermodynamic parameters G, H , and S (the latter
from the relation S = (H G)/T in one experiment.
This makes calorimetry, usually in the form of isothermal
calorimetry (ITC, see Isothermal Titration Calorimetry
in Supramolecular Chemistry, Techniques), one of the
most powerful methods that exists for determining binding
constants in supramolecular chemistry. The main reasons
that it is not used more widely than it is are the equipment
cost and real or imagined issues regarding compatibility
with organic solvents. The sensitivity of the method also
means that special care has to be taken to eliminate or correct for dilution and other interfering environmental factors.
In calorimetry, the heat (Q) formed or absorbed on
adding the guest is measured. This change in the system
is proportional to the concentration of the hostguest complex ([HG]), the total volume of the solution (V ), and the
molar enthalpy for the complex formation (HHG ). We
immediately see that this is just a special case of the absolute concentration depending equations for binding (18) and
(19). For calorimetry titrations with 1 : 1 equilibria, we can
(after correcting for dilution by the guest) therefore use
(38).


1
1
[G]0 + [H]0 +
Q = HHG V
2
Ka





1 2

+ 4[H]0 [G]0 (38)


[G]0 + [H]0 +

Ka
In their work on cooperative multipoint binding interactions, Hunter and coworkers used calorimetry to determine the association constant for certain zinc(II) porphyrinpyridine complexes (Figure 5).19 The reason why
Ka = 3 1010 M1

12
0.52

10

Fluor@406 nm

Fluorescence@406 nm
(104)

10+[B(3,5-(CF3)2C6H3)4 ]

8
6
4

400
200
0

Mole fraction of 10+

1.0

0
0

1.0

2.0

3.0

[10+] nM

Figure 4 Fluorescence intensities of 6 (1 1010 M) at 406 nm in CH2 Cl2 at 293 K (excitation = 395 nm) on addition of 10+ [B(3,5(CF3 )2 C6 H3 )4 ] (0 2.5 equivalents), maintaining the concentration of 6 constant, using a 1 : 1 complexation model. (Inset) Job
plot under the same conditions as in the titration experiment.12 (Reproduced with permission from Ref. 12. American Chemical
Society, 2009.)
Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

10

Techniques

R
NH

O
O

N
N

Zn

N
H

N
Ka

O
N

O
+

R
NH

HN

O
HN

N
Zn
N

N
H
N

HN
HN
O
R

R=

Figure 5

Ka = 790

M1 in

O
R

CHCl3

Ka = 940 M1 in CHCl2CHCl2

The binding strength of the above porphyrin host to a pyridine guest in different solvents as determined by ITC.19

these authors used calorimetry was to overcome problems


due to aggregation-induced NMR spectra broadening of the
host in some of the solvents used.

of the 1 : 2 binding system. The overall stability constant


and the stepwise binding constants are linked by (42).
12 = K1 K2

2.3

The 1 : 2 equilibria

The 1 : 2 equilibria is another very important special case


of (1) with n = 1 and m = 2. There are of course two ways
of viewing it; as a stepwise process of two guest molecules
binding to one host (Figure 1c) or as an overall equilibrium
between the two guest molecules and a host (Figure 1a).
The expressions for the stepwise equilibrium constants with
first = K1 and second = K2 binding constants are shown
in (39) and (40), respectively.
[HG]
[H][G]
[HG2 ]
[HG2 ]
K2 =
=
[HG][G]
K1 [H][G]2
K1 =

(39)

12 =

[HG2 ]
[H][G]2

We further discuss the relationship between K1 and K2


in terms of cooperative binding below but first we define
the key equations that link K1 and K2 to a physical
change in a typical supramolecular chemistry experiment
(e.g., titration).

2.3.1 Basic 1 : 2 equilibria


The problem here is analogous with the situation described
for 1 : 1 complexation. The key issue is that we cannot usually determine the concentration of the species in (39) and
(40) directly, for example, the 1 : 1 intermediate complex
[HG]. We start with the mass balance (43) and (44).

(40)

Here we made use of [HG] = K1 [H][G] from (39) on the


right side of (40). The overall stability constant from (3) is
then defined by (41).
(41)

In some circumstances, it is easier to use the overall


binding constant than the stepwise one; however, it must be
remembered that it is extremely unlikely that a termolecular
complex HG2 can be formed by simultaneous collision of
one host and two guest molecules. Hence, the stepwise
binding constants are a much better physical description

(42)

[H]0 = [H] + [HG] + [HG2 ]

(43)

[G]0 = [G] + [HG] + 2[HG2 ]

(44)

Using (39) and (40), we can write the two latter


terms in (43) and (44) as [HG] = K1 [H][G] and [HG2 ] =
K1 K2 [H][G]. Isolating for [H] and [G], we can now
rewrite (43) and (44) and obtain (45) and (46).
[H] =

[H]0
1 + K1 [G] + K1 K2 [G]2

[G] = [G]0 K1 [H][G] 2K1 K2 [H][G]2

(45)
(46)

We have deliberately chosen two slightly different approaches here to isolate for [H] and [G]. We now substitute

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

Binding constants and their measurement


for [H] in (46) using (37) to obtain after rearranging
terms (47).2, 5

Finally, we substitute for [H] from (45) to obtain (50).


Y =

[G]3 (K1 K2 ) + [G]2 {K1 (2K2 [H]0 K2 [G]0 + 1)}


+ [G]{K1 ([H]0 [G]0 ) + 1} [G]0 = 0
(47)
Equation (47) is a cubic equation containing two known
([H]0 and [G]0 ) variables and two unknowns (K1 and
K2 ). This cubic equation has three solutions that may or
may not include complex numbers. The smallest positive
real solution is the only one of relevance here, and
it can be obtained by certain algorithms in a number
of software packages. Once the concentration of [G] is
known, the concentration of [H] could also be calculated
from (45).
It should be noted that, instead of using (47) to obtain
[G] and then [H] from (45), it is possible to use the method
of successive approximation with (45) and (46). First, an
initial guess of [G] (and of course K1 and K2 ) by some
approximation of (46) is made and the corresponding [G]
is then used to calculate [H] from (45). The new [H] value
would then be used to refine [G] in (46) and the process
then repeated a couple of times or until [H] and [G]
converge to a steady value. Many software packages use
this approach instead of solving the cubic (47); however,
this requires a fairly good initial estimate of the parameters
used in these iteration processes.
To link (43) and (47) to the observed physical changes,
we use a similar approach as used above to obtain (20).
Assuming first that the guest is silent, we can describe
the observed physical properties that depend on the mole
fraction of individual species in solution as a function of the physical property of the host = YH , the 1 : 1
complex = YHG , and the 1 : 2 complex = YHG2 according
to (48).


[H]0 [HG] [HG2 ]


Y = YH
[H]0


[HG2 ]
+ YHG2
[H]0


+ YHG

[HG]
[H]0

YHG K1 [G] + YHG2 K1 K2 [G]2


1 + K1 [G] + K1 K2 [G]2

Y =

[H]0 (YHG K1 [G] + YHG2 K1 K2 [G]2 )


1 + K1 [G] + K1 K2 [G]2

Y = YHG
=

[HG]
[H]0


+ YHG2

(51)

In both (50) and (51), we obtain [G] from the solution


to the cubic (47). Equations (50) and (51) contain two
additional unknown variables, YHG and YHG2 , to the other
two unknown (K1 and K2 ) and known ([H]0 and [G]0 )
from (47). We can obtain these again by carrying out a
titration and fit, resulting in a binding isotherm by nonlinear
regression methods to (47) and (50) or (51) to obtain the
best estimates for the unknown variables. In the case of
statistical 1 : 2 binding, it is also possible to simplify the
process a little further, noting that K1 = 4K2 as seen in the
following discussion on cooperativity and -value.

2.3.2 Cooperativity and -value


How do the stepwise binding constants K1 and K2 relate
to the binding constant for 1 : 1 equilibria Ka ? To explain
this, we try to link the stepwise processes to 1 : 1 equilibria
by relabeling a host (H) with two identical binding sites
with the labels A and B (Figure 6) as they were two hosts
with one binding site each.2 When the first guest (G) binds
to this host, it can bind either to site A to form a HG1A B
(the prime  indicates which site is occupied) complex or
to site B to form HG1AB , with the equilibria described
by the binding constants K1A and K1B , respectively. When
the second G binds to the remaining sites in HG1A B and
HG1AB , the product in both case is HG2A 2B , described by
the binding constants K2A and K2B (Figure 6).

(48)

Defining Y = Y YH , YHG = YHG YH , and


YHG2 = YHG2 YH in (48) and making use of (39) and
(40), we obtain (49).


(50)

If the physical property depends on absolute concentration (e.g., UVvis), the only change we need to make is to
multiply through (50) with [H]0 to obtain (51).

K1A

11

[HG2 ]
[H]0

(49)

HG1AB

HGA1B

K2A

HG2AB

K1B

YHG K1 [H][G] + YHG2 K1 K2 [H][G]2


[H]0

HAB

K2B

Figure 6 A
schematic
explaining
the
microscopic
(K1A , K1B , K2A , and K2B ) association constants involved
in the stepwise formation of a 1 : 2 complex.2 (Reproduced from
Ref. 2. Royal Society of Chemistry, 2011.)

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

12

Techniques

When the binding sites A and B are truly identical,


we cannot distinguish between HG1A B and HG1AB and,
hence, determine K1A and K1B (or K2A and K2B ) as the
physical changes analogous to, for example, YHG in (17)
for sites A and B in HG are identical. We can, however,
relate the binding constants K1A , K1B , K2A , and K2B to the
overall stepwise constants K1 and K2 . From (39), we start
with [HG] = [HG1A B ] + [HG1AB ] to obtain (52).
[HG1A B ] [HG1AB ]
[HG1A B ] + [HG1AB ]
=
+
[H][G]
[H][G]
[H][G]
(52)
= K1A + K1B

K1 =

We now make of use of (1) to show that the equilibrium


constants K2A and K2B can also be used to give [HG1A B ] =
[HG2A B ]/(K2A [G]) and [HG1AB ] = [HG2A B ]/(K2B [G]).
Further, we can see that just as [HG] = [HG1A B ] +
[HG1AB ], [HG2 ] = [HG2A B ]. Combining these facts, we
can now use (40) to obtain (53).

K2 =

(56)

Connors pointed out that (56) can be used to define


cooperativity or the lack of it.4, 23 He defined the interaction
parameter = , according to (57).
K1
4K2

(57)

Alternatively, as Anderson and Hunter pointed out, the


interaction parameter = simply reflects the ratio of the
microscopic binding constants K1m and K2m according
to (58).20

(53)

If the binding sites A and B are identical, then it follows


immediately that K1A = K1B and K2A = K2B . We define
this quantity as the first microscopic binding constant =
K1m with K1m = K1A = K1B . Likewise, it follows that the
second microscopic binding constant can be defined as =
K2m with K2m = K2A = K2B . It is not possible to measure
these microscopic binding constants directly but it is easy
to link them to the macroscopic stepwise binding constants
K1 and K2 using (52) and (53) to obtain (54) and (55).
K1 = 2K1m
K2m
K2 =
2

K1m
K1
K1
K2m
=
=
=
2
2
22
4

[HG2 ]
K2 =
([HG1A B ] + [HG1AB ])[G]
[HG2 ]

=
[HG2A B ] [HG2A B ]
+
[G]
K2A [G]
K2B [G]
[HG2 ]
[HG2A B ](K2A + K2B )
K2A K2B
K2A K2B
=
K2A + K2B

can also explain the statistical factors in (47) and (48) from
symmetry numbers2022 based on the degeneracy of the
intermediates HG1A B and HG1AB .
We are now ready to look at the possible connection
between the macroscopic binding constants K1 and K2 . We
first consider the situation where there is no change in the
empty remaining site in HG1A B or HG1AB when the first
guest G binds and no specific interaction (e.g., electrostatic)
between two molecules of G bound to HG2A B . In this case,
it is clear that K1m = K2m . This situation describes classical
noncooperative binding in a 1 : 2 system. Using (54) and
(55) above and noting that K1m = K2m and K1m = K1m /2
and we obtain (56).

(54)
(55)

The factor of 2 in (55) and (56) can also be explained


on kinetic grounds in relation to (2) as there are two ways
for the guest G to bind to the empty HAB host so that
the observed on-rate (k1 ) appears twice as fast for the first
binding.3 Likewise, there are two ways for the guest G
to come off (k2 ) the complex HG2A B . Alternatively, we

K1m
K2m

(58)

The power of the interaction parameter as defined


by either (57) or (58) is that by determining K1 and K2
(or K1m and K2m ), it is possible to quantify the observed
cooperativity according to three possible scenarios:

If = 1, it follows that K1m = K2m and the binding


is noncooperative.
If > 1, then K1m < K2m and the system displays
positive cooperativity.
If < 1, then K1m > K2m and the system display
negative cooperativity.

It is important to realize the difference between the


macroscopic K1 and K2 and the microscopic binding
constants K1m and K2m in (57) and (58) and that for
noncooperative binding we would expect the latter to be
identical. Because of the stepwise free energy changes upon
1 : 2 binding, it is useful to compare the stepwise binding
constants in terms of free energy changes (G) according
to (59) and (60).24
G1 = RT ln K1m = RT ln(K1 /2)

(59)

G2 = RT ln K2m = RT ln(2K2 )

(60)

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

Binding constants and their measurement


We can now also define cooperativity or the lack of it
in terms of the free energy difference = G12 for the
stepwise binding energies according to (61).24
G12 = G2 G1

(61)

Here, G12 = 0 for noncooperative 1 : 2 binding,


G12 > 0 for negative cooperativity, and G12 < 0
for positive cooperativity.
One consequence of the above discussion is that if
a system is considered to display noncooperativity, the
number of unknown parameters in (47), and hence (50)
and (51), can be reduced by 1 by using K1 = 4K2 .
It is then easy to see from (42) that 12 = K1 /4 =

4K2 and, hence, K1 = 2 12 and K2 = ( 12 )/2. We


can use this to rewrite (47), (50), and (51) as (62)(64)
with (63) describing systems where the observed physical change depends on mole fractions and (64) describing systems where it depends on absolute concentrations.


[G]3 ( 12 ) + [G]2 (2 12 [H]0 12 [G]0 + 12 )



+ [G]{2 12 ([H]0 [G]0 ) + 1} [G]0 = 0

YHG 2 12 [G] + YHG2 12 [G]2


Y =

1 + 2 12 [G] + 12 [G]2

[H]0 (YHG 2 12 [G] + YHG2 12 [G]2 )


Y =

1 + 2 12 [G] + 12 [G]2

(62)
(63)

(64)

It should be noted here that any 1 : 2 equilibria can


be fitted to (62)(64) instead of (47) and (50) or (51),
regardless of cooperativity or lack of it. As discussed
below, it is good practice to fit suspected 1 : 2 equilibria
to both scenarios and then compare the outcomes in
order to determine if a cooperative model with K1 =
4K2 is justified. It is possible to simplify (62)(64) even
further if we assume that the changes in the physical
property Y of interest are additive, that is, YHG2 =
2YHG . In this case, (63) and (64) would be simplified
to yield (65) for systems with Y depending on mole
fractions and (66) for system with Y depending on absolute
concentrations.

YHG2 [G]( 12 + 12 [G])


Y =
(65)

1 + 2 12 [G] + 12 [G]2

YHG2 [H]0 [G]( 12 + 12 [G])


Y =
(66)

1 + 2 12 [G] + 12 [G]2
In principle, one could also apply the simplification
YHG2 = 2YHG in (50) and (51) that explicitly use the
stepwise binding constants K1 and K2 using (67) with Y

13

depending on mole fraction and (68) for Y depending on


absolute concentrations.
Y =
Y =

YHG K1 [G](1 + 2K2 [G])


1 + K1 [G] + K1 K2 [G]2
YHG K1 [H]0 [G](1 + 2K2 [G])
1 + K1 [G] + K1 K2 [G]2

(67)
(68)

Note that here we used YHG instead of YHG2 in (65)


and (66) to simplify the outcome in (67) and (68) further.
It should be noted that this situation is probably rare as
it is quite common for YHG2 = 2YHG in cooperative
systems where K1 = 4K2 . Noncooperative systems (K1 =
4K2 ) also quite often show YHG2 = 2YHG , for example,
in NMR titration experiments where the addition of the
second guest influences the observed chemical shift of
the system in a nonlinear manner due to ring-current or
induction effects, even in the absence of cooperativity.

2.3.3 NMR measurements of 1 : 2 equilibria


Equation (50) provides the foundation for measuring 1 : 2
equilibria by NMR. As with the 1 : 1 equilibria, we first
define HG2 as the difference between the NMR resonances between the 1 : 2 hostguest complex ( HG2 ) and
the host NMR resonance ( H ), that is HG2 = HG2 H .
Using HG = HG H for the change in NMR resonance
for the 1 : 1 complex formation and the observed change in
resonance as  = 0 similar to the 1 : 1 equilibria, we
obtain (69).2, 5
 =

HG K1 [G] + HG2 K1 K2 [G]2


1 + K1 [G] + K1 K2 [G]2

(69)

Here the guest [G] concentration is obtained from (47)


as illustrated in the example below of nitrate binding to a
ditopic pyromellitamide host (Figure 7), where fitting the
titration data into a 1 : 2 binding model gave much better
results than a 1 : 1 binding model.25
In many cases, it is appropriate to simplify the 1 : 2
binding equations in NMR titration by assuming that the
binding is statistical (K1 = 4K2 ), that the NMR resonance
changes are additive ( HG2 = 2 HG ), or both. In these
cases, we can use (70), (71), and (72).

HG 2 12 [G] + HG2 12 [G]2


 =

1 + 2 12 [G] + 12 [G]2

HG2 [G]( 12 + 12 [G])


 =

1 + 2 12 [G] + 12 [G]2
 =

HG K1 [G](1 + 2K2 [G])


1 + K1 [G] + K1 K2 [G]2

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

(70)

(71)

(72)

14

Techniques
O

O O
O

N
H

H
N

O
O O

K1

N
H

H
N

O O
O

O
NO3
O

O O

NO3

K2

NO3

O
O O
O

N
H

H
N

NO3

O3N

O O

(a)

O
8.6

Fitted to a 1 : 2 equilibria:
(Black isotherm)

8.55

K1 = 457 41 M1

dNH

8.5

K2 = 65 3 M1

8.45

a = 0.57

8.4
8.35

Fitted to a 1 : 1 equilibria:
(Red isotherm)
Ka = 131 32 M1

8.3
8.25
(b)

10
20
30
Equivalent NO3 added

40

Figure 7 Comparison of the fitted isotherms for the binding of nitrate to a pyromellitamide host.25 (a) The structures of the
pyromellitamide host and the resulting 1 : 1 and 1 : 2 complexes. (b) The fitted binding isotherms and the resulting binding isotherms for
1 : 1 (red dotted line) and 1 : 2 equilibria (black solid line). Note the difference in estimated uncertainty for the 1 : 1 and 1 : 2 equilibria.

Here (70) describes statistical 1 : 2 binding, (71) statistical binding where the NMR resonances are additive and
(72) where the 1 : 2 binding is not statistical but NMR resonances are additive. For (70) and (71), we use (62) to obtain
the free guest [G] concentration instead of (47).

2.3.4 UVvis measurements of 1 : 2 equilibria


For analysis of 1 : 2 equilibria by UVvis spectroscopy,
(51), (64), (66), and (68) can be applied after minor
modifications. First we define the molar absorptivity change
for the formation of a 1 : 2 complex (HG2 ) as the
difference between the molar absorptivity of the 1 : 2
complex ( HG2 ) and that of the free host (H ) or HG2 =
HG2 H . The change in absorbance is then defined as for
1 : 1 equilibria as A = Aobs AH0 . This gives (73), (74),

(75), and (76).


A =

[H]0 ( HG K1 [G] + HG2 K1 K2 [G]2 )

1 + K1 [G] + K1 K2 [G]2

[H]0 ( HG 2 12 [G] + HG2 12 [G]2 )


A =

1 + 2 12 [G] + 12 [G]2

HG2 [H]0 [G]( 12 + 12 [G])


A =

1 + 2 12 [G] + 12 [G]2
A =

HG K1 [H]0 [G](1 + 2K2 [G])


1 + K1 [G] + K1 K2 [G]2

(73)

(74)

(75)

(76)

In all these cases, the free guest concentration [G]


is obtained from (47). The choice between the above
equations then depends on whether the binding is expected

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

Binding constants and their measurement


to be statistical [(74 and (75)], nonstatistical [(73) and (76)],
and if the molar absorptivities are additive [HG2 = 2 HG
(75) and (76)], or not [HG2 = 2 HG , (73) and (74)]. As
before, we use (62) instead of (47) to obtain the free guest
[G] concentration for (74) and (75).
Ballester and coworkers used UVvis titrations to determine the 1 : 2 binding constants for a dizinc-bisporphyrin
host and a 4-pyridyldiphenylphosphine guest (Figure 8),
concluding that it was statistical ( 1).26

15

the molar absorptivity of the 1 : 2 complex (kHG2 ) and that


of the free host (kH ) or kHG2 = kHG2 kH . The change
in fluorescence then defines on 1 : 1 equilibria as F =
Fobs F0 . This gives (77), (78), (79), and (80).

F =

2.3.5 Fluorescence measurements of 1 : 2 equilibria


Ignoring the complications arising from dynamic quenching, a 1 : 2 equilibria can be analyzed by fluorescence
spectroscopy in a manner similar to the case of UVvis
spectroscopy using the corresponding versions of (51), (64),
(66), and (68). We start by defining the change in molar
proportionality constant (kHG2 ) as the difference between

[H]0 (kHG K1 [G] + kHG2 K1 K2 [G]2 )

(77)

1 + K1 [G] + K1 K2 [G]2

[H]0 (kHG 2 12 [G] + kHG2 12 [G]2 )


F =

1 + 2 12 [G] + 12 [G]2

(78)

kHG2 [H]0 [G]( 12 + 12 [G])


F =

1 + 2 12 [G] + 12 [G]2

(79)

F =

kHG K1 [H]0 [G](1 + 2K2 [G])

(80)

1 + K1 [G] + K1 K2 [G]2

R1
R

N
N

Zn N

Ph

Zn

Ph

R1

Zn

Zn

Ph
=

H
N

+ N

Ph

Ph

K1

Ph
Zn

Zn
N
N Zn
R1

K2

Ph

Ph

O
N
H

Zn

Ph
P
Ph

R1
N

R1 =

N
R

C5H11

(a)
0.4
0.35
Fitted to a 1 : 2 equilibria:

Absorbance

0.3
0.25
0.2

K1 = 6300 M1

0.15

K2 = 1600 M1

0.1

a = 1.01

0.05
0
450
(b)

500

550

600

650

Wavelength (nm)

Figure 8 UVvis spectroscopic determination of a 1 : 2 equilibria. (a) The structures of the dizinc(II) bisporphyrin host and the 4pyridyldiphenylphosphine guest used and the corresponding stepwise equilibria. (b) The UVvis titration of a 5.16 105 M toluene
solution of the dizinc(II) bisporphyrin with 4-pyridyldiphenylphosphine (0135 equivalents).26 (Reproduced with permission from
Ref. 27. American Chemical Society, 2009.)
Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

16

Techniques
(80)], and if the molar absorptivities are additive [kHG2 =
2kHG , (79) and (80)] or not [kHG2 = 2kHG , (77) and
(78)]. For (78) and (79), we again use (62) instead of (47)
to obtain the free guest [G] concentration.

As before, in all the cases, the free guest concentration


[G] is obtained from (47). The choice between the above
equations then depends on whether the binding is expected
to be statistical [(78) and (79)] or nonstatistical [(77) and

O
O

O
O

+ 2Na+

O
O
O

O
O

COPV

(a)

O
O

200

Fitted to a 1 : 2 equilbria:

Intensity (a.u.)

150

K1 = 1.5 107 M1
K2 = 2.5 105 M1
100

a = 0.07

50

0
450

500

(b)

550

600
650
Wavelength (nm)

700

750

Intensity at 518 nm (a.u.)

190

180

170

160

150

140
0.0
(c)

0.5

1.0
1.5
[Na+]/[COPV]

2.0

2.5

Figure 9 Fluorescence titration of a 1 : 2 equilibria. (a) The structure of the chiral ditopic oligo(p-phenylenevinylene) crown ether
(COPV) binding to sodium (Na+ ). (b) Emission spectra (ex = 453 nm) in chloroform of COPV (4.4 106 M1 ) on addition of
NaPF6 . (c) The corresponding binding isotherm showing the change in emission at 518 nm as a function of [Na+ ]/[COPV]. The
resulting binding constants are also shown.27 (Reproduced from Ref. 18. Royal Society of Chemistry, 2007.)
Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

Binding constants and their measurement


The 1 : 2 binding constants for a ditopic biscrown ether
chiral oligo(p-phenylenevinylene) host (COPV) toward
sodium (Na+ ) was determined by a fluorescence spectroscopy titration (Figure 9) in chloroform.27

17

including L-tryptophan, formed a 1 : 2 complex with this


host with considerable positive cooperativity ( > 10).

2.4

The 2 : 1 equilibria

2.3.6 Calorimetric measurements of 1 : 2 equilibria


Analysis of 1 : 2 binding equilibria by calorimetric titration
is based on (51), (64), (66), and (68) in the same fashion
as illustrated above in the case of UVvis and fluorescence
spectroscopy titration, except that we use the enthalpy
changes for the 1 : 1 (HHG ) and 1 : 2 (HHG2 ) complexes
and need to add the total volume (V ) to these equations.
Thus, for calorimetric titrations (after correcting for dilution
effects as mentioned in Section 2.2.4), we obtain (81), (82),
(83), and (84).


[H]0 (HHG K1 [G] + HHG2 K1 K2 [G]2 )
(81)
Q=V
1 + K1 [G] + K1 K2 [G]2

Q=V

[H]0 (HHG 2 12 [G] + HHG2 12 [G]2 )

1 + 2 12 [G] + 12 [G]2
(82)


HHG2 [H]0 [G]( 12 + 12 [G])
Q=V

1 + 2 12 [G] + 12 [G]2


HHG K1 [H]0 [G](1 + 2K2 [G])
Q=V
1 + K1 [G] + K1 K2 [G]2


(83)

(84)

As before, in all the cases, the free guest concentration


[G] is obtained from (47) and the choice between (81), (82),
(83), and (84) depends on whether the binding is statistical
and/or the enthalpy changes are additive or not. Yet again,
we use (62) instead of (47) to obtain the free guest [G]
concentration for (82) and (83).
Kozbia and Poznanski used ITC to determine the
association constants of various amino acids to a
mannonapto-crown-6-ether (Figure 10).28 The ITC measurements showed that several of the amino acids studied,

In principle, it is possible to use the equations discussed


in Section 2.3 for any 1 : 2 interaction, simply by defining
the host as the species that binds to two guests. However,
in supramolecular titrations, the concentration of one component is usually kept fixed while the other is varied. If
the species that we keep at fixed concentration is actually
the one binding two times to the other species, we can
run into some problems if we insist on defining the system
in such a way that we can use the above 1 : 2 equations.
This is because the concentration of [G] in (47), (50), and
(51) would be fixed within a fairly narrow range during the
course of titration. Additionally, there are also situations
where a host could form either a 1 : 2 or a 2 : 1 complex
(and in extremely complicated cases both).5 It is for these
reasons that it is often worthwhile to define the system so
that the host forms a 2 : 1 complex with the guest. An
early illustrative example comes from the work of Taylor
and Anderson on the binding of a zinc(II) porphyrin host
to the ditopic guest DABCO (Figure 11).29
The resulting equations are analogous but not identical
to the one described for 1 : 2 complexation. We only focus
on the key results here and skip most of the details on
how these equations are derived at. We start by defining
C6H13
Si

t-Bu

t-Bu

t-Bu

N
Si

OH

+
H
OH

t-Bu
2

O
O

C6H13

HN

O
O

t-Bu
N

C6H13
Si

N
Zn
N

Mannonaptho-crown-6-ether

OH

H2N (S )

t-Bu

O
L-Tryptophan

N
Zn
N

t-Bu
N

t-Bu
Si

Figure 10 The host (left) and the guest (right) used by Kozbia
and Poznazski. Isothermal calorimetric titration in methanol
gave K1 = 407 162 M1 and K2 = 1285 307 M1 for the
formation of a 1 : 2 complex between the mannonapto-crown-6ether and L-tryptophan.28

C6H13
Porphyrin2 :DABCO

Figure 11 Structure of the 2 : 1 porphyrin hostDABCO guest


complex reported by Taylor and Anderson.29

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

18

Techniques

the stepwise equilibria according to (85) and (86), which


are of course derived from (1).
[HG]
[H][G]
[H2 G]
[HG2 ]
K2 =
=
[HG][H]
K1 [H]2 [G]
K1 =

(85)
(86)

The mass balance equations for a 2 : 1 equilibria are


slightly different from the corresponding equations for the
1 : 2 equilibria as we can see with (87) and (88).
[H]0 = [H] + [HG] + 2[H2 G]

(87)

[G]0 = [G] + [HG] + [H2 G]

(88)

Consequently, although the process is similar to that


described for a 1 : 2 equilibria, we now move from these
equations to develop (89) for the free concentration of the
host in 2 : 1 equilibria. This key cubic equation is similar
to (47) for 1 : 2 equilibria except that we have swapped H
and G around.
[H] (K1 K2 ) + [H] {K1 (2K2 [G]0 K2 [H]0 + 1)}
+ [H]{K1 ([G]0 [H]0 ) + 1} [H]0 = 0
(89)
We now again use the same approach as with 1 : 2
equilibria to develop equations that link (90) to changes
in physical properties. First we need to realize that, for
the formation of the 2 : 1 complex, we define the changes
in the observed physical property as YH2 G = YH2 G 2YH .
For 2 : 1 equilibria, where the physical changes depend on
mole fractions (e.g., NMR), we obtain (90).
3

Y =

or in fact any 2 : 1 binding, we could also use the overall


binding constant 12 to describe the system with (92)(94).

[H]3 ( 12 ) + [H]2 (2 12 [G]0 12 [H]0 + 12 )

(92)
+[H]{2 12 ([G]0 [H]0 ) + 1} [G]0 = 0

2[G]0 12 [H](YHG + YH2 G 12 [H])


(93)
Y =

[H]0 (1 + 2 12 [H] + 12 [H]2 )

[G]0 2 12 [H](YHG + YH2 G 12 [H])


(94)
Y =

1 + 2 12 [H] + 12 [H]2
Here (93) and (94) are for systems with the physical
changes depending on mole fraction and absolute concentration, respectively. Equations (92)(94) correspond to
the similar (62)(64) for 1 : 2 equilibria. Likewise, we can
write (95)(98) instead of (65)(68) for a 2 : 1 equilibria,
assuming that the changes in physical properties are additive with YH2 G = 2YHG

YH2 G [G]0 [H]( 12 + 2 12 [H])


Y =
(95)

[H]0 (1 + 2 12 [H] + 12 [H]2 )


Y =

YH2 G [G]0 [H]( 12 + 2 12 [H])

1 + 2 12 [H] + 12 [H]2

(96)

Y =

YHG K1 [G]0 [H](1 + 4K2 [H])


[H]0 (1 + K1 [H] + K1 K2 [H]2 )

(97)

Y =

YHG K1 [G]0 [H](1 + 4K2 [H])


1 + K1 [H] + K1 K2 [H]2

(98)

YHG K1 [G]0 [H] + 2YH2 G K1 K2 [G]0 [H]2


[H]0 (1 + K1 [H] + K1 K2 [H]2 )

(90)

Comparing this with (50) reveals notable differences as


both [H]0 and [G]0 feature in this equation as well as a
factor of 2 that can be traced back to (87). Likewise, if the
property depends on absolute concentration (e.g., UVvis),
we have (91).
Y =

[G]0 (YHG K1 [H] + 2YH2 G K1 K2 [H]2 )


1 + K1 [H] + K1 K2 [H]2

(91)

These key equations can also be simplified to account for


noncooperative binding and/or assuming that the observed
changes in physical properties are additive, as was done for
1 : 2 equilibria with (62)(68). For noncooperative binding,

It should be noted here that, as with 1 : 2 equilibria,


we can define cooperativity in 2 : 1 equilibria using the
interaction parameter according to (57) and (58). In
this case, an > 1 implies that the formation of the 2 : 1
complex is favorable over the 1 : 1 complex,30 while < 1
suggests that the 1 : 1 dominates, except when [H]0 is in
excess of [G]0 , for example, at the beginning of a titration.
This explains why the so-called 2 : 1 sandwich complexes
are most easily observed when the ratio of host-to-guest is
2 : 1 (0.5 equivalents of guest added).29, 30
For practical examples of 2 : 1 equilibria, we limit the
discussion to NMR and UVvis titrations as most of the
examples in the literature regarding 2 : 1 equilibria are concerned with either of the two of these techniques. We also
limit the NMR and UVvis discussions to nonstatistical
binding where the physical changes are not additive, that
is, excluding (93)(98) from further discussion.

2.4.1 NMR measurements of 2 : 1 equilibria


In a manner analogous to the above discussion with the
1 : 2 equilibria, we use (90) as the starting point for NMR

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

Binding constants and their measurement


analysis of 2 : 1 equilibria. We first need to define H2 G
as the difference between NMR resonances between the
2 : 1 complex ( H2 G ) and the host NMR resonance ( H ),
that is, H2 G = H2 G H . We also use HG = HG H
for the change in NMR resonance for the 1 : 1 complex
and  = 0 for the observed change in resonances as
before with 1 : 1 and 1 : 2 equilibria to give us (99).
HG K1 [G]0 [H] + 2 H2 G K1 K2 [G]0 [H]2
 =
[H]0 (1 + K1 [H] + K1 K2 [H]2 )

(99)

In their seminal paper on porphyrin ladder formation,


Taylor and Anderson used (99) to analyze the formation
of the 2 : 1 porphyrin2 :DABCO complex shown above
(Figure 11).29 It should be noted that as the first binding
constant (K1 ) was too large to be determined by an
NMR, these authors determined K1 independently from
a UVvis titration at a low concentration of the host,
where the concentration of the 2 : 1 complex would be
negligible throughout the titration experiment. They then
used the K1 value obtained from the 1 : 1 fitting of the
UVvis titration in the fitting process with (99) to obtain
(i) PhCH3

0.0

(ii) CHCI3

d (ppm)

0.1

0.2

0.3

0.0

19

K2 from the NMR titration data. In this case, the 2 : 1


complex formation is clearly observed in both toluene
and chloroform from the dip around 0.5 equivalents of
DABCO added (Figure 12).29

2.4.2 UVvis measurements of 2 : 1 equilibria


For analysis of 2 : 1 equilibria by UVvis spectroscopy,
(91) can be applied after defining the molar absorptivity
change for the formation of a 2 : 1 complex (H2 G ) as
the difference between the molar absorptivity of the 2 : 1
complex ( H2 G ) and that of the free host (H ) or H2 G =
H2 G H . As before, the change in absorbance is then
defined as for 1 : 1 and 1 : 2 equilibria as A = Aobs AH0 .
This gives (100).
A =

[G]0 ( HG K1 [H] + 2 H2 G K1 K2 [H]2 )


1 + K1 [H] + K1 K2 [H]2

(100)

In their work on allosteric assembly, Thordarson and


Rowan used (100) to determine the binding constant of
a zinc(II) porphyrin clip (ZnPorClip) host to DABCO in
the presence of viologen (Figure 13) bound to the internal
cavity of ZnPorClip.31
It should be noted that, to obtain a good fit from
(100), the interaction parameter () and hence the ratio
between K1 and K2 was determined by an integration
of the corresponding NMR spectra, as all the species of
interest were observable with the exchange process being
slow on the NMR timescale. Using (101) from Sanders and
coworkers,32 the resulting concentration then yielded the K1
to K2 ratio and hence .
K1
[HG]2
=
K2
[H2 G][G]

0.5
1.0
1.5
2.0
2.5
3.0
Mole ratio (DABCO)/(Porphyrin)

(101)

(a)
Fitted to a 2 : 1 equilibria:
N

Toluene (PhCH3)

Chloroform (CHCI3)

K1: 1.8 106 M1


K2: 9.9 104 M1

2.7 103 M1
4.9 103 M1

a: 0.006

0.20

(b)

O
O

N
N

Figure 12 (a) 1 H NMR titration of porphyrin (Figure 11) with


DABCO in (i) d8 -toluene and (ii) CDCl3 . is the average change
in chemical shift of the two -protons. The smooth curves are
calculated from (99); dashed lines show the HG values of the
1 : 1 porphyrin:DABCO complex. In curve (ii), there are no
points in the range 0 < mole ratio < 0.5 because the system is in
slow exchange in this region. (b) The resulting binding constants
in toluene and chloroform.29 (Reproduced with permission from
Ref. 29. American Chemical Society, 1999.)

N
Zn N
N

N
DABCO

N
N

N
Viologen

ZnPorClip

Figure 13 The zinc(II) porphyrin clip (ZnPorClip) host used


by Thordarson and Rowan to form a 2 : 1 complex with DABCO
in the presence of viologen as an allosteric inducer.31

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

20

2.5

Techniques

The 2 : 2 equilibria and other related cyclic


system

balance (104) and (105).


KF =

Similar approaches can be used to develop equations for


more complex equilibria but not only do these become
computationally difficult (solving a quadric or even quintic
equations) but because of the increased number of unknown
parameters (K1 , K2 , K3 . . .), it also becomes difficult to
get any meaningful results from the fitting process. Here,
simplifications become a necessity. An illustrative example
comes from studies on 2 : 2 complexes29, 30, 33 that have
sometimes been referred to as sandwich complexes30 The
analysis of this system is simplified by assuming that there
are only two key steps: the formation of the 2 : 2 complex
H2 G2 (Figure 1e) and its subsequent breakup when excess
ligand is added (Figure 14).
We start with the defining two binding constants: the
formation constant = KF and breaking constant = KB for
the H2 G2 sandwich formation. These binding constants are
defined by (102) and (103)29 and the corresponding mass

KB =

Zn

[HG2 ]2
[H2 G2 ][G]2

(102)

(103)

[H]0 = [H] + 2[H2 G2 ] + [HG2 ]

(104)

[G]0 = [G] + 2[H2 G2 ] + 2[HG2 ]

(105)

Using (104) and (105) and rearranging (102) and (103)


gives two related quadratic equation for the free host
([H]) and guest ([G]) concentrations according to (106) and
(107), respectively.

[H]2 (2KF [G]2 ) + [H](1 + [G]2 KF KB ) [H0 ] = 0
(106)

[G]2 ([H] KF KB + 2[H]2 KF ) + [G] [G0 ] = 0
(107)

Ar
C6H13 Si

[H2 G2 ]
[H]2 [G]2

Ar
N

Ar

Zn

N
Si C6H13

N
Ar

Ar =

KF

Ar
N
C6H13 Si

N
C6H13 Si

Zn

Ar
N

Zn

Ar
N

Ar
N

Ar

Ar

Zn

Ar

N
N

Zn

Si C6H13

N
N

Si C6H13

Ar
N

2
N

N
C6H13 Si

KB

Ar

Ar

Zn

N
Ar

Zn

N
N

Si C6H13

Ar

Figure 14 The formation of a 2 : 2 sandwich from a dizinc(II) porphyrin host with DABCO. The equilibria for the formation (KF )
and breaking (KB ) constants of the 2 : 2 complex are shown.29
Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

Binding constants and their measurement

2.6

Self-association: from dimerization to


aggregates

The association of a molecular (A) with itself is the


fundamental step in the formation of many self-assembled
materials, including supramolecular gels and polymers.
Even the simplest of these systems, the self-associated
dimer (A2 ), has been and still is of significant interest to
researchers in supramolecular chemistry.
The binding constants for dimerization and selfassociation equilibria can be determined in a manner similar
to the hostguest complexation equilibria discussed above.
We start this discussion by describing how the binding
constants for dimerization are derived at before turning to
0.25
A725

Absorbance

The solutions to (106) and (107) depend on each other


and cannot be solved analytically; however, it is possible
to solve these by the method of successful approximation
starting by making an initial guess for [G] and using that to
solve for [H] in (107). After a few iterative cycles, [H] and
[G] are then used to calculate the expected changes in the
physical property (Y ). In the case of changes that depend on
absolute concentration (e.g., UVvis), we define YH2 G2 =
YH2 G2 2YH , YHG2 = YHG2 YH , and Y = Y YH to
obtain (108).29

Y = YH2 G2 KF [H]2 [G]2 + YHG2 KF KB [H][G]2
(108)
It is not unusual to obtain formation constants (KF )
from (108) in the range of 1016 1020 M3 while the
breaking constant (KB ) is around 103 106 M1 .27, 29, 33
Note the different units between the two constants (M3
vs M1 ). Here, the choice of reference state7, 8 mentioned
earlier (Section 1) becomes crucialas Ercolani pointed
out, the formation of the 2 : 2 sandwich is promoted when
the concentration of the available binding sites is low.11
For a more detailed discussion about this and other
more complex cyclic systems, the reader is referred to the
excellent work of Ercolani11 and a recent review by Hunter
and Anderson.20
For examples of how the equations above can be applied,
we limit the discussion to UVvis as it is the most practical
method that could be applied to this type of equilibria.

0.20

A668
0.15

0.10
0

10

100

25

30

H2(DABCO)2

H(DABCO)2

Mole fraction (%)

60

40

20

0
0

In the above-mentioned paper by Taylor and Anderson,


the UVvis titration data obtained (Figure 15a) was fitted
to (109). They found KF = 5 1019 M3 and K B = 2
105 M1 in toluene. Under these conditions ([H] 106 M),
the mole fraction of the sandwich species H2 G2 is as high
as 0.8 (Figure 15b).29

20

80

A = H2 G2 KF [H]2 [G]2 + HG2 KF KB [H][G]2


(109)

15
(DABCO) / M

(a)

2.5.1 UVvis measurements of 2 : 2 equilibria


We start here by modifying (108) by defining the change
in molar absorptivity for the 2 : 2 complex (H2 G2 ) as
the difference between the molar absorptivity of the 2 : 2
complex ( H2 G2 ) and that of the free host (H ) or H2 G2 =
H2 G2 H . We also define the change in molar absorptivity
for the HG2 complex ( HG2 ) formed on breaking the
H2 G2 as the difference between the molar absorptivity
of the HG2 complex (HG2 ) and that of the free host or
HG2 = HG2 H . Finally, we define A = Aobs AH0
to give us (109).29

21

(b)

Mole ratio (DABCO) / (H)0

Figure 15 UVvis spectrophotometric titration of the dizinc(II)


porphyrin host (H) shown in Figure 14 with DABCO in toluene:
(a) change in absorption at two wavelengths (668 and 725 nm)
fitted to the calculated curve for the equilibria in Figure 14;
(b) mole fractions of H, H2 (DABCO)2 , and H(DABCO)2 during
the course of the titration, calculated from global factor analysis
of binding curves at all wavelengths in the region 350850 nm at
1-nm intervals.29 (Reproduced with permission from Ref. 29.
American Chemical Society, 1999.)

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

22

Techniques

larger aggregates. The discussion here is limited to linear


(noncyclic aggregates), but for self-association involving
cyclic aggregates, the reader is referred again to the work
of Ercolani10, 11 and a review by Hunter and Anderson.20
To derive these binding constants, we initially follow the
approach outlined by Martin.34

properties associated with the monomer as Y and that of


the dimer as Y . We then use this to give us (116).




[A]
2[A2 ]
+ Y
Y = Y
[A]0
[H]0

2.6.1 Dimerization equilibria

In most cases, we do not simplify this equation further


as the physical property of interest for the monomer (Y )
and the dimer (Y ) are usually both unknown. We therefore
substitute for using (115) in (116) to obtain (117).

We start by considering the monomer (A) to dimer (A2 )


equilibrium where the dimerization constant (KD ) is defined
according to (110).34
KD =

[A2 ]
[A]2

(110)

The mass balance equation for this equilibria is then


described by (111), which can be expanded by substitution
from (110).
[A]0 = [A] + 2[A2 ] = [A] + 2KD [A]2 = [A](1 + 2KD [A])
(111)
We now define the mole fraction of (nonaggregated)
monomer as alpha () according to (112), which can be
expanded using (111).
=

1
[A]
[A]
=
=
[A]0
[A] + 2[A2 ]
1 + 2KD [A]

(112)

Dividing by [A]0 through (111) and using the substitution


= [A]/[A]0 from (112) gives us (113).
[A]
2KD [A]2
[A](1 + 2KD [A])
=
+
[A]0
[A]0
[A]0
= + (2KD [A]) = (1 + 2KD [A])

1=

(113)

Finally, we use [A] = [A]0 from (112) and rearrange


(113) to make it equal to zero to obtain (114), which then
can be rearranged to a classical quadratic equation form.
0 = (1 + 2KD [A]) 1 = (1 + 2KD [A]0 ) 1
= 2KD [A]0 2 + 1

= Y + Y () = Y + Y (1 )

(116)


1 + 8KD [A]0
Y = Y
4KD [A]0



4KD [A]0 + 1 1 + 8KD [A]0
+ Y
(117)
4KD [A]0
1 +

If the physical property of the nonaggregated monomer


(Y ) is known (which is rare) before we can simplify the
above equations further, then subtracting Y from both sides
of (116) and rearranging gives us (118).
Y Y = (Y Y )(1 )

(118)

We now define Y = Y Y and YD = Y Y and


substitute for from (115) in (118) to give us (119).



4KD [A]0 + 1 1 + 8KD [A]0
Y = YD
(119)
4KD [A]0
Note that we could also use (118) to write an alternative
version of (117) in the form of (120).



4KD [A]0 + 1 1 + 8KD [A]0
Y Y = (Y Y )
4KD [A]0
(120)
Starting from the mass balance (111), it is also possible
to derive a different set of equations to analyze dimerization
by rearranging it into a quadratic equation in the form
of (121).

(114)

This quadratic equation has only one relevant solution


according to (115).

1 + 1 + 8KD [A]0
=
(115)
4KD [A]0
To link (115) to a change in physical property (Y ) that
depends on the mole fraction (e.g., NMR), we use a
similar approach as in the earlier sections on 1 : 1 and
1 : 2 equilibria. We start by defining the mole fraction of
the dimer () as = 1 . We then define the physical

0 = 2KD [A]2 + [A] [A]0

(121)

This equation has one relevant real solution in the form


of (122).

1 + 1 + 8KD [A]0
(122)
[A] =
4KD
Dividing through (122) with [A]0 would give us A,
taking us back to (115) and eventually to (117), (119), and
(120), which all describe physical changes that depend on
mole fractions (e.g., NMR). Note that it is also possible to

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

Binding constants and their measurement

NMR has been reached (<0.1 mM in some cases). The


change in the observed NMR resonance () of interest
can then be described by a modification of (117), (119),
or (120), depending on individual preferences (or the program used). We define the NMR resonance of the monomer
as and that of the dimer as . For (119), we then define
the change in NMR resonance ( D ) upon dimerization
as D = . Using these definitions, we now obtain
(126), (127), and (128).

derive a quadratic equation from (111) that is expressed in


terms of the concentration of the dimer [A2 ].
To obtain equations that describe changes in physical
properties upon dimerization that depend on absolute concentration (e.g., UVvis), we start by defining (123) for
the change in that physical property that depends on absolute concentration noting that the concentration of the dimer
[A2 ] = [A]0 /2.

 

Y = Y [A] + Y [A2 ] = [A]0 Y + Y


2



1
= [A]0 Y + Y
(123)
2

(124)

Multiplying by [A]0 through (1 )/2 and using (115)


for then gives us (125) from (124).

Chen and coworkers used the NMR dilution method to


determine the dimerization of their amidourea compound
(Figure 16).35 This compound formed strong dimers (KD =
4.4 104 M1 ) in 99 : 1 CDCl3 /DMSO-d6 (v/v) but, as the
percentage of DMSO-d6 was increased, the dimerization
constant dropped to KD = 15 M1 with 20% DMSO-d6
present.

(125)
We limit our discussion to the application of these
equations in NMR and UVvis dilution experiments for
the determination of monomerdimer equilibria.

2.6.2 NMR measurements of dimerization equilibria

2.6.3 UVvis measurements of dimerization


equilibria

The most commonly used method for analyzing dimerization equilibria is the method of NMR dilution. This involves
taking a fairly concentrated sample (>10 mM if possible)
of molecule A, measuring the NMR, and then performing a series of dilution step until the detection limit for
C8H17

C8H17
O

For measurements of a dimerization equilibria by UVvis,


we can apply (102) with slight modifications, that is, after
defining the molar absorptivities of the dimer as and the

H
N

H
N
O

O
N
H

N
H
O

N
H

O
H

H
N

N
O
C8H17

H
N

1 +

(128)

Y ([A]0 Y ) = (2Y + Y )



4KD [A]0 + 1 1 + 8KD [A]0

8KD [A]0

H
N


1 + 8KD [A]0
=
4KD [A]0



4KD [A]0 + 1 1 + 8KD [A]0
(126)
+
4KD [A]0



4KD [A]0 + 1 1 + 8KD [A]0
(127)
 = D
4KD [A]0



4KD [A]0 + 1 1 + 8KD [A]0
= ( )
4KD [A]0

If we now subtract ([A]0 Y ) from both sides of (123) and


rearrange as we did with (118), we get (124).



(1 )
Y ([A]0 Y ) = [A]0 (2Y + Y )
2

23

O
N
H

N
H

O
N
H

R = C6H13
Ka = (4.4 0.5) 104 M1
in 1% DMSO-d6 /CDCl3

O
C8H17

Figure 16 The self-complementary amidourea compound reported by Chen and coworkers forms strong dimers in 99 : 1 CDCl3 /DMSOd6 (v/v).35 (Reproduced with permission from Ref. 35. American Chemical Society, 2010.)
Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

24

Techniques

monomer as . This gives us then (129).


A ([A]0 ) = (2 + )



4KD [A]0 + 1 1 + 8KD [A]0

8KD [A]0
(129)
Kano and coworkers applied (129) to study the dimerization of a cationic 5,15-diphenyl-10,20-bis[4-(N-methyl)
pyridinium]porphyrin compound by a UVvis dilution
study in water in the presence of various salts.36 They
obtained KD = 1.1 106 M1 at 25 C in the presence of
0.01 M KNO3 (aq) for this compound.

2.6.4 Aggregation (linear) equilibria and its


analysis by NMR
We again follow the approach outlined by Martin34 to
derive the necessary equations to describe the linear aggregation of A into 1, 2, 3 . . . i-long stacks of A. We only
consider two cases here (Figure 1f), the equal K model
(EK), whereby all the association constants are equal (KE ),
and the cooperative equal K model (coEK), where the first
association constant (K2 ) differs from all the subsequent
association constants (K3 , K4 , K5 , . . . Ki ), which are all
equalin other words, the aggregation is cooperative once
the initial dimer is formed. Martin has also described other
models for linear aggregation,34 which are not discussed
here.
Ercolani has also derived similar expressions for linear aggregation,22 pointing out the relationship between
K2 and the dimerization constant KD , showing that K2 =
2KD . This can explained on symmetry grounds. Ercolani
also pointed out that the symmetry of the aggregating
molecule is important. If the aggregating molecular has
an XY symmetry, that is, it aggregates by forming an
XY:(XY)n :XY, linear supramolecular polymer, then
the symmetry number ( )20, 22 for aggregation is = 1.22
If the molecule has an XX symmetry and aggregates
by forming an XX:(XX)n :XX, linear supramolecular polymer, then the symmetry number = 4. For
the analysis below, we assume = 1 based on the
XY motif; however, the equations below can be corrected by multiplying KE by (i.e., if = 4, use 4KE
below).
We also limit our discussion to how this equilibria can
be analyzed by NMR as it is probably the most commonly
used method used to analyze this form of aggregation. We
start our discussion by focusing on the coEK model as the
EK model is just a simpler version of the coEK model. We
start by defining for the EK equilibria shown in Figure 1(f),
the association constant KE according to (130).34
KE = K2 = K3 = K4 = K5 = . . . = Ki

(130)

Here, the ith equilibrium constant is defined according


to (131)
Ki =

[Ai ]
[Ai ]
=
[Ai1 ][A]
Ki1 [Ai2 ][A]2

=...=

[Ai ]
(K2 K3 K4 K5 . . . Ki1 )[A]i

(131)

In the case of the coEK model, we modify (130) by


defining the cooperativity constant rho () as = K2 /KE
so that (130) becomes (132).
KE = K2 / = K3 = K4 = K5 = . . . Ki

(132)

We also define two dimensionless quantities x and L


based on KE and the monomer [A] and the total [A]0 concentration of the aggregating molecule according to (133)
and (134).
x = KE [A]

(133)

L = KE [A]0

(134)

We can now write a mass balance equation for the


aggregating molecule according to (135).
[A]0 = [A] + 2K2 [A]2 + 3K2 K3 [A]3
+ 4K2 K3 K4 [A]4 + . . .

(135)

Rearranging (135) gives us (136).


[A]0 = [A](1 + 2K2 [A] + 3K2 K3 [A]2
+ 4K2 K3 K4 [A]3 + . . .)

(136)

After substituting for KE and instead of the stepwise


constants (K2 , K3, . . . Ki ), we get (137).
[A]0 = [A](1 + 2KE [A] + 3KE 2 [A]2 + . . .)
= [A](1 + (2KE [A] + 3KE 2 [A]2 + . . .))

(137)

We now use (133) and (134) for substitution in (137) to


get (138).
L = x(1 + (2x + 3x 2 + 4x 3 + . . .)) = L
= x(1 ) + x(2x + 3x 2 + 4x 3 + . . .) (138)
Provided that x < 1 (as we always find in these systems),
we can now use a Maclaurin series expansion of (138) to
get (139).
L = x(1 ) +

x
(1 x)2

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

(139)

Binding constants and their measurement


In the case of the EK model, where = 1 (no cooperativity), (137) and (138) are simplified to (140) and (141).
[A]0 = [A](1 + 2KE [A] + 3KE 2 [A]2 + . . .) (140)
x
(141)
L=
(1 x)2
We now make use of the quantity defined in
Section 2.6.2 by (112) as the mole fraction of the free
monomer A. Comparing (112) with (133) and (134), we
see that = x/L. We first consider the simple EK model
and substitute x/L by in (141) to get (142).
= (1 x)2 = (1 L)2

(142)

Multiplying from the bracket terms, (142) gives us a


quadratic equation that has one relevant solution in the form
of (143), which can then be expanded by substitution for L
using (134).

2L + 1 4L + 1
2L2

2KE [A]0 + 1 4KE [A]0 + 1


=
2KE 2 [A]0 2

(143)

In the more complex coEK model, substitution of x/L


by into (139) would give us (144).
(1 x)2
(1 L)2
=
(1 x(2 x)(1 ))
(1 L(2 L)(1 ))
(144)
Multiplying from (144) and rearranging gives us the
following cubic (145).
=

3 L2 ( 1) + 2 L(L 2( 1)) (2L + 1) 1 = 0


(145)
Substituting for L using (134) would then give us (146).
3 KE 2 [A]0 2 ( 1) + 2 KE [A]0 (KE [A]0 2( 1))
(2KE [A]0 + 1) 1 = 0

two mole fractions, we make use of the equilibria shown


in Figure 1(f) and (131) to obtain (147) and (148).
=

[A3 ] + 2[A4 ] + 3[A5 ]5 + 4[A6 ]6 + + (i 2)[Ai ]i


[A]0

2[A2 ] + 2[A3 ] + 2[A4 ] + 2[A5 ] + + 2[Ai ]


[A]0

(147)
5

(148)
We now expand these equations in a similar manner
as with (135) and (136) and then use (133) and (134)
for substitution to x and L, followed by Maclarin series
expansion to get (149) and (150), which can be further
simplified using x = KE [A]0 .
=

2 KE 2 [A]20
x 2
=
(1 x)2
(1 KE [A]0 )2

(149)

2 2 KE [A]0
(1 KE [A]0 )

(150)

25

Note that all the parameters we need to solve (149) and


(150) are either experimentally accessible or obtained in
the same fitting process required to obtain from (143) or
(146). For the EK binding model, (149) and (150) can also
be simplified as = 1.
We now define the NMR resonances for the aggregating
molecule A depending on whether it is a free monomer
( ), inside the stack ( ), or at the end of a stack ( ). To
reduce the number of parameters that need to be fitted, we
further assume that the NMR resonance for the molecule at
the end of a stack is the average of the NMR resonances for
the molecule inside the stack and that of the free monomer
according to (151).
=

+ e
2

(151)

We can now write an equation that describes the observed


NMR resonance () according to (152).

(146)

The smallest real solution to cubic (146) is the only one


of interest and it can be obtained by some software packages
as was the case with the cubic (47) in Section 2.3.1.
To apply the above equations for the analysis of linear
aggregation by an NMR dilution study, we start by defining
the mole fractions for the three different environments
within the aggregate that a molecular A can be in: as a
free monomer (), within a stack (), and at the end of
a stack ().34 The solutions of (146) and (143) provide us
with or the mole fraction of the free monomer in solution
for the coEK and EK models, respectively. For the other

= + + e

(152)

In the case of the coEK model, we therefore have four


unknown parameters to deal with; KE , , , and the
first two are required to solve (146), (149), and (150), while
the latter two are needed to complete the solution to (152).
In the case of the EK model, = 1 and we have one less
parameter to fit and we then use (143) instead of (146).
In an NMR dilution study on the aggregation of a
pyromellitamide Pyro in acetone, Thordarson and coworkers found that the data fitted significantly better (Figure 17)
to the coEK model using (152) than the EK model or the

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

26

Techniques

in a positive cooperative manner in acetone are not surprising as Pyro readily forms supramolecular gels in cyclohexane. Interestingly, related pyromellitamide compounds
also show exponential kinetics in terms of aggregation in
cyclohexane.37

O
O O
O

O O
O

Pyro

(a)

d (ppm)

8.2

8.1

8.6
8

8.4

1
[Pyro] (mM)

dN-H

d (ppm)

8.2
8
7.8
7.6
dAr-H

7.4
7.2
(b)

0.5

1.5

2.5

3.5

4.5

Figure 17 (a) The structure of the aggregating pyromellitamide molecule Pyro.25 (b) Data from a 1 H NMR (400 MHz,
acetone-d6 ) dilution study at 300 K showing the change in resonance for the N-H (blue open squares) and Ar-H (red solid
circles) of the pyromellitamide Pyro at different concentrations
of Pyro. Also shown are the calculated aggregation isotherms
for the cooperative (coEK, solid line) and noncooperative (EK,
broken lines), Equal K aggregation models, the latter also being
equal to the simple dimerization binding model. The insert on
the top left corner of (b) shows an enlargement of the region
between 0 and 2 mM for the N-H .25 Fitting the data to the
coEK model gives KE = 232 M1 and = 0.22 (K2 = 51 M1 ),
whereas the EK model gives KE = 10.3 M1 and the dimerization model KD = 5.2 M1 . The quality of fit for the coEK model
was 12 times better than for the EK and dimerization model.
The fitted difference between the Ar-H resonances for the free
monomer (Ar-H ) versus inside the stack (Ar-H ) was 2.98 ppm
according to the coEK model compared to 21.7 ppm from the EK
model and Ar-HD = 21.7 ppm from the dimerization model.
Taken together, the data above suggest that the coEK model is
the best model of the three to describe the aggregation of Pyro in
acetone-d6 . (Reproduced with permission from Ref. 26. American Chemical Society, 2007.)

dimerization model according to (126).25 It is noteworthy


that the coEK model was the only model that gave reasonable values for the NMRs of the Ar-H resonances in Pyro
for the free monomer versus inside the stack (or in the dimer
for the dimerization model). The fact that Pyro aggregates

DETERMINING STOICHIOMETRY

One of the most important questions in studying hostguest


complexation is the determination of stoichiometry for the
system of interest. Prior knowledge of the stoichiometry helps in planning, executing, and analyzing the data
obtained from a titration experiment. In many cases, however, this knowledge is not available and the determination
of stoichiometry can then become a challenge that needs
to be addressed before any further data analysis (or experimentation) takes place.
Unfortunately, there is no magical one-size-fits-all
solution to this challenge. Connors suggested that a good
starting point is to assume a simple 1 : 1 stoichiometry and
then look for evidence to support or dispute that model.4 In
their reviews, Connors4 and Tsukube5 listed several methods that could be used to test the 1 : 1 (or other assumed)
stoichiometry hypothesis including the following2 :
1. The method of continuous variations (Jobs method).
2. The mole ratio method.
3. Consistency with the host structure and available
information on the hostguest complex structure.
4. Comparison of stability constants evaluated by different experimental methods (NMR, UVvis).38
5. Specific experimental evidence such as isobestic
point(s).
6. Constancy of stability concentration as the concentration is varied, that is, the success of a stoichiometric
model to account for the data.

3.1

Jobs method

Jobs method is undoubtedly the most commonly used of


the above listed methods. In fact, it has become so popular that other more appropriate methods are now often
ignored as researchers follow the results obtained from
Jobs method without realizing its considerable limitations
and shortcomings. The method is based around the assumption that the concentration [Hm Gn ] of a hostguest complex
Hm Gn (Figure 1a) is at a maximum when the ratio between
the free host and guest ([H]/[G]) is equal to m/n.3, 5 To
experimentally find this maximum, the mole fraction of the
guest (fG ) is varied while ensuring that the total concentration of the host + guest ([H]0 + [G]0 ) remains constant

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

Binding constants and their measurement


and the concentration of the hostguest complex [Hm Gn ]
is then plotted against fG . This curve then usually shows
a maxima at a certain fG value, which corresponds to
fG = n/(m + n) for the stoichiometry of the system. In
the case of a 1 : 1 stoichiometry, this maxima would be
expected at fG = 1/(1 + 1) = 0.5 (see Figure 4 for an
example). The determination of [Hm Gn ] is, however, not
as straightforward as it sounds as the real concentration of
[Hm Gn ] is usually not directly accessible in supramolecular
chemistry experiments. In most cases, researchers therefore make the assumption that the changes in the physical property of interest (e.g., NMR resonance or increase
in a fluorescence signalFigure 4) are directly proportional to [Hm Gn ]. Crabtree and coworkers, for instance,
applied (153) to plot [Hm Gn ] versus fG .39
[Hm Gn ] =

[H]0
Hm Gn H

(153)

Equation (153) highlights the problem with this approach


as the NMR resonance for the value for the Hm Gn complex
( Hm Gn ) can only be determined directly when the total
concentration of the guest [G]0 approaches infinity. Instead,
we have to make the assumptions about the value of Hm Gn
based on extrapolation of the observed resonances () at
high [G]0 /[H]0 ratios.2
Taking the above into consideration, Jobs method usually works well with only one type of complex present.4
When there are more than one type of complexes present,
it can be unreliable.4, 40 This includes 1 : 2 and 2 : 1 equilibria as they usually include two different type of complexes, for example, HG and HG2 , that have quite different
physical properties. If the physical properties are not additive in a simple manner, that is YHG2 = 2YHG (see also
Section 2.3.1), the assumption that the observed physical
property change (e.g.,  in NMR) is proportional to the
concentration of [Hm Gn ] in a linear manner may break
down. If the host or the guest also aggregate, the assumptions behind Jobs method usually also break down.26

3.2

Other methods for determining


stoichiometry

In general, it is advisable to use a combination of the


methods listed above to determine the stoichiometry. The
mole ratio method (No. 2 above) is very straightforward as
it usually involves inspecting the titration isotherm obtained
upon titration of a guest into a host solution.5 The apparent
linear portions of the beginning and end of the titration
curve are extrapolated. The point where these two lines
intersect usually corresponds to the main inflection point
on the binding isotherm and the corresponding [G]0 /[H]0

27

ratio then corresponds to the n/m ratio of the complex.


This method sometimes works well; there are more than
one type of complexes presenteach inflection point can
then point to the key complexes present. The example
in Figure 9 illustrates this clearly as there are two clear
inflection points at [G]0 /[H]0 = 1 and also [G]0 /[H]0 = 2,
corresponding to the HG and HG2 present in this system.27
It needs to be noted here that, as in all hostguest titrations,
it is important to keep the host concentration constant
throughout the course of the titration to avoid dilution
effects. The most straightforward method to achieve this is
to use the same host solution that will be titrated to make
up the solution of the guest that will then be used in the
titration experiment.2
Of all the methods listed above, simply applying ones
chemical knowledge and intuition based on the structural information about the host and the guest to determine the stoichiometry (Method 3) is perhaps the powerful and straightforward method available. The wealth of
background knowledge and the structural information available from NMR studies (including various two-dimensional
methods) and X-ray structures as well as molecular modeling methods usually allow supramolecular chemistry
researchers to make fairly accurate assumptions about the
expected stoichiometry of the hostguest system of interest.
Another underutilized method is the comparison of
the results obtained from different experimental methods,
including kinetic and solubility studies (Method 4).39 It
is sometimes also possible to look for specific evidence
to rule out certain stoichiometry modelsthe presence of
more than one isobestic points in UVvis titration can
be used to rule out simple 1 : 1 stochiometry (Method
5).3 The converse is not necessarily true, however, as the
absence of more than one isobestic point cannot be used
to rule out more complex stoichiometrys, especially where
cooperative interactions play a role.2
Finally, testing if the expected stoichiometry model fits
to the data at different concentrations is perhaps the most
robust method available to confirm the expected stoichiometry in supramolecular chemistry (Method 6).4 If
the titration data obtained at two or more different concentrations consistently shows the best fit to a particular
binding model (e.g., 1 : 2 equilibria), it strongly suggests
that the underlying assumptions about the stoichiometry
are valid. Note that what the best fit is, needs to be
defined with care; simply looking at the quality of the fit
is not enough as increasing the number of fitting parameters per se will always increase the quality of fit. Other
indications of good fit should be considered, including
the scatter of the residual plot and whether all the fitted parameters (and not just the binding constants) make
physical sense. The example in Section 2.6.4 (Figure 17)

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

28

Techniques

provides a good exampleonly the coEK model gives


sensible results for the changes in the NMRs of interest, suggesting that this model was the best fit to the
data obtained, despite the fact that it had one additional
parameter compared to the other models under consideration.25

DATA ANALYSIS AND SOFTWARE

Analyzing results from supramolecular titration experiments and fitting the data to the binding models detailed
in Section 2 is not always as a simple task as it seems.
There are a number of software programs available that
can aid with this process (see below) but the fact remains
that, without good understanding of these programs and
how they relate to the equations above, it can be difficult to evaluate the results obtainedsimply pressing
a button on a computer program rarely does the trick.
Many factors need to be considered, including the choice
of model to fit the data into, transformation/correction
of the raw data, choice of minimization algorithm, decision on whether and then how to apply global analysis, choosing initial values for the fitting process, adjusting the model, how to display the results, estimation of
uncertainty of the fitting process, and overall evaluation
of the results. Covering all of these topics is beyond the
scope of this chapterhere the focus is therefore on the
following:
1.
2.
3.
4.
5.

4.1

The decision on which model to use versus the number


of parameters to fit.
The advantages of using global analysis methods.
Methods to estimate uncertainty in the fitting progress.
Issues to consider in terms of choosing software for
data analysis.
A short review of a few selected commercial and noncommercial software packages for analyzing binding
data and their relative strengths and weaknesses.

Model selection versus the number of fitted


parameters

We start by discussing the selection of the binding model


versus the number of parameters to fit the data into. In many
circumstances, it is desirable or necessary to fit the data to
more than the binding model and then compare the results
obtained to select the best one. The situation highlighted in
Section 2.3.1 is a case in point1 : 2 equilibria can usually
be fitted to four different types of model depending on
whether one assumes the binding is statistical or not and

whether the physical properties are additive in a simple


manner or not as illustrated by (50), (51), and (63)(68).
Taking NMR titrations as an example, there are four
unknown parameters (K1 , K2 , HG , and HG2 ) to fit in
(69), three ( 12 , HG , and HG2 ) in (70) and (72), and
two ( 12 and HG2 ) in (71). What we need to consider
carefully here is that the more parameters there are to fit,
the better the fit will be in general. Therefore, it would not
be surprising if we find that the best fit of our data is to
(69) as it has the largest number of parameters. The same
would happen if we were to compare a simple binding to
a more complicated one, for example, a 1 : 1 binding to
a 1 : 2 binding modelthe latter would almost certainly
give a better fit. This means that when we cannot rely
solely on comparing the goodness-of-fit parameter given
by most programs to compare different binding models,
especially if they do not contain the same number of fitted
unknown parameters. Instead, we should also compare the
scatter of the residuals, that is, the plot of yfit against
the [G]0 /[H]0 ratio according to (154) where ydata is the
raw data and ycalc is the calculated data according to our
model.2, 41
yfit = ydata ycalc

(154)

This equation provides the starting point for other methods to quantitate the quality of the fit. Most of the software
programs used to fit binding data do so by minimizing the
sum of the squared residuals (ssy ) according to (155).
ssy =

(ydata ycalc )2

(155)

It might be tempting to use the ss y values obtained to


compare different binding models, but this is not advisable
without making some corrections for the number of data
points used (N) and the number of parameters used (k)
in the fitting process. This can be done by defining the
chi-squared ( 2 ) value according to (156), or the slightly
different standard mean of the y estimate (SEy ) value
according to (157).42

2 =

SEy =

(ydata ycalc )2
N k1

(156)

(ydata ycalc )2
N k

(157)

Another quantitative parameter that can be used to


compare the quality of fit is the covariance (cov) of the fit
(covfit ),2, 25 which is defined by (158) as the (co)variance
of residual or yfit from (154), divided by the co(variance)

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

Binding constants and their measurement


of the raw data.
covfit =

cov(yfit )
cov(ydata )

350850 nm at 1-nm intervals to the model based on (108)


for 2 : 2 equilibria.29
(158)

The covfit from (158) is a numerical estimation of how


small and random the scatter plot isthe lower the covfit
value, the smaller and more random the scatter is. The covfit
value is also fairly insensitive to the number of parameters
used in the fitting process.2
It is also important to consider if the fitted parameters
make physical sense (e.g., no negative binding constants
or 1 H NMRs changes of >100 ppm). One also needs
to consider if the models are chemically plausible or fit
with the other information that we may have about the
equilibria. On the other hand, the model with larger number
of parameters can be accepted as the best one provided
the above conditions are fulfilled and that it provides
a significantly better goodness of fit compared to other
competing models.

4.2

Global analysis

Global analysis is perhaps one of the most powerful yet


underutilized tools available in the analysis of binding
data.2, 41 The global analysis method is based on taking
a related set of data and fit it simultaneously to the same
binding model. If we take an NMR titration analysis of
a 1 : 2 equilibria as an example, we often find that more
than proton resonance is significantly affected (shifted) by
the addition of our guests. Taking the pyromelliamide host
titrated with nitrate guest in Figure 7 as an example, both
the NH and ArH resonances are significantly shifted
upon addition of a nitrate.26 Instead of fitting just one of
these resonances, we now fit both the NH and ArH
resonances simultaneously to (69), which now becomes
(159)

NH +ArH =

29

(NHHG + ArHHG )K1 [G]


+(NHHG2 +ArHHG2 )K1 K2 [G]2
1 + K1 [G] + K1 K2 [G]2

(159)
We now have six parameters to fit into two different
data sets (NH and ArH ). We can go on adding more
data sets; each time we can add the corresponding number
of parameters corresponding to the physical changes to
fit but not any additional binding constants. In the case
of a UVvis spectra, it is possible (and frequently done)
to fit the whole UVvis spectra to the binding equation
of interest. In their work on 2 : 2 sandwich formation
(Figures 14 and 15, Section 2.5), Taylor and Anderson
used (160) to fit their UVvis spectra from the region of

n=350850
 nm

n=350850
 nm

j =350 nm

j =350 nm

A(i) = KF [H]2 [G]2

(i)H2 G2

n=350850
 nm

+ KF KB [H][G]2
(i)HG2
j =350 nm

(160)
In (160), we now have 502 parameters to fit (500 
values + KF and KB ) to 500 data sets!
Why would we want to add so many parameters to our
fitting process, especially in light of what was said earlier
about the risks of adding extra parameters as it usually
inflates the goodness of fit (any curve can be fitted to
a polynomial y = 1 + x + x 2 + x 3 . . . if we add enough
terms to it)? We need to make it clear that at the same
time we are increasing the number of parameters in global
analysis as illustrated by (158) and (159) and we have also
increased the number of raw data points in our fitting
process considerably. More importantly, every time we add
another data set to our global analysis process, we do not
double the number of parameters to fitthe key parameters
of interest (e.g., K1 and K2 ) stay the same. For this reason,
global analysis therefore does not give the same result as
that if we had taken two different NH and ArH or 500
different UVvis isotherms in the examples above, fitted
them one-by-one to (69) or (109), and then averaged the
results. On a fundamental level, global analysis tightens the
so-called error surface in our fitting process, compared to
local (simple) analysis as explained in an excellent review
by Beechem.43 This technique is therefore exceptionally
powerful when it comes to fitting more complex equilibria
such as 1 : 2 and 2 : 2 equilibria.
The following simulated example by Thordarson,2 clearly
illustrates the power of global analysis. Two UVvis binding isotherms (with different HG and HG2 values)
for a 1 : 2 equilibria, where K1 = 100 000 M1 and K2 =
10 000 M1 , were generated on the basis of (73). Then, a
random scatter of 1% was added to these calculated A
values (Figure 18a). The two different data sets were then
(i) fitted separately (red and blue curves) by local analysis to (73) and (ii) fitted by global analysis (black curves)
using the corresponding modified version of (73) as shown
in Figure 18(b). The quality of these fits was then compared
by (i) the calculated asymptotic error (see also below)41
and (ii) the standard error of the y estimate (SEy ) according to (130). Comparing the results obtained (Figure 18c)
reveals some striking differences between global and local
analysis.
First, the quality of fit to both data sets is clearly much
better by global analysis than if they are fitted separately,

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

30

Techniques

For both data:


[H]0 = 105 M
K1 = 100 000
K2 = 10 000
(a)

Data 1: eHG = 1 000, eHG = 20000


2
Data 2: eHG = 30 000, eHG = 20 000
2

Global fit
Random error in A = 1%

0.16

data set 2 (which in local analysis mode gave a better fit


than data set 1), K1 is 19% off, and K2 26% off the real
values. Even the calculated changes in molar absorptivities
are closer to the real values by global analysis than they
are when the data sets are fitted individually, highlighting
the power of global analysis in data analysis.

Absorbance change (A)

0.14

4.3

0.12
0.1
0.08
0.06
0.04
0.02
0

(b)

10

Equivalent total guest added ([G]0/[H]0)


Local fit for data 1:
K1 = 21072 (75%), K2 = 22735 (43%),
eHG = 3012 (72%), eHG2 = 14384 (11%), SEy = 6 x 104
Local fit for data 2:
K1 = 118900 (9%), K2 = 7392 (21%),
eHG = 28104 (4%), eHG2 = 26282 (20%), SEy = 1 x 103

(c)

Global fit for data 1 and 2:


K1 = 101800 (5%), K2 = 10492 (9%),
Data 1: eHG = 1009 (9%), eHG2 = 18894 (4%),
Data 2: eHG = 30327 (2%), eHG2 = 18991 (8%),
SEy = 1 x 103

Figure 18 (a) Local versus global fit for a hypothetical UVvis


titration for a 1 : 2 complexation equilibria (a). The initial conditions and the color coding for the two different data and the
global fit. The data points (red and blue squares) are generated
from (73) using the parameters shown here with an additional
random noise of 1%. (b) The resulting binding isotherms with
the local and global fits (solid lines) superimposed. The enlarged
area (red shaded square) shows the similarities of the local and
global fitted isotherms. (c) The results from the local and global
fitting process with the asymptoting standard error in parentheses. SEy = standard error of the y estimatesee (157). See also
text for details.2 (Reproduced from Ref. 2. Royal Society of
Chemistry, 2011.)

based on the SEy and asymptotic error values. Second,


the results obtained by global analysis are not the simple
average of the results obtained from fitting the two data
sets individually. Third and interestingly, the difference
between global and local analysis is not that clear if we
just look at how well the curves fit into the data (see insert
in Figure 18b). Last but not the least, global analysis comes
much closer to returning the initial values used to generate
the two data sets. For instance, by global analysis, K1 is
within 2%, and K2 within 5% of the real value while for

Estimation of uncertainties

Exactly how reliable or accurate are the results we


obtain from analyzing binding data? This is perhaps the
most important question we can ask about the results
obtainedwithout any information about the reliability of
our data, analysis is useless.44 The estimation of uncertainty
can and should be approached from at least two directions
as there are the two components that usually have the largest
contribution to the estimation of uncertainties in binding
studies, namely the experimental repeatability uncertainty
and the data analysis (fitting process) uncertainty. If both
are known, the overall uncertainty of the experiment can
be estimated by comparing and possibly combining them,
albeit this would have to be done in a subjective manner.
The estimation of uncertainty with regards to the repeatability of the experimental data is obtained by repeating the
experiment n times under (near)-identical conditions and
analyzing the data in the same manner. Most researchers
carry out their titration experiments in triplicate and the
variation in the results obtain from the three experiments
gives us some idea of how reliable the results are and
allow a very basic or conservative estimation of the uncertainties. However, to obtain correct quantitative description
about the variation of the data (parameters such as standard
deviation and so on), we should probably carry out the
experiment at least six to eight times.44 This is not always
practical or doable for economic reasons.
Estimation of the uncertainty arising from the data analysis or fitting process itself is equally importantafter
all, data fitting is just an estimate of the real system and cannot therefore be 100% accurate, unless there
is 0% random scatter in the underlying data but that
is not likely to happen. Here we need to remember
that the fitting processes used by most or all modern computer programs are based on nonlinear regression methods. Old fashioned, linear-regression methods
such as BenesiHildebrand,45 LineweaverBurk,46 Scott,47
HanseWoolf,48, 49 or Scathard50 transformation of binding equation were developed at a time when computer
power was scarce or nonexisting. Unfortunately, there are
still some researchers that use them probably not knowing about the underlying fundamental problems associated
with using these transformation. First, as the underlying

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

Binding constants and their measurement


data comes from nonlinear processes, they violate some of
the fundamental assumptions behind linear regression by
distorting the experimental error.41, 51 Second, these methods frequently make highly questionable shortcuts such as
[G]0 [G] or that the final observed physical change Yfinal
equals the one from the fully formed complex; YHG = Yfinal .
For these reasons and with the power of modern computers, there is really no justification for using these lineartransformation methods for determining binding constants2
when nonlinear methods that provide exact solutions to the
equations in Section 2 are now readily available in a number of software packages (including freeware). The only
reason one would use plots such as the Scathard plot according to (161) is to plot the data after it has been calculated
by nonlinear regression methods as it is often easier for the
human eye to spot deviation from straight lines than from
nonlinear plots.41
Y
= Ka [G] + Ka YHG
[G]

(161)

Putting outdated linear regression methods aside, the key


problem with nonlinear regression methods is that there are
no straightforward methods for estimating the uncertainty in
the fitting process. Most of the software packages available
will report an error estimate or the 95% confidence interval
on the results obtained, which in most cases is the socalled asymptotic standard error.41 The asymptotic standard
errors is, in reality, only an approximate as it relies on
a number of assumptions about the nonlinear regression
analysis process, including that the experimental error is
only in the y-estimate, the physical property (Y ) observed.
This is of course not true in most casesin fact, it can
easily be argued that the biggest experimental error is in
the x-estimate, the [G]0 /[H]0 ratio, as the uncertainties
in the concentrations used are often quite considerable.
Another problem is the actual method used to calculate
the asymptotic error. The technical details are beyond the
scope of this chapter but in essence the process involves
a series of matrix manipulations resulting in a so-called
variancecovariance matrix.3 The asymptotic error is then
calculated from this matrix, but this assumes that the
confidence interval is symmetrical around the parameter,
that is, the errors are the same, but this is probably not
true in most cases.41 Additionally, computational rounding
errors can start to play a role here as the numbers
used in these matrix manipulations involve the use of
very small numbers. Sometimes, this results in a unstable
or a singular matrix3 that in turns gives a meaningless
results in terms of the asymptotic errors. In this authors
experience, this problem is particularly noticeable with
larger, more complex equilibria such as the 1 : 2 binding
model.

31

Taking the above into account and provided we do not


have problems with unstable matrix inversions, the asymptotic error often provides a good starting point for our
uncertainty estimation. More intensive but rigorous uncertainty would involve some form of data mapping, including
grid search methods43 and Monte Carlo simulations.2, 41 In
the former, one parameter is kept fixed at a range of values [(e.g., we would vary Ka from 1001000 M1 in steps
of 100 M1 in (26)], while the remaining parameter(s) are
determined from a fitting process in the usual manner. The
resulting residuals ssy according to (155) are then plotted
as a function of the variable that we kept fixed (i.e., Ka
from 1001000 M1 in the example above). Using an F statistics test, we would then determine the range for the
fixed parameters required to increase the fit (ssy ) above
the minimum at a predetermined confidence interval (e.g.,
67%).43 This process can then be repeat for all the other
parameters of interest.
Perhaps the best method for estimating uncertainties
from our fitting process is the Monte Carlo simulation approach.2, 41 This computationally intensive method
involves taking the raw initial data and adding a certain
level (e.g., 1%) of random scatter to it (Figure 18). This
process is then repeated n times and each data set then fitted to the binding equation(s) of interest. This gives us an
n times number of results that we can analyze in terms of
confidence intervals (95%), using tools such as a box plot52
to get a better idea of how reliable the results are. The key
strength of this method is that it allows us to include the
estimated errors for the raw x- and y-data. For instance, in
an UVvis titration, we could add 1% random scatter to
the measured absorbance (A), 2% to our guest concentration ([G]0 ), and 4% to our host concentration ([H]0 ). This
allows us to overcome the key problem with the asymptotic
error estimation method discussed above. The only downside of this method is of course the fact that the magnitude
of the uncertainty we add to the data could be quite subjective. Note that we use a random number generator to
create our n-data; hence, two runs of n-data simulations
will usually not give us exactly the same results.
For an example of a Monte Carlo simulation approach,
consider again the data shown in Figure 18. Focusing
on data set 2, the 95% confidence limits based on the
asymptotic error for K1 = 97 000140 000 M1 (around
the average of 112 980 M1 ), while the Monte Carlo
simulation gave the 95% confidence limits as K1 =
95 000150 000 M1 .2 Note that the confidence limits
obtained with a Monte Carlo simulation are unsymmetrical
around the calculated K1 value. In a more extreme case, a
Monte Carlo simulation on data from (26) with Ka = 1000
larger than 1/[H]0 gave the 95% confidence limits as67%
and +1014 %.2 This is what one would expectif the binding constants is too high for a given concentration, we

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

32

Techniques

could expect a reasonable estimate on the lower (but not


the higher) limit on what the binding constant is.

4.4

Issues to consider when choosing software


for data analysis

Currently, there are a number of software packages available for analyzing binding data. Some of these are reviewed
in more detail below but in general one can divide these into
two classes: standalone programs and those that build on
or use commonly available data software packages such as
Excel , Mathematica , Matlab , or Origin as an underlying engine to perform most of the required functions,
data handling, and plotting. Currently, the preference seems
to be for the latter category with programs using Excel
being among the most popular ones, which is perhaps not
surprising given its ubiquitous use in data handling. Early
legacy programs such as HYPERQUAD53 and EQNMR54
are still in use in some laboratories but the use of these
old programs is now somewhat hampered by compatibility problems with modern operational system. Many of
these old legacy programs are based on MS-DOS or Apple
II Macintosh operational system environments and do not
compile or run well with modern operational systems (we
know of at least one lab that keeps an old Apple II running
just for this purpose!).
When it comes to deciding on which program to use,
there are many factors to consider. Often user interface
(e.g., compatibility with Excel ) and availability (is it
free?) are the deciding factors but other issues should
always be considered. Documentation on how the program
works, what equations are used for each equilibria, and is it
possible the ability to edit these equations are also important
factors when it comes to choosing a program.
First, it needs to be acknowledged that there is probably
no program that covers all the different methods (NMR,
UVvis, fluorescence, calorimetry, and so on) that one
might want to use. This is particularly the cases with
most of the custom-written (free) software packages availableusually they have been written by an enthusiastic
researcher that has only been using some of these techniques herself/himself. Commercial packages may also be
focused on a different application field, for example, biochemistry or pharmakinetics, making it sometimes difficult for supramolecular chemists to adapt them into their
research.
The minimization algorithm used is one factor that
needs to be considered; does the program use steepestgradient methods such as a GaussNewton or Marquardt
algorithm? Or does it also allow or use other methods
such as the Simplex method55 or genetic algorithm?56
The steepest-gradient methods are usually very fast (and

they also give the asymptotic error directly) but for more
complex or difficult problems; the Simplex method or
genetic algorithm methods may be required as the steepestgradient methods may not find the minima as easily. One
related issue here is how to handle the starting values for
the fitting process. Most of the minimization algorithms
require reasonable starting values or at least starting value
boundaries. The steepest-gradient methods are generally
more unforgiven in this respect than the Simplex or genetic
algorithm methods. On the other hand, Simplex usually
does not work with boundary values (e.g., trying to tell
the program to restrict the search for Ka values for positive
values between 0 and 106 M1 ), which can be a problem
(it keeps finding negative Ka values). It is also possible
to use global search methods,57 which create a scattered
grid of starting values, calculate the fit from these points,
and then perform minimization using one of the above
methods (usually steepest-gradient type) from the best
starting point(s).
It is also worth considering if the program uses methods
that directly solve cubic equations such as (47), which are
essential to fit 1 : 2 and other complex equilibria or if the
method of successive approximation is used. Again, neither
should be ruled out as the successive approximation method
is simpler and faster to implementhowever, it may not
always converge to solve the equation(s) of interest.
Finally, it is worth considering whether features such
as global analysis or data simulation are included. For
researchers working mostly with relatively simple 1 : 1
equilibria, they are probably not necessary but for others
these should be considered seriously when choosing (or
writing) software for data analysis.

4.5

Few selected software packages available for


binding data analysis

The short review given below of some available software


packages is by no means comprehensive. To start with, the
legacy program, although still in use in some laboratories
such as HYPERQUAD53 and EQNMR54 are not discussed
here. Current users of these programs will be aware of their
strengths and limitations but given the above-mentioned
problems with legacy program it is not likely that novice
users would want to commit themselves to learning the necessary syntax. The excellent curve-fitting program for NMR
data from Prof. Christopher Hunter58 would also deserve
mention but as it requires out-dated Apple Macintosh computers to run it, it may not have a strong appeal for new
users. The once popular commercial program SPECFIT is
no longer supported or sold as the company owning it seems
to have folded. In some cases, there is also very little documentation available and the programs are only distributed

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

Binding constants and their measurement


by word of mouth between laboratories. Many of the
commercial programs are also of limited use as mentioned
earlier as they have been written for a different user (e.g.,
pharmacology), and they are too rigid to be modified for
use in supramolecular chemistry (many programs ignore
equations based on mole fractions but without them you
cannot analyze NMR data). Here the discussion is therefore limited to two examples of commercial programs and
a handful of noncommercial programs that are well adapted
for analysis of supramolecular binding constants.
GraphPad (www.graphpad.com) is a comprehensive software package for biostatistics. The founder of GraphPad, Dr Harvey Motulsky has written an excellent book
on biostatistics59 and the manual on GraphPad,41 which
is also written by Motulsky, is one of the best references available on data analysis by nonlinear regression.
Strengths: Offers a range of relevant models for 1 : 1 and
1 : 2 binding for supramolecular chemistry. Adding your
own equations (e.g., for NMR) is relatively straightforward. The program offers global analysis functions and
simulations, including a Monte Carlo approach. The documentation is second to none in terms of how the nonlinear
regression works and the different options available (e.g.,
global vs local analysis).
Weaknesses: Being a commercial program, the price
might be an obstacle for some users. It is also a standalone program in terms of data inputthat is, if the
user has the original data in Excel or another program,
it will have to be pasted across (this small problem, however, applies to most software packages). Documentation on
how equations are solved is somewhat sparse but the program seems to use the method of successive approximation
to solve higher-order equations (e.g., cubic equations for
1 : 2 equilibria), which might become a problem is some
complex situations.
ReactLab (www.jplusconsulting.com) is a new global
analysis software package from Dr Peter King and A/Prof.
Marcel Maeder that is based on Matlab but runs as a
standalone addition to Excel using a free downloadable
Matlab component runtime (MCR) module from Matlab
to execute ReactLab. The program is written to allow the
user to do all the data input and adjustment of equations
via the Excel interface.
Strengths: With what was said earlier in mind, the fact
that this program is built around global analysis of spectroscopic data is undoubtedly its strongest feature. The
binding models are simply entered in via an Excel spreadsheet and the program also considers pH measurements
and equilibria when it is desired or necessary. The graphical interface also provides a clear illustration of the global
residual plot and the distribution of components in solution
at different stages of the titration. It also offers simulation

33

features and factor analysis. With its global analysis features, it is exceptionally well suited for the analysis of
complex equilibria by UVvis spectroscopy as illustrated
in a recent paper by Leigh and coworkers on DDDD-AAAA
arrays.60
Weaknesses: The program seems to have been written
focusing around UVvis spectroscopic titration of inorganic complexes. It is unclear if it could handle other equilibria based on absolute concentrations (e.g., fluorescence)
but NMR titrations are not included. The documentation
suggests that the Marquardt steepest-gradient method is
used for optimization, which means that the standard deviation error given is almost certainly the asymptotic error.
The price of the program will undoubtedly also be a barrier
for some users, as is the fact that it really runs as a shell
on top of Excel and Matlab without allowing the users
access to the underlying programs as would be the case
with VBA and Macros in Excel or m-files in Matlab
but this is of course unavoidable for a commercially viable
enterprise.
Gas-Fit (http://gasfit.djurdjevic.org.uk/) is a standalone
freeware written by Dr Dusan Djurdjevic. It is based around
using a genetic algorithm approach for fitting the data. It
offers fitting BET and other surface absorption data as well
as fluorescence and NMR binding.
Strengths: For a freeware, the interface is easy to use
and the equations used are clear to the users. The genetic
algorithm is the key strength of this program as it can
greatly assist in finding difficult solutions, especially
when analyzing 1 : 2 equilibria. The inclusion of both 1 : 1
and 1 : 2 binding equilibria for fluorescence and NMR is
important as these two methods are often ignored in other
packages. Leigh and McNab used this program to analyze
the binding data from the DDD-AAA complex shown in
Figure 4.12
Weaknesses: Apart from being limited to NMR and
fluorescence titrations in its current version, the program
also does not allow the user to modify the existing
equations. The documentation is also sparse as is usually
the case with freeware. The program does not give any
estimation of the errors and it does not offer any simulation
options.
Fittingprogram: (www.chem.unsw.edu.au/research/groups/
thordarson/fittingprogram) from the author of this chapter
accompanying his recent review2 is a free collection of
Matlab m-files for analyzing data from supramolecular
titration experiments. Running within the Matlab package, it uses the Simplex algorithm to solve or simulate 1 : 1,
1 : 2, and 2 : 1 equilibria for NMR and UVvis data (the
author has developed programs for other methods/equilibria
but these are yet to be released).
Strengths: Resting on the power of Matlab to use a
combination of the Simplex algorithm, methods to directly

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

34

Techniques

solve cubic equations and global analysis, this program


has been shown to tackle difficult 1 : 2 and 2 : 1 equilibria.2, 25, 26, 31 In addition to giving the asymptotic error, it
also provides simulation functions (although not for global
analysis) that can also be used to perform Monte Carlo
estimation of the fitting error. A simple program to help in
importing and exporting data from Excel is also provided.
The Matlab m-files can easily be edited by the user.
Weaknesses: Although the documentation provided has
some helpful comments, it would be difficult for users that
do not have some familiarity with Matlab to fully utilize
this programespecially when it comes to troubleshooting.
Matlab is also not as readily available in many laboratories as Excel , and its cost might be a barrier. The
global analysis function is also limited to up to four different data sets (NMRs or wavelengths) as each variation in
the number of data sets requires a separate m-file. The lack
of other methods than NMR and UVvis, such as fluorescence, is also an issue. The documentation regarding the
equations used and how the software works relies on the
above-mentioned review2 and the Matlab manual itself
respectively.
NMR Fit HH and NMR Fit HG: (www.dur.ac.uk/j.m.
sanderson/science/downloads.html) are two related Excel
add-on programs by Dr J. M. Sanderson for analyzing
dimerization and 1 : 1 hostguest complexation and the
combination of these two.
Strengths: The Excel interface means that these programs are easy to use, even for beginners. Another very
powerful feature of these programs is that it allows one to fit
NMR simultaneously to a dimerization and 1 : 1 hostguest
complexation. Additionally this program does global analysis. The documentation is also clear and concise.
Weaknesses: These programs only work for dimerization
and 1 : 1 complexation NMR data. In addition, as they are
written around the Solver routine in Excel , it would
probably be difficult to modify them for more complex
systems. The program does not give any estimation of
the uncertainties of the fitting process. Finally, it should
be noted that it was written for Excel 2004 on Mac
OS XWindows users may experience some problems in
running it (this authors attempts to run it on Excel 2010
on a Win 7 computer were not successful).
Equilibria (www.sseau.unsw.edu.au/Equilibria.zip) is a
standalone program written by Dr Chris Marjo. The program does fitting of NMR data into 1 : 1 hostguest complexation, dimerization, and the combination of these two.
It also does 1 : 1 and 1 : 2 fluorescence titrations as well
as analysis of 1 : 1 equilibria by a fluorescence competition
method (see also Competition Experiments, Concepts).3
Strengths: The program is simple and quite straightforward to use. The 1 : 2 and competition fluorescence options

are quite unique. The search algorithm is quick and fairly


robust. Unusually, it does not require the users to provide
some initial guesses for the parameters that are optimized
in the fitting programs.
Weaknesses: The documentation is very sparseonly the
equations for the equilibria are shown. The minimization
algorithm is not explained but seems to use a combination
of a global search method57 using a grid of initial starting
values, followed by a steepest-gradient method. The starting
values for the grid search are liberal but cannot be edited by
the users, which could cause problems with more complex
systems. The program is limited to NMR and fluorescence
titrations and does not provide any simulation features or
estimation of the errors on the results obtained. Global
analysis options are also not available in this program.

CONCLUSIONS

In this chapter, the author has attempted to give a


broad overview of how binding constants are derived in
supramolecular binding studies and how the data from
supramolecular titration experiments is usually analyzed
focusing on commonly used methods such as NMR,
UVvis, and fluorescence spectroscopy. By explaining in
detail how the equations for the most commonly encountered equilibria such as 1 : 1 and 1 : 2 hostguest complexation and hosthost dimerization are obtained and then
used, the chapter may help the reader to develop their
own equations for more complex equilibria. The more complex or unusual examples and methods covered should also
help the reader to tackle systems with similar complexity
in their own work. In Section 3, the fact that there is no
method (including Jobs method) that provides a bulletproof definition of stoichiometry and that the researchers
own chemical intuition should not be ignored has been
highlighted. Section 4 then highlighted the importance of
choosing and comparing the appropriate binding model(s),
how the uncertainties can be estimated, the power of global
analysis, what factors to consider when choosing a computer program for data analysis, and then concluded with a
review on a few selected software packages.
Looking ahead, it is the authors opinion that the field
is reaching a transition point on at least a couple of
fronts. First, the common availability of software packages
should hopefully once and for all stop users from using
outdated linear transformation methods. There is no excuse
for anyone to use these often-inaccurate methods, and
reviewers should make a note of pointing this out to authors
when reviewing papers. The recent increase in the number
of available software packages that use powerful methods
such as global analysis, genetic algorithm, and global
search methods will also help more researchers in tackling

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

Binding constants and their measurement


more complex systems. Related to this, computational
and simulation methods for estimating the uncertainty of
the results obtained are improving and have been well
implemented in some of the software packages available.
Better treatment of how uncertainties are estimated is one
issue that the community should try to deal with in the
near futureit would be a worthy aim for supramolecular
chemists to try to develop uniform guidelines on how
estimation of uncertainties are obtained and reported on
binding constants.
Finally, as researchers, we should never lose sight of
the fact that chemical intuition and knowledge will never
be substituted by blind data analysis. The vibrant debate
on how to analyze the data from complex cooperative
assemblies10, 11, 21 is a good example of how work in
this area can stimulate research but research on complex
supramolecular assemblies is undoubtedly going to be one
of the key growth areas in the near future. The methods
required to understand and quantify these interactions,
including the necessary mathematical equations and robust
computational approaches for data analysis, are likely to be
the subject of considerable effort as the field advances.

35

7. C. Piguet, Chem. Commun., 2010, 46, 6209.


8. D. Munro, Chem. Br., 1977, 13, 100.
9. A. E. Martell, R. D. Hancock, and R. J. Motekaitis, Coord.
Chem. Rev., 1994, 133, 39.
10. G. Ercolani, J. Phys. Chem. B, 2003, 107, 5052.
11. G. Ercolani, J. Am. Chem. Soc., 2003, 125, 16097.
12. B. A. Blight, A. Camara-Campos, S. Djurdjevic,
J. Am. Chem. Soc., 2009, 131, 14116.

et al.,

13. S. Sarkar, S. Li, and B. B. Wayland, J. Am. Chem. Soc.,


2010, 132, 13569.
14. J. Feeney, J. G. Batchelor, J. P. Albrand, and G. C. K.
Roberts, J. Magn. Reson., 1979, 33, 519.
15. C. S. Wilcox, in Frontiers in Supramolecular Organic Chemistry and Photochemistry, H.-J. Schneider and H. Durr,
Wiley-VCH Verlag GmbH, Weinheim, 1991, pp. 123144.
16. C. S. Wilcox, J. C. Adrian Jr, T. H. Webb, and F. J.
Zawacki, J. Am. Chem. Soc., 1992, 114, 10189.
17. C. Bejger, J. S. Park, E. C. Silver, and J. L. Sessler, Chem.
Commun., 2010, 46, 7745.
18. P. G. A. Janssen, P. Jonkheijm,
J. Mater. Chem., 2007, 17, 2654.

P. Thordarson,

et al.,

19. C. A. Parker and W. T. Rees, Analyst, 1962, 87, 83.


20. E. Chekmeneva, C. A. Hunter, M. J. Packer, and S. M.
Turega, J. Am. Chem. Soc., 2008, 130, 17718.

ACKNOWLEDGMENTS

21. C. A. Hunter and H. L. Anderson, Angew. Chem. Int. Ed.,


2009, 48, 7488.

The author would like to thank the Australian Research


Council (ARC) for support, including for an Australian
Research Fellowship/Discovery Project (DP0666325) and
an ARC Discovery Project (DP0985059). The author would
also like to thank the School of Chemistry at the University
of New South Wales (UNSW) for start-up funding and
continuous support for his work.

22. G. Ercolani, Chem. Commun., 2001, 1416.

REFERENCES
1. J. A. Goodrich and J. F. Kugel, Binding and Kinetics for
Molecular Biologists, Cold Spring Harbour Laboratory Press,
New York, 2007.
2. P. Thordarson, Chem. Soc. Rev., 2011, 40, 1305.
3. K. A. Connors, Binding Constants, John Wiley & Sons, Inc,
New York, 1987.
4. K. A. Connors, in Comprehensive Supramolecular Chemistry, vol. 3, Chapter 6, eds. J. L. Atwood, J. E. D. Davies,
D. D. MacNicol, and F. Vogtle, Pergamon, Oxford, 1996,
pp. 205241.
5. H. Tsukube, H. Furuta, A. Odani, et al., in Comprehensive Supramolecular Chemistry, vol. 8, Chapter 10,
eds. J. L. Atwood, J. E. D. Davies, D. D. MacNicol, and
F. Vogtle, Pergamon, Oxford, 1996, pp. 425482.
6. H.-J. Schneider, R. Kramer, S. Simona, and U. Schneider,
J. Am. Chem. Soc., 1988, 110, 6442.

23. G. Ercolani, C. Piguet, M. Borkovec, and J. Hamacek,


J. Phys. Chem. B, 2007, 111, 12195.
24. K. A. Connors, A. Paulson, and
J. Org. Chem., 1988, 53, 2023.

D. Toledo-Velasquez,

25. P. Thordarson, E. J. A. Bijsterveld, J. A. A. W. Elemans,


et al., J. Am. Chem. Soc., 2003, 125, 1186.
26. J. E. A. Webb,
M. J. Crossley,
P. Turner,
and
P. Thordarson, J. Am. Chem. Soc., 2007, 129, 7155.

27. A. Gonazalez-Alvarez,
A. Frontera, and P. Ballester,
J. Phys. Chem. B, 2009, 113, 11479.
28. M. Kozbia
20, 506.

and J. Poznanski, J. Phys. Org. Chem., 2007,

29. P. N. Taylor and H. L. Anderson, J. Am. Chem. Soc., 1999,


121, 11538.
30. H. L. Anderson, Inorg. Chem., 1994, 33, 972.
31. P. Thordarson, R. G. E. Coumans, J. A. A. W. Elemans,
et al., Angew. Chem. Int. Ed., 2004, 43, 4755.
32. H. L. Anderson, C. A. Hunter, M. N. Meah, and J. K. M.
Sanders, J. Am. Chem. Soc., 1990, 112, 5780.
33. T. E. O. Screen, J. R. G. Thorne, R. G. Denning, et al.,
J. Mater. Chem., 2003, 13, 2796.
34. R. B. Martin, Chem. Rev., 1996, 96, 3043.
35. W.-J. Chu, Y. Yang, and C.-F. Chen, Org. Lett., 2010, 12,
3156.
36. K. Kano, H. Minamizono, T. Kitae, and S. Negi, J. Phys.
Chem. A, 1997, 101, 6118.

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

36

Techniques

37. K. W. K. Tong, S. Dehn, J. E. A. Webb, et al., Langmuir,


2009, 25, 8586.

48. C. S. Hanes, Biochemistry, 1932, 26, 1406.

38. K. A. Connors and J. A. Mollica Jr, J. Pharm. Sci., 1966,


55, 772.

50. G. Scatchard, Ann. N. Y. Acad. Sci., 1949, 51, 660.

39. K. Kavallieratos, S. R. de Gala, D. J. Austin, and R. H.


Crabtree, J. Am. Chem. Soc., 1997, 119, 2325.
40. H.-J. Schneider and A. Yatsimirsky, Principles and Methods
in Supramolecular Chemistry, John Wiley & Sons, Ltd,
Chichester, 2000, 149.
41. H. J. Motulsky and A. Christopoulos, Fitting Models to Biological Data Using Linear and Nonlinear Regression. A Practical Guide to Curve Fitting, GraphPad Software Inc., San
Diego, CA, 2003.
42. E. J. Billo, Excel for Chemists, Wiley-VCH Verlag GmbH,
New York, 1997.
43. J. M. Beechem, Methods Enzymol., 1992, 210, 37.
44. D. B. Hibbert and J. J. Gooding, Data Analysis for
Chemistry, Oxford University Press, New York, 2006.

49. J. B. Haldane, Nature, 1957, 179, 832.


51. H. J. Motulsky and L. A. Ransnas, FABES J., 1987, 1, 365.
52. J. W. Tukey, Exploratory Data Analysis, Addison-Wesley,
Reading, MA, 1977.
53. P. Gans, A. Sabatini, and A. Vacca, Talanta, 1996, 43, 1739.
54. M. J. Hynes, J. Chem. Soc., Dalton Trans., 1993, 311.
55. J. C. Lagarias, J. A. Reeds, M. H. Wright, and P. E. Wright,
SIAM J. Optimizat., 1998, 9, 112.
56. D. E. Goldberg, Genetic Algorithms in Search, Optimization
& Machine Learning, Addison-Wesley, Reading, MA, 1989.
57. Z. Ugray, L. Lasdon, J. C. Plummer, et al., INFORMS J.
Comput., 2007, 19, 328.
58. A. P. Bisson, C. A. Hunter, J. C. Morales, and K. Young,
Chem. Eur. J., 1998, 4, 845.

45. H. A. Benesi and J. H. Hildebrand, J. Am. Chem. Soc., 1949,


71, 2703.

59. H. J. Motulsky, Intuitive Biostatistics: A Nonmathematical


Guide to Statistical Thinking, 2nd edn, Oxford University
Press, New York, 2010.

46. H. Lineweaver and D. Burk, J. Am. Chem. Soc., 1934,


56, 658.

60. B. A. Blight, C. A. Hunter, D. A. Leigh, et al., Nature


Chem., 2011, 3, 244.

47. R. L. Scott, Recl. Trav. Chim. Pays-Bas, 1956, 75, 787.

Supramolecular Chemistry: From Molecules to Nanomaterials, Online 2012 John Wiley & Sons, Ltd.
This article is 2012 John Wiley & Sons, Ltd.
This article was published in the Supramolecular Chemistry: From Molecules to Nanomaterials in 2012 by John Wiley & Sons, Ltd.
DOI: 10.1002/9780470661345.smc018

Вам также может понравиться