of 28
Monod-Wyman-Changeux Analysis of Ligand-Gated Ion Channel
Mutants
Tal Einav
and
Rob Phillips
*,‡
Department of Physics, California Institute of Technology, Pasadena, California 91125, United
States
Department of Applied Physics and Division of Biology and Biological Engineering, California
Institute of Technology, Pasadena, California 91125, United States
Abstract
We present a framework for computing the gating properties of ligand-gated ion channel mutants
using the Monod-Wyman-Changeux (MWC) model of allostery. We derive simple analytic
formulas for key functional properties such as the leakiness, dynamic range, half-maximal
effective concentration ([EC
50
]), and effective Hill coefficient, and explore the full spectrum of
phenotypes that are accessible through mutations. Specifically, we consider mutations in the
channel pore of nicotinic acetylcholine receptor (nAChR) and the ligand binding domain of a
cyclic nucleotide-gated (CNG) ion channel, demonstrating how each mutation can be
characterized as only affecting a subset of the biophysical parameters. In addition, we show how
the unifying perspective offered by the MWC model allows us, perhaps surprisingly, to collapse
the plethora of dose-response data from different classes of ion channels into a universal family of
curves.
Graphical Abstract
*
Corresponding Author: phillips@pboc.caltech.edu. Phone: (626) 395-3374.
ORCID
Tal Einav: 0000-0003-0777-1193
Rob Phillips: 0000-0003-3082-2809
Notes
The authors declare no competing financial interest.
Supporting Information
The following Supporting Information is available: The Supporting Information is available free of charge on the ACS Publications
website at DOI: 10.1021/acs.jpcb.6b12672.-This material is available free of charge via the Internet at
http://pubs.acs.org/
.
Details on aforementioned derivations and calculations (PDF)
Supplementary
Mathematica
notebook from which all of the plots can be reproduced, including the dose-response data for both the
nAChR and CNGA2 systems (ZIP)
HHS Public Access
Author manuscript
J Phys Chem B
. Author manuscript; available in PMC 2017 August 10.
Published in final edited form as:
J Phys Chem B
. 2017 April 20; 121(15): 3813–3824. doi:10.1021/acs.jpcb.6b12672.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
1. INTRODUCTION
Ion channels are signaling proteins responsible for a huge variety of physiological functions
ranging from responding to membrane voltage, tension, and temperature to serving as the
primary players in the signal transduction we experience as vision.
1
Broadly speaking, these
channels are classified on the basis of the driving forces that gate them. In this work, we
explore one such classification for ligand-gated ion channel mutants based on the Monod-
Wyman-Changeux (MWC) model of allostery. In particular, we focus on mutants in two of
the arguably best-studied ligand-gated ion channels: the nicotinic acetylcholine receptor
(nAChR) and the cyclic nucleotide-gated (CNG) ion channel shown schematically in Figure
1.
2
,
3
The MWC model has long been used in the contexts of both nAChR and CNG ion
channels.
4
6
Although careful analysis of these systems has revealed that some details of
ligand-gated ion channel dynamics are not captured by this model (e.g., the existence and
interaction of multiple conducting states
7
,
8
), the MWC framework nevertheless captures
many critical features of a channel’s response and offers one of the simplest settings to
explore its underlying mechanisms. For example, knowledge of both the nAChR and CNG
systems’ molecular architecture and our ability to measure their complex kinetics has only
recently become sufficiently advanced to tie the effects of mutations to key biophysical
parameters. Purohit and Auerbach used combinations of mutations to infer the nAChR
gating energy, finding that unliganded nAChR channels open up for a remarkably brief 80
μ
s
every 15 min.
9
Statistical mechanics has been used to show how changes to the energy
difference between conformations in allosteric proteins translate to different functional
behavior (i.e., how it modifies the leakiness, dynamic range, [EC
50
], and the effective Hill
coefficient),
10
,
11
and we extend this work to find simple analytic approximations that are
valid within the context of ion channels. Using this methodology, we systematically explore
the full range of behaviors that may be induced by different types of mutations. This analysis
enables us to quantify the inherent trade-offs between key properties of ion channel dose-
response curves and potentially paves the way for future biophysical models of evolutionary
fitness in which the genotype (i.e., amino acid sequence) of allosteric molecules is directly
connected to phenotype (i.e., properties of a channel’s response).
Einav and Phillips
Page 2
J Phys Chem B
. Author manuscript; available in PMC 2017 August 10.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
To this end, we consider two distinct classes of mutants which tune different sets of MWC
parameters, either the ion channel gating energy or the ligand-channel dissociation
constants. Previous work by Auerbach et al. demonstrated that these two sets of physical
parameters can be independently tuned within the nAChR ion channel; pore mutations only
alter the channel gating energy whereas mutations within the ligand binding domain only
affect the ligand-channel dissociation constants.
12
Utilizing this parameter independence, we
determine the full spectrum of nAChR phenotypes given an arbitrary set of channel pore
mutations and show why a linear increase in the channel gating energy leads to a logarithmic
shift in the nAChR dose-response curve. Next, we consider recent data from CNGA2 ion
channels with mutations in their ligand binding pocket.
13
We hypothesize that modifying the
ligand binding domain should not alter the channel gating energy and demonstrate how the
entire class of CNGA2 mutants can be simultaneously characterized with this constraint.
This class of mutants sheds light on the fundamental differences between homo-oligomeric
channels composed of a single type of subunit and hetero-oligomeric channels whose
distinct subunits can have different ligand binding affinities.
By viewing mutant data through its effects on the underlying biophysical parameters of the
system, we go well beyond simply fitting individual dose-response data, instead creating a
framework with which we can explore the full expanse of ion channel phenotypes available
through mutations. Using this methodology, we (1) analytically compute important ion
channel characteristics, namely, the leakiness, dynamic range, [EC
50
], and effective Hill
coefficient; (2) link the role of mutations with thermodynamic parameters; (3) show how the
behavior of an entire family of mutants can be predicted using only a subset of the members
of that family; (4) quantify the pleiotropic effect of point mutations on multiple phenotypic
traits and characterize the correlations between these diverse effects; and (5) collapse the
data from multiple ion channels onto a single master curve, revealing that such mutants form
a one-parameter family. In doing so, we present a unified framework to collate the plethora
of data known about such channels.
MODEL
Electrophysiological techniques can measure currents across a single cell’s membrane. The
current flowing through a ligand-gated ion channel is proportional to the average probability
p
open
(
c
) that the channel will be open at a ligand concentration
c
. For an ion channel with
m
identical ligand binding sites (see Figure 2), this probability is given by the MWC model as
(1)
where
K
O
and
K
C
represent the dissociation constants between the ligand and the open and
closed ion channel, respectively,
c
denotes the concentration of the ligand,
ε
(called the
gating energy) denotes the free energy difference between the closed and open
conformations of the ion channel in the absence of ligand, and
where
k
B
is
Einav and Phillips
Page 3
J Phys Chem B
. Author manuscript; available in PMC 2017 August 10.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
Boltzmann’s constant and
T
is the temperature of the system. Wild-type ion channels are
typically closed in the absence of ligand (
ε
< 0) and open when bound to ligand (
K
O
<
K
C
).
Figure 2 shows the possible conformations of the nAChR (
m
= 2) and CNGA2 (
m
= 4) ion
channels together with their Boltzmann weights.
p
open
(
c
) is given by the sum of the open
state weights divided by the sum of all weights. Note that the MWC parameters
K
O
,
K
C
, and
ε
may be expressed as ratios of the experimentally measured rates of ligand binding and
unbinding as well as the transition rates between the open and closed channel conformations
(see Supporting Information section A.1).
Current measurements are often reported as
normalized
current, implying that the current
has been stretched vertically to run from 0 to 1, as given by
(2)
p
open
(
c
) increases monotonically as a function of ligand concentration
c
, with a minimum
value in the absence of ligand given by
(3)
and a maximum value in the presence of saturating levels of ligand given as
(4)
Using the above two limits, we can investigate four important characteristics of ion
channels.
10
,
11
First, we examine the
leakiness
of an ion channel, or the fraction of time a
channel is open in the absence of ligand, namely
(5)
Next we determine the
dynamic range
, or the difference between the probability of the
maximally open and maximally closed states of the ion channel, given by
(6)
Einav and Phillips
Page 4
J Phys Chem B
. Author manuscript; available in PMC 2017 August 10.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
Ion channels that minimize leakiness only open upon ligand binding, and ion channels that
maximize dynamic range have greater contrast between their open and closed states. Just
like
p
open
(
c
), leakiness and dynamic range lie within the interval [0,1].
Two other important characteristics are measured from the normalized current. The
half-
maximal effective concentration
[EC
50
] denotes the concentration of ligand at which the
normalized current of the ion channel equals 1/2, namely
(7)
The
effective Hill coefficient h
equals twice the log-log slope of the normalized current
evaluated at
c
= [EC
50
]
(8)
which reduces to the standard Hill coefficient for the Hill function.
14
The [EC
50
] determines
how the normalized current shifts left and right, while the effective Hill coefficient
corresponds to the slope at [EC
50
]. Together, these two properties determine the approximate
window of ligand concentrations for which the normalized current transitions from 0 to 1.
In the limit
, which we show below is relevant for both the nAChR and
CNGA2 ion channels, the various functional properties of the channel described above can
be approximated to leading order as (see Supporting Information section B):
(9)
(10)
(11)
(12)
Einav and Phillips
Page 5
J Phys Chem B
. Author manuscript; available in PMC 2017 August 10.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
RESULTS
nAChR Mutants Can Be Categorized Using Free Energy
Muscle-type nAChR is a heteropentamer with subunit stoichiometry
α
2
βγδ
, containing two
ligand binding sites for acetylcholine at the interface of the
α
-
δ
and
α
-
γ
subunits.
15
The five
homologous subunits have M2 transmembrane domains which move symmetrically during
nAChR gating to either occlude or open the ion channel.
16
By introducing a serine in place
of the leucine at a key residue (L251S) within the M2 domain present within each subunit,
the corresponding subunit is able to more easily transition from the closed to open
configuration, shifting the dose-response curve to the left (see Figure 3A).
17
For example,
wild-type nAChR is maximally stimulated with 100
μ
M of acetylcholine, while a mutant ion
channel with one L251S mutation is more sensitive and only requires 10
μ
M to saturate its
dose-response curve.
Labarca et al. used L251S mutations to create ion channels with
n
mutated subunits.
17
Figure 3A shows the resulting normalized current for several of these mutants; from right to
left, the curves represent
n
= 0 (wild-type) to
n
= 4 (an ion channel with four of its five
subunits mutated). One interesting trend in the data is that each additional mutation shifts the
normalized current to the left by approximately one decade in concentration (see Supporting
Information section A.2). This constant shift in the dose-response curves motivated Labarca
et al. to postulate that mutating each subunit increases the gating free energy
ε
by a fixed
amount.
To test this idea, we analyze the nAChR data at various concentrations
c
of the ligand
acetylcholine using the MWC model eq 1 with
m
= 2 ligand binding sites. Because the
L251S mutation is approximately 4.5 nm from the ligand binding domain,
18
we assume that
the ligand binding affinities
K
O
and
K
C
are unchanged for the wild-type and mutant ion
channels, an assumption that has been repeatedly verified by Auerbach et al. for nAChR
pore mutations.
12
Figure 3A shows the best-fit theoretical curves assuming all five nAChR
mutants have the same
K
O
and
K
C
values but that each channel has a distinct gating energy
ε
(
n
)
(where the superscript
n
denotes the number of mutated subunits). These gating energies
were found to increase by roughly 5
k
B
T
per
n
, as would be expected for a mutation that acts
equivalently and independently on each subunit.
One beautiful illustration of the power of the MWC model lies in its ability to provide a
unified perspective to view data from many different ion channels. Following earlier work in
the context of both chemotaxis and quorum sensing,
19
,
20
we rewrite the probability that the
nAChR receptor is open as
(13)
where this last equation defines the
Bohr parameter
21
Einav and Phillips
Page 6
J Phys Chem B
. Author manuscript; available in PMC 2017 August 10.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
(14)
The Bohr parameter quantifies the trade-offs between the physical parameters of the system
(in the case of nAChR, between the entropy associated with the ligand concentration
c
and
the gating free energy
βε
). When the Bohr parameters of two ion channels are equal, both
channels will elicit the same physiological response. Using eqs 1 and 13 to convert the
normalized current data into the probability
p
open
(see Supporting Information section A.3),
we can collapse the dose-response data of the five nAChR mutants onto a single master
curve as a function of the Bohr parameter for nAChR,
F
nAChR
(
c
), as shown in Figure 3B. In
this way, the Bohr parameter maps the full complexity of a generic ion channel response into
a single combination of the relevant physical parameters of the system.
Full Spectrum of nAChR Gating Energy Mutants—
We next consider the entire range
of nAChR phenotypes achievable by only modifying the gating free energy
ε
of the wild-
type ion channel. For instance, any combination of nAChR pore mutations would be
expected to not affect the ligand dissociation constants and thus yield an ion channel within
this class (see Supporting Information section A.4 for one such example). For concreteness,
we focus on how the
ε
parameter tunes key features of the dose-response curves, namely, the
leakiness, dynamic range, [EC
50
], and effective Hill coefficient
h
(see eqs 5–12), although
we note that other important phenotypic properties such as the intrinsic noise and capacity
have also been characterized for the MWC model.
10
Figure 4 shows these four
characteristics, with the open squares representing the properties of the five best-fit dose-
response curves from Figure 3A.
Figure 4A implies that all of the mutants considered here have negligible leakiness; the
probability that the wild-type channel (
βε
(0)
= −23.7) will be open is less than 10
−10
.
Experimental measurements have shown that such spontaneous openings occur extremely
infrequently in nAChR,
22
although direct measurement is difficult for such rare events.
Other mutational analysis has predicted gating energies around
βε
(0)
−14 (corresponding
to a leakiness of 10
−6
),
12
but we note that such a large wild-type gating energy prohibits the
five mutants in Figure 3 from being fit as a single class of mutants with the same
K
O
and
K
C
values (see Supporting Information section C.2). If this large wild-type gating energy is
correct, it may imply that the L251S mutation also affects the
K
O
and
K
C
parameters,
though the absence of error bars on the original data make it hard to quantitatively assess the
underlying origins of these discrepancies.
Figure 4B asserts that all of the mutant ion channels should have full dynamic range except
for the wild-type channel, which has a dynamic range of 0.91. In comparison, the measured
dynamic range of wild-type nAChR is 0.95, close to our predicted value.
12
Note that only
when the dynamic range approaches unity does the normalized current become identical to
p
open
; for lower values, information about the leakiness and dynamic range is lost by only
measuring normalized currents.
Einav and Phillips
Page 7
J Phys Chem B
. Author manuscript; available in PMC 2017 August 10.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
We compare the [EC
50
] (Figure 4C) and effective Hill coefficient
h
(Figure 4D) with the
nAChR data by interpolating the measurements (see Supporting Information section C.3) in
order to precisely determine the midpoint and slope of the response. The [EC
50
] predictions
faithfully match the data over 4 orders of magnitude. Because each additional mutation
lowers the [EC
50
] by approximately one decade, the analytic form eq 11 implies that
ε
increases by roughly 5
k
B
T
per mutation, enabling the ion channel to open more easily. In
addition to the L251S mutation considered here, another mutation (L251T) has also been
found to shift [EC
50
] by a constant logarithmic amount (see Supporting Information section
A.4).
23
We also note that many biological systems logarithmically tune their responses by
altering the energy difference between two allosteric states, as seen through processes such
as phosphorylation and calmodulin binding.
24
This may give rise to an interesting interplay
between physiological time scales where such processes occur and evolutionary time scales
where traits such as the [EC
50
] may be accessed via mutations like those considered here.
25
Lastly, the Hill coefficients of the entire class of mutants all lie between 1.5 and 2.0 except
for the
n
= 3 mutant whose dose-response curve in Figure 3A is seen to be flatter than the
MWC prediction. We also note that if the L251S mutation moderately perturbs the
K
O
and
K
C
values, it would permit fits that more finely attune to the specific shape of each mutant’s
data set. That said, the dose-response curve for the
n
= 3 mutant could easily be shifted by
small changes in the measured values, and hence, without recourse to error bars, it is
difficult to make definitive statements about the value adopted for
h
for this mutant.
Note that the simplified expressions eqs 9–12 for the leakiness, dynamic range, [EC
50
], and
effective Hill coefficient apply when
, which given the values of
K
C
and
K
C
for the nAChR mutant class translates to −22
βε
−5. The
n
= 1, 2, and 3 mutants all
fall within this range, and hence, each subsequent mutation exponentially increases their
leakiness and exponentially decreases their [EC
50
], while their dynamic range and effective
Hill coefficient remain indifferent to the L251S mutation. The
βε
parameters of the
n
= 0
and
n
= 4 mutants lie at the edge of the region of validity, so higher order approximations
can be used to more precisely fit their functional characteristics (see Supporting Information
section B).
Hetero-Oligomeric CNGA2 Mutants Can Be Categorized Using an Expanded MWC Model
The nAChR mutant class discussed above had two equivalent ligand binding sites, and only
the gating free energy
βε
varied for the mutants we considered. In this section, we use
beautiful data for the olfactory CNGA2 ion channel to explore the unique phenotypes that
emerge from a hetero-oligomeric ion channel whose subunits have different ligand binding
strengths.
The wild-type CNGA2 ion channel is made up of four identical subunits, each with one
binding site for the cyclic nucleotide ligands cAMP or cGMP.
26
Within the MWC model, the
probability that this channel is open is given by eq 1 with
m
= 4 ligand binding sites (see
Figure 2B). Wongsamitkul et al. constructed a mutated subunit with lower affinity for ligand
and formed tetrameric CNGA2 channels from different combinations of mutated and wild-
type subunits (see Figure 5).
13
Since the mutation specifically targeted the ligand binding
Einav and Phillips
Page 8
J Phys Chem B
. Author manuscript; available in PMC 2017 August 10.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
sites, these mutant subunits were postulated to have new ligand dissociation constants but
the same free energy difference
βε
.
We can extend the MWC model to compute the probability
p
open
that these CNGA2
constructs will be open. The states and weights of an ion channel with
n
mutated subunits
(with ligand affinities
and
) and
m
n
wild-type subunits (with ligand affinities
K
O
and
K
C
) are shown in Figure 5, and its probability to be open is given by
(15)
Measurements have confirmed that the dose-response curves of the mutant CNGA2 channels
only depend on the total number of mutated subunits
n
and not on the positions of those
subunits (for example, both
n
= 2 with adjacent mutant subunits and
n
= 2 with mutant
subunits on opposite corners have identical dose-response curves).
13
Figure 6A shows the normalized current of all five CNGA2 constructs fit to a single set of
K
O
,
K
C
,
, and
ε
parameters. Since the mutated subunits have weaker affinity to
ligand (leading to the larger dissociation constants
and
), the [EC
50
] shifts
to the right as
n
increases. As in the case of nAChR, we can collapse the data from this
family of mutants onto a single master curve using the Bohr parameter
F
CNGA2
(
c
) from eqs
13 and 15, as shown in Figure 6B.
Although we analyze the CNGA2 ion channels in equilibrium, we can glimpse the dynamic
nature of the system by computing the probability of each channel conformation. Figure 7A
shows the 10 possible states of the wild-type (
n
= 0) channel, the five open states
O
j
and the
five closed states
C
j
with 0 ≤
j
≤ 4 ligands bound. Figure 7B shows how the probabilities of
these states are all significantly shifted to the right in the fully mutated (
n
= 4) channel since
the mutation diminishes the channel-ligand affinity. The individual state probabilities help
determine which of the intermediary states can be ignored when modeling. One extreme
simplification that is often made is to consider the Hill limit, where all of the states are
ignored save for the closed, unbound ion channel (
C
0
) and the open, fully bound ion channel
(
O
4
). The drawbacks of such an approximation are 2-fold: (1) at intermediate ligand
concentrations (
c
[10
−7
, 10
−5
] M for
n
= 0 and
c
[10
−4
, 10
−2
] M for
n
= 4) the ion
channel spends at least 10% of its time in the remaining states which results in
fundamentally different dynamics than what is predicted by the Hill response, and (2) even
in the limits such as
c
= 0 and
c
→ ∞
where the
C
0
and
O
4
states dominate the system, the
Hill limit ignores the leakiness and dynamic range of the ion channel (requiring them to
exactly equal 0 and 1, respectively), thereby glossing over these important properties of the
system.
Characterizing CNGA2 Mutants Based on Subunit Composition—
We now turn to
the leakiness, dynamic range, [EC
50
], and effective Hill coefficient
h
of a CNGA2 ion
Einav and Phillips
Page 9
J Phys Chem B
. Author manuscript; available in PMC 2017 August 10.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
channel with
n
mutated and
m
n
wild-type subunits. Detailed derivations for the following
results are provided in Supporting Information section B:
(16)
(17)
(18)
(19)
Note that we recover the original MWC model results, eqs 5–12, for the
n
= 0 wild-type ion
channel. Similarly, the homo-oligomeric
n
=
m
channel is also governed by the MWC model
with
and
. We also show the [EC
50
] and
h
formulas for the
case
to demonstrate the fundamentally different scaling behavior that this hetero-oligomeric
channel exhibits with the MWC parameters.
Figure 8A shows that all of the CNGA2 mutants have small leakiness, which can be
understood from their small
ε
value and eq 16. In addition, the first term in the dynamic
range eq 17 is approximately 1 because the open state affinities are always smaller than the
closed state affinities by at least a factor of 10, which is then raised to the fourth power.
Thus, all of the mutants have a large dynamic range as shown in Figure 8B. Experimentally,
single channel measurements confirmed that the wild-type
n
= 0 channel is nearly always
closed in the absence of ligand; in the opposite limit of saturating cGMP, it was found that
for both the
n
= 0 and
n
=
m
ion channels (see Supporting Information section
C.2).
13
The [EC
50
] and effective Hill coefficient
h
are shown in Figure 8C,D. In contrast to the
nAChR case, where each additional mutation decreased [EC
50
], each CNGA2 mutation
tends to increase [EC
50
], although not by a uniform amount. The effective Hill coefficient
has a particularly complex behavior, first decreasing with each of the first three subunit
mutations and then finally increasing back to the wild-type level for the fully mutated ion
Einav and Phillips
Page 10
J Phys Chem B
. Author manuscript; available in PMC 2017 August 10.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
channel. To explain this decrease in
h
for the hetero-oligomeric channels, we first note that
the wildtype
n
= 0 channel has a sharp response about its [EC
50
]
e
βε
/
m
K
O
while the fully
mutated
n
=
m
channel has a sharp response about
. Roughly speaking,
the response of the hetero-oligomeric channels with 1 ≤
n
≤ 3 will occur throughout the full
range between e
βε
/
m
K
O
and
, which causes the dose-response curves to flatten
out and leads to the smaller effective Hill coefficient. Such behavior could influence, for
example, the response of the hetero-oligomeric nAChR ion channel if the two acetylcholine
binding pockets diverged to have different ligand affinities.
Although we have focused on the particular mutants created by Wongsamitkul et al., it is
straightforward to apply this framework to other types of mutations. For example, in
Supporting Information section B.2 we consider the effect of modifying the
K
O
and
K
C
parameters of all four CNGA2 channels simultaneously. This question is relevant for
physiological CNGA2 channels where a mutation in the gene would impact all of the
subunits in the homo-oligomer, in contrast to the Wongsamitkul constructs where the four
subunits were stitched together within a single gene. We find that when
K
O
and
K
C
vary for
all subunits, the leakiness, dynamic range, and effective Hill coefficient remain nearly fixed
for all parameter values, and that only the [EC
50
] scales linearly with
K
O
as per eq 11. In
order to affect the other properties, either the gating energy
βε
or the number of subunits
m
would need to be changed.
Extrapolating the Behavior of a Class of Mutants
In this section, we explore how constant trends in both the nAChR and CNGA2 data
presented above provide an opportunity to characterize the full class of mutants based on the
dose-response curves from only a few of its members. Such trends may well be applicable to
other ion channel systems, enabling us to theoretically probe a larger space of mutants than
what is available from current data.
First, we note that because the [EC
50
] of the five nAChR mutants fell on a line in Figure 4C,
we can predict the response of the entire class of mutants by only considering the dose-
response curves of two of its members and extrapolating the behavior of the remaining
mutants using linear regression. Experimentally, such a characterization arises because the
L251S mutation acts nearly identically and independently across subunits to change the
gating free energy of nAChR.
12
,
17
,
23
This implies that mutating
n
subunits would yield an
ion channel with gating energy
(20)
where
ε
(0)
is the wild-type gating energy and Δ
ε
is the change in free energy per mutation.
This functional form is identical to the mismatch model for transcription factor binding,
where each additional mutation-induced mismatch adds a constant energy cost to binding.
27
Figure 9A demonstrates how fitting only the
n
= 0 and
n
= 4 dose-response curves (solid
lines) accurately predicts the behavior of the
n
= 1, 2, and 3 mutants (dashed lines). In
Einav and Phillips
Page 11
J Phys Chem B
. Author manuscript; available in PMC 2017 August 10.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript
Supporting Information section D, we carry out such predictions using all possible input
pairs.
We now turn to the CNGA2 ion channel where, once the
K
O
,
K
C
,
, and
ε
parameters
are known, the dose-response curve of any mutant can be predicted by varying
n
in eq 15.
Figure 9B demonstrates that the wild-type ion channel (
n
= 0) and the ion channel with only
one mutated subunit (
n
= 1) can accurately predict the dose-response curves of the remaining
mutants. Supporting Information section D explores the resulting predictions using all
possible input pairs.
MWC Model Allows for Degenerate Parameter Sets
One critical aspect of extracting model parameters from data is that degenerate sets of
parameters may yield identical outputs, which suggests that there are fundamental limits to
how well a single data set can fix parameter values.
25
,
28
This phenomenon, sometimes
dubbed “sloppiness,” may even be present in models with very few parameters such as the
MWC model considered here. Figure 10 demonstrates the relationship between the best-fit
parameters within the nAChR and CNGA2 systems. For concreteness, we focus solely on
the nAChR system.
After the value of
K
O
is fixed (to that shown on the
x
-axis of Figure 10A), the remaining
parameters are allowed to freely vary in order to best fit the nAChR data. Although every
value of
K
O
[10
−11
–10
−9
] M yields a nearly identical response curve in excellent
agreement with the data (with a coefficient of determination
R
2
> 0.96), we stress that
dissociation constants are rarely found in the range
K
O
10
−10
M. In addition, a
dissociation constant above the nanomolar range,
K
O
10
−9
M, cannot fit the data well and
is therefore invalidated by the model. Thus, we may suspect the true parameter values will
fall around the interval
K
O
[10
−10
–10
−9
] M for the nAChR system.
K
O
could ultimately be
fixed by measuring the leakiness eq 5 (and thereby fixing
βε
) for any of the ion channel
mutants.
Two clear patterns emerge from Figure 10A: (1) The value of
K
C
is approximately constant
for all values of
K
O
, and (2) the five free energies all vary as
βε
(
n
)
= 2 log(
K
O
) +
n
(constant). This suggests that
K
C
and e
βε
/2
K
O
are the fundamental parameter combinations
of the system. In Supporting Information section C.4, we discuss how these parameter
combinations arise from our model.
We end by noting that the notion of sloppiness, while detrimental to fixing the physical
parameter values of a system, nevertheless suggests that multiple evolutionary paths may
lead to optimal ion channel phenotypes, providing another mechanism by which allostery
promotes a protein’s capacity to adapt.
29
DISCUSSION
There is a deep tension between the great diversity of biological systems and the search for
unifying perspectives to help integrate the vast data that has built up around this diversity.
Several years ago at the Institut Pasteur, a meeting was convened to celebrate the 50th
Einav and Phillips
Page 12
J Phys Chem B
. Author manuscript; available in PMC 2017 August 10.
Author Manuscript
Author Manuscript
Author Manuscript
Author Manuscript