WorksPhysics and Astronomy
Canonic hemoglobin-oxygen equilibrium: Reassessing the role of cooperativity
Abstract
Hemoglobin-oxygen equilibrium is normally studied within the grand canonical ensemble, which assumes that each hemoglobin is immersed in a reservoir of unbound oxygen molecules of the plasma. We show that this assumption is incorrect inside the RBC (red blood cell or erythrocyte), where the hemoglobin concentration is larger than the concentration of unbound oxygen in plasma. We suggest a better model, where a single hemoglobin and a few oxygen molecules around it reach a canonical equilibrium at a fixed volume (determined from the RBC structure) and a fixed temperature. The basic models of hemoglobin-oxygen equilibrium – Pauling's model and Monod-Wyman-Changeux model – can be reformulated for this canonical situation. They predict cooperative interaction energies, which are significantly lower than predictions of the same models in the grand canonical ensemble. Larger cooperative energies, in particular those predicted by the grand canonical ensemble, lead to instabilities (cascade processes) in oxygen release within the canonical approach. Oxygen-binding fluctuations within this approach are sizably smaller than those in the grand canonical situation. These results suggest that the hopping diffusion of oxygen from one hemoglobin to another may play a role in oxygen diffusion within the RBC.
Introduction
The hemoglobin molecule (Hb) plays several important physiological roles, including oxygen () transport [1, 2]. Hb is also a key example of cooperative interactions in biological physics [1, 2]. Two major scenarios of cooperativity, induced fit and (shifted) conformational equilibrium, were developed in the context of binding to Hb [3–8].
The standard description of the Hb- equilibrium starts with the grand canonical ensemble, where a single Hb molecule is immersed into a particle reservoir of molecules, described by a definite partial pressure [1–8]. The reservoir amounts to many unbound molecules dissolved in the RBC (red blood cell or erythrocyte) cytoplasm or in blood plasma; see Figure 1. Hence, the Hb- equilibrium can be monitored by the mean number of molecules bound to Hb [3–8]. The current understanding of Hb as a cooperative molecule is based on the grand canonical ensemble [1].
Figure 1. (a) The canonical ensemble for the hemoglobin (Hb) molecule (schematically a circle) with (fully bound) O molecules immersed in a thermal bath (water molecules) at temperature . The surrounding volume for (potentially) unbound O molecules is , which is times larger than the proper Hb volume ; see (12) and Table II. Now can refer to a cube with linear size 6.93 nm. The canonical equilibrium is justified since each Hb macromolecule in RBC is surrounded by water molecules; cf. Table II.
Dotted lines refer to Pauling’s model, where bound O molecules interact along the rectangular sides. The yellow bar denotes the RBC cellular membrane, via which the thermalized Hb can exchange particles with the external environment. The arrow indicates that a fully loaded Hb cannot attract more particles, but can release O into blood plasma (or to other Hb); see (29).
(b) The grand canonical ensemble: the Hb macromolecule with volume is immersed in an O reservoir with partial pressure . The reservoir volume holds , and contains number of O molecules. The minimal realization of the particle reservoir with blood plasma is as follows. Given that the concentration of O molecules in the alveolar blood plasma is molecules/, can refer to a cube with the linear size 0.1 m, which will then contain O molecules. However, in RBC, such a contains Hb macromolecules, and not a single Hb; cf. Table II. Section II D shows that the condition per Hb molecule is needed for the grand canonical ensemble.
(c) Two Hb macromolecules with (resp.) and in contact. The O transport is governed by free energies; see (30). As expected, for a moderate cooperativity, the O numbers tend to equalize; see the arrow. This conclusion does not hold for stronger cooperative interactions; see the discussion around (37).
In the context of Hb and its basic function of transport, the grand canonical ensemble has the following drawback. Within the RBC, the concentration of plasmatic is at least 100 times lower than the concentration of bound to Hb [2]; see section II. This is the very reason why the whole machinery of transport via Hb exists (the dissolved in the blood is not sufficient for organisms). Hence, even in arterial blood (where Hb is almost fully -loaded), the majority of Hb molecules do not see any cytoplasmic molecule in their vicinity; each Hb sees only the 4 molecules attached to it. This is far from the standard setting of the grand canonical ensemble, where a single Hb is immersed into a reservoir of molecules; see Figure 1.
We resolve this problem by looking at a single Hb, which together with oxygen molecules (bound or located at its vicinity) is in the canonical equilibrium with a fixed and temperature ; see Table I for the comparison of canonical versus grand canonical ensembles. The canonical ensemble demands an additional parameter: the volume around the Hb available for motion of unbound molecules. The choice of is not free, since Hb molecules are tightly packed in RBC. This demands to be larger but comparable to the proper Hb volume ; see Figure 1. The grand canonical ensemble is recovered as a limiting case for , and , where only enters into pertinent quantities, with being the partial pressure, and is the particle reservoir volume; see section II D.
The main difference between the canonical ensemble and the grand canonical one is as follows. In the latter, the oxygen transport is controlled by the mean number of bound molecules, as a function of the partial pressure of the -reservoir. In the former, the absorption/release of from/to blood plasma is governed by differences of the canonical free energy as a function of . The free energy mechanism puts additional thermodynamical constraints on Hb models (such constraints are not effective for the grand canonical ensemble). It allows to reassess the role of cooperativity in O binding by Hb. In addition, the canonical ensemble allows to describe O transport in between of Hb molecules; this mechanism is absent for the grand canonical ensemble.
Two major models of cooperative oxygen binding to hemoglobin – Pauling’s model [3, 4, 8] and MWC model [5, 6] – are standardly studied within the grand canonical ensemble. These models implement different mechanisms of cooperativity: Pauling’s model assumes interaction energy between the binding sites, while the MWC model introduces two different states for Hb [3–8]. Historically, Pauling’s model was proposed earlier than the MWC model [3, 5]. The current consensus is that the MWC model and its generalizations are a better description of Hb-O equilibrium and related features [10] [4, 9]. However, Pauling’s model is useful for explaining a possible cooperativity scenario [1].
We show that these models can be formulated within the above canonical ensemble. The grand canonical ensemble is a particular case of this more general formulation; see section II. Reassessing Pauling’s model and MWC model, we find that (for both models) the canonical ensemble consistently predicts lower values for the cooperativity parameters. Moreover, we demonstrate that within the canonical approach, higher values of cooperativity parameters will lead to various forms of instability, e.g., a cascade process of O release in capillaries. The identification of these instabilities is a conceptual advantage of the canonical approach: within the grand canonical ensemble, the cooperativity parameters do not have such bounds. As a result, the instabilities provide a suitable test setup for checking the values of cooperativity parameters.
Another important difference between the ensembles is that the physical picture of the O transport function of the Hb macromolecule within the canonical ensemble is machine-like: the O is tightly bound to Hb and is released/accepted in discrete steps nearly without fluctuations. As compared to its mean, the standard deviation of the bound O molecule number is suppressed. This contrasts the grand canonical ensemble, where Hb works as a fluctuating, statistical system with a standard deviation of the same order as the mean number of bound molecules.
Note that there are several pertinent physical situations where previous literature has detected important differences between canonical and grand canonical ensembles. Ref. [11] examined such differences for chemical reactions in the low concentration regime; the approach of Ref. [11] was criticized in Ref. [12]. The differences are closely related to the validity of the mass action law in chemical kinetics [13]. Refs. [14, 15, 16] studied them in the context of quasi-free electron models in solid state physics. An important point of Refs. [14, 15, 16] is that the differences persist even for systems that are macroscopic under the operational meaning of this concept.
The paper is organized as follows. § 2 studies -Hb equilibrium within the canonical ensemble. Thermodynamics of diffusion is discussed in section III. Pauling’s model is studied in section IV A. The obtained results are compared to those found from the grand canonical ensemble; see Appendix A for details. The MWC model is worked out in section IV B. We summarize in the last section, and underline directions for future research, which, in particular, should include non-equilibrium versions of the studied models.
Model for a hemoglobin-oxygen equilibrium
Criticism of the standard approach
The standard approach amounts to describing the Hb-O equilibrium via the grand canonical approach, where a single Hb molecule is immersed in the reservoir of plasmatic (unbound) molecules; see Figure 1. A necessary condition for this is that the concentration of plasmatic is much larger than the concentration of Hb in RBC. (We neglect the equilibrium partial pressure differences between blood plasma and RBC cytoplasm.) This condition does not hold in RBC: ; see Table II. This estimate can be generalized beyond RBC: we can ask for Hb concentration in the blood, which is the concentration of RBC in the blood times the number of Hb’s in a single RBC. Even this number is still larger than : ; cf. Table II. Further details on the applicability of the grand canonical ensemble to the Hb-O equilibrium are discussed in section IID and Appendix § 3.2.
| Notation | Value | Meaning |
| (blood) | molecules/ | Concentration of in alveolar blood plasma at 100 torr |
| (water) | molecules/ | Concentration of in fresh water at 100 torr |
| (air) | molecules/ | Concentration of in air at 100 torr |
| molecules/ | Concentration of in RBC | |
| molecules/ | Concentration of Hb in RBC | |
| The volume of the Hb heme pocket for molecule | ||
| Radius of Hb molecule approximated as a sphere | ||
| cells/ | Concentration of RBC | |
| Number of Hb in RBC | ||
| Volume of RBC | ||
| Diffusion constant of in () | ||
| Diffusion constant of in RBC (+ membrane permeability) | ||
| Diffusion constant of Hb in RBC including crowding effects |
Table II. Quantities relevant for RBC (erythrocyte) and hemoglobin (Hb) [25, 26] under normal conditions.
Canonical ensemble
We consider a single Hb molecule with overall volume and 4 traps. Their volumes are specified below. The free volume (where unbound molecules move) is ; see Figure 1. There are molecules, and no two particles can enter the same trap. Unbound molecules do not interact, while bound molecules do (cooperativity). This interaction is specified below. The equilibrium coordinate density of particles reads [21]:
where is the coordinate part of the canonical partition function, is the inverse temperature (), and where is the potential energy, which is invariant with respect to all permutations of . We note:
where with volume is the space domain of the unbound motion, and where includes the binding domain and its vicinity (). In the integration is restricted to . Putting (2) into (1), we get
where (say) the index in (3) counts the case when the first particle is unbound (), is bound to the first binding volume (), etc. We assume is such that if among there are at least two coinciding non-zero indices, e.g., . This corresponds to two particles being in a single well, which is prohibited. We assume that unbound particles move in volume and do not interact with each other. Bound particles interact with each other. The interaction mechanism depends on the number of bound particles, but (naturally) does not depend on their label. The mechanism is illustrated as follows for :
where only the fourth particle is bound in , is the corresponding binding volume, and is the energy. Neither nor are dependent on which molecule got bound and where it was bound. Likewise,
where the second bound molecule sees a different binding volume and energy . Again, these parameters do not depend on the number of the particle and the number of the trap. The differences and allow for cooperativity within the induced fit approach. (The MWC model employs a different mechanism of cooperativity, but still fits within the present model; see section IV B.)
Hence, if contains () non-zero indices (i.e., bound particles and unbound ones), we get
where is the (effective) free energy. The number of different factors that satisfy (6) equals , where comes from the number of places within , is responsible for permutation of non-zero indices in , and refers to the selection of binding volumes out of 4. We have from (3, 6):
The probability that molecules are bound is found from (8, 7):
The mean number of bound particles reads from (10):
We focused on the coordinate distribution (1) and the corresponding partition function (8). This is sufficient for our purposes. Appendix C recalls the full partition function of the model, also including the momentum part.
Parameters of the canonical distribution
The very structure of RBC, together with our intention of describing a single Hb molecule, estimates the volume around a Hb molecule as [see Table II]:
is the Hb concentration in the RBC, and is the Hb radius (the Hb is approximated as a sphere), and is the Hb volume.
Let us discuss the characteristic relaxation times of the model. It roughly amounts to relaxation of the molecule. The latter relaxation process consists of two steps. First, the molecule diffuses in water to reach the Hb. This time estimates as , where is the diffusion coefficient of in ; see Table II. Second, reaches from the surface of the heme pocket to its interior. This takes longer time [27].
Note that applying the concept of diffusion coefficient requires faster local equilibration, which is satisfied here. The relaxation time of an molecule in the water thermal bath can be estimated from the friction term in the Langevin equation as , where and are (resp.) the mass and radius of , and is the water viscosity. The rationale behind this standard estimate was recently reviewed in Ref. [23], which also discussed that since , the friction in the Langevin equation is not memoryless. This fact does not alter the order of magnitude estimate of the relaxation time presented here [23]. This produces for the fast local relaxation time .
Grand canonical ensemble
The transition to the grand canonical ensemble is carried out as follows. In (8), we assume that , and approximate uniformly . Hence in (10) contain and only in combination (concentration). Hence, from the viewpoint of (10), the Hb molecule is immersed into a particle reservoir with concentration . This is the way the grand canonical ensemble is applied in practice, because the particle reservoir can now be the blood plasma; see Table I. Now is the reservoir volume. It should also be large, but both and can be left unspecified, since only enters into . Using the ideal gas law, we can express via the partial pressure in air, which is controlled externally:
| Canonical ensemble | Grand canonical ensemble |
| Temperature is fixed | Temperature is fixed |
| Number of molecules around a single Hb and the volume available for unbound motion of molecules are fixed; see (9–7) | Partial pressure of reservoir is fixed; see (13) |
| transfer between Hb and plasma is governed by the free energy difference (19) | transfer between Hb and plasma is governed by the number of bound molecules as a function of ; see (15) |
| transfer between Hb molecules is governed by the free energy difference (30) | transfer between Hb molecules is undefined |
Table I. Canonical versus grand canonical ensemble for oxygen-hemoglobin (-Hb) equilibrium.
where is Henry’s constant for air-blood plasma equilibrium. We get from (8, 11) for the partition function and the mean number of bound molecules in the grand canonical ensemble:
Eq. (11) is similar to Adair’s equation [20] [1]. For the grand-canonical situation, in (15) is the basic diagnostic quantity, which, in particular, allows one to decide on cooperativity. For the canonical situation, does not have the same diagnostic meaning. This role is played by the free energy, which is discussed below.
Applicability limits of canonical and grand canonical ensembles are indeed different; cf. Table I. One can ask whether the above results on the grand canonical ensemble can be derived more generally by including into an equilibrium description all molecules and Hb molecules of the RBC. Appendix § 3.2 answers this question negatively.
We mention here another salient feature of the grand canonical ensemble: assume an isolated Hb in alveolar blood. To immerse this Hb into an particle reservoir, we can assume that the characteristic size of this reservoir is . Then Hb will be surrounded by molecules; cf. Table II. Surrounded means that any reservoir particle can reach the Hb by diffusion. Noting the value of the diffusion constant from Table II, we see that the time is s. For shorter processes, the grand canonical ensemble does not apply to the Hb.
Free-energy and thermodynamics of diffusion
O transport with the outer environment
Free energy differences
The free energy of the Hb with molecules reads from (8, 9, 10):
Consider a Hb molecule in contact with the RBC membrane; see Figure 1. We can employ (16) for determining the transport between the Hb and the dissolved in the blood plasma. To this end, we assume that molecules dissolved in the blood plasma make an ideal dilute solution in volume and temperature . The corresponding free energy, and the total free energy read (resp.)
We did not include in (16–18) the partition function coming from the kinetic energy of molecules [see Appendix § 3.1], because it is , and hence it will cancel out from , as seen below.
The redistribution across the membrane tends to minimize the free energy (18). If a single molecule moves from the Hb to the plasma during a quasi-equilibrium process, the free energy change reads
in (19) means that absorption by the Hb is more likely statistically. This is most clearly seen within a non-equilibrium description, where the free-energy difference determines the transition rate from one state to another [1]. In the same sense, means release. Note that when the reasoning of (16–18) is applied to two ideal gases (dilute solutions), we get - consistently with the most usual understanding of diffusion - that the particle has statistical tendency to go from a higher concentration to a lower one.
Bounds for free energy differences in O transport
We want to find a realistic estimate for the parameter in (19). Given the volume of RBC (and approximating it as a sphere) the surface of RBC is estimated as ; see Table II. In the present model, we surrounded each Hb molecule by the cytoplasm volume so that each Hb+ amounts to a cube with side 6.93 nm; cf. Figure 1. Thus, the number of Hb molecules located next to the RBC surface is estimated as [cf. Table II]:
In the blood plasma, the number of O molecules per a RBC can be estimated with the same idea as for (12). For the plasma volume per RBC we get , where is the RBC concentration; see Table II. Thus the number of plasmatic O molecules per RBC reads:
where is the O concentration in alveolar blood; see Table II. Comparing (21) and (22), we see that there are O molecules per one Hb molecule, which is located at the RBC surface. Now we can estimate in (19, 20) for alveolar blood. We choose for the smallest volume which is about to contain O molecules, i.e., and . Together with (12), this produces for the alveolar blood plasma
Eq. (23) does not depend on the concrete value of , provided that . The minimal condition for ensuring that the Hb is fully loaded in lungs is deduced from (19–23):
Table III presents a standard data about Hb saturation. These data is obtained from macroscopic measurements and suggests the following (to some extent conventional, but reasonable) conditions to be applied to a single Hb
where (25) means that the first O molecules is released to the outer environment when in blood plasma decreases from 80 torr to 40 torr. According to (26) the second O is released when in blood plasma changes from 40 to 30 torr. The third O molecules is released when changes between 30 and 20 torr; see (27). Finally, the last O molecule is released when changes between 20 and 7.5 torr, as (27) shows. We emphasize again that the bounds presented in (25–28) are somewhat conventional, because we had to decide how to fit 7 (approximate) numbers to 4 released O molecules. At the present stage of our understanding, such conventionalities are unavoidable.
Note from (8, 20) that for , every term in is smaller than the corresponding term in ; hence
For , contains fewer terms than , and for becomes possible. The message of (29) is that even in alveoles, the filled Hb will not attract more O molecules into its vicinity. Indeed, such an attraction would require , which is impossible due to (29). This conclusion is valid also for a two-molecule O transfer from/to the blood; see Appendix B.
O transport between two Hb molecules
We can apply (16) and the concept of the free energy decrease to the O exchange between two Hb’s with different values of ; e.g. denote by the pair with (resp.) and molecules. We get [cf.(19)]:
Now means stability of state against the spontaneous emergence of the gradient of ; i.e.
The sign of in (31) is not fixed thermodynamically, i.e., the spontaneous emergence of the gradient is possible. The reasoning based on free energy determines statistical trends only. It is silent on characteristic times of O transfer. Determining these times demands dynamic models, which we shall develop in the future based on the present static approach.
Note that the oxygen transport between two Hb molecules is not defined within the grand canonical ensemble.
Different models of cooperativity
The general framework developed above for the hemoglobin-oxygen equilibrium requires accounting for the cooperativity of oxygen binding to the Hb molecule. We will address below the implementation of cooperativity based on two widely known models, proposed (resp.) by Pauling [3] and Monod, Wyman, and Changeux [5]; see [1].
Pauling’s model
This simplest, two-parametric model of Hb was proposed by Pauling [3]. Later on, it was generalized in Ref. [7] within the induced fit (or sequential) approach [4]. A seemingly different approach to hemoglobin was proposed, which is based on the one-dimensional Ising model [8]. However, this approach can be reduced to Pauling’s model [8]. We focus on Pauling’s model, since this is the simplest and clearest example of the induced fit (or sequential) approach.
The model amounts to introducing into (8) a notation and a new parameter : [3]
where is the additional free energy, stabilizing two interacting HbO groups [3]. Now from (8) reads
The logics of introducing the cooperativity parameter into (33) follows Pauling’s idea that the cooperativity in oxygen molecules binding is present only along the dotted lines of the square in Figure 1(a), but not along its diagonals [3]. If only one O molecule is bound, then no cooperativity is present; hence the factor in (33). If all molecules all present then there is full cooperativity along four sides os the square; hence in (33). Next, there are four arrangements of the three molecules, and the cooperativity goes along two sides of the square; hence, the factor . Finally, in (33) is explained by noting that there are 6 possible arrangements of two O molecules on the square. Still, two of them do not lead to cooperativity, since the molecules are arranged along the diagonals of the square. Note that cooperativity is described by ; no cooperativity means . The logic of introducing remains thermodynamic and phenomenological. Mesoscopic features that enable the Hb cooperativity are discussed in [17, 18].
The parameters and of Paulings model are found by calculating from (20, 33) and using (25–28) for fitting the values . For definiteness, (for ) will be fit to the middle point of the corresponding interval in (25–28); e.g. is fit to . The best fit is obtained numerically under the least square criterion. It reads (recall that ):
The obtained values of are shown in Figure 2 as black triangles. It is seen that inequalities (25–28) are satisfied. According to (35), the cooperativity factor is smaller than the prediction found for the grand canonical implementation of Pauling’s model, which is fit to the data of Table III; see around (A5) for more details. Recall that the value of does not depend on the ensemble used; therefore, comparing them is legitimate. Pauling found even a larger value for the grand canonical model [3], but he employed the data by Ref. [19]. We shall see below that the difference between and is indeed physically substantial.
| (torr) | SaO (%) | Interpretation |
| 80 | 95 | Almost full |
| 60 | 90 | Beginning of O release |
| 40 | 75 | Resting tissue O delivery |
| 30 | 55 | Active tissue (e.g. muscle during exercise) |
| 20 | 35 | Very active tissue (vigorous muscle activity) |
| 10 | 15 | Severe hypoxia |
| 5 | 5 | Critical hypoxia |
Table III. Relationship between the partial oxygen pressure in blood plasma [cf. (13)], Hb saturation SaO, and clinical interpretation of oxygenation. At 100 torr (lungs) the O concentration in blood plasma is molecules/cm; see Table II. For other values of , the concentration is recalculated linearly. The above data is taken from Ref. [25]. It is measured at standard physiological conditions: pH 7.4 in blood plasma (pH 7.24 in RBC); the ratio of concentrations [2,3-BPG]/[Hb] 0.9 ([2,3-BPG] = 4.65 mM); = 37 °C. The data for torr is consistent with interpolations of Severinghaus and Hill equations [28].
!Free energy levels: versus for Pauling’s model bbox=0.37,0.07,0.70,0.27
Figure 2. Free energy levels: versus for Pauling’s model; see (20, 33). Horizontal red line: ; cf. (23). Four upper horizontal black lines refer (resp.) to the right-hand sides of (25–28). The lowest horizontal black line denotes ; see (28). Black triangles, magenta points, brown points, and blue points are the values of for (resp.) , , , and ; cf. (35, A5). The latter case with means no cooperativity. Black triangles refer to the best fit (35). They hold the inequalities in (25–28).
Note that for different values of , only the fitted value of changes, while stays nearly intact. For example, for and we get (resp.) and with the same .
The mean number of bound particles found from (11, 34) holds [see Table IV]:
| 4 | 3 | 2 | 1 | |
| 3.984 | 2.998 | 1.996 | 0.997 | |
| 0.125 | 0.097 | 0.077 | 0.0590 |
Table IV. For Pauling’s model under (35), this table shows that , i.e., O molecules are strongly bound to the Hb macromolecule. For the grand canonical ensemble of Pauling’s model, we get indicating on the relatively weak binding; see Figure 4. Note that the data in this table shows that grows with , in contrast to grand canonical behavior of . Also grows with changing from 16.9 for to 31.9 for .
i.e., O molecules are strongly bound to the Hb. In contrast, for the grand canonical ensemble of Pauling’s model, we get indicating on the relatively weak binding; see Figure 4. Another difference between canonical and grand canonical situations is that for the former, grows monotonically with ; see Table IV and compare it with Figure 4.

Figure 4. Predictions of Pauling’s and the MWC model (grand canonical ensemble) versus the partial pressure . Blue and orange curves are (resp.) the mean saturation and the standard deviation calculated from (A3) under parameters (A4) reported by Pauling [3]. Black and red curves are (resp.) the mean saturation and the standard deviation calculated for the MWC model [cf. (A6)] under parameters (A9) reported by Eaton [4]. For the MWC model, we do not show (probability of the R-state), since for the considered range of parameters it nearly coincides with .
Why is cooperativity needed at all? For (no cooperativity) and we can satisfy (25), but (26–28) cannot be satisfied, because the corresponding values of , and are too big; see Figure 2. Hence, without cooperativity the Hb can be loaded in lungs, but will serve only resting tissues; see Table III.
What can happen for such large values of ? Figure 2 shows for and . The latter value is chosen such that inequality (25) is satisfied. It is seen that the basic condition (24) does not hold for , i.e., such a model Hb cannot be loaded in lungs. We can take somewhat smaller value of such that (24) is satisfied: see Figure 2 for and . But now we get a cascade process of O release, which is described via
This means that once the fourth bound O molecule is released – e.g., for , as in (25) – the other three molecules will be released automatically. This is not an acceptable scenario, because it will lead to O poisoning. Figure 3 shows for Pauling’s model the free energy difference defined in (30). It is seen that for all studied values of parameters, the state of the Hb pair is the most stable one against the spontaneous gradient formation: assumes the largest value, which is almost parameter independent; see Figure 3. However, for strongly cooperative situations with and , the states and are instable: spontaneously moves to , while moves to . Though we do not involve detailed dynamical calculations in this paper, we note that such spontaneous gradient formations will impede the O diffusion across the RBC.
!Plot of versus for Pauling’s model bbox=0.35,0.27,0.71,0.53
Figure 3. versus for Pauling’s model; see (30, 33). Black, magenta, and brown points are the values of for (resp.) , , and ; cf. (35, A5). Now means that the Hb pair with O molecules will spontaneously move (due to cooperativity) to ; cf. (31).
Monod-Wyman-Changeux (MWC) model within the canonical ensemble
We return to (8) and recall that this model describes cooperativity via direct coupling between binding sites. Within (8), no cooperativity means that does not depend on , and then . In addition to this assumption of no direct cooperativity, MWC model introduces two different states (R and T), which have different affinity to oxygen [4–6]; see (43) below. Now depends on the state: , . The mechanism of cooperativity proposed by MWC allows thermal transitions between T and R, and chooses such that these transitions are influenced by oxygen binding. In this way, MWC model explains cooperativity via allostery [10]. Like Pauling’s model, MWC is based on statistical thermodynamics. According to Perutz, the MWC model is based on conformational features (quaternary and partially tertiary) of Hb: R (T) refers to oxyhemoglobin (deoxyhemoglobin); see [24] for a review.
The canonical partition function of the MWC model is found by analogy to (8):
where the energy controls relative weights of R and T states. Working out (38) we get
where choosing does not contain any loss of generality. The probabilities of bound O molecules follow from (39) and refer to separate terms in :
Likewise, the probability of the -state reads ()
For three limiting cases: , , and we get from (38), (42) the (simplest) non-cooperative version of the Hb-O equilibrium with a single parameter . More generally, the model has three independent parameters: , , and , and the deviation from the non-cooperative equilibrium is not characterized by a single parameter.
We now calculate via (39), (20). The best fit of parameters , , and to the data in Table III [see before (35) for details] produced:
For the grand-canonical model the fit to the data of Table III led to a sizable difference: and ; see (A9, A10) in Appendix A. The latter value is comparable with the value reported by Eaton [4]. As compared to (43), these grand-canonical values indicate a stronger deviation from the non-cooperative situation.
In contrast to Pauling’s model, where there is a single parameter characterizing direct interaction between binding sites, the cooperativity in the MWC model is an emergent feature of the equilibrium shift due to O binding. Eqs. (42, 43) show that for we have , while for we have due to
which is satisfied for (43). Table V shows results found from the MWC model under (43). The behavior of and is very similar to what Table IV predicts for Pauling’s model. Again, we witness a clear difference with the grand canonical situation. As expected, increases from with one O binding to when all four O molecules are bound. Given the deterministic behavior of the number of bound O molecules [see Table V], and indeterministic behavior of , it is legitimate to say that the O binding indeed drives the state change. This contrasts the grand canonical situation, where the number of bound particles and the state are two correlated random variables.
| 4 | 3 | 2 | 1 | |
| 3.985 | 2.991 | 1.993 | 0.997 | |
| 0.123 | 0.095 | 0.082 | 0.057 | |
| 0.922 | 0.722 | 0.358 | 0.107 |
Table V. For MWC model under (43), this table shows , , and versus . We see that , i.e., O molecules are strongly bound to the Hb macromolecule. For the grand canonical ensemble of MWC model, we get indicating on the relatively weak binding; see Figure 4. The first two lines of this table are close to the first two lines of Table IV. It is seen that the R-state dominates (i.e., ) after binding the third oxygen molecule. Note that for the grand canonical implementation of the MWC model, and are very close to each other; cf. Figure 4. In contrast, for the canonical situation presented in this table, is visibly larger than .
The behavior of and for the MWC model under the best fit (43) is again similar to Pauling’s model under parameters (35) (which is the best fit for Pauling’s model); see Figure 2 and Figure 3. Hence, we do not present them in a separate figure.
Instability for large cooperativity
The instability effects noted for Pauling’s model in the stronger cooperativity regime are also present here; cf. (37). We illustrate those instabilities as follows. We apply the same best fit that led to (43), but now we fix to a value , which is found from the grand-canonical ensemble; see (A10) in Appendix A. Now we find
which satisfies (44). Note that the ratio deduced from (45) (recall that this ratio characterizes the deviation from the non-cooperative situation) is still lower than the ratio obtained as the best to the grand canonical ensemble; see (A10) in Appendix A.
Parameters (45) lead to stability conditions for ; see (31). However, (45) does not satisfy (26) due to , but still . Hence, under (45), the third bound O molecule is released almost simultaneously with the fourth one. Conditions (25, 27, 28) are satisfied with (45).
For parameters , , and , we get , which is similar to the instability in (37). Also, for this situation, there is another sign of instability: , while , and are negative.
Thus, also for MWC model, parameters inferred from the grand canonical situation lead to an instability in the canonic setting.
The influence of ligands (within MWC model)
Hb ligands are of two types: heme ligands (e.g., O, CO, or NO) bind to the heme pocket. Non-heme ligands (or heterotropic ligands; e.g. protons H, 2,3-BPG, and CO) bind outside of the heme pocket. Normally, binding of heme ligands decreases the affinity for non-heme ligands and vice versa [10]. A related empiric rule is that heme ligands bind with higher affinity to the R state, while non-heme ligands usually bind with higher affinity to T [10].
Heterotropic ligands 2,3-BPG and proton () play important roles in maintaining and regulating transport functions of Hb [10, 24] [9]. In particular, they facilitate release in capillaries [4]. The concentration of 2,3-BPG in RBC changes under various physiological and clinical conditions, since this concentration regulates binding [24, 29] [9]. Increasing , which is frequent under anemia and hypoxia, facilitates the unbinding from Hb. Decreasing , which happens after blood conservation, strengthens the binding to Hb [29]. The proton binding to Hb is central for Bohr’s effect [10].
The influence of 2,3-BPG and on Hb- equilibrium was already studied within the grand-canonical MWC model [4]. They bind to the T-state and thereby influence mostly parameters and [24]. Ref. [5] initially suggested that only is affected, but this assumption turned out to be too restrictive [4]. Within MWC model, these ligands bias the allosteric equilibrium of the model towards the T state and hence facilitate release [4].
We now show that the canonical MWC model is consistent with the expected behavior of the Hb- equilibrium under the influence of heterotropic ligands 2,3-BPG and . Note that ligands were already taken into account in the data of Table III by stating standard physiological conditions. Therefore, we mean conditions that go beyond physiological ones.
We want to understand how [see (39, 20)] behaves under changing and (while is fixed). Table VI shows that upon increasing , monotonically increases, i.e. the release is facilitated. This is the same conclusion, as reached under increase of in the grand canonical ensemble; see Appendix A 2. The behavior under increasing is more convoluted, as Table VII shows: increases ( release facilitation), changes non-monotonically, whereas and decrease thereby stabilizing the binding more. This behavior does not contradict results obtained from the grand canonical ensemble; see Appendix A 2.
| 4 | 3 | 2 | 1 | |
| -5.67 | -6.18 | -6.48 | -7.22 | |
| -5.62 | -6.10 | -6.43 | -7.21 | |
| -5.54 | -5.98 | -6.37 | -7.19 | |
| -5.43 | -5.86 | -6.31 | -7.18 |
Table VI. For MWC model under and [cf. (43)], this table shows under various values of .
| 4 | 3 | 2 | 1 | |
| -5.67 | -6.18 | -6.48 | -7.22 | |
| -5.40 | -6.11 | -6.74 | -7.65 | |
| -5.38 | -6.30 | -7.07 | -8.03 |
Table VII. For MWC model under and [cf. (43)], this table shows under various values of .
Summary
Our main message is that the distribution of oxygen () and hemoglobin (Hb) in RBC (red blood cells) require changing the approach of statistical physics to the Hb- equilibrium. The usual approach is based on the grand canonical ensemble, which models a single Hb immersed into a reservoir of many molecules, where the reservoir is described by the concentration (or partial pressure) of [1]. As we argued in section I, this is not the situation realized in RBC, where each Hb most probably sees molecules. A better model is to assume that a single Hb with a few molecules surrounding it form a canonical ensemble; see Table II.
In this model, thermodynamic features of oxygen transport are determined via free energy differences discussed in section III. The free energy difference is conceptually distinct from the mean number of bound molecules, which (as a function of the partial pressure) determines the Hb- equilibrium within the grand canonical ensemble; see Table I for a detailed comparison between the two ensembles. The above distinction between the ensembles leads to contrasting predictions. We examined them in two basic models that describe the Hb- equilibrium and provide different explanations for cooperativity: Pauling’s model and MWC model. It turns out that the cooperativity values predicted by the grand canonical ensemble are not suitable for the canonical situation, because they will lead to various forms of instability discussed in Figure 2 and Figure 3, as well as around (45). Also, Hb behaves differently in each ensemble. Within the grand canonical ensemble, the number of bound molecules is a fluctuating quantity, whose mean follows the saturation data of Fig. III. Within the canonical approach, every available molecule is bound, and once Hb cannot bind anymore, it also does not attract them; see the discussion after (29).
Anticipating future developments, we note that our results on the thermodynamics of diffusion indicate that can be transported within RBC via hopping diffusion along with the usual diffusion; for recent results in oxygen diffusion inside RBC see [26]. To design nonequilibrium models of this type, one should fill the RBC volume with cubes of volume , each one containing a single hemoglobin; cf. Figure 1. Hopping diffusion means that molecules move from one cube to another also taking into account effective thermodynamic forces that arise due to free energy differences; see section IIIB. The hypothesis should be corroborated in future dynamical studies, also taking into account both rotational and translational diffusion of Hb macromolecules; for reviews on hopping diffusion see [30, 31, 32].
ACKNOWLEDGMENTS
We thank Arik Avagyan for discussions. The work was supported by the HESC of Armenia, grant No. 21AG-1C038.
CONFLICT OF INTEREST STATEMENT
The authors have no conflicts to disclose.
AUTHOR CONTRIBUTIONS (CREDIT TAXONOMY)
AEA: Conceptualization, Formal analysis, Writing – Original Draft; SGS: Conceptualization, Writing – Review & Editing; AH: Data Curation, Methodology, Software, Writing – Review & Editing; YShM: Validation, Investigation, Writing – Review & Editing; VS: Conceptualization, Formal analysis, Writing – Review & Editing;
DATA AVAILABILITY STATEMENT
The data are within the article.
Grand canonical ensemble
Here we review the grand canonical formulation of Pauling’s and MWC model, and determine their parameters by fitting to data of Table III.
Pauling’s model
We apply the procedure described in the beginning of section II D to (33), and introduce a new parameter ( is defined before (32)):
Eq. (A1) illustrates the essence of the grand canonical ensemble: for , only enters observable quantities. We get from (33) for the mean number of bound O molecules, and the variance [3]:
The mean number in (A3) as a function of (with parameters and ) is the main outcome of the grand canonical approach, because versus can be compared with saturation curves [3]. This determines and from curve-fitting [3]. Pauling employed the data by Ferry and Green [19] and found from the best fit [3]:
We used the coarser data from Table III, which represents as a function of , and fitted it to in (A3) to find parameters and . The best fit values are:
The values of in (A4) and (A5) are numerically comparable to each other. Though the values of in (A4) and (A5) are numerically different, they produce qualitatively the same results if employed for the canonical model; see Figure 2. Recall in this context that the parameter has the same form and the same meaning for both canonical and grand canonical ensembles.
MWC model
For we get from (39)
which is the standard partition function of the MWC model [4]. We carry out the same procedure as before (A1):
and employ (A3) for MWC model. We get for the probabilities of T and R states:
Eqs. (A6–A8) are studied within conditions (44), where , as seen from (A7). Conditions (44) ensure the following scenario. For low O concentration we get from (A8, 44): and . This means that for low O concentration, no O is bound and the T-state is dominant. This is consistent with the fact is negative, while is zero; see (39). For higher , we get from (A8, 44): and . Now the R-state is dominant and the Hb is fully loaded. Between high and low regimes, the mean number of bound O molecules changes via a S-shaped curve.
Note that the values of and for the grand canonical situation can be compared to (resp.) and for the canonical ensemble. Eaton reports the following values found from the best-fit to several datasets [4]:
We used the data from Table III and found from the best fit a larger value for , but a smaller value for the ratio :
Both (A9) and (A10) satisfy (44).
Let us now see how the basic diagnostic quantity of the grand canonical ensemble MWC model, i.e. reacts on changes of and . We restrict ourselves by these two parameters, because heterotropic ligands (2,3-BPG and protons) specifically bound in the T-state, and are standardly assumed to influence only and [24]. Note that from (A3, A6)
which means that in the considered regime , increasing will decrease .
The response of to changes of is more convoluted: increasing increases for smaller values of , but decreases it for larger values of ; see Figure 5. Put differently, increasing facilitates O release, but only for larger values of . In this context, one frequently involves , which the partial O pressure, where . Figure 5 shows that changes non-monotonically upon increasing .
Figure 5. For MWC model we plot versus the partial pressure (measured in torr) for and ; see (A9). Black curve , red , green , blue . It is seen that (the partial pressure, where ) changes non-monotonically upon increasing .
Free energy differences when transferring more than 2 particles
Transfer of molecules
We can generalize (19, 20) for the case when particles are transferred:
where in (B1) we assumed that . Now in (19) means that O molecules are absorbed by the Hb, while means that O molecules are released by the Hb.
We saw in (29) that one O transfer from the blood to the fully loaded Hb molecule is not possible; i.e. in notations of (B2). It turns out that 2-oxygen transfer is also impossible. For Pauling’s model, we get employing (35):
Even though now , it is not sufficiently negative to satisfy (24). Changing parameters in (35) does not affect (B3) at least within the presented order of magnitude. Eq. (B3) holds also for MWC model, where it is also nearly parameter independent.
Transfer of one molecule in a larger gradient
(30) can be generalized to one molecule transfer, when the gradient between two Hb molecules amounts to 3 molecules:
where means tendency towards gradient decreasing. For a strong cooperativity, shows the same type of pathologies as . For example, in Pauling’s model with and , we get and , while . For and , the same picture holds but with smaller absolute values of positive terms: and , and . For (35) we get indicating on thermodynamic stability.
Full (coordinate plus momentum) partition function and free energy
In this Appendix, we determine the momentum part of the partition function to be added to the coordinate part (8). For particles the momentum (Maxwell’s) distribution leads to the following total partition function and the corresponding free energy [21]:
For pressure and the mean energy we have from (C1, 8):
Both these formulas have transparent meaning: (C2) means that the pressure is created by unbound particles (as it should), while (C3) means that the mean energy is determined from plus the mean kinetic energy.
Possible generalizations of the grand canonical ensemble
The above derivation of the grand canonical ensemble (presented in § 2 D) assumes one Hb macromolecule and many O molecules. Given that the ensemble does not apply to RBC, one wonders whether the grand canonical ensemble can still apply to cases in which many Hb molecules interact with many O molecules. To answer this question, let us now assume that we have for molecules of Hb and molecules of O. This situation is realized in RBC; see Table II. We now demonstrate that the grand canonical ensemble (discussed in § 2 D and in Appendix A) does not emerge even in the limit .
To keep things simple in this demonstration, we also neglect cooperativity. Now instead of (8) we find
Assuming (but still ) does not allow us to approximate uniformly in (D1) , and hide and in the concentration parameter , which is the essence of the grand canonical ensemble. Using the confluent hypergeometric function we get
Employing a known formula
we find
Once is the main quantity of the grand canonical ensemble, (D4) can be used to compare the present calculation with this ensemble. In the limit , we now show that neither nor depends on and only via (this feature is the hallmark of the grand canonical ensemble). We have from (D4):
where in (D6) we increased and two times. So if were a function of only, it had to stay constant. Eqs. (D5), (D6) are also inconsistent with being a function of only.
References
- [10]A. Bellelli, Non-Allosteric Cooperativity in Hemoglobin, Current Protein and Peptide Science, 19, 1-16 (2018).DOI
- [11]L. A. Blumenfeld, A. Yu. Grosberg, and A. N. Tikhonov, Fluctuations and mass action law breakdown in statistical thermodynamics of small systems, J. Chem. Phys. 95, 7541-7547 (1991).DOI
- [12]A. V. Sokirko, Does Mass Action Law Breakdown Occur in Small Thermodynamic Systems?, J. Chem. Soc. Faraday Trans. 90, 2353-2358 (1994).DOI
- [13]D. Brogioli, Violation of the mass-action law in dilute chemical systems, J. Chem. Phys. 139, 184102 (2013).DOI
- [14]H. Reiss, Chemical effects due to the ionization of impurities in semiconductors, J. Chem. Phys. 21, 1209-1217 (1953).DOI
- [15]M.R.A. Shegelski, The chemical potential of an ideal intrinsic semiconductor, Am. J. Phys. 72, 676-678 (2004).DOI
- [16]T.A. Kaplan, The chemical potential, J. Stat. Phys. 122, 1237-1260 (2006).
- [17]J. J. Hopfield, Relation between Structure, Co-operativity and Spectra in a Model of Hemoglobin Action, J. Mol. Biol. 77, 207-222 (1973).pubmed.ncbi.nlm.nih.gov/4765360
- [18]S. G. Gevorkian, A. E. Allahverdyan, D. S. Gevorgyan and Chin-Kun Hu, Thermal-induced force release in oxyhemoglobin, Sci. Rep. 5, 13064 (2015).DOI
- [19]R. M. Ferry and A. A. Green, Studies in the chemistry of hemoglobin. III. The equilibrium between oxygen and hemoglobin and its relation to changing hydrogen ion activity, J. Biol. Chem. 81, 175-203 (1929).
- [20]G.M. Adair, The hemoglobin system VI: The oxygen dissociation curve of hemoglobin, J. Biol. Chem. 63, 529-545 (1925).
- [21]L.D. Landau and E.M. Lifshitz, Statistical Physics, Part 1, Second Edition: Volume 5 (Course of Theoretical Physics), (Reed, Oxford, 1989).
- [23]A. Matevosyan and A. E. Allahverdyan, Nonequilibrium, weak-field-induced cyclotron motion: A mechanism for magnetobiology, Phys. Rev. E, 104, 064407 (2021).DOI
- [24]M.F. Perutz, Species adaptation in a protein molecule, Mol. Biol. Evol. 1, 1-28 (1983).
- [25]R.K. Dash and J.B. Bassingthwaighte, Erratum to: Blood HbO₂ and HbCO₂ dissociation curves at varied O₂, CO₂, pH, 2,3-DPG and temperature levels, Ann. Biomed. Eng. 38, 1683–1701 (2010).DOI
- [26]S. L. Richardson, A. Hulikova, M. Proven, R. Hipkiss, Magbor Akanni, N. B. A. Roy, and P. Swietach, Single-cell O₂ exchange imaging shows that cytoplasmic diffusion is a dominant barrier to efficient gas transport in red blood cells, PNAS, 117, 10067–10078 (2020).DOI
- [27]B. Alpert, R. Banerjee, and L. Lindqvist, The kinetics of conformational changes in hemoglobin, studied by laser photolysis, Proceedings of the National Academy of Sciences, 71, 558-562 (1974).pubmed.ncbi.nlm.nih.gov/4521822
- [28]J.W. Severinghaus, Simple, accurate equations for human blood O2 dissociation computations, Journal of Applied Physiology, 46, 599-602 (1979).pubmed.ncbi.nlm.nih.gov/35496
- [29]J. Duhm and E. Gerlach, On the mechanisms of the hypoxia-induced increase of 2,3-diphosphoglycerate in erythrocytes, Pflügers Arch. 326, 254–269 (1971).DOI
- [30]P.M. Richards, Theory of one-dimensional hopping conductivity and diffusion, Phys. Rev. B 16, 1393 (1977).DOI
- [31]B. Derrida, An exactly soluble non-equilibrium system: the asymmetric simple exclusion process, Phys. Rep. 301, 65-83 (1998).DOI
- [32]R. Collins, S.R. Carson, and J.A. Matthew, Diffusion equation for one-dimensional unbiased hopping, American Journal of Physics, 65, 230-237 (1997).DOI

