The activity coefficient is largely considered an empirical parameter that was traditionally introduced to correct the non-ideality observed in thermodynamic systems such as osmotic pressure. Here, the activity coefficient of free-solvent is related to physically realistic parameters and a mathematical expression is developed to directly predict the activity coefficients of free-solvent, for aqueous protein solutions up to near-saturation concentrations. The model is based on the free-solvent model, which has previously been shown to provide excellent prediction of the osmotic pressure of concentrated and crowded globular proteins in aqueous solutions up to near-saturation concentrations. Thus, this model uses only the independently determined, physically realizable quantities: mole fraction, solvent accessible surface area, and ion binding, in its prediction. Predictions are presented for the activity coefficients of free-solvent for near-saturated protein solutions containing either bovine serum albumin or hemoglobin. As a verification step, the predictability of the model for the activity coefficient of sucrose solutions was evaluated. The predicted activity coefficients of free-solvent are compared to the calculated activity coefficients of free-solvent based on osmotic pressure data. It is observed that the predicted activity coefficients are increasingly dependent on the solute-solvent parameters as the protein concentration increases to near-saturation concentrations.
Citation: McBride DW, Rodgers VGJ (2013) Predicting the Activity Coefficients of Free-Solvent for Concentrated Globular Protein Solutions Using Independently Determined Physical Parameters. PLoS ONE 8(12): e81933. https://doi.org/10.1371/journal.pone.0081933
Editor: Danilo Roccatano, Jacobs University Bremen, Germany
Received: August 29, 2013; Accepted: October 27, 2013; Published: December 4, 2013
Copyright: © 2013 McBride, Rodgers. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: Devin W. McBride was supported by the National Science Foundation (NSF) IGERT Video Bioinformatics Training Grant and was supported by an NSF IGERT Video Bioinformatics Fellowship (grant number DGE 0903667). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Many cells contain macromolecular crowded protein environments (mixed proteins with total concentrations between 50 – 400 g/L), and therefore, the crowded environment is an essential component of cells , . One feature of macromolecular crowding is the deviation of the osmotic pressure from ideality called crowded protein osmotic pressure. The significance of the osmotic pressure due to these crowded proteins is that it may play a critical factor in intracellular flux as well as impact the reactive environment.
Although crowded protein environments are abundant and naturally occurring, many studies focus on single protein solutions for studying and understanding the effect(s) of crowded environments. These concentrated solutions, in which a single macromolecule is examined at high concentrations, are more convenient than crowded solutions; they can yield information about the effects of excluded volume (volume which is occupied by the macromolecule) on various phenomenon, such as reaction kinetics and thermodynamics .
Generally, to correct for the deviations from ideal models in crowded environments, an activity coefficient is introduced which accounts for the various interactions responsible for observations. Until now, there has been no rigorous assessment of how the activity coefficient of free-solvent is related to the solute and solution properties. Recently, the free-solvent model, introduced by van Laar  and developed by Yousef et al. – as be shown to give excellent predictability of the osmotic pressure of single and binary protein solutions up to near saturation. Once more, the model developed by Yousef et al. ,  used only physically realistic and independently determinable parameters in making these excellent predictions. In this work, the free-solvent model is used to directly couple the activity coefficient of free-solvent to these parameters, thus providing, for the first time, a fundamental basis for the concentration dependency of the solution activity coefficients.
Definition of the Activity Coefficient
Historically, the activity coefficient model for relating concentrations to chemical potential was developed to correct for non-idealities observed in many equilibrium systems. Recall, that the chemical potential for species can be related to the species relative activity, , as,
For an ideal system (with no attractive interactions), the relative activity is proportional to a composition variable, (such as , , , , etc.). For observed non-ideal behavior, an activity coefficient, , is introduced to ‘correct’ for the deviation
Relationship of the Activity Coefficient to Osmotic Pressure
When a two-chamber osmometer, containing diffusible species on one side and diffusible and non-diffusible species on the other side, is separated by a semi-permeable membrane, an osmotic pressure develops which directly corresponds to the chemical potential of the diffusible species across the membrane.
Denoting the chamber containing proteins denoted as compartment II, and the chamber containing only solvent and diffusible ions denoted as compartment I, the chemical potential of species in Chamber II at pressure , , and the chemical potential of species in Chamber I at pressure , , are related such that
At equilibrium, assuming that the chemical potential of species in Chamber I remains unchanged, that the temperature and pressure are constant, and that the number of diffusible species crossing the membrane is constant, then
where , the osmotic pressure, is the increase in the pressure required to satisfy chemical potential equivalence of species in the two chambers.
and the specific volume, , be defined as
Therefore it follows that
Finally, the osmotic pressure, in terms of the solvent activity, is
In an ideal system, the activity coefficient is unity, thus the activity is linearly related to the composition variable, and therefore the osmotic pressure is expressed as
However, for a non-ideal system, such as observed in a concentrated or crowded protein environment, using the mole fraction of the solvent, , as the composition variable, the osmotic pressure is related to mole fraction as
Assuming that the activity coefficient of free-solvent in compartment I (non-protein solution) is unity, , Eqn. 5 can be used to fit the experimental osmotic pressure data to determine the values of the activity coefficient of free-solvent at each protein concentration. In this work, we will reexamine the activity coefficient of free-solvent in terms of the free-solvent parameters.
As early as 1916, Frazer and Myrick  analyzed the non-idealities in concentrated, aqueous solutions of sucrose using a free-solvent model understanding that the mole fraction of water is affected by the hydration of sucrose. When the water that interacts with sucrose was removed from the total water available in the system, the free-solvent model provided an excellent prediction of the osmotic pressure data.
More recently, the free-solvent model was revised for aqueous protein solutions in which ion binding occurs, in addition to hydration . Essentially, the free-solvent model treats the protein with all associated water and salt ions as a unique species, the hydrated macromolecule. In effect, this approach renders the solution ideal with respect to the remaining, diffusible solvent species that have no attractive interactions. The modified mole fraction of the free water, , considers the hydrated macromolecule as the impermeable solute. The free-solvent model with the mole fraction of the free water, , as the composition variable is
where the mole fraction of free water is the remaining moles of solvent that are not bound to the protein. Assuming the solutions is made up of n distinct species and p proteins, and letting species 1 be the solvent, species 2 through be the proteins, and species through n be the remaining diffusible species, the initial total moles of the solution in compartment II is , where denotes each species. The final total moles of free-solvent in chamber II, after protein-solvent interactions, is , where denotes the moles of protein in solution and is the number of moles of species interacting with protein to make the hydrated protein. Then, the mole fraction of free-solvent in chamber II is(7)
while in chamber I, the mole fraction of free-solvent is
For a single protein species in a monovalent salt aqueous solution, the free-solvent model reduces to
Robustness of the Physical Parameters in the Free-Solvent Osmotic Pressure Model
The parameters of the free-solvent model have been shown to be remarkably robust and well-within independently determined values when regressed relative to measured osmotic pressure for highly concentrated protein solutions –, . As an example, the regressed hydration number, , for all globular proteins measured was found to be well within the 17O NMR approximation of 1 g H2O/g globular protein  but more precisely determines the value to be a monolayer of water with 0.6% when compared to the solvent accessible surface area (SASA) of each protein. Thus, the free-solvent model is likely to provide an excellent prediction of the activity coefficient of free-solvent that is developed from only independently determined physical parameters.
Coupling the Activity Coefficient of Free-Solvent to the Free-Solvent Model
Using the free-solvent model (Eqn. 6), the activity coefficient of free-solvent can be determined based on the ratio of the mole fractions of total water, , and the ratio of the mole fractions of free water, . Using and as the mole fractions of total water, and and as the mole fractions of free water, setting Eqns. 5 (with ) and 6 equal yields(10)
For a single protein in an aqueous solution with a single monovalent salt, the activity coefficient of free-solvent becomes
Eqn. 12 gives the relationship of the activity coefficient of free-solvent to the protein-solvent interactions and moles of species in solution. Note that, again, only measurable physical properties are necessary to determine the activity coefficient and there are no arbitrary parameters.
Materials and Methods
The activity coefficients of free-solvent were predicted based on protein-solvent interactions (Eqn. 12) and compared to the activity coefficients of free-solvent calculated using osmotic pressure data (Eqn. 5) for two proteins: bovine serum albumin (BSA) in 0.15 M NaCl, 25°C at pH 4.5, 5.4, and 7.4 and sheep hemoglobin (Hb) in 0.1 M KCl, 0°C, pH 7.43. The model is also used to predict the activity coefficients of free-solvent for sucrose in water at 30°C.
The calculated (osmotic pressure-based) activity coefficients of free-solvent were computed at each protein concentration by solving Eqn. 5, with , using the osmotic pressure data by Vilker et al.  for the concentrated BSA solutions (0.15 M NaCl at 25°C, pH 4.5, 5.4, and 7.4), to the osmotic pressure data by Adair, published by Dick , for concentrated Hb in 0.1 M KCl, 0°C, pH 7.43, and to the osmotic pressure by Frazer and Myrick  for concentrated sucrose solutions.
The Activity Coefficients of Free-Solvent Based on Independently Measurable Parameters
Using the model developed for the activity coefficient of free-solvent based on protein-solvent interactions (Eqn. 12), the activity coefficients of free-solvent were predicted, for each macromolecule, using available literature values for the hydrations and ion bindings.
Since the value of hydration can vary depending on the experimental method used, here the solvent accessible surface area (SASA) was used to determine the value of hydration . The SASA, computed using five molecular modeling software as previously described to compute hydration , was used to determine the value of hydration assuming 15.2 molecules per nm2 of surface area . The five molecular modeling software used are Swiss-Pdb Viewer , MOLMOL , UCSF Chimera , VegaZZ , and GETAREA . For Swiss-Pdb Viewer and MOLMOL a quality and precision of 6 were used, respectively, for calculating the SASA.
For BSA, three molecular structures are available (two in the Protein Data Bank (PDB: 3V03  and 4F5S ) and a homology model (based on human serum albumin (PDB: 1BM0 ) ). Here, the hydration values used are determined from the SASA using the molecular structure obtained from homology modeling. The ion binding values of BSA were those based on the two-site model by Scatchard et al. , .
Similarly, for Hb the hydration value used was that of the hydration computed from the SASA of the molecular structure (PDB: 2QU0 ) and the ion binding value was determined by De Rosa et al. .
Eqn. 12 was used to estimate the activity coefficients of free-solvent for three separate macromolecules in aqueous solutions up to near-saturation concentrations. Figures 1, 2, 3, 4, 5 show the calculated activity coefficients of free-solvent (Eqn. 5), based on the osmotic pressure data, and the activity coefficients of free-solvent based on protein-solvent interactions (Eqn. 12) applied to three BSA solutions, one Hb solution, and one sucrose solution using only the physical parameters available in literature (Table 1).
The calculated activity coefficients of BSA in 0.15(closed circles) are shown. The predicted activity coefficients (Eqn. 12) are plotted using the physical parameters available in literature for BSA in 0.15 M NaCl, pH 4.5 ( mol NaCl/mol BSA ): g H2O/g BSA (solid curve) and g H2O/g BSA (dotted curve).
The calculated activity coefficients of BSA in 0.15(closed circles) are shown. The predicted activity coefficients (Eqn. 12) are plotted using the physical parameters available in literature for BSA in 0.15 M NaCl, pH 5.4 ( mol NaCl/mol BSA ): g H2O/g BSA (solid curve) and g H2O/g BSA (dotted curve).
The calculated activity coefficients of BSA in 0.15(closed circles) are shown. The predicted activity coefficients (Eqn. 12) are plotted using the physical parameters available in literature for BSA in 0.15 M NaCl, pH 4.5 ( mol NaCl/mol BSA ): g H2O/g BSA (solid curve) and g H2O/g BSA (dotted curve).
The calculated activity coefficients of Hb in 0.1(closed circles) are shown. The predicted activity coefficients (Eqn. 12) are plotted using the physical parameters available in literature for Hb ( mol KCl/mol Hb ). The hydration values, determined from the SASA, are: g H2O/g Hb (solid curve), g H2O/g Hb (dotted curve), g H2O/g Hb (dash-dot curve), and g H2O/g Hb (dash-dot-dot curve).
The calculated activity coefficients of sucrose in water (closed circles) are shown. The predicted activity coefficient (Eqn. 12) is plotted using the literature values of hydration ( g H2O/g sucrose (solid curve) and g H2O/g sucrose, (dotted curve)).
For all solutions, as the solute concentration increases, the activity coefficient of free-solvent decreases from unity as expected; the activity coefficient of free-solvent for a pure water solution should be unity. The calculated activity coefficients of free-solvent follow this trend for most of the solutions studied; however there is some deviation, which is most likely due to experimental error. The predicted activity coefficients of free-solvent based on protein-solvent interactions decrease from unity as the protein concentration increases for all five solutions studied.
The activity coefficients of free-solvent predicted based on protein-solvent interactions are compared to the calculated activity coefficients of free-solvent for BSA in 0.15 M NaCl, pH 4.5, 5.4, and 7.4 (Figures 1, 2, 3). The ion binding values are 11.59 mol NaCl/mol BSA, 10.62 mol NaCl/mol BSA, and 8.81 mol NaCl/mol BSA for BSA in 0.15 M NaCl, pH 4.5, 5.4, and 7.4, respectively . The homology model SASA was computed using four molecular modeling software. The SASA are 28,065 Å2 (Swiss-Pdb Viewer ), 28,188 Å2 (MOLMOL ), 27,985 Å2 (VegaZZ ), and 27,746 Å2 (GETAREA ) and the corresponding hydration values are 1.157 g H2O/g BSA (Swiss-Pdb Viewer ), 1.162 g H2O/g BSA (MOLMOL ), 1.154 g H2O/g BSA (VegaZZ ), and 1.144 g H2O/g BSA (GETAREA ). The activity coefficient of free-solvent is predicted using the minimum and maximum hydration values (1.144 g H2O/g BSA and 1.162 g H2O/g BSA) and the corresponding ion binding values for each BSA solution. The predicted activity coefficients of free-solvent for all three BSA solutions follows the same trend as the calculated activity coefficients of free-solvent; the predicted activity coefficients are in excellent agreement with the calculated activity coefficients for BSA in 0.15 M NaCl, pH 5.4 and 7.4, and in good agreement for BSA in 0.15 M NaCl, pH 4.5.
The SASA from the molecular structure of Hb (PDB: 2QU0) for four of the molecular modeling software are 24,304 Å2 (Swiss-Pdb Viewer ), 24,981 Å2 (MOLMOL ), 26,100 Å2 (UCSF Chimera ), and 24,759 Å2 (GETAREA ) and the corresponding hydration values are 0.955 g H2O/g BSA (Swiss-Pdb Viewer ), 0.981 g H2O/g BSA (MOLMOL ), 1.025 g H2O/g BSA (UCSF Chimera ), and 0.973 g H2O/g BSA (GETAREA ). The activity coefficients of free-solvent were predicted for all four values of hydration using the literature value for ion binding, 6 mol KCl/mol Hb . The predicted activity coefficients of free-solvent using the SASA from three of the molecular modeling software are in excellent agreement with the calculated activity coefficients, and the activity coefficients of free-solvent predicted using the SASA from UCSF Chimera  is in good agreement with the calculated values.
The predicted activity coefficients of free-solvent for sucrose are compared to the calculated activity coefficients of free-solvent from experimental osmotic pressure data (Figure 5). Many studies have determined the hydration of sucrose, with the most notable being those by Frazer and Myrick , Scatchard , and Einstein . The range of sucrose hydration values is 5 – 6 mol H2O/mol sucrose , , . The activity coefficients of free-solvent was predicted using the minimum and maximum values (within the range) of sucrose hydration: 0.184 g H2O/g sucrose (3.5 mol H2O/mol sucrose) and 0.316 g H2O/g sucrose (6 mol H2O/mol sucrose) , , .
The activity coefficient of free-solvent has now been given a physiological basis. Here, the activity coefficients of free-solvent are predicted for two macromolecules based on hydration and ion binding. As expected, the activity coefficients of free-solvent for all solutions decrease from unity as the protein concentration increases. Using the Gibbs-Duhem relationships, the activity coefficients of free-solvent can be used to determine the activity coefficients of the protein or the salt based on physically realistic parameters.
Independently Determining the Physical Parameters of the Free-Solvent Model
The free-solvent model, which relies only on hydration and ion binding, has been shown to be remarkably robust due to the use of only physically realistic and independently measureable parameters.
The hydration of macromolecules has been extensively studied for several decades, including Einstein’s estimation of sucrose hydration from viscosity data . Hydration can be determined using various methods, including 17O NMR , x-ray solution scattering , and small angle neutron scattering . Furthermore, if structural information is known, the hydration value can be calculated assuming a monolayer of water surrounds the macromolecule .
The interaction between ions and proteins have also been examined using several techniques, electromotive force (EMF), distribution method, and isopiestic method, , – as well as mathematical models based on surface residues .
The developed model for the activity coefficient of free-solvent which relies on these two fundamental physical parameters is highly robust since, in addition to the independent methods for estimation of the values, protein hydration and protein-ion binding are also unique and do not rely on each other. Protein hydration is primarily dependent on the solvent accessible surface area, while protein-ion binding depends on surface residue charge, their location and neighboring residues, as well as the net charge of the protein.
Crystallization Solution Properties of Bovine Serum Albumin
Herein, the BSA molecular structure, based on the homology model, was used for calculating the SASA due to the experimental conditions used for the crystallization of the molecular structures available in the Protein Data Bank for BSA (PDB: 3V03 and 4F5S). Both of the structures were crystallized at pH 6.5 in polyethylene glycol (PEG) solutions: 20% (w/v) PEG 3350 (PDB: 3V03) and 20 – 24% (w/v) PEG monomethyl ether (MME) 5000 (PDB: 4F5S). In addition, 200 mM Ca acetate and 100 mM Tris-HCl were used in the crystallization of BSA by Majorek et al. ; 150 – 300 mM NH4Cl and 100 mM 2-(N-morpholino)ehanesulfonic acid (MES) were used in the crystallization of BSA by Bujacz . In the former case, the authors mention that monoclinic crystals were observed ; however, in the latter case, the authors state that the crystals were poor .
The effect of these solutions on the SASA compared to the SASA obtained from osmotic pressure for 0.15 M NaCl solutions is unknown. To investigate this, the osmotic pressure-based SASA can be determined for BSA in the crystallization solution properties.
Furthermore, the crystallization process, dehydrating the molecules, may have effects on the molecular structure due to charge repulsion. This is a very likely problem with the crystallization of BSA since it is a very negatively charged molecule in both of the crystallization solutions.
Limitations of the Activity Coefficients of Free-Solvent Based on Protein-Solvent Interactions
Herein, the activity coefficient of free-solvent was only developed for protein solutions in which only solute-solvent interactions occur. However, for solutions in which protein-protein interactions occur, while Eqn. 12 is correct, the mole fraction of free-solvent in each compartment can be revised to include protein-protein interactions in order to determine the closed-form solution of the activity coefficients of free-solvent. This modification of the free-solvent model which accounts for protein-protein interactions in addition to the protein-solvent interactions, has been recently developed .
A model for the activity coefficient of free-solvent was developed based on solute-solvent interactions. Unique about this approach is that this model uses no adjustable parameters and is based only on the independently determined physical parameters associated with protein hydration and ion binding. The closed-form solution for the single macromolecule, monovalent salt system activity coefficient of free-solvent is provided, and the predicted activity coefficient of free-solvent based on physical parameters from literature for three single macromolecule solutions, up to near-saturation concentrations, is shown.
Conceived and designed the experiments: DWM VGJR. Performed the experiments: DWM. Analyzed the data: DWM VGJR. Contributed reagents/materials/analysis tools: DWM VGJR. Wrote the paper: DWM VGJR.
- 1. Ellis RJ (2001) Macromolecular Crowding: Obvious but Underappreciated. Trends in Biochemical Sciences 26: 597–604.
- 2. Minton AP (2001) The Influence of Macromolecular Crowding and Macromolecular Confinement on Biochemical Reactions in Physiological Media. Journal of Biological Chemistry 276: 10577–10580.
- 3. Minton AP (1983) The Effect of Volume Occupancy Upon the Thermodynamic Activity of Proteins: Some Biochemical Consequences. Molecular and Cellular Biochemistry 55: 119–140.
- 4. van Laar JJ (1894) On the Exact Formulas for the Osmotic Pressure, for the Changes to the Solubility, for Freezing Point - and Boiling Point - Changes, and for the Solutions - and Dilutions - Warm with Bodies Dissociated in Solution. Z Phys Chem 15: 457–497.
- 5. Yousef MA, Datta R, Rodgers VGJ (1998) Free-Solvent Model of Osmotic Pressure Revisited: Application to Concentrated IgG Solution Under Physiological Conditions. J Colloid Interface Sci 197: 108–118.
- 6. Yousef MA, Datta R, Rodgers VGJ (1998) Understanding Nonidealities of the Osmotic Pressure of Concentrated Bovine Serum Albumin. J Colloid Interface Sci 207: 273–282.
- 7. Yousef MA, Datta R, Rodgers VGJ (2001) Confirmation of Free Solvent Model Assumptions in Predicting the Osmotic Pressure of Concentrated Globular Proteins. J Colloid Interface Sci 243: 321–325.
- 8. Yousef MA, Datta R, Rodgers VGJ (2002) Model of Osmotic Pressure for High Concentrated Binary Protein Solutions. AIChE Journal 48: 913–917.
- 9. Yousef MA, Datta R, Rodgers VGJ (2002) Monolayer Hydration Governs Nonideality in Osmotic Pressure of Protein Solutions. AIChE J 48: 1301–1308.
- 10. Frazer JCW, Myrick RT (1916) The Osmotic Pressure of Sucrose Solutions at 30 Degrees. J Am Chem Soc 38: 1907–1922.
- 11. McBride DW, Rodgers VGJ (2012) Obtaining Protein Solvent Accessible Surface Area (SASA) when Structural Data is Unavailable using Osmotic Pressure. AIChE J 58: 1012–1017.
- 12. Sedykh LG, Sedykh NV (1967) Mathematical Model of Phenomenon of Hydration of Biopolymers in Solution. Biofizika 12: 936–938.
- 13. Vilker VL, Colton CK, Smith KA (1981) The Osmotic Pressure of Concentrated Protein Solutions: Effect of Concentration and pH in Saline Solutions of Bovine Serum Albumin. J Colloid Interface Sci 79: 548–566.
- 14. Dick DAT (1967) An Approach to the Molecular Structure of the Living Cell by Water Flux Studies. In: Reeve EB, Guyton AG, editors. Physical Bases of Circulatory Transport: W. B. Saunders Co., Philadelphia. pp. 217–234.
- 15. Guex N, Peitsch MC (1997) SWISS-MODEL and the Swiss-Pdb Viewer: An Environment for Comparative Protein Modeling. Electrophoresis 18: 2714–2723.
- 16. Koradi R, Billeter M, Wüthrich K (1996) MOLMOL: A Program for Display and Analysis of Macromolecular Structures. J Mol Graphics 14: 51–55.
- 17. Pettersen EF, Goddard TD, Huang CC, Couch GS, Greenblatt DM, et al. (2004) UCSF Chimera-A Visualization System for Exploratory Research and Analysis. Journal of Computational Chemistry 25: 1605–1612.
- 18. Pedretti A, Villa L, Vistoli G (2002) VEGA: A Versatile Program to Convert, Handle and Visualize Molecular Structure on Windows-Based PCs. J Mol Graphics 21: 47–49.
- 19. Fraczkiewicz R, Braun W (1998) Exact and Efficient Analytical Calculation of the Accessible Surface Areas and Their Gradients for Macromolecules. J Comput Chem 19: 319–333.
- 20. Majorek KA, Porebski PJ, Dayal A, Zimmerman MD, Jablonska K, et al. (2012) Structural and Immunologic Characterization of Bovine, Horse, and Rabbit Serum Albumins. Mol Immunology 52: 174–182.
- 21. Bujacz A (2012) Structures of Bovine, Equine and Leporine Serum Albumin. Acta Crystallographica Section D-Biological Crystallography 68: 1278–1289.
- 22. Sugio S, Kashima A, Mochizuki S, Noda M, Kobayashi K (1999) Crystal Structure of Human Serum Albumin at 2.5 Angstrom Resolution. Protein Engineering 12: 439–446.
- 23. McBride DW (2013) Further Understanding the Physical Phenomena of Crowded Protein Osmotic Pressure and Its Application to Medical Devices: University of California, Riverside.
- 24. Scatchard G, Scheinberg IH, Armstrong SH (1950) Physical Chemical of Protein Solutions .IV. The Combination of Human Serum Albumin with Chloride Ion. J Am Chem Soc 72: 535–540.
- 25. Neelagandan K, Moorthy PS, Balasubramanian M, Ponnuswamy MN (2007) Crystallization of Sheep (Ovis aries) and Goat (Capra hircus) Haemoglobins Under Unbuffered Low-Salt Conditions. Acta Crystallographica Section F - Structural Biology and Crystallization Communications 63: 887–889.
- 26. De Rosa MC, Castagnola M, Bertonati C, Galtieri A, Giardina B (2004) From the Arctic to Fetal Life: Physiological Importance and Structural Basis of an "Additional" Chloride-Binding Site in Haemoglobin. Biochemical Journal 380: 889–896.
- 27. Scatchard G (1921) The Hydration of Sucrose in Water Solution as Calculated from Vapor-Pressure Measurements. Journal of the American Chemical Society 43: 2406–2418.
- 28. Einstein A (1956) Investigations on the Theory of Brownian Movement: Dover Publications.
- 29. Svergun DI, Richard S, Koch MHJ, Sayers Z, Huprin S, et al. (1998) Protein Hydration in Solutions: Experimental Observation by X-Ray and Neutron Scattering. PNAS 95: 2267–2272.
- 30. Carr CW (1952) Studies on the Binding of Small Ions in Protein Solutions with the use of Membrane Electrodes .I. The Binding of the Chloride Ion and Other Inorganic Anions in Solutions of Serum Albumin. Archives of Biochemistry and Biophysics 40: 286–294.
- 31. Carr CW (1953) Studies on the Binding of Small Ions in Protein Solutions with the use of Membrane Electrodes .II. The Binding of Calcium ions in Solutions of Bovine Serum Albumin. Archives of Biochemistry and Biophysics 43: 147–156.
- 32. Carr CW (1955) Competitive Binding of Calcium and Magnesium with Serum Albumin. Proceedings of the Society for Experimental Biology and Medicine 89: 546–549.
- 33. Scatchard G, Coleman JS, Shen AL (1957) Physical Chemistry of Protein Solutions .VII. The Binding of Small Anions to Serum Albumin. Journal of the American Chemical Society 79: 12–20.
- 34. Bull HB, Breese K (1968) Protein Hydration .I. Binding Sites. Archives of Biochemistry and Biophysics 128: 488–496.
- 35. Reboiras MD, Pfister H, Pauly H (1978) Activity-Coefficients of Salts in Highly Concentrated Protein Solutions .1. Alkali Chlorides in Isoionic Bovine Serum-Albumin Solutions. Biophysical Chemistry 9: 37–46.
- 36. Reboiras MD, Pfister H, Pauly H (1986) Activity-Coefficients of Salts in Highly Concentrated Protein Solutions .2. Potassium-Salts in Isoionic Bovine Serum-Albumin Solutions. Biophysical Chemistry 24: 249–257.
- 37. McBride DW, Rodgers VGJ (2013) Development of the Free-Solvent Model for the Osmotic Pressure of Multi-Component Solutions Considering Protein-Protein Interactions. Journal of Applied Mathematics Submitted.