Biotechnol. Prog. 1991, 7, 116-124

116

Solvent Selection Strategies for Extractive Biocatalysis Lynda J. Bruce and Andrew J. Daugulis* Department of Chemical Engineering, Queen’s University, Kingston, Ontario, Canada K7L 3N6

This report follows the development of systematic solvent screening strategies for the identification of superior pure solvents and introduces techniques for the identification of effective coextractants. Specifically, methods to predict the biocompatibility and extractant capability of solvents are discussed. Biocompatibility is predicted by using heuristic data or the correlations between bioactivity and the logarithm of the partition coefficient of the solvent or the concentration of solvent in the cell membrane. A computer program, known as the extractant screening program or ESP, has been developed to effectively predict the behavior of virtually any product in any solvent/ aqueous system. I t is demonstrated that a biocompatible yet poor solvent can be mixed with a toxic solvent that has better extractant properties to yield a mixture with improved solvent characteristics that is still biocompatible. The fact that solvents do not mix in an ideal manner is exploited by using ESP to identify solvent mixtures that are still biocompatible a t relatively high concentrations of toxic solvent.

Introduction The use of an organic solvent phase in biocatalytic applications can provide both kinetic and thermodynamic advantages. When a product is continuously removed by a solvent phase, a biocatalyst that is subject to decreased productivity due to end-product inhibition will show an increase in activity. With the correct choice of solvent, it is no longer necessary to work with dilute solutions, and waste-water treatment and product recovery costs can be reduced. The equilibrium position of biocatalyzed equilibrium reactions may also be affected by the introduction of a suitable solvent. Many equilibrium reactions are driven by providing large excesses of the reactants, but by continuously removing the products with a suitable solvent, the reaction may be “pulled”toward completion and large excesses of reactants are no longer required. Although this review will focus on the use of living cells in extractive biocatalysis, enzyme-catalyzedbioreactions have also been shown to benefit from the use of an extraction solvent as well as many of the solvent selection techniques that will be discussed (Eggers et al., 1989; Halling, 1990;Martinek et al., 1981; Reslow et al., 1987). Advances in fermentation process design (Daugulis et al., 1987; Finn, 1966; Kollerup and Daugulis, 1986, 1987; Minier and Goma, 1981, 1982; Roffler et al., 1988) have shown that reaction and product recovery can be made to occur simultaneously in a single processingunit. By cycling a water-immiscible solvent through the culture medium, the product is selectively extracted, and if the product normally acts to decrease the process reaction rate, improved productivity can result. A great deal of effort goes into solvent selection, as the solvent must perform optimally for both the reaction and product recoverysteps. Table I is a general list of the strict requirements that govern the choice of a suitable solvent. Specific biocatalytic systems may have additional criteria. The distribution coefficient is a measure of the solvent’s capacity for the product and is defined as the ratio of the

* To whom correspondence should be addressed. 8756-7938/9 ~3007-0116$02.50/0

Table I. Desirable Solvent Characteristics. favorable distribution coefficient for product high selectivity low emulsion-forming tendency low aqueous solubility chemical and thermal stability favorable properties for product recovery nonbiodegradability nonhazardous inexpensive available in bulk quantitites biocompatibility From Daugulis (1988).

product concentration in the solvent to the product concentration in the aqueous culture medium, at equilibrium. As the distribution coefficient increases, less solvent is required for effective extraction. Selectivity describes the ability of a solvent to preferentially remove the product over water. A solvent that does not tend to form emulsions increases the ease of separation of the two phases, and choosing a solvent with low aqueous solubility minimizes solvent loss. Chemical and thermal stability are necessary over the course of long or continuous fermentationssince the solvent may be recycled many times. Solvent properties that affect the ease of product recovery include density, viscosity, and the solvent boiling point. The solvent should be nonbiodegradable so that the organism does not use it as a substrate. Ideally it should also be nonhazardous, inexpensive, and available in large quantities. Finally, there is the very important criterion that the solvent must be completely biocompatible as it will be in direct contact with the biocatalyst. Whereas the other desirable solvent attributes are relative conditions, the requirement of biocompatibilityis a particularly restrictive criterion as it is an absolute condition. A method to predict a solvent’s ability to meet these strict requirements would be of great practical use, since experimental data are seldom available for a specificsystem and there are many thousands of solvents from which to choose. As the use of a systematic screening procedure

0 1991 American Chemical Society and American Institute of Chemical Engineers

117

Biotechnol. Rog., 1991, Vol. 7, No. 2

for the identification of potentially superior solvents for the extraction of a product would significantly decrease the amount of experimental work required, much effort has been directed at developing strategies and tools for the selection of organic solvents for use in extractive biocatalytic systems. In this paper we trace the evolution of these strategies and extend the work by providing new theoretical and experimental information for the selection of appropriate organic solvents for coextractant systems.

Liquid-Liquid Equilibrium Predictions The employment of computing power is logically one of the first steps in the establishment of a systematic screening procedure. Kollerup and Daugulis developed a computer program that enabled them to systematically screen the performance of over 1500 pure solvents for extracting ethanol from an aqueous phase with the aim of identifying suitable solvents for use in an ethanol extractive fermentation process (Kollerup and Daugulis, 1985,1986). The program, known as the extractant screening program or ESP, utilizes the UNIFAC group contribution method for predicting liquid-liquid equilibrium data. Each solvent’s UNIFAC parameters were entered by identifying each of its functional groups. Additional solvent information such as physicochemical properties, availability, price, and any existing toxicity data (as LD50) were added to the ESP database. Parameters, such as the distribution coefficient, selectivity, and aqueous solubility of the solvent, were calculated and ESP, possessing database features, ranked the solvents by these calculated values or other measures, such as toxicity and price. Experimental work aimed at confirming suitable extractants was then significantly reduced by selecting those solvents ranked highest on the basis of their predicted behavior as extractants. The distribution coefficients predicted by ESP were compared to the experimentally determined values in order to evaluate the quality of the predicted liquid-liquid equilibrium data. Although the exact numerical predictions were found to be lacking in accuracy, an appreciable agreement was found especially in the ranking within solvent classes (e.g., alcohols). Relationships between the molecular structure of solvents and their predicted performance as extractants can also be examined with ESP by “creating” molecules with functional group components. The first version of ESP has now been significantly improved so that the product can be virtually any compound, not just ethanol, and the behavior of quaternary systems in addition to ternary systems can be predicted. The use of ESP for predicting the behavior of quaternary systems will be described later in this paper.

Predicting Biocompatibility The ability to predict the biocompatibility of a pure solvent would be very useful, as it would reduce the amount of time-consuming, but necessary, biocompatibility studies. Kollerup and Daugulis established quantitative heuristic criteria based on literature data to predict solvent biocompatibility (Kollerup and Daugulis, 1985). They noted that solvents biocompatible to yeast cells tend to have low distribution coefficients, selectivities, and aqueous solubilities, and high LD50 values, molecular weights, densities, and normal boiling points. They further showed that there are correlations between these values and the level of biocompatibility associated with any solvent. Brink and Tramper attempted to relate the observed increase in toxicity of a pure solvent with increased solvent polarity,

Activity Retention (%) -

100 I2O

I

~

.

~

I

40

t

20

1 I

0 0

--e1

2

I -

1

3

4

5

I

I

1

6

7

8

-1

9

,10

11

log P

Figure 1. Resulting relationship between the activity retention and epoxidizing cells exposed to various organic solvents and log P. [Data from Brink and Tramper (1985) as presented by Laane et al. (1985).]

represented by the Hildebrand solubility parameter, and decreased molecular weight (Brink and Tramper, 1985). Shortly later, Laane and co-workers demonstrated that there is a stronger relationship between bioactivity and a different measure of polarity, the logarithm of the partition coefficient (logP ) of the solvent (Laane et al., 1985).Figure 1shows the relationship that Laane et al. found between the log P of the solvents and the biocompatibility data of Brink and Tramper. The partition coefficient is a dimensionless value that is arbitrarily measured in a standard octanol-water twophase system and is defined as being

where, in this case, the solvent under investigation is the solute. The relationship between log P and bioactivity is based on the assumption that the octanol-water system provides a sufficient description of hydrophobic and transport interactions of a structure when it is introduced into a biological system. Although partition coefficients can be determined experimentally, log P is even more useful as an indicator of bioactivity, as it can be predicted for any substance from the knowledge of its molecular structure. Rekker’s hydrophobic fragmental constant method, which is a group contribution method similar to the UNIFAC method used by ESP, is often used (Rekker, 1977;Rekker and de Kort, 1979). Log P has conventionally been used in the pharmaceutical and medicinal chemistry fields as a part of drug activity studies. As a further test of the applicability of the log Pstrategy for predicting biocompatibility of pure solvents, we have calculated the log P values for various solvents which have been tested in our laboratory with three different organisms: Saccharomyces cereuisiae (Kollerup, 1986), Clostridiumacetobutylicum (Barton, 1990),and Zymomonus mobilis (Bruce, 1990). The estimated log P values of the solvents were plotted against a measure of the resulting bioactivity of the cells after exposure to the solvent, as shown in Figures 2, 3, and 4. It is generally true that as the size of the solvent molecule increases, the accuracy of the estimated log P value decreases. The solvents used in each of Figures 2, 3, and 4 are listed in Table I1 with their corresponding log P values. The above figures show a sigmoidal shape similar to the results of Laane et al. (1985) depicted in Figure 1. The figures indicate that, excluding the transition region where the inflection point is located and a solvent can be inhibitory, a solvent that has a log P value lower than the

Biotechnoi. Prog., 1991, Vol. 7, No. 2

118 Percent Metabolic Activity

120-

----

_______

1

Understanding Solvent Toxicity

log P

Figure 2. Resulting relationship between the percent metabolic activity of S. cereuisiae cells after exposure to various organic solvents and log P. [Data from Kollerup (1986).] Percent Metabolic Activity

lZor----

I

m,

2'3

60

20 L-+*

0 -1

0

1

-++ ~+& I - A

1 + * -

3

2

5

4

7

6

8

9

10

log P

Figure 3. Resulting relationship between the percent metabolic activity of C. acetobutylicum cells after exposure to various organic solvents and log P. [Data from Barton (1990).] Percent Metabolic Activity 120

-

~~

--

40 I

0 0

x:

,-

---J

I

2

3

L -

4

Although there are a few exceptions to the relationship between log P and bioactivity, it appears to hold true for the majority of solvents tested to date. This provides convincing support for the use of log P values to predict the biocompatibility of pure solvents. It is now advantageous to combine the use of log P as an indicator of biocompatibility with ESP. Once the ESP database has been used to rank the solvents by the best physical characteristics for extraction, the solvents with a log P value higher than the inflection point for the biocatalyst of interest can be selected for further study with a high level of confidence that they will be biocompatible.

l d -

5

6

7

8

9

IO

log P

Figure 4, Resulting relationship between the percent metabolic activity of 2. mobilis cells after exposure to various organic solvents and log P. [Data from Bruce (1990).]

inflection point is usually toxic and a solvent that has a log P value higher than the inflection point is usually biocompatible. Bioactivity was measured by a variety of different methods and does not appear to affect the shape of the curves. The location of the inflection point is dependent on the type of organism, probably due to differences in types of cell wall (Laane et al., 1987) and the degree of agitation (Hocknull and Lilly, 1987). Increased agitation causes the log P curve to shift to the right, causing a decrease in the stability of the cells as a result of the improved transfer of the solvent through the aqueous phase. A compromise exists between good transfer of the product into the solvent and the loss of biological activity of the cells.

The mechanisms of solvent-caused toxicity are not well understood. It was noted that certain reagents were found not to be toxic unless present at levels in excess of that needed to saturate the aqueous phase (Playne and Smith, 1983). Bar (1987) divided the toxic action of solvents into two major classes, physical and dissolved toxicity, which is essentially the difference between toxicity at the phase level and the molecular level. Dissolved toxicity describes the effect of a solvent when it is present in the aqueous phase at levels below saturation. Lilly and co-workers hypothesized that the lipid membranes of the cell absorb some of the dissolved solvent, causing a modification of membrane permeability that may result in enzyme inhibition, protein deactivation, or a breakdown of transport mechanisms (Lilly et al., 1987). The additional effect of physical toxicity is felt when enough solvent is present to form a separate phase. Bar has documented that a solvent coat can form about a cell, which could block nutrient diffusion or disrupt the cell wall, causing extraction of inner cellular components (Bar, 1987). A recent publication by Osborne and co-workers has provided additional insight into the mechanismsof toxicity in cells as well as a possible solvent selection procedure (Osborne et al., 1990). They demonstrated that there is a correlation between the lla-hydroxylase activity of Rhizopus nigricans and the concentration of solvent in the cell membrane. This information agreed with what was already known about the action of anesthetics, as anesthetic potency is related to the concentration of compounds in the cell membrane but is not dependent on the type of compound (Seeman 1972). Osborne and co-workers separated the effects of dissolved toxicity from phase toxicity by using solvents at subsaturation levels. By performing a series of biocompatibility studies on different solvents and R. nigricans, they showed that the concentration of solvent in the cell membrane that caused complete loss of bioactivity was approximately 200 mM for most solvents tested. The concentration of solvent in the cell membrane that caused complete loss of bioactivity was called the critical membrane concentration. Solvents that remained biocompatible up to the saturation concentration in the aqueous phase were found to be incapable of attaining this critical membrane concentration. Any activity loss observedwhen there is enought of such a solvent present to form a separate phase can be attributed to phase toxicity. These findings relate well with the correlation between log P and activity retention, as derived from the experimental data of Seeman (1972). The following equation relates the partitioning behavior of a solute, in this case

Biotechnol. Prog., 1991, Vol. 7, No. 2

119

Table 11. Log P Values for the Solvents Used in Figures 2,3, and 4. for Figure 2 solvent acetic acid, ethyl ester nonanoic acid acetic acid, isopropyl ester aldehyde (2-18 acetic acid, butyl ester 2-heptanone heptanal acetic acid, pentyl ester aldehyde C-14 2-octanone pentanoic acid, pentyl ester 1- heptanol glycerol, tributanoate 1-octanol 2,6-dimethyl-4-heptanone dibutyl ether 1-nonanol geraniol heptane 1-decanol 10-undecen-1-01 hexanedioic acid, dibutyl ester phthalic acid, dibutyl ester 2-dodecanone 1-undecanol octane dodecanal dodecanoic acid, nitrile 1-dodecanol CO-1214 fatty alcohol farnesol Epal 1214 9,12-octadecadienoic acid acetic acid, dodecyl ester Epal 1218 decanedioic acid, dibutyl ester dodecanoic acid, butyl ester Eutanol G-16 oleyl alcohol Adol 66 phytol isophytol Eutanol G

for Figure 3

.

log P 0.7

solvent diisopropyl L-tartrate 1,7-heptanediol 3-methyl-2,4-heptanediol 2-ethyl-1,3-hexanediol ethyl DL-mandelate dibutyl L-tartrate aldehyde (2-14 linalool nonanoic acid 1-nonanol geraniol pentyl valerate dibutyl adipate dibutyl phthalate 1-dodecanol Neodol 23 lauryl alcohol CO-1214 fatty alcohol farnesol Epal 1214 Neodol25 dodecyl acetate Epal 1218 dibutyl sebacate butyl laurate Eutanol G-16 linoleic acid corn oil soy oil olive oil oleyl alcohol Adol 66 bis(2-ethylhexyl) sebacate bis(2-ethylhexyl) adipate isophytol phytol Eutanol G bis(3,5,5-trimethylhexyl)phthalate

1.2 1.2

1.3 1.7 1.8 2.1 2.2 2.3 2.4 2.4 2.4 2.8 2.9 2.9 2.9 3.4 3.5 4.0 4.0 4.1 4.1 4.3 4.4 4.5 4.5 4.7 4.9 5.0 5.2 5.4 5.5 5.7 5.9 6.0 6.2 6.9 7.0 7.5 8.0 9.1 9.1 9.1

for Figure 4 log P 0.4 0.7 1.2 1.2

1.3 1.5 2.3 3.2 3.4 3.4 3.5 3.8 4.1

4.3 5.0 5.0 5.0 5.2 5.4 5.5 5.7 5.9 6.0 6.2 6.9 7.0 7.3 7.4 7.4 7.5 7.5 8.0 8.2 8.2 9.1 9.1 9.1 9.4

solvent hexanol aldehyde C-14 heptanol octanol octanoic acid nonanol geraniol decanal decanol dibenzyl ether undecanol dodecanal dodecanol Neodol23 2-tridecanone CO-1214 fatty alcohol Epal 1214 Neodol25 Epal 1218 dibutyl sebacate methyl myristate bis(2-ethylhexyl) maleate butyl laurate ethyl myristate Eutanol G-16 isopropyl myristate oleic acid trihexyl amine bis(2-ethylhexyl) sebacate isopropyl palmitate hexadecane decyl decanoate isophytol

log P 1.8 2.3 2.4 2.9 2.9 3.4 3.5 3.7 4.0 4.0 4.5 4.7 5.0 5.0 5.2 5.2 5.5 5.7 6.0 6.2 6.4 6.8 6.9 6.9 7.0 7.4 7.7 7.8 8.2 8.5 8.7 9.0 9.1

For sources of industrial solvents, see Appendix.

the solvent, between two systems: Y

(2) Pmembrane = XPoctanol is the partition coefficient for the solute in a membrane/aqueous buffer system, and Poctanol refers to the aqueous/octanol system. X and Y refer to the product and power terms. Since

Pmembrane

(3) eq 2 can be rewritten with the critical solvent concentration in the aqueous phase required to cause total loss of activity, [~olvent,,,~], and the corresponding critical solvent concentration in the membrane, [SdVentmemb,~],to give [solventmemb] = solvent,,] (4) The value of X,the product term, was determined by Seeman (1972) to be close to 0.19 for a variety of mammalian membrane systems. Osborne et al. (1990) reasoned that the value of the product term for a microbial membrane would also be close to 0.19, although this may not hold true for all solvent types. Seeman found Y , the power term, to have a value of 0.98, again for mammalian membrane systems. A series of bioactivity studies performed by Osborne and co-workers, involving R. nigri-

cam and a series of alcohols, showed the value to be 0.84 for this microorganism. The difference in the values of the product term for the two different membrane systems was explained by their structural and compositional differences. The value of the power term can be determined for any microorganism by plotting the logarithm of experimentally determined critical solvent concentrations in the aqueous phase versus the logarithm of 1/ (Pwtanol of the solvent) for a number of solvents. The power term will have the value of the slope of the resulting straight line. The y-intercept of this line will yield the logarithm of the critical membrane concentration. Osborne and co-workers suggest that increased membrane fluidity, due to high solvent concentrations in the cell membrane, may be the primary cause of the loss of bioactivity. When the solvent concentration in the membrane exceeds the critical level, the high level of membrane fluidity prevents the maintenance of protein complexes. This increased knowledge of the toxic action of solvents can be used in a solvent selection technique in much the same way as the log P method but with greater insight into the mechanisms involved. If the aqueous solubility of a solvent is known and its log P value can be calculated, eq 4 can be used to determine the maximum concentration the solvent can reach in the cell membrane. If the critical

Biotechnol. Prog., 1991, Vol. 7, No. 2

120

solvent concentration cannot be attained, the solvent will not show dissolved toxicity. As well, the critical aqueous concentration for any solvent can be estimated.

Mixtures of Solvents The previous discussion has focused on prediction and selection techniques for pure (single) solvents. Clearly, significant progress has been made in this area, and researchers, through the judicious selection of several of the above strategies, are in a good position to identify effective solvents for extractive biocatalysis. Biocompatible solvents, however, tend to possess physical characteristics that make them poor extractants. The motivation to use more effective extractants is provided by the fact that product extraction by a superior extractant requires a smaller solvent volume to extract the product to the same extent as an inferior solvent. It has been hypothesized that a biocompatible yet poor solvent could be mixed with a toxic or inhibitory solvent that has a better extracting capability to yield a mixture with improved solvent characteristics that is still biocompatible. In accordance with equilibrium thermodynamics, the addition of a biocompatible solvent to a toxic solvent in a two-liquid phase system decreases the concentration of the toxic solvent dissolved in the aqueous phase. Evans and Wang have hypothesized that a solvent mixture could still be biocompatible if bioactivity is not inhibited below a specific concentration of solvent in the aqueous phase (Evans and Wang, 1988). The research by Osborne et al. (1990) indicates that this should be so. We have recently focused our attention (Bruce, 1990) on developing a strategy for predicting suitable cosolvents for extractive biocatalysis. Our intent was to include consideration of extraction capabilities as well as biocompatibility of the solvent mixtures. As a model system we have considered the bacterium 2. mobilis, and the in situ extraction of the ethanol product of the organism in connection with improving on the ethanol extractive fermentation process. An initial set of biocompatibility studies was performed with solvent mixtures to investigate whether a simple relationship between the log P values of the mixture and the resulting biocompatibility exists. The log P values of the mixtures were calculated on the basis of molar and mass ratios. Log P values based on volume ratios are almost the same as those based on mass ratios since most of the solvents investigated had similar densities. As shown in Figure 5, a good correlation was not apparent in either case. I t was also observed that, compared to biocompatibility studies involving pure solvents, studies with solvent mixtures gave far more results in the inhibitory region and there was no longer a sharp cutoff between toxicity and biocompatibility. The log P strategy appears not to be sufficient for predicting the biocompatibility of solvent mixtures. Figure 6 illustrates how all solvents do not mix in an ideal manner. These curves are typical of those which can be generated by ESP, providing a good indication of the shape of the actual equilibrium curves and showing the wide range in variation from ideal behavior. Ideal behavior is represented by a straight "curve", and solvents that mix in an increasingly nonideal manner are more highly curved. I t is apparent from the shape of the curves that some toxic and biocompatible solvent combinations would be better than others. For example, Figure 6a illustrates how solvent combinations involving hexadecane and heptanol would not be expected to be as good as ones involving dodecanal and heptanol. A greater

a

Percent Metabolic Act~vity

1207-

~-

__. . -

-

~

100 -

3

P

1

so

1

9

60 ; 40 F2

1

20

c

I 0'

~

5

4

6

log

7

--_

_.

9

8

10

P (molar ratios)

Percent Metabolic Activity ~

G

100

I

o c

I

I

i

so 1

0

.

4

5

6

-~ 7

log P (mass

I

~~

S

9

10

ratios)

Figure 5. (a)Relationship between the percent metabolic activity

of 2.mobilis cells and the log P values of mixtures, based on

molar ratios. (b) Relationship between the percent metabolic activity of 2.mobilis cells and the log Pvalues of mixtures,based on mass ratios. amount of heptanol, the toxic solvent, in the organic phase could be used with a mixture containing dodecanal before the critical concentration of toxic solvent in the aqueous phase is reached and a loss of bioactivity occurs. It is anticipated that, a t a constant fraction of toxic solvent in the organic phase, the order of measured bioactivity should be dodecanal, Ado1 85 NF, bis(2-ethylhexyl) maleate, and then hexadecane. The mixture of dodecanal and heptanol is the most biocompatible. A greater amount of the superior extracting solvent is also expected to result in better extracting abilities for the solvent mixture. The physical characteristics of the resulting mixture would, however, depend on the characteristics of both of the coextractants. Four sets of experiments were subsequently performed to demonstrate that the order of measured bioactivity can be predicted by the equilibrium curves. It was assumed that toxicity is dependent solely on the toxic solvent. The critical aqueous concentration of any toxic solvent corresponding to the critical membrane concentration can be calculated so that, with the use of the equilibrium curves, the concentration of toxic solvent in the cell membrane up to which it will be biocompatible can be estimated. The toxic solvents used in these studies, as shown in panels a, b, c, and d of Figure 6, were heptanol, octanol, 4-methylpentan-2-one, and octanoic acid. These solvents were chosen to represent a variety of types of solvents and to illustrate the order of measured bioactivity for mixtures of each toxic solvent with various biocompatible solvents. These toxic solvents have maximum membrane saturation concentrations greater than 200 mM but not so great that the loss of bioactivity would occur at very low or high organic volume fractions, where it is difficult to distinguish

Biotechnol. Prog., 1991, Vol. 7, No. 2

a ~

121

b

Mole Fraction Heptanol in Aqueous Phase

o

(Predicted by ESP)

o

o

6

-

~

-

-

~

-

~

~

-

-

Mole Fraction Octanol In Aqueous Phase (Predicted by ESP)

0.00020,

-

1

1

7

I

ic

oooozl 1

1

1

Y'

00001

I

C

Dd2-elhylhexyl) maleate

A I

l

I

'

i

1

0

1

09

d

Mole Fraction 4-Methyl Pentan-2-one in Aqueous Phase

___

r

Hexadecane

+ D~(Z-ethylhexyl) maleate

3

ic Adel 65 NF -6 Dodecanal

02 0.3 04 05 06 07 08 Volume Fraction Octanol in Organic Phase

01

09

1

Mole Fraction Octanoic Acid in Aqueous Phase (Predicted by ESP)

(Predicted by ESP) 0 0030---

-

,

Dodecanal

02 03 04 05 06 07 OS Volume Fraction Heptanol in Organic Phase

01

,

000005'

* L

oL 0

T

Hexadecane

-i- Ado1 8 5 NF

c' a

1

-

' 0

4

t

0 00035-

I

0 00030

00025-

0 00025 0 0020 0 00020 00015t I

II

_-

000015 I

00010

+

E

ooooto

?

--c 000051

1

'8

I80prol)yI Palmitate

5

000005

I8Oohy101

ri

a'

+ Hexadecane + Ado1 66

t

, 0

01 02 03 04 05 0.6 0.7 0.6 0.9 Volume Fraction 4-Methyl Pentan-2-one in Organic Phaae

O*

1

-_-i

0

01 02 03 04 05 06 07 0.8 09 Volume Fraction Octanoic Acid i n Organic Phase

1

Figure 6. (a) Equilibrium curves predicted by ESP for mixtures of heptanol and four biocompatible solvents, hexadecane, bis(2ethylhexyl) maleate, Adol 85 NF, and dodecanal. (b) Equilibrium curves predicted by ESP for mixtures of octanol and four biocompatible solvents, hexadecane, bis(2-ethylhexyl)maleate, Adol 85 NF, and dodecanal. (c) Equilibrium curves predicted by ESP for mixtures of 4-methylpentan-2-oneand four biocompatible solvents, hexadecane, isopropyl palmitate, isophytol, and tridecanone. (d) Equilibrium curves predicted by ESP for mixtures of octanoic acid and four biocompatible solvents, dibenzyl ether, hexadecane, Adol 66, and ethyl myristate.

between the different solvent mixtures. The solvents used were picked to show loss of bioactivity in the region where the difference between the various solvent mixtures was greatest, so that differentiating between the response of the microorganisms to the various mixtures would be easier. By using ESP to generate equilibrium curves, it was possible to select for biocompatible solvents that gave a wide range in variation from ideal behavior. As seen in Figure 6, the biocompatible coextractants used in these studies were chosen to show a good spread so that the results for the different mixtures could be easily differentiated. The biocompatible solvents were also chosen to represent a range of solvent types. The horizontal line on each of the graphs represents the aqueous mole fraction of the toxic solvent corresponding toa cell membrane concentrationof 200 mM. The aqueous solubilities of the toxic solvents required for this calculation were taken from Riddick et al. (1986). The aqueous solubilities of the toxic solvents predicted by ESP are used in conjunction with the actual solubilities to position the horizontal line representing the critical aqueous mole fraction on the graphs. Although the predicted values for aqueous solubility can vary from the actual values by over 300 , the purpose of ESP in this case was merely to establish the shape of the equilibrium curves. The value of 200 mM was the critical membrane concentration determined by Osborne and co-workers to

be the concentration at which complete loss of bioactivity occurred with R. nigricans. It was not anticipated that loss of activity found with these biocompatibility studies would correspond exactly to this concentration for two reasons. First, 2. mobilis is a significantly different microorganism, an anaerobic bacterium as opposed to a fungus. Second, during the present experiments the solvent was always present at concentrations greater than saturation. The solvent layer may have caused additional loss of activity in the form of the not-well-understoodphase toxicity. It was therefore expected that loss of activity would occur at a lower critical membrane concentration. Figure 7 shows the results of the biocompatibility studies. The results for heptanol, shown in Figure 7a, confirm that the order of bioactivity of the mixtures containing the various biocompatible solvents is the same as the order predicted by the order of the curves in Figure 6a. That is, mixtures containingdodecanal were the most biocompatible, followed by Adol 85 NF (oleyl alcohol), bis(Zethylhexy1)maleate, and finally mixtures containing hexadecane,which were the least biocompatible. The same pattern is shown in Figure 7b by the mixtures containing octanol. The mixtures containing 4-methylpentan-2-one illustrate the expected order of bioactivity, as shown by Figure 7c; however, the results for the middle two extractants are close together. From Figure 6c, it was expected that the bioactivity results for these mixtures would be close.

Biotechnol. hog., 1991, Vol. 7, No. 2

122

a

f

e'ac, c

.,=tli

t 7

I---

1

1

3 Hexadecane f

DlLP-elhylhexyll maleale I

+ Ado1 85

\

-

60 -

I

NF

Dodecanal

I ~~

\

I

\

40

-

80 \

I

I

-

I

'

\ \

I

F

01

2

0

01 d3

02

03

-

20

,

-

04

0.5

ume Frac' on rieptanol

0.6

07

11 O.c_an

1 -

-

+ I

-

0.8

0.9

1

01 0

~-

.u

01

02

03

04

05

06

07

0.8

1

09

c Phase

Hexadecane 180propyl Palmitate

,

I

loot ',

Isophytol

I 80-

Tridecanone

+

1

Dibenzyl Ether

I f Hexadecane

Ado1 6 6

- Ethyl M y r i i t a t a

,

~

I I

i

60

I 40 1

401

-\

i 20 I 00

& 0.1

0.2

0.3

0.4

I

0.5

0.6

0.7

0.8

0.9

1

'JoILme F r a c : i w 4-7ethyl pe?tap-2-or~eI * OrGarlc F i a s n

Figure 7. (a) Resulting percent metabolic activity of heptanol and four biocompatible Solvents. (b) Resulting percent metabolic activity of octanol and four biocompatible solvents. (c) Resulting percent metabolic activity of 4-methylpentan-2-oneand four biocompatible solvents. (d) Resulting percent metabolic activity of octanoic acid and four biocompatible solvents.

The bioactivity study involving octanoic acid mixtures also gave slightly unexpected results, as seen in Figure 7d. The solvent coextractant systems of Adol 66 and ethyl myristate appear to be in an order reversed to what was predicted by the equilibrium curves. It is not known at this point how or if phase toxicity resulting from the different mixtures might affect the bioactivity of the microorganisms to varying degrees. Despite these slight variations, it can be concluded that the equilibrium curves can be used to predict the order of bioactivity that will result from the use of cosolvents. ESP is therefore a valuable tool for the selection of superior coextractants. The distribution coefficients of the pure solvents and the solvent mixtures were determined experimentally to show that adding the toxic coextractant to the biocompatible solvent did in fact improve its extractant capabilities. The distribution coefficients predicted by ESP, DESP,are shown in Table IT1along with the experimentally obtained distribution coefficients, Dexp.The distribution coefficients predicted by ESP are tested over a wide range of solvent type and composition. In Table I11 the solvent mixtures are grouped by toxic solvent type and are arranged in order of increasing distribution coefficient. In some instances the predicted values are remarkably accurate; however, this accuracy cannot always be assured. The ESP parameters are derived from experimental UNIFAC data obtained from the literature, and where data are lacking, the program will obviously not give predictions as accurate.

It was mentioned previously that the solvent mixture that remained biocompatible to the highest fraction of the toxic solvent in the organic phase may not result in the optimal solvent combination. This is because the extraction properties of the mixture, such as the distribution coefficient, are dependent on the physical characteristics of both the solvents in the mixture and the way in which they interact. Experimental work by Munson and King (1983) and Mitchell et al. (1987) has been performed to investigate how the distribution coefficients and the selectivity of solvent mixtures behave. As a first approximation, it may be tempting to assume that these values could be predicted by simple interpolation of the values for the pure solvents; however, it was found that the measured distribution coefficients and selectivity values differed substantially from the values estimated by interpolation. Munson and King attribute the failure of the predictions to the following factors: preferential association between the components of a solvent mixture, the degree of solvation of a complexed form of the solute by the solvent mixture, and/or the ability of multiple solvent constituents to interact separately or synergistically with multiple functional groups on the solute molecule. Mitchell and co-workers add that the variability in the deviation between the predicted and observed results illustrates how difficult it is to employ methods based on the manipulation of ternary system data to predict the behavior of mixed solvents. Further examination of Table I11 shows the lack of accuracy of the numerical predictions in many cases;

Biotechnol. Prog., 1991, Vol. 7, No. 2

123

Table 111. Distribution Coefficients for Ethanol for P u r e and Mixed Solvents at 30 OC toxic solvents 4-methylpentan-2-one octanol octanoic acid heptanol

Dexp 0.34 0.52 0.57 0.61

biocompatible solvents hexadecane bis(2-ethylhexyl) maleate isopropyl palmitate ethyl myristate dibenzyl ether tridecanone Adol 85 NF dodecanal Adol 66 isophytol solvent mixtures (volume ratios) heptanol/ hexadecane (694) heptanol/ bis(2-ethylhexyl) maleate (12:88) heptanol/Adol 85 NF (1882) heptanol/dodecanal (42:58) octanol/ hexadecane (9:91) octanol/ bis(2-ethylhexyl) maleate (1882) octanol/Adol 85 NF (25:75) octanol/dodecanal (4258) 4-methylpentan-2-one/ hexadecane (15:85) 4-methylpentan-2-one/ isopropyl palmitate (33:67) 4-methylpentan-2-one/ tridecanone (45:55) 4-methylpentan-2-one/ isophytol (26:74) octanoic acid/hexadecane (15:85) octanoic acid/ethyl myristate (21:79) octanoic acid/dibenzyl ether (15:85) octanoic acid/Adol 66 (23:77)

DESP 0.62 0.91 0.62 1.08

Deip 0.07 0.12 0.16 0.19 0.23 0.23 0.24 0.29 0.30 0.30

DESP 0.07 0.27 0.17 0.19 0.12 0.22 0.34 0.63 0.30 0.30 DeXn 0.09 0.20

DESP 0.17 0.55

0.33 0.47 0.08 0.22

0.57 0.72 0.17 0.50

0.34 0.36 0.07

0.52 0.65 0.11

0.17

0.24

0.23

0.34

0.24

0.39

0.07 0.17 0.27 0.28

0.10 0.22 0.20 0.34

however, ESP is still of considerable use. The order by which the distribution coefficients of pure and mixed solvents are ranked by ESP often corresponds to what is found experimentally. ESP is able to predict trends among combinations of both type and composition. As well, when the distribution coefficient values of mixtures of two pure solvents are plotted against the composition of the mixtures, ESP can be used to show the nonlinear variations in the distribution coefficient values.

Summary The purpose of a solvent screening strategy is to find the best solvent for any extractive biocatalytic process with minimal experimental work. There are a few ways by which the amount of experimental testing can be reduced. A superior, biocompatible, pure solvent for extracting any product is easily selected by using ESP to screen and rank the extraction behavior of all solvents with a log P value greater than the inflection point corresponding to the biocatalyst used. The top-ranked solvents could then be subjected to the additional criteria listed in Table I. The selection of a superior solvent mixture is slightly more complicated. The biocompatible coextractants are selected in the same way biocompatible pure solvents are. The toxic or inhibitory coextractants are chosen by employing both knowledge of their aqueous solubilities and log P values. Equation 4 is then used to determine

the maximum concentration of the toxic solvents in the cell’s membrane, and from this it can be determined which solvents are too toxic to bother screening. ESP is then used to generate equilibrium curves for all toxic and biocompatible coextractant pairs as was done for Figure 6. Since the critical membrane concentration for the cell can be found, the corresponding critical aqueous concentration can be determined for each toxic or inhibitory solvent. The ratio of the biocompatible solvent to the toxic solvent in the organicphase can then be calculated to correspond to the critical concentration of the toxic solvent in the aqueous phase, and ESP can be used again to calculate the distribution coefficient of this solvent mixture. With a modest amount of work on the computer using ESP, the most promising solvent mixtures can be selected and a significant amount of experimental work can be avoided. We have, in the past, assisted a variety of researchers and companies in the selection of solvents for numerous applications by means of our ESP database. We would welcome additional collaborations of this nature.

Appendix: Sources of Industrial Solvents Adol 85 NF and Adol 66 were obtained from Sherex Chemical Co. Inc., Dublin, OH, and Aldehyde C-14 and Aldehyde C-18 came from Givaudan Co., Clifton, NJ. Procter and Gamble, Industrial Chemicals Division, Cincinnati, OH, supplied CO-1214 fatty alcohol, and Ethyl Co., Baton Rouge, LA, provided Epall214 and Epal1218. Eutanol G-16 and Eutanol G were obtained from Henkel Canada Ltd., Hamilton, ON, Canada, and Neodol23 and Neodol 25 were obtained from Shell Chemical Co., Houston, TX. The log P values for these solvents were calculated by using molar ratios of their components.

Literature Cited Bar, R. In Biocatalysis in Organic Media; Laane, C., Tramper, J., Lilly, M. D., Eds.; Elsevier: Amsterdam, 1987; p p 147-156. Barton, W. E. M.Sc. Thesis, Queen’s University, Kingston, Ontario, Canada, 1990. Brink, L. E. S.; Tramper, J. Optimization of Organic Solvent in Multiphase Biocatalysis. Biotechnol. Bioeng. 1985,27,12581269. Bruce, L. J. M.Sc. Thesis, Queen’s University, Kingston, Ontario, Canada, 1990. Daugulis, A. J. Integrated Reaction and Product Recovery in Bioreactor Systems. Biotechnol. Prog. 1988, 4 , 113-122. Daugulis, A. J.; Swaine, D. E.; Kollerup, F.; Groom, C. A. Extractive Fermentation-Integrated Reaction and Product Recovery. Biotechnol. Lett. 1987, 9, 425-430. Eggers, D. K.; Blanch, H. W.; Prausnitz, J. M. Extractive Biocatalysis: Solvent Effects on Equilibria of Enzymatic Reactions in Two-Phase Systems. Enzyme Microb. Technol. 1989,11, 84-89. Evans, P. J.; Wang, H. Y. Response of Clostridium acetobutylicum to the Presence of Mixed Extractants. Appl. Biochem. Biotechnol. 1988,16, 175-192. Finn, R. K. Inhibitory Cell Products-Their Formation and Some New Methods of Removal. J. Ferment. Technol. 1966, 44, 305-310. Halling, P. J. Solvent Selection for Biocatalysis in Mainly Organic Systems: Predictions of Effects on Equilibrium Position. Biotechnol. Bioeng. 1990, 35, 691-701. Hocknull, M. D.; Lilly, M. D. In Biocatalysis in Organic Media; Laane, C., Tramper, J., Lilly, M. D., Eds.; Elsevier: Amsterdam, 1987; pp 669-674. Kollerup, F.; Daugulis, A. J. Screening and Identification of Extractive Fermentation Solvents Using a Database. Can.J. Chem. Eng. 1985,63, 919-927.

124

Kollerup, F.; Daugulis, A. J. Ethanol Production by Extractive Fermentation-Solvent Identification and Prototype Development. Can. J. Chem. Eng. 1986,64,598-606. Kollerup, F.; Daugulis, A. J. Process Development of a Prototype Extractive Fermentation System. Ann. N . Y. Acad. Sci. 1987, 506,478-491. Kollerup, F. Ethanol Production by Extractive Fermentation. Ph.D. Thesis, Queen’s University, Kingston, Ontario, Canada, 1986. Laane, C.; Boeren, S.; Vos, K. On Optimizing Organic Solvents in Multi-Liquid Phase Biocatalysis. Trends Biotechnol. 1985, 3, 251-252. Laane, C.; Boeren, S.; Vos, K.; Veeger, C. In Biocatalysis in Organic Media; Laane, C., Tramper, J., Lilly, M. D., Eds.; Elsevier: Amsterdam, 1987; pp 65-84. Lawford, H. G. In Fuel Ethanol; Noyes Publications Inc.: Park Ridge, NJ, 1991; in press. Lilly, M. D.; Brazier, A. J.; Hocknul, M. D.; William, A. C.; Woodley, J. M. In Biocatalysis i n Organic Media; Laane, C., Tramper, J., Lilly, M. D., Eds.; Elsevier: Amsterdam, 1987; pp 3-20. Martinek, K.; Semenov, A. N.; Berezin, I. V. Enzymatic Synthesis in Biphasic Aqueous-Organic Systems I. Chemical Equilibrium Shift. Biochim. Biophys. Acta 1981,658,76-89. Minier, M.; Goma, G. Production of Ethanol by Coupling Fermentation and Solvent Extraction. Biotechnol. Lett. 1981, 3,405-408. Minier, M.; Goma, G. Ethanol Production by Extractive Fermentation. Biotechnol. Bioeng. 1982, 24, 1565-1579. Mitchell, R. J.;Arrowsmith, A.; Ashton, N. Mixed Solvent Systems for Recovery of Ethanol from Dilute Aqueous Solution by Liquid-Liquid Extraction. Biotechnol. Bioeng. 1987,30,348351.

Biotechnol. frog., 1991, Vol. 7, No. 2

Munson, C. L.; King, C. J. Factors Influencing Solvent Selection for Extraction of Ethanol from Aqueous Solutions. Ind. Eng. Chem. Process Des. Deu. 1983,23, 109-115. Osborne, S. J.; Leaver, J.; Turner, M. K.; Dunnill, P. Correlation of Biocatalytic Activity in an Organic-Aqueous Two-Liquid Phase System with Solvent Concentration in the Cell Membrane. Enzyme Microb. Technol. 1990, 12, 281-291. Playne, M. J.; Smith, B. R. Toxicity of Organic Extraction Reagents to Anaerobic Bacteria. Biotechnol. Bioeng. 1983, 25, 1251-1264. Rekker, R. F. T h e Hydrophobic Fragmental Constant; Nauta, W. T., Rekker, R. F., Eds.; Elsevier: Amsterdam, 1977. Rekker, R. F.; de Kort, H. M. The Hydrophobic Fragmental Constant; an Extension to a 1000 Data Point Set. Eur. J. Med. Chem. 1979,14,479-488. Reslow, M.; Aldercreutz, P.; Mattiasson, B. Organic Solvents for Bioorganic Synthesis. 1. Optimization of Parameters for a Chymotrypsin Catalyzed Process. Appl. Microbiol. Biotechnol. 1987, 26, 1-8. Riddick, J. A.; Bunger, W. B.; Sakano, T. Techniques of Chemistry; Weissberger, A., Ed.; John Wiley and Sons: New York, 1986; p 2. Roffler, S. R.; Blanch, H. W.; Wilke, C. R. In Situ Extractive Fermentation of Acetone and Butanol. Biotechnol. Bioeng. 1988,31, 135-143. Seeman, P. The Membrane Actions of Anesthetics and Tranquilizers. Pharmacol. Rev. 1972, 24, 583-655. Accepted January 2, 1991.

Solvent selection strategies for extractive biocatalysis.

This report follows the development of systematic solvent screening strategies for the identification of superior pure solvents and introduces techniq...
1MB Sizes 0 Downloads 0 Views