- Email: [email protected]
High speed centrifugal separator for rapid on-line sample clarification in biotechnology
Journal of Biotechnology 49 (1996) 1 ll- 118
ELSEVIER
High speed centrifugal separator for rapid on-line clarification in biotechnology P. Richardson’,“, “The Aduanced
J. Molloyb,
R. Ravenhall’,“,
I. Holwill”T*, M. Hoare”,
Centre ji)r Biochemical Engineering, Department of‘ Chemical and Biochemical Engineering, London, Torrington Place, London WClE 7JE, UK bFisons Applied Sensor Technology, Saxon Way, Bar Hill, Cambridge, CB3 8SL, UK
sample
P. Dunnill” Unioersity
College
Received 13 February 1996; revised 7 March 1996; accepted 3 April 1996
Abstract Sample clarification is a common operation in biochemical analytical methods for removing interfering or unwanted particulates from an analyte sample. Filtration provides one option for removal of particulates. However, in many cases the loss of soluble protein due to filter adsorption is unacceptable and an alternative must be sought. In this paper a microcentrifuge designed to automatically sample, spin, deliver supernatant to an analyser and wash out solids from the bowl is described. The performance of the system is assessed in terms of its clarification efficiency and the time required to achieve satisfactory clarification. Additionally, the effects of different protein precipitating agents on yeast homogenate samples separated using the microcentrifuge are studied where the system is used to deliver supernatant to a flow injection analyser. The paper demonstrates that the microcentrifuge may be used to separate rapidly such samples on a time scale between IO&60 s depending upon the type and size of sample and be successfully used as a component of an at-line monitoring system. Keywords:
Microcentrifuge;
At-line;
Monitoring;
Automated;
Bioprocess
Abbreciations: F, fraction of precipitate in supernatant (-); OD, optical density at 650nm (-); OD,,, optical density of supernatant (-); OD,, optical density prior to centrifugation (-); dp, particle-
1. Introduction
fluid density.difference (kg m - ‘); w, angular velocity of rotor (rad s- ‘); t, time (s); p, fluid viscosity (N s m -*); PEG, polyethylene glycol; RCF, relative centrifugal force. * Corresponding author. ’ Currently employed at: Zeneca Pharmaceuticals, Mereside, Alserley Park, Macclesfield, Cheshire, SKI0 4TG, UK. z Currently employed at: Industrial Research Limited, Gracefield Research Centre, PO Box 31-310, Lower Hutt, New Zealand.
The production, recovery and purification of biological products involves a series of often complex multi-phase unit operations. Sophisticated processing equipment such as intermittent discharge disc-stack centrifuges and cross flow filtration devices have evolved to facilitate concentration and enrichment by phase separa-
0168-1656/96/$15.00
0 1996 Elsevier
PII SO168-1656(96)01533-7
Science
B.V. All rights
reserved
112
P. Richardson
et al. :I Joumcrl
tion. Solid-liquid separation of large volumes of suspension during cell harvesting and washing from a reactor is an example; however, the technology for analytical scale on-line sampling of these unit operations is still limited. This is partially because monitoring of biological processes has traditionally been by manual sample preparation and analysis and partly due to the difficulty of effecting accurate and consistent on-line sample preparation, given the unpredictable and variable nature of bioprocesses. In addition there has been a lack of suitably robust on-line analytical equipment available to justify the development of such devices. A major requirement in the analysis of biological fluids is that measurements interfered with by even small quantities of contaminating solids phase must be eliminated or there will be spurious and erratic results. If the measurements form part of a feedback control loop, the deviations from the set point due to sample contamination might be indicated when in reality the sample preparation is at fault. Sample microfiltration has been used successfully in a number of applications such as fermenter broth clarification (Recktenwald et al., 1985) for nutrient analysis and monitoring the efficiency of cell homogenisation (Kroner and Kula, 1984). The system has been developed into a patented continuous sampling device for bioprocess monitoring (Braun-Melsungen, 1986). However in many instances microfiltration is unsuitable because blinding of the filter surface or adsorption of protein to the filter material leads to a fall in transmission of protein through the membrane, resulting in discrepancies between measured and true process levels of analyte. Centrifugation is an alternative means of sample preparation where solid-liquid separation is required which, although a mechanically complex process compared with filtration, can cope more effectively with high solids loadings and heavily blinding suspensions that might foul a microfilter. In this paper we describe the operation and separation characteristics of a recently developed high-speed, small-volume centrifugal separator, suitable for the rapid at-line sampling of biological processes. The application of the device for facilitating the measurement of protein solubility during fractional precipitation of
of Biotechnolog_v
49 (1996)
I I I-l
18
proteins from yeast homogenate is described. Here it is the recovery of a representative sample of the remaining soluble protein for rapid measurement which is of importance and issues such as precipitate carry-over, precipitate redissolution or soluble protein retention must be addressed.
2. Materials
and methods
2.1. Cen trifugul separutor A schematic view of the centrifugal separator is shown in Fig. 1. The prototype device was designed and constructed by Fisons Scientific Instruments (Uxbridge, Herts, UK). The centrifuge consists of a tulip shaped, 20-mm diameter bowl rotated at speeds up to 60000 rev./min by an air-driven turbine conventionally used for cutting tools. The speed of the bowl is controlled by the inlet air pressure to the turbine; a speed of _ 60000 rev./min being generated by 6 bar gauge. The operation and timing of the various items making up the centrifuge were sequenced by a bench top microcomputer via a digital output port. The inset in Fig. 1. shows the definition of r0 and ri the outer wall radius and sample surface when spinning respectively. Y, is 10 mm and Y, ranges from 9.1 mm to 7.6 mm for sample volumes of 0.3 ml to 0.9 ml. As the sample load to the centrifuge increases, r, is reduced and the ratio rJr, increases. At sample initiation, the feed pump is switched on for a pre-determined length of time. At the same time the vacuum solenoid valve is opened so that any sample entering the bowl is immediately removed to waste. This has the effect of clearing the sample line of the preceding sample and ensuring that the sample delivered to the centrifuge is representative of that at the point of sampling. Once the sample line has been cleared, the vacuum solenoid valve is closed and the feed pump continues to deliver sample to the bowl for a fixed length of time (fixed sample volume). The air pressure valve to the turbine is then opened, and the bowl rapidly gains maximum speed. The sample is spun for the required length of time causing the solids to sediment in the extremes of the bowl
P. Richardson
et al. / Journal of Biotechnology
after which the air flow to the turbine is stopped and the friction brake solenoid is activated. As the bowl stops rotating, the clarified supernatant flows to the base of the bowl while the solids pellet remains on the bowl wall. The sample-out pump is then activated for a predetermined time
Table 1 Sequence tion
of events and timings
Sequence
time (s)
0 5
sample in I clarified sample out
” vacuum
-Protective
shroud
sample volume - 1 ml
6.5 6.5 + T 26.5 + T 41.5 + T 51.5 + T
turbine
113
49 (1996) 111- 118
for the microcentrifuge
opera-
Action Sample in pump and vacuum solenoid on Vacuum solenoid off, sample is delivered to bowl Sample in pump off and turbine drive valve open Turbine drive valve closed and friction brake on Sample out pump on to deliver sample to analyser Wash cycle stop
pressurised air
T, centrifugation
time.
T
pump control lines
Fig. 1. A schematic of the microcentrifuge sampling system. Pumps and valves are operated under microcomputer control. The sample is pumped in and out and wash water directed at the sides of the bowl by means of peristaltic pumps. The bowl is cleaned out using a vacuum suction line and rotates at speeds up to 60000 rev./min driven by a turbine powered by air flow with input pressures regulated up to 6 bar. A friction brake is included to slow the bowl down after the sample centrifugation time has elapsed. The bowl is housed in a moulded rigid plastic protective cover and the entire unit ruggedised to enable operation in a pilot plant environment. The bowl dimensions are approximately 2 cm diameter (wall to wall) and a typical sample volume would be 1 ml. An entire cycle for sample preparation takes approximately 30 s plus the centrifugation time. The inset shows the position of the sample in the bowl as it is spinning and the definition of Y,, the radius of the wall of the centrifuge and Y, the radius of the sample surface.
to deliver a set volume of clarified material to the sample loop of the associated analyser, a flow injection analyser in this instance. To clean the centrifuge, wash water is pumped into the bowl and against the wall as a high pressure jet through two small bore delivery tubes. Simultaneously vacuum is applied to remove the waste washings. The turbine is also given a number of rapid blasts of air to turn the bowl ensuring solids are removed from all parts. The centrifuge is then ready for the next sample. The complete cycle takes approximately 50 s plus the described clarification time, the sample being delivered to the analyser after approximately 30 s plus the clarification time. Typical timings for these various sequences are given in Table 1. 2.2. Yeast homogenate
preparation
Blocks of bakers yeast Saccharomyces cerevisiae (Distillers Co. Ltd., Crawley, West Sussex, UK) were suspended in phosphate buffer (0.067 M KH,PO,, 0.033 M K,HPO,, pH adjusted to 6.5 with 2.5 M NaOH) to a final concentration of 750 kg rn-’ packed weight. The yeast was mechanically disrupted using an APV Gaulin high pressure homogeniser (model 15M, APV Crawley, W. Sussex, UK) operating continuously at 500 bar and 5°C for the equivalent of four passes of the suspension volume. Cell debris was removed by
114
P. Richrrrdson et al.
/ Journal of Biotrclznolo~y
centrifuging the homogenate at 53000 average RCF for 1 h (Pegasus Ultracentrifuge, MSE Crawley, UK using 6 x 032 head at 21000 rev./min). 2.3. Precipitant
preparation
Ammonium sulphate precipitation agent (Analar, BDH, Poole, UK) was made up to 3.9 M solution in phosphate buffer (0.067 M KH,PO,, 0.037 M K,PO,, pH adjusted to 6.5 with 2.5 M NaOH). At a temperature of 20°C this corresaturated solution. sponded to a 98% Polyethylene glycol (PEG) precipitating agent (Biochemical grade, BDH, Poole, UK) was made up to 250 kg rnp3 in phosphate buffer, pH 6.5. 2.4. Precipitate
suspension
49 (1996) 111~ 118
3. Results and discussion Figs. 2 and 3 illustrate examples of the settling curves for precipitated yeast homogenate using the centrifugal sampler. The fraction of precipitate remaining in suspension, F, was estimated from measurements of optical density at 650 nm: OD ~ OD,,,,
F=
OD,, -
(1)
OD,,,
OD,, was determined from supernatant samples following high speed batch centrifugation at 10000 RCF for 0.5 h. OD is the optical density after clarification and OD, the optical density prior to clarification.
preparation
The precipitate suspension was prepared in a continuous precipitation device which has been described in detail elsewhere (Richardson et al., 1989). In summary, it consisted of a pair of variable speed gear pumps feeding yeast homogenate and precipitant into a T-piece, followed by a needle valve to achieve intimate mixing of the two streams. Following the needle valve, another T-piece formed the sampling point. The main process flow continued to waste while the sampling centrifuge intermittently pulled a small side flow off from the branch.
0.0
-
-0.5 -
3 G-1.0
-
z
-1.5 -
2.5. Assays The clarified samples delivered from the centrifuge were analysed manually for ADH activities and total protein concentration. The use of the centrifuge in conjunction with an on-line flow injection analyser is described elsewhere (Richardson et al., 1996). ADH activity was measured by following the absorption change at 340 nm of NAD + to NADH accompanying the conversion of ethanol to acetaldehyde (Bergmeyer et al., 1983). Total protein concentration was determined by the dual UV wavelength method of Groves et al. (1968) as modified by Ehresmann et al. (1973) using bovine serum albumin as standard.
Pelletingtime (s) Fig. 2. Settling curves for two different microcentrifuge rotor speeds (0 30 000 rev./min), (m 60000 rev.,‘min) at a precipitant saturation of 70%. Linear fits to each data set are also shown (. .). At the higher speed, clarification to the same fraction
F =
of precipitate
remaining
in sus-
pension occurs apL&oxima;eiy three times faster, in line with theoretical calculations (t, = 97.9 k 29.2 s: t, = 260.27 i 26.2 s). Times were measured for clarification levels to reach 99% as a residual suspension turbidity was always observed with ammonium sulphate used as the precipitant.
P. Richardson et al. 1 Journal of Biotechnology 49 (1996) 1 I1 - I18
spin time (s) Fig. 3. The effect of precipitant and sample size on yeast protein precipitate settling characteristics. Key points are the rapid separation of the precipitates formed with 20% w/v PEG6000 (0.5 ml sample volume) (- V -) and 80% w/v ethanol (0.5 ml sample volume) (- 0 -) compared with those formed with ammonium sulphate and the increase in the separation times of these precipitates with correspondingly increased sample volumes; (- 0 -) 0.3 ml, (- 0 -) 0.5 ml, (- A -) 0.9 ml 80% saturation ammonium sulphate as precipitant. The centrifuge speed for ail these samples was 30000 rev.imin.
From centrifugation theory, the time required for a particle of diameter d to move from the outer radius of the sample (v,) to the radius of the bowl (Y,) is given by:
where Ap is the particle-fluid density difference, w is the angular velocity and ,LLis the suspending fluid viscosity. Theoretically the time required to pellet the precipitated protein completely is equivalent to the time for the smallest protein aggregate present to move from r. to r, assuming the precipitated protein has a small volume fraction. The true pelleting time may be obtained from settling plots. In the case of ammonium sulphate precipi-
115
tation, however, a small residual turbidity was always observed which was greater than OD,, making the true pelleting time difficult to estimate. This residual turbidity was shown to be due to suspended precipitated protein, i.e. it easily redissolved on lowering the ammonium sulphate concentration. To characterise the centrifuge performance, the nominal pelleting time was assigned as the time required to clarify 99% of the precipitate. This value was measured by plotting the ordinate as a logarithmic scale to emphasise the smaller optical density measurements. Fig. 2 illustrates the effect of rotor speed on clarification of a 0.5-ml sample of yeast homogenate precipitated at 70% saturated ammonium sulphate. As the rotor speed is increased from 33 000 to 60000 rev./min, Eq. (1) predicts that the pelleting time should be reduced by a factor of 3.3 ( = 602/332). From Fig. 2, it is estimated that the settling time reduced by a factor of 2.7 + 0.84 confirming that attainment of high velocity is a prerequisite for achieving rapid sampling frequency. Fig. 3 presents data showing the effect of sample size on clarification speed for clarified yeast homogenate insolubilised with the precipitants: saturated ammonium sulphate, 25% w/v PEG and absolute ethanol. In each case, the clarified homogenate was taken to 80% by volume of precipitant solution. Previous work (Richardson, 1987) has indicated that at this concentration the fraction of protein precipitated was the same for each precipitant. The figure clearly indicates the ease with which the PEG or ethanol precipitated material was clarified compared with that of the ammonium sulphate, the former taking place within 10 s for a 0.5-ml sample. This illustrates one of the difficulties often encountered in processing precipitated protein at high concentrations of ammonium sulphate. At these concentrations, the density difference between precipitated protein and suspending phase becomes very small ( < 100 kg mp3) and thus slow sedimentation velocities result. In this instance however the effect is in part beneficial since it allows the settling curve to be constructed. For PEG and ethanol, clarification was so rapid that the settling curves could not be accurately followed, which is a desirable situation for high frequency sampling.
P. Richurdson et al.
116
/ Journal of Biotechnology 49 (1996) I I I-I 18
As the sample size added to the rotor increases, the settling time is elongated since this has the effect of increasing the effective settling distance, and therefore In (r,/r,). From an accurate engineering drawing of the rotor, the value of In (r,/r,) was calibrated against the sample volume. The pelleting time for the three sample sizes was estimated as before. In Fig. 4 pelleting time is plotted against In (YJY,) according to Eq. (1). Although the data are limited there is good correspondence with Eq. (1) in that a linearity that passes through the origin exists and a statistically good fit is achieved. From Figs. 2 and 4, it is apparent that changes in pelleting time in the centrifugal sampler due to variations in spin speed or sample volume can be predicted satisfactorily. This is important since choosing a maximum angular velocity and small sample volume to minimise set-
5-
0 60
20
40
Ammonium sulphate concenttatia,
60
80
(% satuation)
Fig. 5. Protein solubility profile measured using sampling microcentrifuge versus equilibrium solubility curve for ammonium sulphate precipitated protein. The discrepancy, which remains approximately constant across the range of precipitant concentrations, is due to the unsedimented or resuspended protein precipitated.
20
10
: .’ -
,. :
:
ol
0.00
*
’
0.E
’
’
0.10
’
’
0.15
*
’
020
m
’
025
’ 0.30
In(r&)
Fig 4. The time required to pellet ammonium sulphate precipitated protein as a function of ln(r,/r,) where r, is the inner sample radius and rO the rotor bowl radius. The rotor speed was 30000 rev./min. A good linear least squares fit is apparent with a x2 value of 3.64 which is entirely compatible with a linear fit given the error estimates on the data points. This allows prediction of the spin-time for a given sample volume in order to optimise the sampling time for the analysis cycle.
tling time may not always be consistent with the intended use of the sampler. For example an on-line analyser sample loop downstream of the centrifuge may require a minimum volume to fill and purge it. Similarly an overly high spin speed could lead to over-dewatering of the sedimented material, making water jet cleaning of the bowl difficult and leading to a cumulative build up of solids. Sample size and bowl velocity are therefore best optimised for each individual application. Figs. 5 and 6 show the solubility of total protein precipitated with a range of concentrations of ammonium sulphate and PEG respectively as measured by clarification for 60 s using the sampling centrifuge, or by using batch centrifugation at 10000 RCF for 0.5 h i.e. the ‘true’ equilibrium solubility. As suggested by the preceding settling curves, the sampling centrifuge totally clarifies the PEG precipitated material and this is indicated by the fact that the solubility
P. Richardson
et al. / Journal of Biotechnology
curve generated from the sampling centrifuge corresponds to the true solubility curve. For the ammonium sulphate precipitated material however, the solubility measured following sampling centrifugation is slightly higher than the equilibrium solubility over the range of ammonium sulphate concentrations. This results in the slight residual turbidity for the higher spin time (Fig. 2). The concentration offset between measured and equilibrium solubility is effectively constant over the entire profile, thus this difference is independent of the equilibrium solubility. The reason for the incomplete clarification of the ammonium precipitated protein is unclear. The offset is not an inherent characteristic of the machine since complete sedimentation was observed with PEG and ethanol produced precipitates (Fig. 3). Neither is the offset due to insufficient settling time being allowed as the residual level is rapidly reached (well within 3 min) often with no further clarification achieved.
49 (1996) I II- 118
117
A potential cause for this behaviour could be in the particular properties of ammonium sulphate precipitated protein. It has been shown (Chan et al., 1986) that the protein aggregates produced by ammonium sulphate precipitation are small compared with other precipitants extending well into the submicron range. Whilst the preponderance of small aggregates will undoubtedly be sedimented given sufficient time, it is suggested that they form a loosely parted layer on top of the truly sedimented pellet. As the rotor is slowed down following a run the shear induced in this surface layer is sufficient to disrupt the loosely captured smaller aggregates thereby resuspending them in the bulk suspension. It is likely that these fine particles will not be recovered in a continuous centrifuge typically used for the industrial scale recovery of the precipitate and hence the sampling centrifuge probably gives a better measure than the laboratory batch centrifuge of recoverable precipitate.
4. Conclusion
0
I,
I
2
4
I
I
6
I
I
6
PEG concentration
I1
II
.I.
10
12
14
(% VA)
Fig. 6. Protein solubility profiles measured using the sampling microcentrifuge and the equilibrium solubility for PEG precipitated protein. The equilibrium and microcentrifuge data are identical indicating the relative ease with which PEG precipitates are separated in comparison to ammonium sulphate precipitates. This plot indicates that the offset shown in Fig. 5 is not a feature of the microcentrifuge but peculiar to ammonium sulphate precipitates.
The centrifugal sampling device represents a novel and potentially very useful solution to the problem of rapid on-line sampling of bioprocesses. Unlike filtration devices, it is not prone to variability with time caused by proteinaceous filter fouling, the separation being achieved by robust mechanical means. The centrifugal sedimentation curves generated may be theoretically well characterised in terms of angular velocity and settling distance (sample size). Practical clarification results obtained from precipitated yeast homogenate show agreement with the predicted theoretical trends. Spun homogenate precipitated with PEG or ethanol is particularly well clarified while spun homogenate precipitated with ammonium sulphate is satisfactorily but not completely clarified. This latter effect is ostensibly due to resuspension of the smaller aggregates produced by ammonium sulphate which are resuspended when the rotor decelerates after a run. In all cases a sample suitable for at-line analysis can be produced within 30-60 s.
Acknowledgements UCL is the Biotechnology and Biological Sciences Research Council sponsored Advanced Centre for Biochemical Engineering and the Council’s support is gratefully acknowledged.
References Bergmeyer, H.U., Grabl. M. and Walter, H.L. (1983) Reagents for enzymatic analysis. In: Bergmeyer, H.U. (Ed.), Methods of Enzymatic Analysis. Vol. 2. 3rd edn. Verlag Chemie, Weinheim, Ch. 2. pp. 1399140. Braun-Melsungen, A.G. (1986) Das Probeentnahmegerat fur die chemische und insbesondere die biochemische Proresstechnik.... Patentschrift DE 3520489 Cl. Chart, M.Y.Y., Hoare. M. and Dunnill, P. (1986) The kinetics of protein precipitation by different reagents. Biotechnol. Bioeng. 28. 387-393. Ehresmann, P.. Imbaull, P. and Weil. J.H. (1973) Spectroscopic determination of protein concentration in cell ex-
tracts containing tRNA’s and rRNA’s. Anal. Biochem. 54. 454 463. Groves. W.E.. Davies Jr., F.C. and Sells, B.H. (1968) Spectroscopic determination of microgram quantities of protein without nucleic acid interference. Anal. Biochem. 22. 195 210. Kroner. K.H. and Kula. M.R. (1984) On-line measurement of extracellular enzymes during fermentation by using membrane techniques. Anal. Chim. Acta 163. 3 15. Recktenwald. A., Kroner, K.H. and Kula. M.R. (1985) Online monitoring of enzymes in downstream processing by flow injection analysis. Enzyme Microb. Technol. 7. 607 612. Richardson. P. (1987) A Biochemical Engineering Study of Fractional Protein Precipitation. Ph.D. Thesis. University of London. Richardson, P., Hoarc, M. and Dunnill, P. (1989) Optimisation of fractional precipitation for protein purification. Chem. Eng. Res. Des. 67, 273-227. Richardson. P.. Ravenhall. R., Flanagan. M.T.. Holwill, 1.. Molloy. J.. Hoarc, M. and Dunnill., P. (1996) Monitoring and optimisation of fractional protein precipitation by flow injection analysis. Proc. Cont. Qual., submitted.
ELSEVIER
High speed centrifugal separator for rapid on-line clarification in biotechnology P. Richardson’,“, “The Aduanced
J. Molloyb,
R. Ravenhall’,“,
I. Holwill”T*, M. Hoare”,
Centre ji)r Biochemical Engineering, Department of‘ Chemical and Biochemical Engineering, London, Torrington Place, London WClE 7JE, UK bFisons Applied Sensor Technology, Saxon Way, Bar Hill, Cambridge, CB3 8SL, UK
sample
P. Dunnill” Unioersity
College
Received 13 February 1996; revised 7 March 1996; accepted 3 April 1996
Abstract Sample clarification is a common operation in biochemical analytical methods for removing interfering or unwanted particulates from an analyte sample. Filtration provides one option for removal of particulates. However, in many cases the loss of soluble protein due to filter adsorption is unacceptable and an alternative must be sought. In this paper a microcentrifuge designed to automatically sample, spin, deliver supernatant to an analyser and wash out solids from the bowl is described. The performance of the system is assessed in terms of its clarification efficiency and the time required to achieve satisfactory clarification. Additionally, the effects of different protein precipitating agents on yeast homogenate samples separated using the microcentrifuge are studied where the system is used to deliver supernatant to a flow injection analyser. The paper demonstrates that the microcentrifuge may be used to separate rapidly such samples on a time scale between IO&60 s depending upon the type and size of sample and be successfully used as a component of an at-line monitoring system. Keywords:
Microcentrifuge;
At-line;
Monitoring;
Automated;
Bioprocess
Abbreciations: F, fraction of precipitate in supernatant (-); OD, optical density at 650nm (-); OD,,, optical density of supernatant (-); OD,, optical density prior to centrifugation (-); dp, particle-
1. Introduction
fluid density.difference (kg m - ‘); w, angular velocity of rotor (rad s- ‘); t, time (s); p, fluid viscosity (N s m -*); PEG, polyethylene glycol; RCF, relative centrifugal force. * Corresponding author. ’ Currently employed at: Zeneca Pharmaceuticals, Mereside, Alserley Park, Macclesfield, Cheshire, SKI0 4TG, UK. z Currently employed at: Industrial Research Limited, Gracefield Research Centre, PO Box 31-310, Lower Hutt, New Zealand.
The production, recovery and purification of biological products involves a series of often complex multi-phase unit operations. Sophisticated processing equipment such as intermittent discharge disc-stack centrifuges and cross flow filtration devices have evolved to facilitate concentration and enrichment by phase separa-
0168-1656/96/$15.00
0 1996 Elsevier
PII SO168-1656(96)01533-7
Science
B.V. All rights
reserved
112
P. Richardson
et al. :I Joumcrl
tion. Solid-liquid separation of large volumes of suspension during cell harvesting and washing from a reactor is an example; however, the technology for analytical scale on-line sampling of these unit operations is still limited. This is partially because monitoring of biological processes has traditionally been by manual sample preparation and analysis and partly due to the difficulty of effecting accurate and consistent on-line sample preparation, given the unpredictable and variable nature of bioprocesses. In addition there has been a lack of suitably robust on-line analytical equipment available to justify the development of such devices. A major requirement in the analysis of biological fluids is that measurements interfered with by even small quantities of contaminating solids phase must be eliminated or there will be spurious and erratic results. If the measurements form part of a feedback control loop, the deviations from the set point due to sample contamination might be indicated when in reality the sample preparation is at fault. Sample microfiltration has been used successfully in a number of applications such as fermenter broth clarification (Recktenwald et al., 1985) for nutrient analysis and monitoring the efficiency of cell homogenisation (Kroner and Kula, 1984). The system has been developed into a patented continuous sampling device for bioprocess monitoring (Braun-Melsungen, 1986). However in many instances microfiltration is unsuitable because blinding of the filter surface or adsorption of protein to the filter material leads to a fall in transmission of protein through the membrane, resulting in discrepancies between measured and true process levels of analyte. Centrifugation is an alternative means of sample preparation where solid-liquid separation is required which, although a mechanically complex process compared with filtration, can cope more effectively with high solids loadings and heavily blinding suspensions that might foul a microfilter. In this paper we describe the operation and separation characteristics of a recently developed high-speed, small-volume centrifugal separator, suitable for the rapid at-line sampling of biological processes. The application of the device for facilitating the measurement of protein solubility during fractional precipitation of
of Biotechnolog_v
49 (1996)
I I I-l
18
proteins from yeast homogenate is described. Here it is the recovery of a representative sample of the remaining soluble protein for rapid measurement which is of importance and issues such as precipitate carry-over, precipitate redissolution or soluble protein retention must be addressed.
2. Materials
and methods
2.1. Cen trifugul separutor A schematic view of the centrifugal separator is shown in Fig. 1. The prototype device was designed and constructed by Fisons Scientific Instruments (Uxbridge, Herts, UK). The centrifuge consists of a tulip shaped, 20-mm diameter bowl rotated at speeds up to 60000 rev./min by an air-driven turbine conventionally used for cutting tools. The speed of the bowl is controlled by the inlet air pressure to the turbine; a speed of _ 60000 rev./min being generated by 6 bar gauge. The operation and timing of the various items making up the centrifuge were sequenced by a bench top microcomputer via a digital output port. The inset in Fig. 1. shows the definition of r0 and ri the outer wall radius and sample surface when spinning respectively. Y, is 10 mm and Y, ranges from 9.1 mm to 7.6 mm for sample volumes of 0.3 ml to 0.9 ml. As the sample load to the centrifuge increases, r, is reduced and the ratio rJr, increases. At sample initiation, the feed pump is switched on for a pre-determined length of time. At the same time the vacuum solenoid valve is opened so that any sample entering the bowl is immediately removed to waste. This has the effect of clearing the sample line of the preceding sample and ensuring that the sample delivered to the centrifuge is representative of that at the point of sampling. Once the sample line has been cleared, the vacuum solenoid valve is closed and the feed pump continues to deliver sample to the bowl for a fixed length of time (fixed sample volume). The air pressure valve to the turbine is then opened, and the bowl rapidly gains maximum speed. The sample is spun for the required length of time causing the solids to sediment in the extremes of the bowl
P. Richardson
et al. / Journal of Biotechnology
after which the air flow to the turbine is stopped and the friction brake solenoid is activated. As the bowl stops rotating, the clarified supernatant flows to the base of the bowl while the solids pellet remains on the bowl wall. The sample-out pump is then activated for a predetermined time
Table 1 Sequence tion
of events and timings
Sequence
time (s)
0 5
sample in I clarified sample out
” vacuum
-Protective
shroud
sample volume - 1 ml
6.5 6.5 + T 26.5 + T 41.5 + T 51.5 + T
turbine
113
49 (1996) 111- 118
for the microcentrifuge
opera-
Action Sample in pump and vacuum solenoid on Vacuum solenoid off, sample is delivered to bowl Sample in pump off and turbine drive valve open Turbine drive valve closed and friction brake on Sample out pump on to deliver sample to analyser Wash cycle stop
pressurised air
T, centrifugation
time.
T
pump control lines
Fig. 1. A schematic of the microcentrifuge sampling system. Pumps and valves are operated under microcomputer control. The sample is pumped in and out and wash water directed at the sides of the bowl by means of peristaltic pumps. The bowl is cleaned out using a vacuum suction line and rotates at speeds up to 60000 rev./min driven by a turbine powered by air flow with input pressures regulated up to 6 bar. A friction brake is included to slow the bowl down after the sample centrifugation time has elapsed. The bowl is housed in a moulded rigid plastic protective cover and the entire unit ruggedised to enable operation in a pilot plant environment. The bowl dimensions are approximately 2 cm diameter (wall to wall) and a typical sample volume would be 1 ml. An entire cycle for sample preparation takes approximately 30 s plus the centrifugation time. The inset shows the position of the sample in the bowl as it is spinning and the definition of Y,, the radius of the wall of the centrifuge and Y, the radius of the sample surface.
to deliver a set volume of clarified material to the sample loop of the associated analyser, a flow injection analyser in this instance. To clean the centrifuge, wash water is pumped into the bowl and against the wall as a high pressure jet through two small bore delivery tubes. Simultaneously vacuum is applied to remove the waste washings. The turbine is also given a number of rapid blasts of air to turn the bowl ensuring solids are removed from all parts. The centrifuge is then ready for the next sample. The complete cycle takes approximately 50 s plus the described clarification time, the sample being delivered to the analyser after approximately 30 s plus the clarification time. Typical timings for these various sequences are given in Table 1. 2.2. Yeast homogenate
preparation
Blocks of bakers yeast Saccharomyces cerevisiae (Distillers Co. Ltd., Crawley, West Sussex, UK) were suspended in phosphate buffer (0.067 M KH,PO,, 0.033 M K,HPO,, pH adjusted to 6.5 with 2.5 M NaOH) to a final concentration of 750 kg rn-’ packed weight. The yeast was mechanically disrupted using an APV Gaulin high pressure homogeniser (model 15M, APV Crawley, W. Sussex, UK) operating continuously at 500 bar and 5°C for the equivalent of four passes of the suspension volume. Cell debris was removed by
114
P. Richrrrdson et al.
/ Journal of Biotrclznolo~y
centrifuging the homogenate at 53000 average RCF for 1 h (Pegasus Ultracentrifuge, MSE Crawley, UK using 6 x 032 head at 21000 rev./min). 2.3. Precipitant
preparation
Ammonium sulphate precipitation agent (Analar, BDH, Poole, UK) was made up to 3.9 M solution in phosphate buffer (0.067 M KH,PO,, 0.037 M K,PO,, pH adjusted to 6.5 with 2.5 M NaOH). At a temperature of 20°C this corresaturated solution. sponded to a 98% Polyethylene glycol (PEG) precipitating agent (Biochemical grade, BDH, Poole, UK) was made up to 250 kg rnp3 in phosphate buffer, pH 6.5. 2.4. Precipitate
suspension
49 (1996) 111~ 118
3. Results and discussion Figs. 2 and 3 illustrate examples of the settling curves for precipitated yeast homogenate using the centrifugal sampler. The fraction of precipitate remaining in suspension, F, was estimated from measurements of optical density at 650 nm: OD ~ OD,,,,
F=
OD,, -
(1)
OD,,,
OD,, was determined from supernatant samples following high speed batch centrifugation at 10000 RCF for 0.5 h. OD is the optical density after clarification and OD, the optical density prior to clarification.
preparation
The precipitate suspension was prepared in a continuous precipitation device which has been described in detail elsewhere (Richardson et al., 1989). In summary, it consisted of a pair of variable speed gear pumps feeding yeast homogenate and precipitant into a T-piece, followed by a needle valve to achieve intimate mixing of the two streams. Following the needle valve, another T-piece formed the sampling point. The main process flow continued to waste while the sampling centrifuge intermittently pulled a small side flow off from the branch.
0.0
-
-0.5 -
3 G-1.0
-
z
-1.5 -
2.5. Assays The clarified samples delivered from the centrifuge were analysed manually for ADH activities and total protein concentration. The use of the centrifuge in conjunction with an on-line flow injection analyser is described elsewhere (Richardson et al., 1996). ADH activity was measured by following the absorption change at 340 nm of NAD + to NADH accompanying the conversion of ethanol to acetaldehyde (Bergmeyer et al., 1983). Total protein concentration was determined by the dual UV wavelength method of Groves et al. (1968) as modified by Ehresmann et al. (1973) using bovine serum albumin as standard.
Pelletingtime (s) Fig. 2. Settling curves for two different microcentrifuge rotor speeds (0 30 000 rev./min), (m 60000 rev.,‘min) at a precipitant saturation of 70%. Linear fits to each data set are also shown (. .). At the higher speed, clarification to the same fraction
F =
of precipitate
remaining
in sus-
pension occurs apL&oxima;eiy three times faster, in line with theoretical calculations (t, = 97.9 k 29.2 s: t, = 260.27 i 26.2 s). Times were measured for clarification levels to reach 99% as a residual suspension turbidity was always observed with ammonium sulphate used as the precipitant.
P. Richardson et al. 1 Journal of Biotechnology 49 (1996) 1 I1 - I18
spin time (s) Fig. 3. The effect of precipitant and sample size on yeast protein precipitate settling characteristics. Key points are the rapid separation of the precipitates formed with 20% w/v PEG6000 (0.5 ml sample volume) (- V -) and 80% w/v ethanol (0.5 ml sample volume) (- 0 -) compared with those formed with ammonium sulphate and the increase in the separation times of these precipitates with correspondingly increased sample volumes; (- 0 -) 0.3 ml, (- 0 -) 0.5 ml, (- A -) 0.9 ml 80% saturation ammonium sulphate as precipitant. The centrifuge speed for ail these samples was 30000 rev.imin.
From centrifugation theory, the time required for a particle of diameter d to move from the outer radius of the sample (v,) to the radius of the bowl (Y,) is given by:
where Ap is the particle-fluid density difference, w is the angular velocity and ,LLis the suspending fluid viscosity. Theoretically the time required to pellet the precipitated protein completely is equivalent to the time for the smallest protein aggregate present to move from r. to r, assuming the precipitated protein has a small volume fraction. The true pelleting time may be obtained from settling plots. In the case of ammonium sulphate precipi-
115
tation, however, a small residual turbidity was always observed which was greater than OD,, making the true pelleting time difficult to estimate. This residual turbidity was shown to be due to suspended precipitated protein, i.e. it easily redissolved on lowering the ammonium sulphate concentration. To characterise the centrifuge performance, the nominal pelleting time was assigned as the time required to clarify 99% of the precipitate. This value was measured by plotting the ordinate as a logarithmic scale to emphasise the smaller optical density measurements. Fig. 2 illustrates the effect of rotor speed on clarification of a 0.5-ml sample of yeast homogenate precipitated at 70% saturated ammonium sulphate. As the rotor speed is increased from 33 000 to 60000 rev./min, Eq. (1) predicts that the pelleting time should be reduced by a factor of 3.3 ( = 602/332). From Fig. 2, it is estimated that the settling time reduced by a factor of 2.7 + 0.84 confirming that attainment of high velocity is a prerequisite for achieving rapid sampling frequency. Fig. 3 presents data showing the effect of sample size on clarification speed for clarified yeast homogenate insolubilised with the precipitants: saturated ammonium sulphate, 25% w/v PEG and absolute ethanol. In each case, the clarified homogenate was taken to 80% by volume of precipitant solution. Previous work (Richardson, 1987) has indicated that at this concentration the fraction of protein precipitated was the same for each precipitant. The figure clearly indicates the ease with which the PEG or ethanol precipitated material was clarified compared with that of the ammonium sulphate, the former taking place within 10 s for a 0.5-ml sample. This illustrates one of the difficulties often encountered in processing precipitated protein at high concentrations of ammonium sulphate. At these concentrations, the density difference between precipitated protein and suspending phase becomes very small ( < 100 kg mp3) and thus slow sedimentation velocities result. In this instance however the effect is in part beneficial since it allows the settling curve to be constructed. For PEG and ethanol, clarification was so rapid that the settling curves could not be accurately followed, which is a desirable situation for high frequency sampling.
P. Richurdson et al.
116
/ Journal of Biotechnology 49 (1996) I I I-I 18
As the sample size added to the rotor increases, the settling time is elongated since this has the effect of increasing the effective settling distance, and therefore In (r,/r,). From an accurate engineering drawing of the rotor, the value of In (r,/r,) was calibrated against the sample volume. The pelleting time for the three sample sizes was estimated as before. In Fig. 4 pelleting time is plotted against In (YJY,) according to Eq. (1). Although the data are limited there is good correspondence with Eq. (1) in that a linearity that passes through the origin exists and a statistically good fit is achieved. From Figs. 2 and 4, it is apparent that changes in pelleting time in the centrifugal sampler due to variations in spin speed or sample volume can be predicted satisfactorily. This is important since choosing a maximum angular velocity and small sample volume to minimise set-
5-
0 60
20
40
Ammonium sulphate concenttatia,
60
80
(% satuation)
Fig. 5. Protein solubility profile measured using sampling microcentrifuge versus equilibrium solubility curve for ammonium sulphate precipitated protein. The discrepancy, which remains approximately constant across the range of precipitant concentrations, is due to the unsedimented or resuspended protein precipitated.
20
10
: .’ -
,. :
:
ol
0.00
*
’
0.E
’
’
0.10
’
’
0.15
*
’
020
m
’
025
’ 0.30
In(r&)
Fig 4. The time required to pellet ammonium sulphate precipitated protein as a function of ln(r,/r,) where r, is the inner sample radius and rO the rotor bowl radius. The rotor speed was 30000 rev./min. A good linear least squares fit is apparent with a x2 value of 3.64 which is entirely compatible with a linear fit given the error estimates on the data points. This allows prediction of the spin-time for a given sample volume in order to optimise the sampling time for the analysis cycle.
tling time may not always be consistent with the intended use of the sampler. For example an on-line analyser sample loop downstream of the centrifuge may require a minimum volume to fill and purge it. Similarly an overly high spin speed could lead to over-dewatering of the sedimented material, making water jet cleaning of the bowl difficult and leading to a cumulative build up of solids. Sample size and bowl velocity are therefore best optimised for each individual application. Figs. 5 and 6 show the solubility of total protein precipitated with a range of concentrations of ammonium sulphate and PEG respectively as measured by clarification for 60 s using the sampling centrifuge, or by using batch centrifugation at 10000 RCF for 0.5 h i.e. the ‘true’ equilibrium solubility. As suggested by the preceding settling curves, the sampling centrifuge totally clarifies the PEG precipitated material and this is indicated by the fact that the solubility
P. Richardson
et al. / Journal of Biotechnology
curve generated from the sampling centrifuge corresponds to the true solubility curve. For the ammonium sulphate precipitated material however, the solubility measured following sampling centrifugation is slightly higher than the equilibrium solubility over the range of ammonium sulphate concentrations. This results in the slight residual turbidity for the higher spin time (Fig. 2). The concentration offset between measured and equilibrium solubility is effectively constant over the entire profile, thus this difference is independent of the equilibrium solubility. The reason for the incomplete clarification of the ammonium precipitated protein is unclear. The offset is not an inherent characteristic of the machine since complete sedimentation was observed with PEG and ethanol produced precipitates (Fig. 3). Neither is the offset due to insufficient settling time being allowed as the residual level is rapidly reached (well within 3 min) often with no further clarification achieved.
49 (1996) I II- 118
117
A potential cause for this behaviour could be in the particular properties of ammonium sulphate precipitated protein. It has been shown (Chan et al., 1986) that the protein aggregates produced by ammonium sulphate precipitation are small compared with other precipitants extending well into the submicron range. Whilst the preponderance of small aggregates will undoubtedly be sedimented given sufficient time, it is suggested that they form a loosely parted layer on top of the truly sedimented pellet. As the rotor is slowed down following a run the shear induced in this surface layer is sufficient to disrupt the loosely captured smaller aggregates thereby resuspending them in the bulk suspension. It is likely that these fine particles will not be recovered in a continuous centrifuge typically used for the industrial scale recovery of the precipitate and hence the sampling centrifuge probably gives a better measure than the laboratory batch centrifuge of recoverable precipitate.
4. Conclusion
0
I,
I
2
4
I
I
6
I
I
6
PEG concentration
I1
II
.I.
10
12
14
(% VA)
Fig. 6. Protein solubility profiles measured using the sampling microcentrifuge and the equilibrium solubility for PEG precipitated protein. The equilibrium and microcentrifuge data are identical indicating the relative ease with which PEG precipitates are separated in comparison to ammonium sulphate precipitates. This plot indicates that the offset shown in Fig. 5 is not a feature of the microcentrifuge but peculiar to ammonium sulphate precipitates.
The centrifugal sampling device represents a novel and potentially very useful solution to the problem of rapid on-line sampling of bioprocesses. Unlike filtration devices, it is not prone to variability with time caused by proteinaceous filter fouling, the separation being achieved by robust mechanical means. The centrifugal sedimentation curves generated may be theoretically well characterised in terms of angular velocity and settling distance (sample size). Practical clarification results obtained from precipitated yeast homogenate show agreement with the predicted theoretical trends. Spun homogenate precipitated with PEG or ethanol is particularly well clarified while spun homogenate precipitated with ammonium sulphate is satisfactorily but not completely clarified. This latter effect is ostensibly due to resuspension of the smaller aggregates produced by ammonium sulphate which are resuspended when the rotor decelerates after a run. In all cases a sample suitable for at-line analysis can be produced within 30-60 s.
Acknowledgements UCL is the Biotechnology and Biological Sciences Research Council sponsored Advanced Centre for Biochemical Engineering and the Council’s support is gratefully acknowledged.
References Bergmeyer, H.U., Grabl. M. and Walter, H.L. (1983) Reagents for enzymatic analysis. In: Bergmeyer, H.U. (Ed.), Methods of Enzymatic Analysis. Vol. 2. 3rd edn. Verlag Chemie, Weinheim, Ch. 2. pp. 1399140. Braun-Melsungen, A.G. (1986) Das Probeentnahmegerat fur die chemische und insbesondere die biochemische Proresstechnik.... Patentschrift DE 3520489 Cl. Chart, M.Y.Y., Hoare. M. and Dunnill, P. (1986) The kinetics of protein precipitation by different reagents. Biotechnol. Bioeng. 28. 387-393. Ehresmann, P.. Imbaull, P. and Weil. J.H. (1973) Spectroscopic determination of protein concentration in cell ex-
tracts containing tRNA’s and rRNA’s. Anal. Biochem. 54. 454 463. Groves. W.E.. Davies Jr., F.C. and Sells, B.H. (1968) Spectroscopic determination of microgram quantities of protein without nucleic acid interference. Anal. Biochem. 22. 195 210. Kroner. K.H. and Kula. M.R. (1984) On-line measurement of extracellular enzymes during fermentation by using membrane techniques. Anal. Chim. Acta 163. 3 15. Recktenwald. A., Kroner, K.H. and Kula. M.R. (1985) Online monitoring of enzymes in downstream processing by flow injection analysis. Enzyme Microb. Technol. 7. 607 612. Richardson. P. (1987) A Biochemical Engineering Study of Fractional Protein Precipitation. Ph.D. Thesis. University of London. Richardson, P., Hoarc, M. and Dunnill, P. (1989) Optimisation of fractional precipitation for protein purification. Chem. Eng. Res. Des. 67, 273-227. Richardson. P.. Ravenhall. R., Flanagan. M.T.. Holwill, 1.. Molloy. J.. Hoarc, M. and Dunnill., P. (1996) Monitoring and optimisation of fractional protein precipitation by flow injection analysis. Proc. Cont. Qual., submitted.