Biotechnol. Prog. 1991, 7, 28-32

28

Rapid Immunoassay Technique for Process Monitoring of Animal Cell Fermentations E. Jervis and D. G. Kilburn* Biotechnology Laboratory, University of British Columbia, Vancouver, British Columbia, Canada V6T 1W5

A calibration and quality control technique suited t o process monitoring with immunoassay is demonstrated. T h e particle concentration fluorescence immunoassay (PCFIA) is shown t o provide a sensitive and rapid method for the quantification of specific biomolecules in cell cultures. Smoothing of linear calibration parameters is performed by forming weighted averages of standard points as the run progresses. These estimates are then used t o determine slope and intercept values for improved calibration. T h e nonuniformity of the fluorescent signal variance is also considered, and a weight model is developed t o describe the relationship between signal fluorescence and signal variance for weighted linear curve fitting. Pooling calibration results over the process run improves overall assay performance as determined by using standard control chart analysis. This method is suitable for semicontinuous monitoring of animal cell fermentations and has been used here to measure cell-associated and culture supernatant concentrations of monoclonal antibody (Ab) from hybridoma cells. T h e cell-associated Ab concentration correlates with cell-specific production rate. Assay times on the order of 10 min for supernatant and 25-30 min for cell-associated Ab concentrations can be achieved, making this procedure suitable for process monitoring and control. Under these conditions the assay has a detection limit of approximately 10ng/mL, providing a sensitive and specific method for t h e quantification of cell culture constituents.

Introduction Mammalian cell culture is widely used for the production of proteins important as pharmaceuticals or diagnostic agents. The production process is an expensive operation, in part because the desired products represent only a small fraction of the overall spent culture medium. This low product concentration also greatly increases product purification costs downstream of the fermenter. There is therefore a great need to develop strategies that permit dynamic optimization of cell culture conditions to maximize product concentrations in the recovered culture media. One class of protein products of immediate significance is monoclonal antibodies (MAbs). MAbs are antibodies produced from specially constructed cells (hybridoma cells) that produce a single antibody of known specificity (Kohler and Milstein, 1975). Analysis of hybridoma cultures a t the cellular level has shown that cell-specific productivities and Ab secretion kinetics vary considerably during a batch growth cycle (Meilhoc et al., 1989; Altshuler et al., 1986). Other work has demonstrated the influence of cell regulatory mediators on cell productivity and growth (Dalili and Ollis, 1988). These investigations suggest that some level of dynamic optimization of fermenter productivity is possible by direct manipulation of the cells’physiological “state”. The control and optimization of fermentation processes is typically performed by monitoring secondary parameters (eg. temperature, pH, dissolved oxygen) and relating these measurements to primary parameters of interest. This is necessary because primary variables, such as the concentration of cell protein products, are often difficult to measure quickly and accurately, particularly in the minute

concentrations typical of most bioprocesses. Underlying this approach is the assumption that the secondary response variable can be related directly to the primary response variable with known or negligible space/time dependencies. In most systems, however, there are complex interactions between the various parameters used to describe the system and simple relationships do not hold. Thus, off-line analysis of primary parameters by standard analytical procedures must be used to determine the concentrations of interest. Unfortunately, standard analytical techniques often introduce significant “dead time” in the control loop. This measurement lag makes optimization difficult or impossible unless appropriate models are available to predict the time course of the desired parameter between process measurements. Most methods presently used for quantifying protein concentration [e.g., enzyme-linked immunosorbent assay (ELISA)] have relatively long analysis times, making these techniques inappropriate for investigating dynamic systems when feedback interaction or control is desired. A rapid immunofluorescent technique, “particle concentration fluorescence immunoassay” (PCFIA),was introduced in 1984 (Jolleyet al., 1984). This method employs capture Ab bound to submicrometer polystyrene spheres. This “activated” solid phase acts as a specific adsorbent for the analyte. A fluorescently labeled second antibody, also specific for the analyte, is then incubated with the solid capture phase to form a complex whose fluorescent signal is proportional to the analyte concentration (Figure 1). The object of this work was to test the feasibility of using this technique for on-line monitoring of mammalian cell cultures. A rapid “off-line” method using the PCFIA system is presented in which repeated assays are used to monitor supernate and cell-associated antibody concen-

8756-7938/91/3007-0028$02.50/0 0 1991 American Chemical Society and American Institute of Chemical Engineers

Biotechnol. Prog., 1991, Vol. 7, No. 1

29 U

1. Ab ACTIVATE S O L D PHASE

2. INCUBATE W E

3. INCUBATE WITH

ANALYTE

LABELED Ab

4. VACUUM AND WASH

5. VACUUM TO CONCENTRATE PARTICLES

6. READ FLUORESCENCE

Figure 1. FCA protocol: (1)“Capture”Ab is adsorbed onto the polystyrene solid phase, (2) the analyte is incubated with the solid-phase adsorbant, (3) a second, fluorescently labeled Ab is incubated with the sample, forming an Ab “sandwich”with the analyte in the middle, (4) unbound second Ab (i.e., the labeled Ab was added in excess) is separated from the solid phase by vacuum filtration in the special Pandex 96-well plate, (5) the solid phase is concentrated at the bottom of the well by vacuum, and (6) the fluorescent signal, proportional to the amount of adsorbed analyte, is determined. trations using a hybridoma culture. Samples were collected from cell cultures and assayed for protein product with emphasis on maintaining calibration accuracy and rapid sample processing. Calibration and assay quality control are considered, and results demonstrate that this system is well suited to on-line process monitoring applications.

Materials and Methods ( 1 ) Cell Culture. The murine hybridoma cell line 2 E l l (Biomedical Research Center, UBC) produces murine immunoglobulin (IgG1) directed against murine IL-3 receptors. These cells were cultured in RPMI medium (Gibco Laboratories, Grand Island, NY) with 10% fetal calf serum (FCS) (Gibco Laboratories, Grand Island, NY) in 800-mL Nunclon T-flasks (A./S. Nunc, Roskilde, Denmark). Cells were incubated a t 37 OC in 5% COz. Cell concentrations were determined by counting cells on a hemocytometer. Cell viability was determined by trypan blue exclusion (Kaltenbach et al., 1958). Glucose concentration was determined by using an enzymatic glucose electrode (Glucose Analyzer 2, Beckman Instruments, Fullerton, CA). Cells in the late exponential phases of growth [Le., (7-9) X 105 cells/mL] were used to seed individual runs. (2) IgG Assay. Analyses are performed in wells on specially designed 96-well microplates (Baxter/Pandex Healthcare Corp., Mundelein, IL). Pandex does not recommend reuse of their 96-well plate. However, since only a few wells were used for each process measurement, repeated use of the plate allowed a significant reduction in the cost of the assay system. Changes in the 96-well plate during the course of process monitoring introduced a significant drift in calibration. To compensate for this drift a linear smoothing algorithm was developed and applied to all measurements. To ensure acceptable assay performance, standards and control samples were included with each analysis. The Ig concentration in the culture supernatant and cell extract samples was determined by using the Pandex

fluorescent concentration analyzer (FCA) (protocol shown in Figure 1). A 20-pL aliquot of sample, standard (IgG1, Southern Biotechnology ASSOC.,Birmingham, AL), or control at appropriate dilution was added to wells in the FCA 96-well plate. Ten wells were used per process measurement (four standards, two controls, and two each of supernatant and intracellular samples), permitting a single plate to be used for 9 process measurements. Standards a t 0.5,0.25,0.1, and 0.01 pg/mL were used for calibration curve preparation. Controls a t 0.3 and 0.07 pg/mL were used for quality control analysis. A 20-pL aliquot of goat anti-mouse Ig coated polystyrene spheres (0.7 pm, 0.25 5% v/v, Baxter/Pandex Healthcare Corp., Mundelein, IL) was added to each well containing a sample. The contents of the wells were gently mixed by tapping the side of the 96-well plate, and the mixtures were incubated for 10 min a t room temperature (21 “C). Following the first incubation, 20 pL of secondary Ab, goat anti-mouse Ig-fluorescein isothiocyanate (FITC) conjugate (BRL Life Technologies Inc., Bethesda, MD) was added to each sample and incubated for 5 min, at room temperature, in the dark, after gentle mixing. The plate was then evacuated by using the Pandex FCA. Wells were washed three times with phosphate-buffered saline (PBS) containing 0.1 % sodium azide (Sigma, St. Louis, MO) and read with the 485/535 filter pair a t 25X gain. The addition of azide to the wash buffer prevents bacterial growth in unused wells during storage of the plate (at 4 “C) between process sampling runs. (3) Sample Preparation. Cells were counted and sufficient volume was collected to yield 2.5 X lo6 cells after washing. The sample was centrifuged (5 minat 200g) and the cell-free supernatant collected and set aside for subsequent assay. The cell pellet was washed with 4 mL of RPMI containing 2% FCS and 0.1% azide. The cell wash supernate was discarded and the cells were resuspended at 1 X l o 7 cells/mL. A 100-pL sample of the concentrated cell suspension was then added to 200 pL of ice-cold lysis buffer [ 150 mM sodium chloride (Sigma, St. Louis, MO), 50 mM tris(hydroxymethy1)methylamine (BDH Chemicals, Poole, England), and 1%Triton X-100 (BCH Chemicals, Poole, England)] and mixed gently for 5 min. The cell extract was diluted 20X in fresh Dulbecco’s modified Eagle’s medium (DMEM) containing 5 70 FCS and 0.1 % azide. The sample was assayed as outlined above.

Results ( 1 ) Assay Incubation Time. The linear range and sample throughput of the assay system were investigated. Experiments were designed to determine the minimum incubation times necessary for satisfactory assay performance. Calibration curves were produced for primary and secondary incubation times ranging from 0 to 10 min. Generally it was found that the assay sensitivity increased with increasing incubation time. However, the minimum required incubation time varies depending upon analyte and antibody used. Figure 2 shows the effect of incubation time on calibration linearity and assay sensitivity (i.e., slope) for analysis of human tranferrin. Analysis of the results for the transferrin assay example show that acceptable assay performance can be obtained with a total incubation time of 6 min. A similar analysis for murine IgG demonstrated acceptable performance with the 15min incubation schedule used here. (2) Rapid Cell Lysis. Dilute detergent solutions are frequently used for the lysis of animal cells to release intracellular protein (Harlow, 1988). Nonionic detergents,

Biotechnol. Prog., 1991, Vol. 7, No. 1

30 6E6

LT!

7

I

5E6--

30.0 n

3 LL

a

W

25.0

*

4E6--

-*

W

W

03E6--

0 W

20.0

-

W

2E6--

0

w 0

3

--1 1 E 6 - -

LL

15-0

Z - . 0.0

:M / 3.5 4.0 4.5 5.0 5 . 5 6.0

3

,

LOG(FLU0RESCENCE)

W 0.1

TRANSFERRIN

0

c (ug/ml)

0:2

v,

10.0

W

CY

3

5.0

1 L

0.0

0.0

0.1

0.2 0 . 3

[Ab1

Figure 2. Determination of an appropriate incubation time schedule for a human transferrin immunoassay. The y-axis is presented with the power transform determined to be optimal for the transferrin assay. The average % CV is the ratio of the standard deviation of any point (from triplicates) and its associated mean fluorescent signal. such as Triton X-100, do not denature proteins or disrupt antigen-antibody interactions even at fairly high concentrations (Helenius and Simmons, 1975). Because rapid sample preparation was required for this procedure, experiments were conducted to investigate the effects of lysis times upon sample recovery. Cell lysis incubation times ranging from 2 to 40 min were used as outlined under Materials and Methods. No significant time effect was found (ANOVA analysis, 9 5 5 confidence limits) over the range investigated. Microscopic examination of samples showed that all cells were lysed (asjudged by their inability to exclude trypan blue stain). Five-minute lysis times were therefore adopted to permit rapid sample throughput. (3) Calibration. The linear range of the assay using the 10- and 5-min incubation times was investigated. A linear calibration was used to minimize the standard curve sample requirements (i.e., minimizing fitted parameters). Linearity was best in the range from 0.001 to 0.07 pg/mL. Figure 3 shows fluorescence as a function of antibody concentration for the assay using 10- and 5-min incubation times. Tests to demonstrate parallelism between the standard and unknown sample titration curves showed that good parallelism existed only in the range from 0.01 to 0.5 pg/mL. This range was selected for calibration. The use of a power-transformed calibration equation of the form (fluorescence)“ = m(concentration) + b where m and b are the slope and intercept, respectively, corrected for slight nonlinearity in the standard curve. The power transformation value n was determined by using likelihood function analysis (Box et al., 1978) from data pooled from several standard curves. A value of n = 1.1 was found to yield optimum linearity for the murine IgG analysis, with typical R2 values of 0.999. The nonuniformity of the fluorescence variance further

0.4

0.5

bs/ml>

Figure 3. Typical standard curve fcir the mouse IgG assay. The y-axis is presented with the power transform value determined to be optimal for the mouse IgG assay. The inset shows the data weighting function developed by pooling results from several individual assays. Weighting values are normalized such that n Z W , = n. i=l

complicates calibration. The use of a power transformation amplifies this effect. Failure to consider nonuniform variance in regression calculations has been shown to invalidate the calculation of confidence regions for predicted concentrations (Schwartz, 1979). To faciltiate estimation of sample concentration variance, a model was developed for the regression weighting function by using results pooled from standard curves obtained over several weeks. The inset in Figure 3 shows the weighting function describing the relationship between measurement variance (normalized) and fluorescent signal over the calibration range used. The weighting function is described by weight = 11708 exp[-2.2 log (fluorescence)] This equation was used in regression calculations as well as in the determination of appropriate weighting values for the estimation of calculated sample concentration confidence intervals (Rodbard et al., 1987). (4) Calibration Changes withTime. Calibrationdrift is a common problem with on-line assay systems. Figure 4 shows the variations observed in standard curve points over a 5-day culture period. Substitution of a fresh plate at run 10 (118 h) caused the sharp change in standard curve points noted in Figure 4. It has been suggested that assay calibrations may be improved by using recursive estimation algorithms. Application of linear filtering techniques to drifting calibration systems has been discussed in several recent papers (Vecchia et al., 1989;Brown, 1986; Thijssen et al., 1984). These algorithms combine present calibration parameter estimates with appropriately weighted past estimates to produce “optimal” values of regression parameters. Such recursive techniques are generally modified Kalman filters. To minimize calibration error for this system, a linear averaging algorithm was used for standard curve point estimation. Smoothing was accomplished by forming a weighted average of the current and past two standard point fluorescence measurements. Each corrected cali-

Biotechnol. Prog., 1991, Vol. 7, No. 1

31

7.0 n

LL

6

+

W

n0 0.35 '40

6.0

2 0.30

5.0

3

*

W

7

0.04

n

4.0

n Q

w

z

rY

0.25

W

0

T

=t

3

0.02

0.20

U

3.0 -I

Id

0.00

0.15

w

I

.0.02=I

0

m

E 0.10Z 0.05 0

2.0

+

w 0

3 1.0 _I

G

0.00

0.0

25

50

75

CULTURE

100

125

CULTURE TIME

TIME (hr)

represent measured values , and solid symbols show calculated 0.25; A,0.1; and 0 , O . O l pg/mL). smoothed estimates (.,0.5; ., The discontinuity at point 10 was a result of the substitution of a fresh plate.

L 25 50 75 100 125 150

150

Figure 4. Calibration time course results: open symbols

0

(hr)

Figure 5. Control chart generated from the demonstration run for the hybridoma culture time course (0, 0.30, and +, 0.075 pg/mL). No overall bias in control concentration estimation is evident in the CUSUM plot.

c 200 U

bration point is calculated by smoothed estimate = 0.54(current) + 0.32(last) + 0.14(2nd last) The smoothing action of this equation is demonstrated in Figure 4. This direct approach has the advantage of being simpler to formulate than the Kalman filter for the limited data set available for each run (i.e. 9 calibrations per plate). The weighting parameters used in the linear smoothing equation were selected on the basis of two criteria: (1) minimization of the sum of the regression error estimates and (2) minimization of the control sample concentration biasing as indicated by the slopes of the cumulative error sum (CUSUM) plots (see Figure 5 ) (Box and Drapier, 1987). These control plots are obtained by summing the errors in estimating the known control concentrations as the run progresses. A steadily increasing or decreasing CUSUM plot indicates biasing in the predicted concentration estimates. (5) Antibody Production. The Ab assay results for the 2 E l l hybridoma culture are shown in Figures 6 and 7. Each process measurement was obtained within 30 min of sampling. Assay quality control was maintained by including positive controls at two concentrations for each time point run. Figure 5 shows the time course for the control samples. For the example run presented, controls were consistently within 5 % of the known control concentrations. Figure 6 shows the batch growth curve patterns for the 2E11 hybridoma. The production kinetics presented are representative of most cultures. The cell-specific rate of antibody production (i.e,, /.tg of Ab produced cell-l h-l) shown in Figure 7 is typical of 80-9092 of all hybridoma lines (Merten, 1989). The implications of this production pattern are presently unclear, but it does suggest that continuous cultures based on high dilution (growth) rates would have relatively lower cell specific productivities. The correlation shown in Figure 7 between cell specific productivity and intracellular Ab concentration is quite striking. These results indicate that cell specific produc-

\

E

175

n

-\*

h

I

m

0

a >

25

50

75

I/

I

i u 100 125 150

CULTURE TIME

(hr)

Figure 6. Hybridoma batch cell culture time course results: cell count; glucose; 0 , mouse IgG.

+,

W,

tivity peaks correspond very closely to peaks in intracellular Ab concentration. Experiments are currently being performed to further examine this relationship.

Discussion The successful engineering of bioreactor systems requires the application of improved and rapid sensing techniques for the monitoring of specific biomolecules. The current work demonstrates the feasibility of adapting an existing immunoassay technology to process monitoring. Rapid sample throughput has provided estimates of hybridoma cell culture Ab levels in the cell-associated and extracellular product pools. Greatly decreased measurement dead times should facilitate the development of

Biotechnol. Prog., 1991, Vol. 7, No. 1

32 5.5

3.5

n

A

L

1 1

1

T

W

0 5.0

3.0

2.5

7

\ 4.5

7

m

a

2.0

W

4.0

0

1.5 F 0

Q

3

n

3.5

F

1.0

.Q 0

2

a a

0 3.0 cn m

0.5

9

.cl

L.il

-2 Z

Q

-A

i

0

W

w

i

w

r.

n

\

L 0 0.0 w

2.5

Acknowledgment This work was supported by the Province of British Columbia through the STDF-AGAR program of the Science Council of British Columbia. Literature Cited

~

25

50

75

1 0 0 125

150

,~

U

Immunoassays for continuous process monitoring will require calibration and quality control to be addressed sequentially as the run progresses. We have shown that the FCA system provides acceptable performance when assays are processed in sequenced batches. This method should offer advantages when coupled to flow techniques because process measurements are performed in “sets”, in which calibration and quality control samples may be included as required. This should offer benefits over schemes in which calibration, quality control, and measurement occur individually, as would be the case in single sample processing systems. Work is in progress to automate sample handling for on-line application of the system.

C U L T U R E TIME

Figure 7. Hybridoma batch cell culture tiome course results: 0 ,cell-associated Ab; .-,mean cell-specific Ab productivity (rate of appearance of Ab in the supernatant on a per cell basis).

closed-loop control schemes for the manipulation of cell culture variables (e.g., glucose or serum concentrations) to optimize culture productivities. Furthermore, by directly measuring the desired product concentrations, assumptions concerning the relationships between secondary variables, such as glucose consumption rate, and product production rates are not required. The additional measurement of other cell-associated proteins may provide better indicators of overall cell condition. Calibration quality was enhanced by pooling results obtained over several analyses. This approach has two main advantages. First, the calibration models developed become relatively insensitive to “outliers” in any single analysis. This permits reliable calibration with minimum replication of calibration samples. Second, assay quality control becomes an integral part of the calibration process. For example, if results are pooled to develop transform models, trends or drifts in calibration become quite evident when the model is refitted. Analysis of the residuals from fitting the model will clearly show any time-dependent shifts in calibration. The pattern of cell growth and antibody production observed in this work is typical of that seen for most hybridomas: antibody accumulating rapidly in the stationary and declining phases of cell growth (Merten, 1989). Presumably this is a reflection of the physiological state of the cells during these phases of the culture. Measurements of cell-associated Ab may thus provide a convenient method for diagnosis of cell physiological state. With the rapid assay method presented here, an estimate of the cell culture’s conditions could be available within 30 min of sampling. This possibility offers a unique opportunity for feedback control to optimize culture conditions for maximum cell productivity.

Altshuler, G.; et al. Hybridoma Analysis a t the Cellular Level. Biotech. Bioeng. Symp. 1986, 17, 725-736. Box, G.; Drapier, N. Response Surface Methodology, John Wiley & Sons: New York, 1987. Box, G.; Hunter, S.; Hunter, W. Statisticsfor Experiments;John Wiley & Sons: New York, 1978. Brown, S. The Kalman Filter in Analytical Chemistry. Anal. Chim. Acta 1986, 181, 1-35. Dalili, M.; Ollis, D. The Influence of Cyclic Nucleotides on Hybridoma Growth and Monoclonal Antibody Production. Biotechnol. Lett. 1988, 10 ( l l ) , 781-796. Harlow, E., Lane, D., Eds. Antibodies: A Laboratory Manual; Cold Spring Harbor Laboratory: Cold Spring Harbor, NY, 1988. Helenius, A,; Simmons, K. Solubilization of Membranes by Detergents. Biochim. Biophys. Acta 1975, 415, 29-79. Jolley, M.; et al. Particle Concentration Fluorescence Immunoassay (PCFIA): a New, Rapid, Immunoassay Technique with High Sensitivity. J . Immunol. Methods 1984,67,21-35. Kaltenbach, J. P.; et al. Nigrosin as a Dye for Differentiating Live and Dead Ascites Cells. E x p . Cell Res. 1958,15,112-117. Kohler, G.; Milstein, C. Continuous cultures of fused cells secretingantibody of predefined specificity. Nature 1975,256, 495-497. Meilhoc, E.; et al. Application of Flow Cytometric Measurement of Surface IgG in Kinetic Analysis of Monoclonal Antibody Synthesis and Secretion by Murine Hybridoma Cells. J. Immunol. Methods 1989, 121, 167-174. Merten, 0. Culture of Hybridomas-A Survey. In Aduanced Research on Animal Cell Technology;Miller, A., Ed.; Kluwer Academic Publishers: New York, 1989. Rodbard, D.; et al. Statistical Aspects of Radioimmunoassay. In Handbook of Experimental Pharmacology, Vol. 82, Radioimmunoassay in Basic and Clinical Pharmacology; Patrono, R., Peskar, B., Eds.; Springer-Verlag: New York, 1987; pp 193-212. Schwartz, L. Calibration Curves With Nonuniform Variance. Anal. Chem. 1979, 1 (6),723-727. Thijssen, P.; et al. A Kalman Filter for Calibration, Evaluation of Unknown Samples and Quality Control in Drifting Systems. Anal. Chim. Acta 1984,156, 87-101. Vecchia, D.; et al. Calibration with Randomly Changing Standard Curves. Technometrics 1989, 31 (11,83-90. Accepted October 31, 1990.

Rapid immunoassay technique for process monitoring of animal cell fermentations.

A calibration and quality control technique suited to process monitoring with immunoassay is demonstrated. The particle concentration fluorescence imm...
610KB Sizes 0 Downloads 0 Views