Modeling and optimization of a continuous bead milling process for bacterial cell lysis using response surface methodology

Abstract

Efficient cell lysis for intracellular protein recovery is a major bottleneck in the economics and commercial feasibility of any biotechnological process. Grinding of cells with abrasive beads, also known as bead milling remains a method of choice, as it can handle a large volume of cells. Bead mills when operated in a continuous mode substantiate to be economical, and more productive as compared to a batch mode process. In this study, the recovery of recombinant cholesterol oxidase (COD) was investigated and optimized using response surface methodology (RSM) based on Central Composite Design (CCD) in a continuous bead milling process. Process parameters, viz. slurry feed rate (A), bead loading (B), cell loading (C) and process time (D) were found to be significant during the continuous bead milling process. A polynomial model was developed to correlate the participating factors for efficient cell disruption. Optimized conditions yielded 3.20 g L⁻¹ (∼90%) of COD with A = 300.6 mL h⁻¹, B = 77.5% (v/v), C = 69.9 (OD_{600 nm}) and D = 29.7 (min), when compared to existing batch mode operations (3.56 g L⁻¹). This is the very first study that attempts to optimize a continuous bead milling process using RSM to maximize the intracellular protein (COD in this case) recovery with minimum inputs to make the process economical and scalable to industrial levels. The developed model in this study can be scaled-up to large-scale for efficient recovery of intracellular proteins in similar expression systems.

---

Introduction

Recombinant proteins are generally expressed intracellularly in Escherichia coli (E. coli) systems in research laboratories all over the world. Complete disruption of the microbial cells becomes indispensable for the maximum release of an intracellular protein(s) of interest that ultimately facilitates the product's recovery and subsequent purification.^1–3 A variety of cell disruption processes are available at the laboratory scale, as samples generally are in low volumes, easier to handle and process. These processes lack scalability, productivity and yield requirements. Various cell disruption methods have been reviewed extensively in the past.^4–9 On the other hand, for a process intended to be commercially feasible on the industrial scale, disruption efficiency, process duration, power required, final recovery and productivity are critical parameters.^10,11 The choice of disruption method depends on the process parameters, i.e., the nature of the product, its thermo-stability, activity, half-life, tolerance towards range of pH and ionic concentrations, etc., and the intentional use of the product released. Unfortunately, the commonly used cell lysis methods (chemical, physical or enzymatic) prove efficient only at the laboratory scale,^10,12 but fail to address the needs of large-scale cell disruption processes.

Mechanical methods, on the other hand, are promising as these can be scaled-up with minimum or no pre-treatments/chemical additives and can be used under batch or continuous mode. As mechanical methods are milder, these result in lower degradation of proteins as compared to chemical lysis methods with competent efficiency. Bead milling is one of the most preferable mechanical cell lysis methods at the industrial scale due to its ease of operation, controllability and ability to load concentrated cell slurry.¹⁰ On the other hand, downtime associated with cleaning and sanitization of beads/bead mills severely affects the overall process cost and productivity. Therefore, a higher capacity bead mill is required to process larger amount of sample per batch. Batch operation of a large bead-mill is cumbersome, while smaller bead mills with continuous feeding of fresh cell slurry offer a promising alternative. Laboratory scale bead mills, e.g. Dynomill® (WAG, Switzerland) mimicking large-scale bead mills are generally expensive at lab scale and hence, not routinely preferred. Lab-scale instruments utilizing bead milling principle have a significantly different design (e.g., Bead beater®, Biospec Inc., USA) and are not feasible at larger scale.¹³ Bead mill generates significant heat during the process, thus it is necessary to maintain the temperature, shorten the run time and reduce bead loading.¹⁰

Response Surface Methodology (RSM) is apt for optimization studies where the process variables interact with each other and need to be varied simultaneously to estimate the interaction effects between any two selected variables at one point of time.¹⁴ RSM combines mathematical as well as statistical tools in one complete package to optimize (maximize or minimize), improve and enhance product recoveries in multi-factor (multi-variable) systems.¹⁵ RSM has been successfully exploited for process development and its improvement, development of new formulations, minimization of potentially unimportant or toxic entities involved in the process.¹⁶ RSM not only indicates the effect of one variable on the other but also provides a discrete quantification of the effects of individual factors on each other. Based on the experimental design, the results obtained are used to develop a best-fit mathematical equation, taking into account ANOVA, lack of fit, F-value and p-values for assessing the statistical significance. The equation is used further to validate the results at different level of the process variables. Due to these advantages over conventional ‘one-factor-at-a-time’ optimization processes, RSM has become a leading methodology for most of the laboratories and industries working on bioconversions of biological products.^17–20 Earlier reports suggest that bead loading, bead size, agitator velocity, cell slurry concentration are significant parameters affecting cell lysis in bead milling under batch-mode,^10,21 but none deal with scalable continuous cell lysis optimization process. In general, the bead milling process under batch mode suffers some drawbacks mainly due to limited capacity, longer downtime between the runs for cleaning and reloading, batch-to-batch variations etc. Whereas, continuous bead milling process overcomes the above limitations by continuously feeding larger volumes of the cell slurry resulting in lower ‘batch-to-batch’ variation and less number of cleaning/sanitization cycles, making the process more economical and feasible for scale-up.

In the present study, cholesterol oxidase (COD) was taken as an example for modeling and optimization of various parameters viz. bead fraction, cell concentration, bead-milling time and cell-slurry feed rate in continuous bead milling process to maximize bacterial cell lysis and efficient recovery leading to scalable production.

COD is an important enzyme that can be potentially used in the analysis of clinical samples, determination of cholesterol levels in food and serum samples, and development of biosensors.^22,23 COD can participate in the synthesis of steroid based drugs and it works as a larvicidal protein that can be used as an insecticide.²⁴ In addition, COD acts as a potential virulence factor uniquely against bacteria that can represent a novel molecular target for the discovery of new antibiotics.^25,26 COD may act as signaling protein for the biosynthesis of polyene macrolide pimaricin in some Streptomyces sp.^24,27 COD belongs to the class of alcohol dehydrogenases [EC 1.1.3.6]. It catalyzes the oxidation and isomerization of cholesterol.

In this study, COD was expressed intracellularly in bacterial system and the extent of cell lysis was measured by the amount of recovered activity of COD in cell lysate as compared to the standard ultra-sonication process. This is the very first report on modeling and optimization leading to scalable cell lysis by continuous bead milling process employing RSM for enhanced recovery of the protein (i.e., COD in this case). The schematic representation of the entire modeling and optimization process of bacterial cell lysis is given in Graphical Abstract of the manuscript.

Results

Bacterial culture and COD recovery

COD expression studies were successfully performed in E. coli cells at shake flask and fermenter levels as reported.²⁸ The bacterial cells were separated by centrifugation and analysis of the extracellular supernatant showed no COD activity, confirming that it expressed intracellularly. The intracellular COD production was found to be ∼58.98 μg per mL per OD_{600 nm} by directly lysing the cells and subjecting the lysate to SDS PAGE analysis (data not shown). The estimated molecular weight of COD was found to be ∼46.5 kDa.

Evaluation of cell lysis

The bacterial cells were lysed using ultra sonication method for the maximum recovery of COD enzyme per unit of cell mass. The cell slurry was diluted to a final OD_{600 nm} of 10 and 10 mL of this suspension was subjected to sonication for 25 min. The maximum COD activity was found to be ∼91.42 units (58.89 μg COD per mL per OD_{600 nm}) after 18 min of sonication and no further increment was observed (Fig. 1). The maximum COD recovered was ∼58.9 μg per 10 mL per OD_{600 nm} and it was considered as 100% COD recovery for further yield calculations.

Fig. 1 Release of intracellular COD using ultra-sonication.

Bead size and bead milling time

Three sets of runs were performed with three different bead sizes (0.25–0.50 mm; 0.50–0.75 mm, 0.75–1.00 mm) separately, in the bead mill loaded at 80% (v/v) for 30 min. The remaining volume of the grinding chamber was filled with cell slurry (OD_{600 nm} 50). Bead sizes ranging between 0.50–0.75 mm were found to be most effective as it resulted in a total recovery of 10 406 units of COD (50.86 μg mL⁻¹ OD_{600 nm}). The cell lysis efficiency was found to be ∼86.3% in comparison with ultra-sonication method. COD recoveries with bead sizes in the range of 0.25–0.5 mm and 0.75–1.00 mm were 8428.82 (41.19 μg per mL per OD_{600 nm}) and 7493.04 units (36.62 μg per mL per OD_{600 nm}), respectively (Fig. 2). The yield increased almost steadily with time, slowing down after 30 min reaching a maximum at 35 min. Therefore, the bead milling time was fixed at a maximum of 30 min for further experiments.

Fig. 2 Release of intracellular COD using bead milling with different bead sizes, 0.25–0.50 mm, 0.50–0.75 mm, 0.75–1.00 mm.

Design of experiments

Central composite design under RSM was used resulting in a combination of 21 experiments each with a different level of the participating parameters as given in Tables 1 and 2. The amount of COD obtained was averaged and fed into the combination matrix of the generated design against each run (Table 2). The response ranges from a minimum of 0.761 g L⁻¹ obtained in Run no. 20 to a maximum of 3.282 g L⁻¹ COD observed in Run no. 21. The ratio of the maximum response to the minimum was found to be 4.31. The developed model had 12 and 4 degrees of freedom for lack of fit and ‘Pure’ error respectively. The F-value of the model was 455.16 implying that there was only a 0.01% chance that the model F-value this large, could occur due to noise. The model interaction terms A, B, C, D, AB, AC, BD, CD, A², B², C², and D² were found to be significant. The terms AD and BC had p-value higher than the prescribed value of 0.1, indicating that both the terms did not contribute significantly to support the hierarchy of the model. Hence, these terms were considered ‘not significant’ and were removed from the model for simplification and to make better curve fitting. All the other terms had p-values lower than 0.1, and were therefore included in the model. The F-value for feed rate, cell loading and run-time were found to be 1087.64, 563.86 and 301.31, respectively, suggesting that these parameters have remarkable effect on cell lysis for COD recovery. The effect of bead loading was lower than the other selected variables (F-value = 57.25). The coefficient of determination (R²) for the model was 0.998 and adjusted R² was 0.996 (Table 3). For a valid statistical model, signal to noise ratio (also called adequate precision) must be more than 4, and in this model, adequate precision was found to be 75.92, that confirms the validity and aptness of the model to be used for navigating the design space. The simplified polynomial equation for the recovery of COD is depicted as:

Y = +2.28 − 0.58 × A + 0.086 × B + 0.27 × C + 0.31 × D + 0.075 × AB − 0.19 × AC + 0.14 × BD + 0.049 × CD − 0.19 × A² − 0.12 × B² − 0.062 × C² − 0.19 × D² (1)

where, A = feed rate, B = bead loading, C = cell loading, and D = run-time. The interactions between the individual factors are indicated as AB, AC, BD and CD. The degree of precision of the experiments is demonstrated by CV (%) values and in the present study, it was found to be 2.19%, which indicates the accuracy of the responses. Above said responses and outcomes proved the accuracy and sustainability of the model to be used for predicting the responses at various design points. The developed model also showed high correlation between the predicted and the actual responses across all the design points (Fig. 3). The model ensured a good-fit between the observed and the predicted responses. Based on the model and the interaction terms, the predicted responses have been plotted on z-axis in the form of a 3-dimensional (3D) graph called as ‘response surface plot’. The yield is plotted as contours showing the variation in the response obtained over all the combinations of the selected process variables. The contours provide a bird's eye view and a quicker way to analyze the response and to get an idea of the trend followed by the response across the selected two variables in x- and y-dimensions. The developed model was validated by performing bead milling with the predicted optimized condition in triplicate.

Table 1 The experimental design parameters and constrains of design space for participating factors

Factor	Name	Units	Type	Low actual	High actual	Low coded	High coded	Mean
A	Feed rate	mL h^−1	Numeric	300.00	500.00	−1.000	1.000	400.0
B	Bead loading	(%, v/v)	Numeric	60.00	80.00	−1.000	1.000	70.0
C	Cell loading	(OD600 nm)	Numeric	50.00	70.00	−1.000	1.000	60.0
D	Run time	min	Numeric	20.00	30.00	−1.000	1.000	25.0

---

Table 2 Design of experiments and recovered cholesterol oxidase (g L⁻¹) in each run

Run no.	Feed rate (mL h^−1)	Bead loading (%, v/v)	Cell loading (OD600 nm)	Run time (min)	COD recovered (g L^−1)
1	400.0	70.0	60.0	25.0	2.282
2	300.0	80.0	50.0	30.0	2.266
3	231.8	70.0	60.0	25.0	2.720
4	300.0	60.0	70.0	20.0	2.510
5	500.0	60.0	70.0	30.0	1.232
6	500.0	60.0	50.0	30.0	1.066
7	400.0	70.0	43.2	25.0	1.584
8	400.0	70.0	60.0	33.4	2.248
9	400.0	70.0	60.0	16.6	1.217
10	400.0	70.0	76.8	25.0	2.603
11	400.0	70.0	60.0	25.0	2.282
12	400.0	86.8	60.0	25.0	2.082
13	500.0	80.0	50.0	20.0	0.842
14	500.0	80.0	70.0	20.0	0.882
15	400.0	70.0	60.0	25.0	2.282
16	400.0	70.0	60.0	25.0	2.282
17	300.0	60.0	50.0	20.0	1.761
18	400.0	70.0	60.0	25.0	2.282
19	400.0	53.2	60.0	25.0	1.801
20	568.2	70.0	60.0	25.0	0.761
21	300.0	80.0	70.0	30.0	3.282

---

Table 3 ANOVA for cholesterol oxidase recovery (g L⁻¹)

Source	Sum of squares	DOF^a	Mean square	F-Value	p-Value
a DOF – degree of freedom.
Model	9.6382	12	0.80318	455.1686	<0.0001
A – feed rate (mL h^−1)	1.9192	1	1.91923	1087.644	<0.0001
B – bead loading (%)	0.1010	1	0.10103	57.25632	<0.0001
C – cell loading (OD600 nm)	0.9950	1	0.99496	563.8566	<0.0001
D – run time (min)	0.5317	1	0.53168	301.3109	<0.0001
AB	0.0188	1	0.01882	10.66984	0.0114
AC	0.3036	1	0.30357	172.039	<0.0001
BD	0.0669	1	0.06693	37.93347	0.0003
CD	0.0192	1	0.01920	10.88532	0.0109
A^2	0.5221	1	0.52212	295.891	<0.0001
B^2	0.2008	1	0.20082	113.81	<0.0001
C^2	0.0574	1	0.05739	32.52566	0.0005
D^2	0.5380	1	0.53804	304.9135	<0.0001
Residual	0.0141	8	0.00176
Lack of fit	0.0141	4	0.00352
Pure error	0	4	0
Cor total	9.652284	20
Std. dev.	0.0420		R-Squared (R^2)	0.9985
Mean	1.9175		Adj R-squared	0.9963
CV%	2.1907		Pred R-squared	0.9787
			Adeq precision	75.920

---

Fig. 3 Graph showing actual vs. predicted responses.

Optimization and validation

To study the effect of individual process variables and their mutual interactions on the recovery of intracellularly expressed COD protein, a statistical model was developed after running 21 experiments for obtaining the yields at different selected points/combinations of variables as suggested by the design of experiments (DoE) study. The yields obtained were fed into the respective combination and the data obtained by ANOVA analysis was further employed to develop a reduced quadratic model to navigate the design space within the specified boundaries or restrictions as taken into account before designing the process. The restrictions that are defined by the upper and the lower values of the participating variables are shown in Table 1. The response surface model efficiently calculated all the possible combinations of all the levels of the selected parameters and presented with a continuous response surface plot (Fig. 4a–d). The numerical optimization gives control over the selection criteria of the response (i.e., COD yield); it was employed for the optimization of the obtained responses. The numerical optimization of the model suggested a combination of A : B : C : D as 300.6 : 77.5 : 69.9 : 29.7 as feed rate (mL h⁻¹), bead loading (%, v/v), cell loading (%, v/v) and run time (min), respectively for obtaining 3.283 g L⁻¹ of COD. The predicted combination of parameters when tested experimentally resulted in 3.21 g L⁻¹ of COD (average of three experimental runs) suggesting that the model can be successfully used for the optimization and understanding the interactions of different process parameters involved in the cell lysis process.

Fig. 4 Response surface plot. (a) Effect of feed rate (mL h⁻¹) vs. bead loading (%, v/v) on the recovery of COD. (b) Effect of cell loading (OD_{600 nm}) vs. run time (min) on the recovery of COD. (c) Effect of run time (min) and bead loading (%, v/v) on the recovery of COD. (d) Effect of feed rate (mL h⁻¹) and cell loading (OD_{600 nm}) on the recovery of COD.

The observed responses based on the experimental data and the predicted response, indicated that the release of COD increased with decrease in the feed rate (A), e.g., Run no. 3 showed a recovery of 2.72 g L⁻¹ of COD (feed rate = 231.8 mL h⁻¹) which was higher as compared to Run no. 1 (feed rate = 400 mL h⁻¹; COD yield: 2.28 g L⁻¹) considering same combination of B, C and D. In the case of Run no. 20, where the feed rate was maximum at 568 mL h⁻¹, the recovery of COD went down close to 0.76 g L⁻¹ indicating a negative correlation of the feed rate versus COD recovery. Likewise, the model equation also suggests a negative value of A, (eqn (1)) suggesting an inverse relationship of feed rate with COD recovery. Increased bead loading (B) in Run no. 18 and 19 resulted in better recovery of COD and suggests a positive correlation of bead loading at 53% and 70%. However, when the bead loading was increased to 86% the yield of COD decreased reflecting a negative effect of bead-loading beyond a certain limit between 70 and 86% (v/v).

Cell loading (C) indicates the concentration of the cells in the cell slurry. Increased cell loading results in an increase in the number of cells per unit volume, containing COD available for lysis. As the concentration of COD increases in the feed, the amount of COD also increases in the lysate that leads ultimately to higher yields. An increase in the cell concentration in Run no. 6 and Run no. 5 suggested increase in COD yield from 1.06 g to 1.23 g, respectively. By maintaining a constant cell concentration, an increase in run time normally increases the recovery of COD as indicated by Run no. 9 and 18 with 1.21 g L⁻¹ and 2.28 g L⁻¹, respectively; COD recovery plateaus around 30–33 min with 2.24 g L⁻¹ recovered in Run no. 8. ANOVA showed that the feed rate had maximum influence on COD recovery, whereas, bead loading had comparatively lower but significant effect on COD yield. Similarly, the interaction effects of AC (feed rate and cell loading) were distinct than any other interaction terms. The optimized condition accounted for 3.2 g L⁻¹ (∼90%) as compared to the maximum COD recovered (3.56 g L⁻¹) by batch bead milling process.

Economic analysis of bead milling process (batch vs. continuous mode)

In-house economic analysis of batch and continuous process was performed for COD recovery. The grinding process was started as per the optimized conditions obtained by RSM. On the basis of static and variable costs incurred in the bead milling process, the process was divided into three phases; viz. (A) preparation phase, (B) running phase, (C) cleaning & sanitization phase. Bead loss was termed as ‘D’. The cost incurred in the preparation phase of the bead milling process (i.e., A) and cleaning/sanitization (C) remains constant. However, variations occur only in the running phase (i.e., B). Bead mill was run for 30 min in batch mode with cell slurry filled in the grinding chamber with beads loaded at 77.5% (v/v), while in the continuous mode, cell slurry was fed at 300.6 mL h⁻¹ till 1200 mL of slurry passed through the grinding chamber. The batch mode was continued for 30 min, while the continuous mode took 240 min for the completion of the cell lysis process. Analysis revealed that the total cost incurred in the running phase (B) was USD$ 0.28 for the batch mode and USD$ 2.245 for the continuous mode of cell lysis for COD recovery (pricing shown in the Table 4 is based on the conversion of INR to USD as per the international rate 67.71 as on Jan 18^th 2016). The productivity of the batch process for COD recovery was found to be 6.56 mg COD per USD per day, whereas the productivity of the continuous process was 26.16 mg COD per USD per day. The continuous process was found to be 3.99 times more productive than batch process (Table 4). The continuous bead milling process can run for a longer time, processing larger volumes, saves on the preparation and cleaning/sanitization phase as it has to be done only once for any volume of cell slurry (we run 1200 mL in this case). This increases the overall process productivity, even with 90% COD recovery per unit volume of the cell slurry compared to the batch mode, making it useful for commercial applications.

Table 4 Economic analysis of bead milling process for bacterial lysis (e.g. COD recovery)^a

Steps in bead milling process		Batch cell lysis process (total volume = 274 mL)			Continuous cell lysis process (total volume = 1200 mL)
		Description	Requirement	Approx. cost^b (INR)	Description	Requirement	Approx. cost^b (INR)
a The calculation doesn't take into account the fixed costs (bead mill, glass beads, cooling water bath, feed pump, incubator. Misc. items include: glass tray, measuring cylinders, tubing, beakers, funnel, etc.).b The prices shown here are indicative only and subject to change as per the country norms of local taxes, freight charges, VAT, etc., therefore, the final cost may not remain the same.c Currency conversion 1 USD = 67.61 INR, as on Jan 18^th, 2016.
Preparation and assembly	Step 1: washing and assembly	Manpower required	1 day	1200.00	Manpower required	1 day	1200.00
	Step 2: measuring and filling of beads	Glass beads required	465 mL	—	Glass beads required	465 mL	—
	Step 3: cell slurry preparation/filling	Tris–HCL buffer	1 L	794.00	Tris–HCL buffer	1 L	794.00
	Step 4: initiation of cooling water bath	Cooling (2 kW h)	30 min	10.00	Cooling (2 kW h)	30 min	10.00
		(A) Preparation cost		2004.00	(A) Preparation cost		2004.00
Running of bead mill	Step 5: running of bead mill, feeding of cell slurry (continuous mode)	Electricity for milling (1.8 kW h)	30 min	9.00	Electricity for milling (1.8 kW h)	240 min	72.00
	Step 6: collection of cell lysate (misc.)	Electricity for cooling (2 kW h)	30 min	10.00	Electricity for cooling (2 kW h)	240 min	80.00
		Collection of cell lysate	Glass tray	Misc.	Collection of cell lysate	Glass tray	Misc.
		(B) Running cost		19.00	(B) Running cost		152.00
Cleaning/sanitization	Step 7: bead removal (decantation)	Removal of beads	Glass tray	Misc.	Removal of beads	Glass tray	Misc.
	Step 8: cleaning/sanitization	Water for cleaning of beads	10 L	6450.00	Water for cleaning of beads	10 L	6450.00
	Step 9: drying of beads at 80 °C in oven	Sodium hydroxide for sanitization	1 L	1229.00	Sodium hydroxide for sanitization	1 L	1229.00
		Drying of beads in oven (2 kW h)	10 h	200.00	Drying of beads in oven (2 kW h)	10 h	200.00
		(C) Cleaning/sanitization		7879.00	(C) Cleaning/sanitization		7879.00
		Total volume of slurry processed	274 mL		Total volume of slurry processed	1200 mL
		Total COD recovered	975.44 mg		Total COD recovered	3936.10 mg
		Total process time (approx.)	11 h		Total process time (approx.)	14.5 h
		Bead loss (D)	∼1%	150.00	Bead loss (D)	∼1%	150.00
		Total cost (A + B + C + D)		10 052.00	Total cost (A + B + C + D)		10 185.00
		Total cost of running in USD^c		148.67	Total cost of running in USD^c		150.42
		Productivity (mg COD per USD per day)		6.56	Productivity (mg COD per USD per day)		26.16
		Fold increase in COD productivity continuous (1200 mL) vs. batch cell lysis (274 mL) = 26.16/6.56 = 3.99

---

Discussion

Intracellular proteins of interests are generally produced via expression into bacterial host systems, especially in E. coli. Therefore, maximizing the recovery of the protein of interest becomes indispensable if the process is aimed to be scaled-up for industrial applications.

Currently available cell lysis processes are generally designed for laboratory scale and suffer from a number of drawbacks. Ultra-sonication/chemical lysis is said to be more effective, rapid, and easy to apply, but it can handle only low volumes and its effectiveness is compromised at higher cell densities that results in lower yields. It also generates significant heat in the process, which again is detrimental to the expressed protein(s). Ultra-sonication is slow and requires constant ‘ON’ and ‘OFF’ cycles, which increases process time, thereby making the process uneconomical at larger scales. Chemical processes are milder but applicable only at smaller scale. These chemical processes employ harsh chemicals, are effective at only lower cell concentrations, interfere in downstream processing and require expensive effluent treatment if piloted at commercial scale. All these limitations discourage the use of chemical lysis at industrial scale.

Cell lysis by bead milling is a preferred method to lyse the cells at commercial scale. A number of variations are available in bead mill designs, still the basic principle remains the same, i.e., lysis of cells by grinding between two beads that transfer their kinetic energy to break the cell continuum. The studies till date have not documented the effect of interactions in between the participating parameters on product recovery in bead milling process, e.g., feed rate affects the cell lysis in a negative manner, whereas process time and cell loading influence the lysis process in a positive manner. Increased bead loading decreases the space between the individual beads that leads to more collision, therefore results in better grinding. The cells undergoing the lysis process encounter collision with the beads at a higher rate and ultimately result in better recoveries. Cell loading increases the number of cells per unit volume loaded in the grinding chamber, increasing the amount of available protein to be recovered; however, it also affects the process negatively by increasing the viscosity of the slurry. Release of intracellular components, for e.g., DNA, RNA etc., further increase the viscosity of the cell slurry. Smaller beads have lower mass and therefore gain lower momentum when agitated at constant velocity as compared to bigger beads. Smaller size leads to fluidization thereby reducing collision and reducing the shear stress created by the beads.¹⁵ Small beads are also significantly affected by viscosity under high cell loads. Therefore, at higher viscosities, small beads result in lower protein yields as these break fewer cells and are mostly useful when the cells concentrations are low. As the cells are believed to disrupt in the contact zones of the beads by shearing action, higher energy transfer, and compaction, the higher recovery obtained in the case of 0.5–0.75 mm size beads can be attributed to stronger impact, higher shear stress, and better transfer of kinetic energy to the cell mass. Bigger beads (0.75–1.0 mm or higher) have higher momentum, but their bigger size creates larger void spaces and result in reduced number of collisions, reducing their effectiveness when compared to the beads of intermediate size. Higher viscosities reduce the impaction between the beads, which results in incomplete lysis. To overcome this limitation, either the speed of the impeller has to be increased or beads have to be changed with denser material with higher specific gravity than glass, for e.g., zirconium. Cell lysis in bead mill generally follows first order kinetics when OFAT approach is applied²⁹ as depicted:

ln[R_m/(R_m − R)] (2)

where, R is the amount of observed protein release and R_m is the maximum protein that can be obtained. When multiple parameters are changed simultaneously, the situation becomes more complex and a number of statistical parameters need to be calculated.

In the present study, COD recovery increases with increase in bead as well as cell loading and reaches a plateau at higher side of the design matrix. Process time, though not much important at laboratory scale, it becomes increasingly crucial during scale-up. Bead milling requires a lot of power to drive the agitator shaft as it has to move the battery of beads against gravity, overcome the inertia and friction of beads continuously. Higher bead loading increases the inertial mass; therefore, more power is required to keep the agitator rotating and the beads in motion. With increase in bead loading, the interstitial space between the beads also reduces, bead packing becomes dense, thus reducing the total volume available for cell slurry to be loaded. This also necessitates more power to be delivered to the agitator shaft. The reduction in volume affects the residence time of the cell slurry and volumetric productivity negatively. The reduction in the residence time results in reduced/incomplete lysis of the cells, which reduces the overall COD yield. Under-loading of beads below optimum volume creates more interstitial spaces between the beads. This reduces the transfer of kinetic energy from the impaction of the beads to break the cell continuum. Therefore, it is evident that bead loading needs to be optimized very carefully. Based upon the findings, cell loading and process time are two very important parameters that effect and interact with the bead loading parameter. The F-values for the bead loading, cell loading and runtime were 57.26, 563.86, and 301.31, respectively. Interaction terms involving bead loading/feed rate (AB) and bead loading/run time (BD) had a low p-value, indicate that there is significant interaction among the building and other parameters involved in the experiment. Likewise, high F-value (172.04) indicates good interaction between the feed rate and the cell loading (AC). Numerical optimization as generated by targeted maximization of COD recovery yields a maximum of 3.283 g L⁻¹ COD. The COD recovery was validated by running the predicted conditions by RSM. The COD recovery was found to be 3.21 g L⁻¹, which was very close to the predicted recovery by RSM design.

Use of bead mill along with its benefits also has few major drawbacks; especially when used under the batch mode. After the batch run concludes, beads have to be taken out of the grinding chamber, cleaned, sanitized and dried at 80 °C, adding considerable downtime time before the next batch can be initiated. Loss of beads becomes inevitable with each cleaning cycle. In order to minimize the downtime, higher capacity bead mills are required that can process the larger volumes of the cell slurry per batch. Increasing the volume of the grinding chamber makes the bead mill bulky, requires more man-power, needs higher electrical power to drive the beads against the inertial forces, compromises on cooling efficiency, and results in the formation of temperature gradients inside the grinding chamber. In order to avoid these limitations, relatively smaller bead mills with a provision to feed the cell slurry continuously or intermittently by a feed pump is generally preferred. Smaller bead mills with smaller foot print are easy to operate, can process larger volumes of cell slurry in continuous mode, have efficient cooling, suffer lower bead losses due to less frequent changeovers. In addition, power and labor requirements are lower which adds to the process economics. Since, the cell slurry is continuously added, feed rate becomes a critical factor for the optimization. A faster feed may negatively affect the productivity due to incomplete lysis of cells (cells spend less time in the grinding chamber), while a slow feed prolongs the grinding process unnecessarily wasting time, power and valuable resources, thus making it inefficient and economically less viable. Therefore, an optimum balance is necessary between the feed rate and other participating parameters, i.e., bead loading (%, v/v), cell loading (OD_{600 nm}), and process time (min) for higher productivity. Economic analysis of the batch process (274 mL cell slurry) vs. continuous (1200 mL cell slurry) bead milling process, done in our lab is shown in Table 4. The continuous bead milling results into 3.99-fold higher COD productivity when compared to the batch process. The in-house comparative economic analysis indicated the suitability and productivity of the continuous bead milling process over the batch mode of operation.

Currently, our group and collaborators are aggressively working on the application of the proposed model given in this study at large-scale using various bead mill designs (e.g., different agitator configuration, agitation speeds. etc.) and inclusion of more factors (e.g., specific gravity of beads, namely zirconium beads, ceramic beads, etc.). We are also under the process of applying artificial intelligence techniques, e.g., genetic algorithm (GA), artificial neural networks (ANN) etc., for the optimization and scale-up of continuous bead milling process for COD recovery and we will report the findings in our successive publication.

In conclusion, the present study proves the significance of RSM in large-scale cell lysis and product recovery (COD in this case) in optimization of continuous bead-milling process for the very first time. It also establishes and quantifies the interactions between the participating parameters affecting the recovery of intracellularly expressed recombinant protein. The COD recovery reached up to ∼90% as compared to that achieved in batch process under optimized conditions. As the bead mills can be run for longer duration in the continuous mode with constant feed of fresh cell slurry, the overall productivity of the process exceeds much higher than that obtained under the batch mode. Our findings provide a promising strategy for large-scale continuous cell lysis by bead milling process leading to enhanced recovery of intracellular protein(s) with similar expression patterns.

Materials and methods

Bacterial strain and growth conditions

All chemicals, biochemical, bacterial culture media or their components were obtained from RFCL (India), HiMedia Laboratories (Mumbai, India) and BDH (India). The recombinant protein (COD) was expressed in host E. coli BL21 (DE3) pLysS cells (Invitrogen) transformed with recombinant plasmid pET24b(+) harboring COD gene, under IPTG inducible lambda phage promoter following the protocol of Volante et al., (2010).²⁸ Starter cultures were initiated with a single colony of recombinant E. coli in Luria Bertani (LB) medium containing kanamycin (30 μg mL⁻¹) at 37 °C under a rotary orbital shaker at 180 rpm. The bacterial cells were initially grown at shake flask and subsequently scaled up to fermenter level in Terrific broth (TB) supplemented with glycerol (12 g L⁻¹ Bacto-tryptone, 24 g L⁻¹ yeast extract, 8 mL L⁻¹ glycerol, 16 mM KH₂PO₄ and 54 mM K₂HPO₄). Baffled Erlenmeyer flasks (500 mL) containing 80 mL of the medium was inoculated with the starter culture (initial OD_{600 nm} 0.6) and the cells were grown at 37 °C for 14 h at 180 rpm. The cells were induced for protein expression after 6 h of incubation.

Batch culture

TB medium added with 20 mL L⁻¹ glycerol was used as a production medium in a glass reactor with 5 L working volume (Biostat® C, Sartorius AG, Germany). Bacterial cultivation in the fermenter was performed at 37 °C with 300–700 rpm stirrer speed and at 2–5 L min⁻¹ aeration rate cascaded with pure oxygen. Foaming was controlled by using Hodag antifoam agent maintained through an antifoam sensor. The starter culture was grown in LB medium for overnight, afterwards it was diluted to an initial OD_{600 nm} of 0.6. After 6 h of incubation IPTG was added at a final concentration of 1 mM for the induction of recombinant COD expression. Cell samples were collected at different time points throughout the run. All experiments were done in duplicate each with three sets throughout the study.

Determination of enzymatic activity

The activity of the enzyme was determined at 25 °C using 1 mM cholesterol in 100 mM potassium phosphate buffer at pH 7.5 in the presence of 1% propanol and 1% Thesit® (v/v, final concentration) by evaluating H₂O₂ production using the horseradish peroxidase assay (4 μg mL⁻¹, 0.3 mg mL⁻¹ o-dianisidine) as per previous reports published.^30,31 One unit of enzyme was defined as the amount of enzyme that produces one μmol of H₂O₂ per minute at 25 °C. Recovered lysate was centrifuged at 10 000 rpm and remaining supernatant was used for calculating enzyme activity.

Cell lysis using sonication

The cell pellet was washed two times with Tris–HCl buffer (pH 7.5) and resuspended in the same buffer. The OD of the cell suspension was adjusted to OD_{600 nm} 50. Ten mL of the cell suspension was diluted five times (final OD_{600 nm} 10; used as control) and subjected to sonication (15 s ON; 30 s OFF cycle), using 25 kHz, 65 W cell sonicator (Heat Systems Inc., New York, USA) equipped with titanium micro-tip probe with tip diameter of 6.4 mm. In order to prevent heating of the slurry, the cell suspension was kept under ice-bath and 200 μL of the sample was withdrawn at specific time intervals for 25 min for the quantification of the released COD. The amount of COD released from the cell suspension per unit cell mass under sonication, was considered as 100% cell lysis, and it was taken as a reference for further experiments.

Cell lysis using bead-mill

Continuous lysis of the cells was done using bead mill (KDL type), also called as Dynomill® from W. A. Bachofen AG, Switzerland. The grinding chamber was loaded with glass beads (0.25–0.5 mm) at 70% v/v. The cell slurry was adjusted to OD_{600 nm} 50 and was fed in the grinding chamber. The total capacity of the grinding chamber used was 600 mL and the temperature was maintained at 4 °C using circulating chilled water around gaskets and ball bearings. The mill was run at 3000 rpm for different time intervals upto 40 min under continuous mode of operation. The cell slurry was fed at different rates in the grinding chamber with the help of a pre-calibrated peristaltic pump. The samples were collected at different time intervals as per the experimental design for the estimation of the released COD.

Selection of bead diameter for cell lysis

The cell lysis was done with beads size ranging from 0.2–0.5 mm, 0.5–0.75 mm, and 0.75–1.0 mm, at 70% (v/v) bead loading for 30 min. The OD_{600 nm} of the cell slurry was adjusted to 50. The cell slurry was subjected to grinding chamber and the bead mill was run at 3000 rpm for 30 min. Samples were withdrawn at specific intervals to quantify the amount of COD released.

Statistical design and analysis

Based on the observations from previous literature (Middleberg, 1995) four factors, i.e., feed rate, bead loading percentage, cell loading, and run time were found crucial for bead milling process for the optimal lysis of the bacterial cells for enhanced recovery of intracellular proteins. Central composite design (CCD), a subset of Response Surface Methodology (RSM) was employed with fractional factorial design, considering the above-mentioned four significant factors for enhanced recovery of COD from recombinant E. coli. A four factor, five level design was employed for the optimization and interaction studies. These bead milling factors (i.e., feed rate, bead loading percentage, cell loading, and run time) were considered to work independently and were designated as A, B, C, and D, respectively. These variables were varied at five levels (−2, −1, 0, +1, +2). The experimental design and constrains of design space are depicted in Table 1. The model was developed as a design matrix that included factorial points, center points, and axial (star) points. As a result, the experimental design was created with 21 experimental runs (Table 2). An average of three runs performed in duplicate was considered as a response for each run and inserted in the experimental matrix (Table 2). Values of mean and standard deviations were calculated to confirm the reproducibility of the experimental data. In each run, the amount of COD recovered was measured and filled in as a ‘response’ by the quantification method as stated earlier. A reduced quadratic model was developed by analyzing the data and the model was depicted as:

Y = a₀ + a₁A + a₂B + a₃C + a₄D + a₅AB + a₆BC + a₇CD + a₈AD + a₉AC + a₁₀BD + a₁₁A² + a₁₂B² + a₁₃C² + a₁₄D² (3)

The experimental ‘responses’ obtained were used to perform statistical analysis including analysis of variance (ANOVA). Statistical significance of the model was determined by calculating Fischer's test value (F-value), p-value and fitness of the data using the coefficient of determination (R²). Individual factors with p-values ≤ 0.05 and interaction terms with p-value ≤ 0.1 were considered significant. The developed model was represented by 2-dimensional contour plots and 3-dimensions response surface plots as generated by Design Expert software program (Statease Inc., Minneapolis, USA). These plots were used to navigate the design space and to find the ‘optimum’ for COD recovery using best-fit combination of all the variables under consideration. Bead milling experiments were performed in triplicate as per the predicted conditions to validate the developed model.

Conflict of interest

The authors declare no conflict of interest exists.

Acknowledgements

We thank Jamia Millia Islamia, New Delhi (India), University of Kwazulu-Natal, Durban (South Africa), and Jazan University (Saudi Arabia), for providing the necessary facilities for this research work.

References

1. H. M. Hsiung, A. Cantrell, J. Luirink, B. Oudega, A. J. Veros and G. W. Beeker, Bio/Technology, 1989, 7, 267–271.
2. C. Kato, T. Kobayashi, T. Kudo, T. Furusato, Y. Murakami, T. Tanaka, H. Baba, T. Oishi, E. Ohtsuka, M. Ikehara, T. Yanagida, H. Kato, S. Moriyama and K. Horikoshi, Gene, 1987, 54, 197–202.
3. P. Yu and K. Y. San, Biotechnol. Prog., 1992, 8, 25–29.
4. D. E. Hughes, J. W. T. Wimpenny and D. Lloyd, The disintegration of micro-organisms, in Methods in Microbiology, ed. J. R. Norris and D. W. Ribbons, Academic press, New York, 1971, pp. 1–54.
5. S. Morein, D. Henricson and L. Rilfors, Anal. Biochem., 1994, 216, 47–51.
6. B. Schmitt, Biochimie, 1976, 58, 1405–1407.
7. C. De Virgilio, N. Biirckertm, T. Boiler and A. Wiemken, FEBS Lett., 1991, 291, 355–358.
8. F. Bunge, M. Pietzsch, R. Mfiller and C. Slydatk, Chem. Eng. Sci., 1992, 47, 225–232.
9. J. X. Feliu and A. Viilaverde, Biotechnol. Tech., 1994, 8, 509–514.
10. A. P. J. Middleberg, Biotechnol. Adv., 1995, 13(3), 491–551.
11. N. M. Fish and M. D. Lilly, Bio/Technology, 1984, 2, 623–627.
12. A. R. Byreddy, A. Gupta, C. J. Barrow and M. Puri, Mar. Drugs, 2015, 13, 5111–5127.
13. R. S. Singh, Indian J. Exp. Biol., 2013, 51, 997–1003.
14. K. Dubey, A. Jawed and S. Haque, Eng. Life Sci., 2011, 11(6), 598–606.
15. I. Belo, A. L. Santos, J. M. S. Cabral and M. Mota, Biotechnol. Prog., 1996, 12, 201–204.
16. S. R. Azar, R. Naiebi, A. Homami, Z. Akbari, A. Kianmehr, R. Mahdizadehdehosta and F. Najafzadeh, Indian J. Biochem. Biophys., 2015, 52(1), 68–74.
17. V. Akbari, H. M. Sadeghi, A. Jafarian-Dehkordi, C. P. Chou and D. Abedi, Res. Pharm. Sci., 2015, 10(1), 75–83.
18. A. A. Piccolomini, A. Fiabon, M. Borrotti and D. De Lucrezia, Biotechnol. Appl. Biochem., 2015 DOI:10.1002/bab.1459.
19. A. R. Choudhary, P. Saluja and G. S. Prasad, Carbohydr. Polym., 2011, 83, 1547–1552.
20. S. Lotey and R. S. Singh, Braz. Arch. Biol. Technol., 2010, 53, 1005–1013.
21. Y. J. Tam, Z. N. Allaudin, M. A. M. Lila, A. R. Bahaman, S. J. Tan and M. A. Rezaei, BMC Biotechnol., 2012, 12, 70.
22. W. Richmond, Clin. Chem., 1973, 19(12), 1350–1356.
23. K. Hino, M. Nakamura, K. Nakanishi and M. Manabe, Clin. Chem., 1996, 42, 296–299.
24. J. MacLachlan, A. T. L. Wotherspoon, R. O. Ansell and C. J. W. Brooks, J. Steroid Biochem. Mol. Biol., 2000, 72, 169–195.
25. G. Gadda, G. Wels, S. Pollegioni, S. Zucchelli, M. Ambrosius, M. S. Pilone and S. Ghisla, Eur. J. Biochem., 1997, 250(2), 369–376.
26. L. Pollegioni, L. Piubelli and G. Molla, FEBS J., 2009, 276, 6857–6870.
27. J. Kreit and N. S. Sampson, FEBS J., 2009, 276, 6844–6856.
28. F. Volante, L. Pollegioni, G. Molla, L. Frattini, F. Marinelli and L. Piubelli, BMC Biotechnol., 2010, 10(33), 1472–6750.
29. Y. Chishti and M. Moo-Young, Enzyme Microb. Technol., 1986, 8, 194–203.
30. L. Motteran, M. S. Pilone, G. Molla, S. Ghisla and L. Pollegioni, J. Biol. Chem., 2001, 276, 1824–1830.
31. L. Pollegioni, G. Wels, M. S. Pilone and S. Ghisla, Eur. J. Biochem., 1999, 264, 140–151.

---