Use of response surface methodology to optimize protease synthesis by a novel strain of Bacillus sp. isolated from Portuguese sheep wool


  • A.C. Queiroga,

    1.  CBQF/Escola Superior de Biotecnologia, Universidade Católica Portuguesa, Rua Dr. António Bernardino de Almeida, Porto, Portugal
    2.  Instituto Superior de Agronomia, Universidade Técnica de Lisboa, Tapada da Ajuda, Lisboa, Portugal
    Search for more papers by this author
  • M.E. Pintado,

    1.  CBQF/Escola Superior de Biotecnologia, Universidade Católica Portuguesa, Rua Dr. António Bernardino de Almeida, Porto, Portugal
    Search for more papers by this author
  • F.X. Malcata

    1.  ISMAI – Instituto Superior da Maia, Avenida Carlos Oliveira Campos, Castêlo da Maia, Avioso S. Pedro, Portugal
    2.  CEBAL – Centro de Biotecnologia Agrícola e Agro-alimentar do Baixo Alentejo e Litoral, Rua Pedro Soares, Beja, Portugal
    3.  CICECO – Centre for Research in Ceramics & Composite Materials, Universidade de Aveiro, Aveiro, Portugal
    Search for more papers by this author

Francisco Xavier Malcata, ISMAI – Instituto Superior da Maia, Avenida Carlos Oliveira Campos, Castêlo da Maia, P-4475-690 Avioso S. Pedro, Portugal. E-mail:


Aims:  To investigate the influence of yeast extract, peptone, temperature and pH upon protease productivity by Bacillus sp. HTS102 – a novel wild strain isolated from wool of a Portuguese sheep breed (Merino).

Methods and Results:  A 24 full factorial, central composite design together with response surface methodology was used to carry out the experiments and analyse the results, respectively. Among the individual parameters tested, temperature and peptone concentration produced significant effects upon protease productivity. A high correlation coefficient (R= 0·994, < 0·01) indicated that the empiric second-order polynomial model postulated was adequate to predict said productivity, with the optimum loci characterized by: temperature of 43°C, peptone content of 1·4 g l−1, pH of 5·1 and yeast extract concentration of 10·0 g l−1.

Conclusions:  Protease synthesis depends chiefly on temperature and peptone level. The maximum protease activity was more than twice that obtained with the basal medium, so the experimental design and analysis undertaken were effective towards process optimization.

Significance and Impact of the Study:  Rational choice of processing conditions for maximum protease productivity will be relevant if an economically feasible fermentation process based on Bacillus sp. HTS102 is intended.


Proteolytic enzymes have been routinely used in the detergent industry; this large-scale application has backed up their commercial development, thus promoting fundamental and applied research focused thereon (Reddy et al. 2008). However, proteolytic enzymes have also found numerous applications in other industrial sectors, for example, food and feed, leather, textile, pharmaceutical, fine chemistry, and effluent and waste treatment (Gupta et al. 2002).

Proteases are ubiquitous in Nature, yet those from microbial sources possess competitive advantages in that micro-organisms can be cultured to high densities and exhibit short doubling times – besides their methods of fermentation being well established; this has led to more and more abundant and regular supply of those enzymes (Seong et al. 2004). From a production standpoint, microbial extracellular proteases are particularly convenient owing to their obvious ease of extraction from the medium and purification afterwards, relative to their intracellular counterparts: large productivities have in particular been observed with specific species of the Bacillus genus (Navaneeth et al. 2009). As a consequence, there has been a renewed interest on wild Bacillus spp. that can synthesize and excrete proteases (Puri et al. 2002; Seong et al. 2004; Queiroga et al. 2007; Wang et al. 2008).

The growing array of proteases available commercially has prompted expansion of the portfolio of their uses; at present, comprehensive screening for new strains of bacteria has aimed at proteases with better features and more appropriate for industrial processes, that is, featuring lower costs and higher throughputs (Huang et al. 2008). Such efforts have necessarily to be complemented with optimization of the processing conditions themselves: significant differences in protease synthesis rates when a given micro-organism is subjected to distinct culturing conditions indicate that a few essential nutrients are to be supplied, besides providing narrow ranges of environmental parameters (Wilson 1930). On the other hand, the overall costs of enzyme production arising from downstream purification have remained a major obstacle to successful white biotechnology (Gupta et al. 2002), so bioprocess optimization upstream entailing chiefly extracellular enzymes is of central importance in this field (Reddy et al. 2008).

In view of the above, many biochemists and process engineers have focused on how to improve microbial protease yield. Classical approaches involving change of ‘one-variable-at-a-time’ are time-consuming and thus expensive when many variables are to be considered simultaneously (Rao et al. 2007); furthermore, they are intrinsically unable to pinpoint interactions among processing parameters. Hence, more powerful statistical experimental designs have been proposed (Puri et al. 2002). One successful example is response surface methodology (RSM) (Myers and Montgomery 2002), which allows building of models able to accurately approximate the true response function within a region around the optimum, using processing parameters as independent factors (Puri et al. 2002). Several studies are indeed available pertaining to application of RSM to optimize fermentation medium composition and physical processing factors, taking protease productivity as objective function (Bhaskar et al. 2008; Oskouie et al. 2008; Reddy et al. 2008; Wang et al. 2008; Cai and Zheng 2009).

The current research effort came along these lines and was aimed at optimizing protease productivity by a wool-associated, novel strain of Bacillus sp. (HTS 102) – via a 24 full factorial, central composite design, using effect of yeast extract, peptone, temperature and pH as regressors. Said wild strain was the outcome of a comprehensive screening programme among 158 isolates from Portuguese Merino sheep.

Materials and methods

Curvature of starting polynomial model

Previous work in our group, using a 2VI6−1 fractional factorial design, unfolded non-negligible two-way interactions among factors – thus implying existence of model curvature. Therefore, the original experimental design was expanded so as to also encompass three centre points; the values of each factor at said centre point were considered as the average setting of the design and are depicted in Table 1.

Table 1.   Expansion of 2IV6−1 fractional factorial design with three centre points, to check for process stability and possible curvature: definition of processing parameters (in original and coded format) and range tested
Coded parametersDescriptionActual values corresponding to coded values
X1Yeast extract (g l−1)57·510
X2Peptone (g l−1)234
X3Inoculum level (%, v/v)123
X4Agitation (rev min−1)050100
X5Temperature (°C)343740

Optimization via full factorial central composite design

Only four of 24 parameters previously screened exhibited a significant effect upon protease productivity by Bacillus sp. strain HTS102: yeast extract level, peptone content, temperature and pH. Hence, these were selected for optimization via a 24 full factorial, central composite design: the levels of each factor used for centre points were those leading to the best response in the preliminary 2VI6−1 fractional factorial design and are tabulated in Table 2. Said type of design with four factors encompasses 30 runs (with eight star points and six centre points); the associated details are conveyed by Table 3.

Table 2.   24 Full factorial central composite design, to enhance protease productivity: definition of processing parameters (in original and coded format)
Coded parametersDescriptionActual values corresponding to coded values
X1Temperature (°C)3035404550
X2Peptone (g l−1)1·01·52·02·53·0
X4Yeast extract (g l−1)89101112
Table 3.   24 Central composite design, to optimize protease productivity: design matrix (in coded format) and results of trials (observed and predicted)
Exp.Run orderCoded parameterProtease activity (U ml−1)
 3 2−11−1−116·6917·56
 5 3−1−11−120·3120·25
 8 7111−136·4536·24
 9 5−1−1−1126·7127·77
12 811−1122·8223·72
14 41−11149·5349·28
15 6−111121·1521·16
17 900003·343·46
18 1000026·7126·63

Micro-organism source and fermentation conditions

The strain employed was Bacillus sp. HTS102 (GenBank accession number HQ698269; reference LMG 26323, from BCCM/LMG Bacterial Culture Collection, Ghent, Belgium), a wool-associated micro-organism exhibiting strong proteolytic features (Queiroga et al. 2007). The basal fermentation medium contained 0·5 g l−1 meat extract (Merck, Darmstadt, Germany) and 1·0 g l−1 NaCl. Erlenmeyer flasks (250 ml) containing 25 ml of growth medium were inoculated with 1% (v/v) fresh inoculum (corresponding to an optical density of 0·6 at 600 nm). The extracellular protease activity was quantitated after a 36-h incubation period without stirring (as a nonsignificant statistical effect of stirring was eventually found, and given that the most economical processing conditions were under scrutiny).

Protease activity assay

The cell-free supernatants (sterilized by filtration through a 0·45-μm filter) were assayed for protease activity using colorimetry (660 nm), based on the extent of casein breakdown as assessed via Folin-Ciocalteu’s reagent (Sigma-Aldrich, St Louis, MO). One unit (U) of proteolytic activity was defined as the amount of enzyme able to hydrolyse casein at pH 7·5 at 37°C, so as to produce an absorbance variation per min equal to that produced by 1·0 μmol of tyrosine.

Protein content assay

Throughout the 36-h incubation period, aliquots were withdrawn for total protein assay (by measuring absorbance at 562 nm), which resorted to the BCA™ Protein Assay kit (Pierce, Rockford, IL, USA).

Statistical analyses

The statistical software package Design-Expert® ver. 8.0.3 (Stat-Ease, Minneapolis, MN, USA) was used to set up and analyse the central composite design at stake. The statistical significance of the underlying model equation and corresponding terms was assessed via Fischer’s F-tests; and the quality of the fit was ascertained by the coefficient of determination (R2) and an adjusted R2. The fitted polynomial equation was represented as response surface plots, for the two-factor interactions – to better illustrate their underlying relationships and effects upon the response.


Parameter interaction and curvature, and response surface analysis

The design matrix, with observed and predicted values, that underlined the trials following the 2IV6−1 fractional factorial design (expanded with three centre point runs) is depicted in Table 4a.

Table 4.   2IV6−1 Fractional factorial design, to check for process stability and possible curvature: (a) design matrix (in coded format) and results of trials (observed and predicted); and (b) results of analysis of variance encompassing parameters with highest individual linear effects upon protease productivity (R= 0·891; adj R= 0·871; pred R= 0·834; adeq precision = 22·142)
Exp.Run orderCoded parameterProtease activity (U ml−1)
  1. SS, sum of squares; df, degrees of freedom; MS, mean square; F, F ratio; P, probability value; *, statistically significant (P < 0.05).

Lack of fit4·64260·18502·790·0020*
Pure error7·093 × 10−423·546 × 10−4  
Cor total42·6334   

After taking into account a set of reasonable considerations, a 24 full factorial, central composite design was attempted – using the levels of yeast extract and peptone as nutritional factors, and temperature and pH as processing factors; this is described in Table 3. This design contains an imbedded factorial design with centre points, expanded with a group of axial (star) points; hence, a total of 30 runs resulted that encompass several combinations of temperature (X1), peptone level (X2), pH (X3) and yeast extract level (X4).

The optimum levels of the factors and of how interactions among them affect protease productivity were ascertained based on the results of the experiments run according to the aforementioned central composite design. The values for protease activity varied widely – from 3·34 to 51·75 U ml−1, among the various combinations considered (see Table 3). Following normalization of the data by a square root transformation, these results were tackled by regular analysis of variance; the outcome of this analysis is depicted in Table 5. The production of protease was significantly affected by temperature (X1) and peptone (X2), at < 0·001, and also by pH (X3), at < 0·05.

Table 5.   24 Central composite design, to optimize protease productivity: results of analysis of variance (R= 0·994; adj R= 0·989; pred R= 0·970; adeq precision = 56·056)
  1. SS, sum of squares; df, degrees of freedom; MS, mean square; F, F ratio; P, probability value; *, statistically significant (P < 0.05).

Model49·7114 3·55188·10<0·0001*
X27·111 7·11376·64<0·0001*
X30·121 0·126·51 0·0221
X4 0·0111 0·0110·60 0·4520
X1X21·851 1·8598·24<0·0001*
X1X30·101 0·105·44 0·0341
X1X40·401 0·4021·34 0·0003
X2X30·411 0·4121·83 0·0003
X2X40·111 0·115·59 0·0319
X3X4 0·0281 0·0281·49 0·2411
inline image20·63120·631093·13<0·0001*
inline image4·231 4·23224·24<0·0001*
inline image 0·0341 0·0341·80 0·2001
inline image1·311 1·3169·42<0·0001*
Residual0·2815 0·019  
Lack of fit0·2610 0·0264·59 0·0532
Pure error 0·0285 5·562 × 10−3  
Cor total49·9929   

A second-order polynomial function was consequently fitted to the experimental results, viz.


The geometric nature of the second-order model is displayed in Fig. 1; it unfolds the effect of every pair of variables upon protease productivity, while the remaining variables are held at their zero level (i.e. the centre point). In Fig. 1a,c,f, the centre of the system (i.e. the stationary point) corresponds to a point of maximum response. Inspection of Fig. 1b,d,e, one concludes that the type of surface defined by the second-order model is a ‘rising ridge’. The optimum levels of the four variables under study were estimated along such a surface, using the Solver function of Microsoft Excel; they were found to be a temperature of 43°C, a peptone level of 1·4 g l−1, a pH of 5·1, and a yeast extract content of 10·0 g l−1– with a predicted maximum activity of 56·76 U ml−1.

Figure 1.

 Response surfaces and corresponding contours of protease productivity by Bacillus sp. HTS 102, in terms of interaction between (a) temperature and peptone level, (b) temperature and pH, (c) temperature and yeast extract level, (d) pH and yeast extract level, (e) peptone level and pH and (f) peptone level and yeast extract level.

Model validation

The optimum conditions for protease production by Bacillus sp. HTS102, as predicted by the second-order polynomial model, were experimentally checked. The actual protease activity was 54·57 U ml−1, whereas its predicted value was 56·76 U ml−1; this entails a prediction accuracy of our model better than 95%– which in general validates our approach.


The F-value of 45·66, related to the trials following the 2IV6−1 fractional factorial design – associated with an R2 value very close to unit and an adequate precision of 22·142 (see Table 4b), imply that there is no statistically significant reason to doubt the functional form of the model postulated.

Based on examination of the results of the analysis of variance presented in Table 4b, one unfolded a significant ‘lack of fit’– thus indicating the relevance of curvature; hence, the first-model equation (represented in terms of coded factors), given by


where X1, X2, X5 and X6 denote yeast extract level, peptone level, temperature and pH, respectively, was proven inadequate (Myers and Montgomery 2002). Addition of centre point runs allowed testing of that curvature but did not permit estimation of the individual quadratic effects of each factor (Tamhane 2009). Remember that when neither the lack of fit nor the curvature is significant, the optimization process should follow the path of steepest ascent – that is along the linear direction of response enhancement, until no further increase in response is detected or else significant curvature (or lack of fit) is found. As the lack of fit tested after adding centre point runs proved significant, one had to expand the design with axial points – so that the response would be fitted to by a second-order model, via estimation of the individual quadratic effects of the factors under an RSM approach.

Recall that addition of axial points to a fractional (or full) factorial design containing centre runs gives rise to a central composite design. For such a design encompassing six factors (i.e. yeast extract level, peptone level, inoculum level, stirring rate, temperature and pH), the value of α for the axial points would be ±2·378; this implies that their actual values would lie beyond the limits specified for the factor settings. On the other hand, the factors stirring rate and inoculum level were found not significant upon protease productivity, so they were dropped out as variables; surprisingly, the favourable effects of stirring towards aeration of the culture and inoculum size towards total viable numbers were too marginal to be classified as statistically significant – at least at the (arbitrary) level of significance selected. They were accordingly fixed at their lowest levels for economic considerations, that is, further optimization experiments used an inoculum level of 1% (v/v) and incubation without agitation.

In view of the considerations above, a 24 full factorial, central composite design was attempted (Table 3), so the second-order polynomial function labelled as Eqn (1) was fitted to the experimental results. After inspection of the magnitude of the parameter estimates in Eqn (1), one concluded that peptone content produces the maximum linear positive effect (< 0·01) towards protease productivity, followed by the linear positive effect of yeast extract. The positive quadratic effects of temperature, peptone and yeast extract were all significant (< 0·01). Among the two-factor interactions, the negative effect of the temperature × peptone interaction is to be outlined (< 0·01); this means that the extent of the temperature effect upon protease production depends also on the actual level of peptone used. The linear effect of yeast extract was negligible, yet its quadratic effect was significant (< 0·01) – so a low level of yeast extract has a marginal impact, which increases fast when higher levels are provided.

The regression and determination coefficients associated with said model are given in Table 5; they unfold a high significance, as derived from the large F-value and the R2 value very close to unit, coupled with an ‘adequate precision’ as high as 56·06. The lack-of-fit test in Table 5 (P = 0·053) suggests that the quadratic model postulated is adequate.

The response surface plots depicted in Fig. 1 facilitate analysis of the response surface curves of two-factor interactions – and prediction of the response, or estimation of the mean response at a particular point in the process variable space (Myers and Montgomery 2002). Detection of the nature and location of any stationary point is also an important issue of such a second-order analysis (Myers and Montgomery 2002), which obviously depends on the signs and magnitudes of the coefficients in Eqn (2).

The ‘rising edge’ in Fig. 1b,d,e indicates that extrapolation beyond the experimental region should, for precaution, rely on additional experimentation. In any case, several directions will lead to improvement, so there is no single (correct) direction; however, our fitted model will aid in identifying treatment combinations that would likely lead to better results. The predicted maximum activity was more than twice that obtained under the basal conditions; this realization emphasizes the usefulness of our optimization strategy.

Despite the bacterial strain selected for this study having been isolated from a moderately alkaline environment (pH 9·5), the optimum levels of its protease productivity were found at an acidic pH (pH 5·1). It is not at all uncommon that, for a given micro-organism, the optimum processing conditions for growth differ from the optimum processing conditions for specific metabolite production thereby; for example, Bacillus polymyxa produces enzymes optimally at pH 5 (Castro et al. 1992), whereas it is typically cultivated for growth purposes at pH values in the vicinity of 7·0.

This work is relevant towards eventual development of an economically feasible fermentation process based on Bacillus HTS101 strain. The protease production was dependent chiefly on temperature and peptone level, and on pH and yeast extract level to a lesser degree; the optimum loci were 5·1 for pH, 43°C for temperature, 1·4 g l−1 for peptone level and 10·0 g l−1 for yeast extract content. Under these optimum conditions, a 2·3-fold increase in protease productivity was possible – using the basal medium as reference.


Author A.C. Queiroga received a PhD fellowship (ref. SFRH/BD/19121/2004), granted by Fundação para a Ciência e a Tecnologia (Portugal) and supervised by author F.X. Malcata. Availability of laboratory premises by CBQF for performance of a part of the experimental work described is hereby acknowledged.