Biosorption of Methylene Blue by De-Oiled Algal Biomass: Equilibrium, Kinetics and Artificial Neural Network Modelling

The main objective of the present study is to effectively utilize the de-oiled algal biomass (DAB) to minimize the waste streams from algal biofuel by using it as an adsorbent. Methylene blue (MB) was used as a sorbate for evaluating the potential of DAB as a biosorbent. The DAB was characterized by SEM, FTIR, pHPZC, particle size, pore volume and pore diameter to understand the biosorption mechanism. The equilibrium studies were carried out by variation in different parameters, i.e., pH (2–9), temperature (293.16–323.16 K), biosorbent dosage (1–10 g L−1), contact time (0–1,440 min), agitation speed (0–150 rpm) and dye concentration (25–2,500 mg L−1). MB removal was greater than 90% in both acidic and basic pH. The optimum result of MB removal was found at 5–7 g L−1 DAB concentration. DAB removes 86% dye in 5 minutes under static conditions and nearly 100% in 24 hours when agitated at 150 rpm. The highest adsorption capacity was found 139.11 mg g−1 at 2,000 mg L−1 initial MB concentration. The process attained equilibrium in 24 hours. It is an endothermic process whose spontaneity increases with temperature. MB biosorption by DAB follows pseudo-second order kinetics. Artificial neural network (ANN) model also validates the experimental dye removal efficiency (R2 = 0.97) corresponding with theoretically predicted values. Sensitivity analysis suggests that temperature and agitation speed affect the process most with 23.62% and 21.08% influence on MB biosorption, respectively. Dye adsorption capacity of DAB in fixed bed column was 107.57 mg g−1 in preliminary study while it went up to 139.11 mg g−1 in batch studies. The probable mechanism for biosorption in this study is chemisorptions via surface active charges in the initial phase followed by physical sorption by occupying pores of DAB.


Introduction
Treatment of wastewater streams has always been an important and challenging area of research and several physico-chemical and biological processes exist for colour removal from various effluents. Most of the industrial effluents i.e. of textile, rubber, paper and printing, cosmetics, food, leather, pharmaceutical, plastics etc. contain large amount of toxic dyes which are mutagenic and carcinogenic to life forms [1].
Common methods for colour removal are coagulation and flocculation [2], biological oxidation and chemical precipitation [3] and activated carbon adsorption [4]; the latter is the preferred method for removing, recovering and recycling of dyes from wastewaters due to its simplicity. However, it is not without its limitations due to the costs involved as well as difficulties in regeneration of the substrate [5,6]. Therefore, recent interest has shifted to low-cost materials that range from waste products of other industries to naturally abundant biomass such as peanut hull [7], rice husk [2], water hyacinth roots [8], hexane-extracted spent bleaching earth [9], raw and activated date pits [10], guava seeds [11], macroalgae Sargassum muticum [12], Parthenium plants [13], bacteria and fungi [14] among many others.
Algae are one of the most promising sources of biomass, biofuel, organic fertilizer, protein supplements and high value nutraceutical products due to their benefits like, ability to utilize fresh, marine or wastewater, non-requirement of fertile land, do not compete with food crops and reduce green house gas from the environment [15,16]. Recent research has been focusing on practicality and sustainability of algal biofuels [16,17]. Algal biodiesel production generates a large amount of waste de-oiled algal biomass (DAB), which is still the subject of numerous studies including its use as a substrate for bioethanol, biogas in addition to be used as animal/poultry/fish feed, fertilizer, and remediation of dyes & heavy metals [18,15].
Methylene blue (MB) has been used in the present study to model the dye adsorption capacity of the de-oiled algal biomass. Main objective of the present study is to evaluate the effects of different physicochemical parameters like pH, agitation speed, temperature, dye-biosorbent contact time, adsorbent concentration and initial dye concentration with the performance as predicted through different kinetic models including pseudo-first and second order kinetics and intra particle diffusion. We have also analysed the whole data set using artificial neural network (ANN) to validate the experimental as well as predicted values generated which were found to agree very well with each other.

DAB preparation and characterization
The naturally occurring floating algal biomass dominated by Microspora sp. (ATCC PTA-12197) was collected from the coastal lagoons of Gujarat (site coordinates 20u 42.3919 N and 70u 54.9599 E) and was first de-oiled through Soxhlet extraction using hexane as the solvent [19]. The algal biomass collected was not an endangered/protected species. No specific permission was required from any authority for collection of said biomass. DAB was oven dried for 48 hours at 80uC to remove any residual hexane and ground for further characterization and experiments. Surface topology of the DAB was observed using a scanning electron microscope (LEO 1430VP, Zeiss, Germany), whereas BET (Brunauer-Emmett-Teller) surface area and total pore volume was obtained using Micromeritics ASAP 2010 V5.02, USA. The particle size distribution was calculated based on differential pore volume of Barrett-Joyner-Halenda (BJH) adsorption and desorption. FT-IR spectrum of DAB was obtained using the KBr disc method (Spectrum GX, Perkin Elmer, USA) with a resolution of 4 cm 21 in the range of 400-4000 cm 21 region for the examina-tion of surface open functional groups. The point of zero charge (pH PZC ) was determined by salt addition method [20].
The samples were withdrawn from each flask at 5 minutes interval till an hour and every hour upto 5 hours and the final reading was taken at 24 hours. The samples were centrifuged and MB concentration in the supernatant was determined at 665 nm using UV-Visible Spectrophotometer (Varian Cary-50 Bio, Varian Inc., USA). Except column experiment, all the experiments were carried out in duplicate with a working volume of 200 mL in 500 mL Erlenmeyer flasks at room temperature (2762uC) and under static conditions unless specified otherwise.

Adsorption thermodynamics
The changes in enthalpy (DHu), entropy (DSu), Gibbs free energy (DGu) as well as the equilibrium constant (K c ) for MB biosorption over DAB was calculated using van't Hoff equation (Equation 3).
The change in free energy is related to the equilibrium constant by the following relationship: where R is the gas constant (8.314 J mol 21 K 21 ) and T is the absolute temperature in K. According to the Gibbs' free energy equation: Combining equation 1 and 2, we get: A plot between ln K c and 1/T gives a linear graph whose intercept and slope yields DSu/R and DHu/R, respectively.

Parameters of Kinetic studies
The MB adsorption q t (mg g 21 ) at time t, was calculated using the following equation: where q t is the dye uptake (mg g 21 ), C 0 (mg L 21 ) is the initial dye concentration and C t (mg L 21 ) is the dye concentration at time t in the solution, V is the solution volume (L) and M is the mass of the biosorbent (g). The percentage of MB removal was calculated using the following equation: Different kinetic models such as pseudo-first order (Equation 6), pseudo-second order (Equation 7) and intra-particle diffusion models (Equation 8) were assessed in order to study the rate of MB adsorption over DAB [21,22,23].
where q e and q t are the amount of MB adsorbed per unit mass (mg g 21 ) at the equilibrium and at time t, respectively while k 1 (min 21 ) , k 2 (g mg 21 min 21 ) and k int (mg g 21 min 21/2 ) are the rate constants whereas C i represents the intercept (mg g 21 ).

ANN modeling
Artificial neural network (ANN) modeling is important to identify the complex input-output relationship and to develop a model to predict the output of dependent variables from the given set of independent variables [24]. ANN has been used to evaluate the performance of various parameters for dye and metal adsorption using mean square error and regression [25,26,27]. The present study uses Neural Network Toolboxof MATLAB R2013a software with a total of 527 experimental data sets and ranges of input variables like pH (2-9), initial dye concentration (25-125 mg L 21 ), adsorbent concentration (1-10 g L 21 ) temper- The data sets were normalized in the range 0.1-0.9 using the following equation: where min(X i ) and max(X i ) are the extreme values of variable X i [24]. All data were divided into training (70%), validation (15%), test (15%) subsets and the network was trained with Levenberg-Marquardt back-propagation algorithm. Optimum number of hidden nodes was 10 for the present model. The performance of network was measured by mean squared error (MSE) using following equation: where y i , pred and y i , exp are the values predicted by the neural network and obtained by experiments, respectively. N is the number of data points and i is an index of data. Relative importance of different input variables on adsorption efficiency was determined by sensitivity analysis [28] calculated by following formula: where I j is the relative importance of the j th input variable on the output variable, N i and N h are the numbers of input and hidden neurons, respectively. W is connection weight, the superscripts 'i', 'h' and 'o' refer to input, hidden and output layers, respectively, and the subscripts 'k', 'm' and 'n' refer to input, hidden and output neuron numbers, respectively.  Column experiment For the column experiment, a sintered glass disc column (1.2615 cm) was packed with a known quantity (0.3 g) of DAB biosorbent upto a height of 1 cm. A column feeding solution containing 100 mg L 21 MB at pH 7 was loaded into the column at a flow rate of 0.37 ml min 21 . The fractions were collected at an interval of one hour and concentration of MB was analyzed in each fraction as detailed above. The graph plotted between C/C 0 and time (t) is termed as the breakthrough curve. Breakthrough (t b ) and exhaustion (t e ) times represent the time at which the dye concentration in the collected fractions reached 5% and 95% of feed dye concentration respectively. The length of mass transfer zone (L m ), mass of dye removed (M r ) and dye uptake (q e ) at exhaustion time were calculated using the following formulae [29,30].
where L (cm) is total bed depth, V ex (L) throughput volume at column exhaustion, C 0 (mg L 21 ) influent dye concentration, V n (L) throughput volume at n th reading, V n+1 throughput volume at (n+ 1) th reading, C n (mg L 21 ) the effluent dye concentration at n th reading, C n+1 (mg L 21 ) the effluent dye concentration at (n+1) th reading and M (g) is the weight of biosorbent used to prepare the column bed. The length of unused (LUB) and used bed (UB) at breakthrough point were determined by following equations [31].

Characterization of the biosorbent
The texture of the DAB surface was observed by SEM image (Figure 1) which reveals different irregularities leading to surface roughness that plays an important role in the surface adsorption of the dye. Lipid extraction of biomass through solvent leads to harsh effects like breaking of cell wall which may cause such surface topology.
FT-IR analysis of the DAB (Figure 2) verifies the functional groups present on its surface. Several peaks like 3429 cm 21 , 2929 cm 21 , 1461 cm 21 , 1638 cm 21 , 1544 cm 21 and 1253 cm 21 corresponds to the O-H stretching vibrations of glucose, -NH groups of proteins, C-H aliphatic stretching vibrations, C-H scissoring, amide I (carbonyl group stretching), amide II and amide III (NH bending and CN stretching), respectively, confirming the presence of polysaccharides, proteins, etc. [32,33]. These molecules have their own functional groups such as amino, sulfhydryl, phosphate, carboxylic and thiol groups that can bind various ions [34]. These surface functional groups of biosorbent possibly play an important role in MB sorption.
The particle size distribution of the sample was such that the diameter of 90 and 50% of DAB was less than 1482.57 and 242.75 mm, respectively. The surface area, pore size and pore volume of particles were determined by BET and BJH methods and the values are presented in Table 1. BET and BJH surface area of DAB are 1.46 and 1.57 m 2 g 21 respectively. The BET and BJH pore size of DAB are 11.29 and 24.09 nm respectively, which indicates the porous structure of DAB which aid in MB (size 0.93 nm) sorption [35].

Adsorption experiments
Effect of pH. The experiments were carried out at various pH from 2 to 9 ( Figure 3). The results were best with approximately 98% MB removal after 24 hours at pH 3 and 4 while MB removal was reduced from 98 to 86% at pH 5-7. MB removal improved from 86 to 93% at pH 7-9. These results prove that MB removal was above 90% for both acidic and basic pH. MB removal was 86% at neutral pH, a condition of natural water bodies, so apparently there is no need to change the pH for further scale up and all further experiments were also conducted at pH 7. Similar trends were observed in MB removal by the alga Spirogyra [36], acid activated carbon [37], activated carbon [4,38].
The point of zero charge (pH PZC ) is an important parameter for biosorbent to characterize the sensitivity to the pH and their surface charges. The pH PZC of DAB was found to be 9.05. Generally, adsorption of cations is more favoured at pH .pH PZC while for anions at pH ,pH PZC [20]. However, the present study emphasizes the point that electrostatic interactions are not the only reason for this adsorption process. When the pH ,pH ZPC , the repulsion of similar positively charged DAB surface and MB leads to the entry of MB into the interlayer surface of DAB without being adsorbed at the outer surface [39,40]. In highly alkaline conditions (pH .9), methylene blue forms aggregates due to its zwitterionic structure in aqueous solutions leading to larger conformation like dimeric and trimeric forms which hinders their entry in the porous surface of the biomass [37]. Other possible reasons for such observations include dye-adsorbent interactions due to hydrogen bonding and hydrophobic-hydrophobic interaction mechanisms. Hence, the pore size and surface area, which remain unaffected by pH changes, also play an important role during the process [4,38].
Effect of contact time. Figure 4 represents the effect of contact time on MB removal efficiency and adsorption capacity of DAB. Nearly 75% of MB was removed from the solution by DAB under neutral pH conditions in 5 minutes. The rate of removal was found to decrease gradually with 81 and 86% sorption after an hour and 24 hours respectively. The amount of adsorbed dye (q e ) increases with an increase in the contact time for up to five hours after which the rate of removal, which is dependent on the number of active sites in the biomass, decreases. A higher concentration of active sites during the initial stages of the process accelerates the removal of the dye from the solution. However, with passage of time the number of active sites get occupied with the dye molecules and hence, the rate of dye removal decreases until attaining equilibrium after 24 hours.
Effect of initial methylene blue concentration. The effect of initial MB concentration on its removal efficiency by DAB is represented in Figure 5a. The percentage MB removal increases with decrease in initial MB concentration. The equilibrium capacity of DAB linearly increases with increasing MB concentration up to 2000 mg L 21 after which the q e value was nearly constant even when the dye concentration was increased upto 2500 mg L 21 (Figure 5b). The maximum monolayer adsorption capacity at equilibrium was 139.11 mg g 21 at 2000 mg L 21 . The availability of adsorption sites is the rate-limiting factor which influences the equilibrium concentration of the dye. Higher initial dye concentration in the solution leads to saturation of the available sites much earlier which results in a higher dye content in the solution at equilibrium. Such trends are also observed for algae Sargassum sp. [12,15] and Galidium sp. [41] where they have reported a maximum monolayer biosorption capacity of 107.5 and 104 mg g 21 respectively.
Effect of adsorbent dosage. Figure 6a represents the influence of DAB concentration on the extent of biosorption. The extent of MB removal increased initially with an increase in the DAB concentration. The graph between the dye removal (%) vs. contact time (min) shows that after 24 hours, 94.58% of the dye was removed when we used the adsorbent at a concentration of  The dye adsorption capacity decreased with increasing biosorbent dosage (Figure 6b). The decreasing value of q e can be attributed to the abundant active sites available for dye sorption when we start increasing the DAB concentration while keeping the concentration of the dye constant. Thus, a low dye: adsorbent ratio results in a lower adsorption capacity. Various other reasons for reduction in q e value have been reported in literature like decrease in the effective surface area of the adsorbent due to partial aggregation of adsorbent particles which forms a protective outer layer preventing the interaction of dye with the biomass and formation of a concentration gradient between the dye solution and the biosorbent [42,43,44].
Effect of agitation speed. The distribution of dye molecule in solution is affected by agitation speed and it affects the uptake of dye molecules by disrupting the film resistance surrounding the adsorbent particles [45]. The effect of agitation speed on the rate of MB adsorption is presented in Figure 7. As the speed was increased, the time required to achieve equilibrium was also increased. This phenomenon may be due to the absence of aggregation of the biomass that ultimately increases the available surface area of DAB leading to rapid adsorption of MB. The efficiency of dye removal in 24 hours increases from 86% to 99% as the speed is increased from 0 to 150 rpm. The removal efficiency in 1 hour at 150 rpm is significantly higher than at other speeds supporting the theory of reduced aggregation of biomass at higher speeds.

Thermodynamic studies
The effect of temperature on MB adsorption under different temperature regimes (293.16-323.16 K) is depicted in Figure 8. The values for DGu, DHu and DSu are shown in Table 2. As evident from the positive values of DHu and DSu, the process is spontaneous and endothermic in nature [46]. The negative values of DGu also depict the spontaneity of the process. As the temperature is increased (293. 16-323.16 K), the values of DGu become more negative, indicating that the process becomes more spontaneous. The low value of DHu also indicates that physical binding forces may be involved in the process [47]. Increasing the temperature increase the rate of diffusion of dye molecule into the internal pores of the biosorbent [48].

Adsorption kinetics
The kinetic data obtained from the effect of initial dye concentration on equilibrium biosorption capacity of biosorbent was utilized for testing different kinetic models. Their graphical presentations are shown in Figure 9. Using the slopes and intercepts, different kinetic parameters i.e. equilibrium biosorption capacity (q e, cal ) and correlation coefficient (R 2 ) values were calculated and are presented in Table 3. Figure 9a shows t vs. ln  (q e -q t ) plot for pseudo-first order kinetics. As is evident, the value of the rate constant k 1 decreases with an increase in initial MB concentration. Most of the literatures cite pseudo second order model as a best fit only at the preliminary stages of adsorption instead of the whole data range [49]. This is evident from the differing experimental and predicted q e values obtained from this model. The t vs. t/q t plot given in Figure 9b represents the pseudo-second order kinetics with the slope equal to 1/q e . Decreasing slope of the plot with increasing initial dye concentration indicates increased adsorption capacity at equilibrium (q e ), whereas the rate constant k 2 for the process decreases. The pseudo-second order kinetic is more relevant for the present study which is evident by nearly similar values of experimental and calculated values of q e with the high degree of correlation (R 2 = 0.99). Based on the pseudo second order model, it is assumed that the rate limiting step may be chemical phenomena involving the exchange of electrons between the sorbate and the sorbent [22]. Intra-particle diffusion kinetic model is represented in Figure 9c which depicts a linear relationship between the amounts of dye adsorbed vs. square root of the contact time. The lines do not pass through the origin, denoting the fact that adsorption process is not limited by intra-particle diffusion alone [10]. The adsorptions of MB on DAB occurs by chemisorption via surface exchange reaction until all the active sites are occupied by the dye, which then diffuses into available pores and any channels present in the biomass. Figure 10 shows the regression analysis between the experimental and predicted values of adsorption efficiency using a neural network model. The correlation coefficient of this comparison plot was 0.97 which shows that ANN model fits the experimental values of adsorption efficiency (%) and is well reproduced in this system. Similar results were reported for algae Chara sp. [27], walnut husk [50] and Penicilium sp. [51] with a correlation coefficient ranging from 0.97 to 0.99. The minimum mean squared error for test was 0.0024 at epoch 117.

ANN modeling
The relative importance and ranking of the input variables for the dye removal efficiency (%), shown in Table 4, follows the trend: Temperature.agitation speed.contact time.pH.Initial dye conc..Adsorbent dose. The temperature and agitation speed are most influential parameter in biosorption which is supported by literature [52,53] while contact time and pH have moderate influence on biosorption i.e.18.80% and 15.06% respectively. Such influence was supported by experimental results obtained in sections of respective parameters.

Column experiments
Adsorption of dye on a fixed bed column is widely used to scale up the process in actual conditions. In the column, when the dye solution enters the bed, it comes in contact with the first few layers of DAB particles and fills up the available active sites. As soon as the adsorbent in the uppermost layer is saturated and the dye solution penetrates further into the bed, the adsorption region shifts down. The bed is considered ineffective when the unabsorbed dye emerges into the effluent. The breakthrough curve of MB adsorption on DAB is represented in Figure 11. The breakthrough time t b and exhaustion time t e were observed to be 7 and 18.5 hours, respectively. The length of mass transfer zone (L m ), mass of dye removed (M r ) and dye uptake (q e ) at exhaustion time were 0.90 cm, 32.27 mg, and 107.57 mg g 21 respectively. The length of unused and used beds at breakthrough is 0.62 and 0.38 cm, respectively. In comparison to batch biosorption, column biosorption capacity is much higher because of greater dye concentration gradient at the interface zone of dye solution and biomass bed [54].

Conclusion
The maximum MB removal was found at both acidic and basic pH but it was also good at neutral pH with 86% dye removal which represents conditions of natural water bodies. The DAB (10 g L 21 ) removes 86% of the dye (50 mg L 21 ) in 5 minutes under static condition and nearly 100% in 24-hours with agitation at 150 rpm, which indicates the MB removal by DAB in static conditions is more feasible as it requires lesser energy input. The optimum DAB concentration was 5-7 g L 21 for effective MB removal at 100 mg L 21 MB concentration. The highest adsorption capacity value of DAB is 139.11 mg g 21 at 2000 mg L 21 initial MB concentration while 107.57 mg g 21 for column study at 100 mg L 21 MB concentration in preliminary study. The kinetic studies of MB adsorption were best fit with the pseudo-second order model. Based on the study of different parameters and kinetic models it can be concluded that no single rate limiting step is involved in biosorption mechanism. It involved different bonding of surface active sites functional group with dye, physical adsorption of dye in micro-pores and macro-pores of DAB surface and monolayer adsorption. ANN modelling for each parameter predicts the experimental value with high correlation coefficient value (R 2 = 0.97) and sensitivity analysis suggests that temperature and agitation speed are the most influencing parameter in this biosorption experiment. Present study supports the practical valorisation of de-oiled algal biomass, obtained during algal biofuel process, for the treatment of coloured effluents and economically balances the algal bio-refinery. The advantages of the present work include the utilization of DAB as a biosorbent for removal of hazardous dyes, utilizing the waste generated from algal biofuel as well as reducing algal blooms.