ANN-GA based biosorption of As(III) from water through chemo-tailored and iron impregnated fungal biofilter system

The iron impregnated fungal bio-filter (IIFB) discs of luffa sponge containing Phanerochaete chrysosporium mycelia have been used for the removal of As(III) from water. Two different forms of same biomass viz. free fungal biomass (FFB) and modified free fungal biomass (chemically modified and iron impregnated; CFB and IIFB) have been simultaneously investigated to compare the performance of immobilization, chemo-tailoring and iron impregnation for remediation of As(III). IIFB showed highest uptake capacity and percentage removal of As(III), 1.32 mg/g and 92.4% respectively among FFB, CFB and IIFB. Further, the application of RSM and ANN-GA based mathematical model showed a substantial increase in removal i.e. 99.2% of As(III) was filtered out from water at optimised conditions i.e. biomass dose 0.72 g/L, pH 7.31, temperature 42 °C, and initial As(III) concentration 1.1 mg/L. Isotherm, kinetic and thermodynamic studies proved that the process followed monolayer sorption pattern in spontaneous and endothermic way through pseudo-second order kinetic pathway. Continuous mode of As(III) removal in IIFB packed bed bioreactor, revealed increased removal of As(III) from 76.40 to 88.23% with increased column height from 5 to 25 cm whereas the removal decreased from 88.23 to 69.45% while increasing flow rate from 1.66 to 8.30 mL/min. Moreover, the IIFB discs was regenerated by using 10% NaOH as eluting agent and evaluated for As(III) removal for four sorption–desorption cycles, showing slight decrease of their efficiency by 1–2%. SEM–EDX, pHzpc, and FTIR analysis, revealed the involvement of hydroxyl and amino surface groups following a non-electrostatic legend exchange sorption mechanism during removal of As(III).

Arsenic contamination of groundwater is a serious global environmental concern affecting a large number of populations, especially in developing countries. Chronic arsenic toxicity causes severe consequences to human health, such as skin lesions, circulatory and nervous disorders, black foot disease, melanosis, keratosis, non pitting oedema, Bowen's disease gangrene and cancers of different organs 1,2 . Thus, the removal of arsenic from potable water is an essential requirement.
In India, as per the reports of CGWB, GOI, in year 2009, seven states namely, West-Bengal, Jharkhand, Bihar, Uttar Pradesh in the flood plain of Ganga River; Assam and Manipur in the flood plain of Brahamaputra and Imphal rivers and Rajnandgaon village in Chhattisgarh state have so far been reported affected by arsenic contamination in groundwater above the permissible limit of 50 µg/L. At certain location tube wells were found to have Arsenic concentration of 300 µg/L. The present study confirms treatment of more than highest concentration of Arsenic observed in groundwater samples. People in these affected states have chronically been exposed to arsenic drinking arsenic contaminated hand tube-wells water. With every new survey, more arsenic affected Scientific Reports | (2022) 12:12414 | https://doi.org/10.1038/s41598-022-14802-w www.nature.com/scientificreports/ 10 flasks (250 mL) were arranged. First set of flasks was containing 100 mL of growth media while second set of flaks was containing 100 mL of growth media along with 5 prior washed luffa sponge discs of 2 cm in diameter and 2 mm in thickness. The media of each flask in both sets was inoculated with 0.5 mL of spore suspension followed by incubation at 25 °C and 100 rpm in an incubator shaker for 8 days. A schematic flow-sheet describing comprehensive methodology for the production of different forms of biomass was provided as supplementary material (Supplementary data: Figure S1). After 8 days, the flasks of set one produced the free fungal biomass (FFB), which was harvested through filtration. Whereas second set of flasks produced immobilized fungal biomass (IFB) over luffa sponge discs (step 1). Immobilization of the fungus onto luffa sponge has been performed according to our previous study 21 . FFB and IFB produced in step 1 were used for examining the effect of immobilization by comparing their efficiencies for As(III) removal.
In step 2, dried FFB and IFB were chemo-tailored by treating them with 0.1 N HCl. For chemical modification, 5 g of both the dried biomasses were individually suspended in 100 mL of 0.1 N HCl and incubated at the agitation condition of 100 rpm at 60 °C for 6 h. The resulting chemo-tailored FFB and IFB biomasses were collected and washed thoroughly with deionised double distilled water till pH of rinsed water was stabilized. The effect of chemo-tailoring on As(III) removal efficiency of biomasses was again analysed by keeping FFB as control. The chemo-tailored biomasses were modified through iron impregnation at pH range from1 to 9 as in step 3 in order to further improve the removal capacity of biomasses following the method described by 22 . Iron impregnation approach has been well implemented to enhance the sorption capacity and affiliated removal efficiency for As(III) in different studies 22,23 . After iron impregnation, chemo-tailored FFB provides chemo-tailored and iron-impregnated free fungal biomass (CFB) while chemo-tailored IFB yields immobilized and iron-impregnated fungal biomass (IIFB). The amount of impregnated iron per unit mass of CFB and IIFB was quantified by acid digestion method (U.S. EPA 3050B). Further, the As(III) removal was studied using these two biomasses (i.e. CFB and IIFB) by keeping FFB as control to comprehend the parallel effect of all the modifications i.e. immobilization, chemo-tailoring and iron impregnation.
Batch biosorption studies in shake flasks. The parameters affecting biosorption over the biomass viz.
pH, biomass dose, initial metal concentration, temperature and contact time were investigated by conducting individual biosorption experiments 24 . Preliminary experiments were conducted with FFB, CFB and IIFB in order to get the optimal values of aforesaid parameters, selecting them arbitrarily by following one factor at a time approach, in the range of pH 2 to 10, biomass dose 0.1 to 1.0 g/L, temperature 20 to 50 °C and metal ion concentration 0.1 to 2.0 mg/L. However, all the experiments were re-performed by varying one factor, while keeping others at their optima for examining real behavior of a parameter in relation to others, reproducibility of results and actual difference in performances of FFB, CFB and IIFB for As(III) removal. The final results, thus obtained, have been taken in to account and are discussed here. Experiments were performed in 250 mL conical flasks along with 100 mL of As(III) solution and desired amount of biomass followed by incubation under agitating condition at 100 rpm in rotary shaking incubator until the establishment of equilibrium. Samples were examined at regular time interval to evaluate the concentration of As(III) in the solution. All the experiments were conducted in triplicates and average values were taken for the data analysis.
In order to justify the non-electrostatic exchange mechanism during sorption of As(III), the zero point charge of FFB and IIFB was measured with the help of solid addition method discussed by 25 . Bio-filtration studies in up-flow packed bed bioreactor. Bio-filtration experiments in continuous mode were carried out in borosilicate glass bioreactor (diameter 2 cm, length 30 cm). IIFB discs were stacked between two supporting layers of glass wool of thickness 1 cm to attain certain bed heights. Approximately 2 cm length of bioreactor was packed with glass beads on both sides of glass wool for dispersing the flow and holding the glass wool in position. Metal solution of different initial concentration was pumped in an up-flow mode within the bioreactor at desired flow rate using a peristaltic pump (Miclins PP-10). Extracted samples were analyzed for residual As(III) concentration at regular basis. After the exhaustion of bed, 250 mL of deionized double distilled water was passed through bioreactor to wash out the un-adsorbed metal from the bioreactor. For desorption of attached metal and recharging the column, 10% NaOH was passed as eluting agent in an up-flow mode through the bioreactor at a flow rate of 1.66 mL/min. The effluent solution was collected and analyzed for elution efficiency.

Analysis of arsenic in aqueous solution.
Arsenic analysis in the sample solution was conducted using atomic absorption spectrophotometer (AAS, Shimadzu AA-6300). In order to produce the hydrides of the metal ions, basic unit of continuous flow hydride vapor generator (HVG, Shimadzu HVG-1) was coupled to the AAS. In order to examine arsenic, the hollow cathode lamp was employed as sole light source, adjusted at 1937 Aº wavelength, 12 mA lamp-current and 7 Aº slit width with deuterium lamp for background correction. For the generation of flame for the AAS instrument grade (98%) acetylene was supplied at 4.0 L/min at a pressure of 0.9 kg/cm 2 , along with compressed air delivered with a flow rate of 17.5 L/min and gas pressure of 3.5 kg/cm 2 . High purity (99.99%) argon was used as purge gas at a 70.0 mL/min flow rate and 3.2 ± 0.2 kg/cm 2 supply pressure.
Optimization of adsorption parameters. Response surface design (RSM). RSM is a common statistical tool for building quadratic models for providing the statistical relationship among the variables and determining the optimal process conditions. The four process variables viz. biomass dose, pH, temperature and initial As(III) concentration were chosen for experimental design to optimize As(III) biosorption by applying where β 0 is the constant, n represents the number of variables, β i is the slope or linear effect of the input variable X i , β ii denotes the quadratic effect of input factor X i and βij stands for linear by linear interaction effect between the input variable X i and X j .
Artificial neural network (ANN). ANN is experimentally proved advanced and more precise modeling technique in comparison to RSM as it comprises the nonlinearities of a fitness function significantly in a unique way 26 . Thus, a neural network with feed forward and back propagation architecture (FFBP) was employed by utilizing the experimental dataset generated through CCD to optimize the process parameters more precisely (Fig. 1). The MATLAB software (Version R2014b, Mathworks, Inc., USA) was applied for ANN-GA based modeling and data simulation. The topology for neural network was composed of three layers of neurons: an input, a hidden and an output layer. All these layers were interlinked together through a weighted connection in the forward direction. The experimental data of CCD was subdivided into two sets, training (24 experiments) and testing set (7 experiments). ANN algorithm was trained utilizing the training data, while the ANN performance was estimated on test data sets. The input layer of the network just supplied input data to the nodes of hidden layer through some weights. In hidden layer, sum total weight of inputs was estimated as by the equation (Eq. 2): Here x i (i = 1, n) denotes for input dataset, w i stands for the correlated weight and θ represents bias. Thereafter, the weighted output was further transmitted by hidden neurons through an activation function. Primarily weight and bias values were selected arbitrarily; but, in successive training steps, the weights and biases in hidden and output layers were synchronized with a convergence standard to obtain resemblance in training and testing experimental outcome. The logistic output function used in this simulation can be shown by the following equation (Eq. 3): www.nature.com/scientificreports/ The learning of the neural network was conducted initially until it approached to an optimal value of coefficient of correlation (R 2 ). The goodness of fit of created model can be represented significantly by R 2 value. The statistical difference between experimental and predicted values of parameters through computational model is further revealed by mean square error (MSE) values.
GA optimization. After building up a generalized form of feed forward ANN model, its output values (i.e. weights and bias) was optimized by genetic algorithm (GA) as the following output function (Eq. 4): where weights 0 represents weight on connections between hidden and output layer. However, weights H expresses the weight on connections between input and hidden layer. GA creates an initial population of arbitrarily built chromosomes (expressed as series of binary digits) and pursues the analogy of natural selection to resolve the non-linear equations 26 . Further, in order to produce new populations of chromosomes from pre-existing one, three genetic operations: selection, crossover and mutation were introduced (Fig. 1). Selection operator facilitates to pick the best chromosomes as breeding parents. Crossover adopts parental-pairs of selection step and helps them to produce offspring. While, the mutation operator slightly modifies the parental string so as to diversify the generation within a population 27 . The GA process went on continuously until it approached to an approximate optimum solution 28 . The GA operations reiterated several times by varying the input space parameters until the accomplishment of suitable results 19 .

Theory and calculations
Expressions for the evaluation of uptake capacity of biomass, percentage removal of As(III), reaction kinetics, sorption isotherms, thermodynamic constants and the parameters for packed bed bioreactor have been enlisted in Table 1 21,29,30 .

Results and discussion
Effect of Immobilization biomass growth and As(III) biosorption. FFB and IFB produced in step 1 as described in material and method section were investigated for the effect of immobilization on growth of biomass. After comparison of dry weighs of FFB and IFB, it was revealed that biomass yield was 18% higher for IFB. It was because luffa sponge provided good support and prevented compact pellet formation resulting into better nutrient supply and rapid growth of fungal mycelia over it.
FFB and IFB were then investigated for the effect of immobilization on As(III) biosorption at arbitrarily fixed experimental condition. The comparative biosorption studies have illustrated that the immobilization had increased the As(III) removal by 4.9% and uptake capacity by 0.05 mg/g in case of IFB (Supplementary data: Table S1). It occurred due to that biomass, after immobilization, becomes less compact as compared to FFB pellets and so the effective surface area for sorption increases. In addition to this, the reduced compactness also increases the accessibility of binding sites over fungal surface which results in higher uptake of the metal 31 .
Effect of chemical-modification on As(III) biosorption. The chemo-tailored FFB and chemo-tailored IFB produced in step 2 (Supplementary data: Figure S2) were investigated for analysing the effect of chemotailoring on percentage As(III) removal and uptake capacity of biomass by keeping unmodified biomasses (FFB and IFB) as control. The experimental conditions were fixed arbitrarily. As in results, removal of As(III) and uptake capacity was increased by 6.2% and 0.07 mg/g in case of chemo-tailored FFB, whereas these were 6.8% and 0.07 mg/ in case of chemo-tailored IFB respectively (Supplementary data: Table S1). The increment in uptake capacity of chemo-tailored biomasses might be because of generation of additional binding sites over fungal surfaces by acid mediated cleaning of the surface by replacing the pre-bounded ionic species with surface groups 32 . Moreover, mild acid hydrolysis of chitin present in fungal cell wall produces purer amino sugars (D-glucosamine) onto the fungal biomass surface which increases the sorption of metal anions, as mentioned in different reports 21,32 . Figure S2 (Supplementary data) shows the impregnation efficiency of iron on CFB and IIFB as function of pH. From the results, it was clear that impregnated quantity of iron onto the biomasses was strongly dependent on the initial solution pH. The highest impregnation efficiencies of CFB and IIFB were 10.1 and 10.5 g/kg of biomass, respectively at pH 2. Further rise in pH from 2 to 9, the impregnation efficiency abruptly reduced in both the cases. This might be due to high solubility and high diffusion efficiency of soluble iron from bulk solution to biomass surface at lower pH 22 . Higher impregnation efficiency of IIFB was due to its low compactness as compared to CFB, which results in better diffusion of soluble iron on to the biomass surface.

Quantification of impregnated iron and its effect on As(III) biosorption.
The effect of iron impregnation over As(III) biosorption was investigated by comparing the percentage As(III) removal and uptake capacities of iron impregnated biomasses (CFB and IIFB) with that of their respective chemotailored biomasses at arbitrarily fixed experimental conditions. The As(III) removal as well as uptake capacities were found to be increased by 32.7% and 0.32 mg/g in case of CFB, whereas those were 33.0% and 0.33 mg/g  Table S1).

Effect of pH and sorption mechanism.
The effect of pH on the biosorption of As(III) was observed by varying pH from 1.0 to 10.0 and maintaining other factors constant at; temperature at 40 °C, initial As(III) concentration of 1.0 mg/L, biomass dose 0.7 g/L and contact time of 135 min. The results thus obtained represented in Fig. 2a showed that the optimum pH for maximum percentage removal and maximum uptake of As(III) for FFB was 6.0 and for CFB and IIFB was 7.0. Figure 2a represents that FFB showed maximum percentage removal 58.1% and maximum uptake of 0.83 mg/g at pH 6.0, whereas CFB and IIFB represent maximum percentage removal of 87.5% and 92.4% along with maximum uptake 1.25 and 1.32 mg/g respectively at pH 7.0. It was also found that in these cases, the percentage removal increased initially with rising pH from 1.0 to optimum value, thereafter it was decreased quickly with rise in pH from optimum to 10.0. This profile can be clarified through the aqueous chemistry of As(III), non-electrostatic exchange mechanism and zero point charge (pHzpc) of biomass. It has been reported in aqueous chemistry of arsenic that, As(III) exists in two forms, non-ionic 'H 3 AsO 3 0 ' and anionic 'H 2 AsO 3 − ' at pH range 3-11, as described in following equation (Eq. 5) 33 . Formulae for calculation of batch biosorption parameters and models 21 Uptake capacity Percentage removal Ci × 100 q t : Uptake capacity (mg/g) at time t C i : Initial metal ion concentration in solution (mg/L) C t : Remaining metal ion concentration in solution (mg/L) V: Volume of solution (L) W: Weight of biosorbent (g)

Sorption kinetics
Pseudo first-order model Pseudo second-order model Initial sorption rate (h) k s : Equilibrium rate constant (min -1 ) k 2 ': Equilibrium rate constant (g/mg/min) h: Initial sorption rate (mg/g/min) www.nature.com/scientificreports/ Both of these As(III) species are found to be sorbed on the surface of sorbent through non electrostatic exchange mechanism, wherein hydroxyl ions act as an exchangeable legend for H 3 AsO 3 0 and water molecule for H 2 AsO 3 -. Both of the species ultimately bounded with amino and hydroxyl groups present over the surface of biomass. Similar mechanism has already been established by other workers also 33,34 .

Calculation of fixed bed bioreactor parameters
Further to justify that non electrostatic exchange mechanism was involved during sorption of As(III), experiments were conducted to measure the zero point charge of FFB and IIFB with the help of solid addition method 25 . The results of zero-point charge (Fig. 2b) showed that pHzpc of FFB was 6.3 and pHzpc of CFB was 7.2, which were very close to optimum pH for maximum sorption. These results indicated that neutral surface favors maximum sorption. Zero-point charge also explained that with increase in pH from 1.0 to optimum (6.0 or 7.0) the surface protonation decreases and this decrease in protonation leads towards lowering of surface charge, which will support non-electrostatic legend exchange interaction between sorbate and sorbent. Further increase in pH sifted from optimum to alkaline range, the biomass surface attains repulsive negative charge leading to decrease in sorption. In addition to this, at alkaline pH hydroxyl anions becomes dominant, which will compete with anionic species of As(III) (H 2 AsO 3 ) resulting towards lower uptake of As(III) 35 .
It is worth mentioning that major component of fungal cell walls is chitin, which plays a major role during the sorption process. Probable non-electrostatic ligand exchange mechanism for arsenic sorption onto the virgin and iron impregnated fungal biomass surface (chitin) is also illustrated by schematic diagram as shown in Fig. 3. FTIR study of unmodified biomass, modified biomass prior and after the sorption, also strengthens the proposed mechanism.
Effect of biomass dose. The effect of FFB, CFB and IIFB biomass dose on As(III) uptake was examined at different biomass doses varying from 0.1 to 1.0 g/L and maintaining other factors constant at; 1.0 mg/L initial metal ion concentration, 40 °C of temperature, contact time of 135 min and pH 6 for FFB while pH 7 for CFB and IIFB.
Results shown in Fig. 4a clearly depicts all forms of biomass i.e. FFB, CFB and IIFB that when biomass dose increased from 0.1 g/L to 1.0 g/L, the uptake capacity of FFB, CFB and IIFB decrease from 2.05 to 0.62 mg/L, 4.0 to 0.91 mg/g and 4.2 to 0.95 mg/g respectively. While percentage metal removal was increased from 20.5% to 58.1% for FFB, 39.6% to 87.5% for CFB and from 42.0% to 92.4% for IIFB respectively.
This can be explained that the uptake capacity of biomass depends upon metal-to-binding sites ratio. As the biomass dose increases, aforesaid ratio decreases leading in depletion of uptake capacity of biomass. However, at www.nature.com/scientificreports/ higher biomass dosage the available metal ions becomes inadequate to cover all the vacant sites entirely present on the biomass surface, as a result fewer metal ion gets adsorbed at per unit weight of biomass finally lowering the uptake capacity of biomass. In contrast, the percentage As(III) removal increases with increase in biomass dose because the efficiency of As(III) removal shows direct correlation with figure of binding sites i.e. greater the number of available binding sites, greater is the percentage removal. However, at excessive biomass dose, the adsorption is regulated by equilibrium and not by available binding sites. Therefore, beyond a certain limit of biomass dose, the percentage removal curve attains a plateau which represents the optimum biomass dose, in this study it was found to be 0.7 g/L.
Effect of contact time. The time of equilibrium for current sorption system was evaluated by measuring the residual As(III) concentration in solution until equilibrium achieved i.e. up to accomplishment of constant metal concentration in solution. These experiments were performed by regular screening of the change in metal ion concentration in the solution at initial pH 6.0 for FFB, 7 for CFB and IIFB, initial metal ion concentration of 1.0 mg/L, biomass dose 0.7 g/L and temperature 40 °C. This was observed from results ( Fig. 4b) that primarily rapid sorption occurs and equilibrium is accomplished after 120 min for CFB and IIFB however equilibrium is established after 135 min for the case of FFB. It clearly indicates that the sorption process occurs faster for CFB and IIFB as compared to untreated biomass (FFB). This property of CFB and IIFB is due to availability of greater numbers of binding sites over these biomass surfaces in comparison to FFB and thus iron impregnation improves the affinity of biomasses for arsenic removal.  Figure 4c illustrates that at very low metal ion concentrations the As(III) removal was 100% whereas the uptake capacity of biomass was minimum at corresponding concentrations. In contrast to this, as initial metal ion concentration increased, the percentage removal progressively decreased while uptake capacity of biomass augmented to maxima attaining a state of equilibrium. Further increase in metal ion concentration did not show effect on uptake capacity since at optima all the binding sites present on biomass surface are fully saturated with metal ions and having no availability for excess metal ions. The optimum value of metal ion concentration was found to be 0.9 mg/L for FFB and 1.0 mg/g for CFB and IIFB, beyond which the uptake capacity of biomass did not increase. It clearly illustrated that numeric value of binding sites on FFB was lower than that of CFB and IFB, which got saturated at comparatively lower concentration.  Table 2). Based on the experimental outputs of CCD, a polynomial equation was generated showing the correlation of maximum percentage As(III) removal (Y) with the process parameters, viz. biomass dose (X 1 ), pH (X 2 ), temperature (X 3 ), and Initial As(III) concentration (X 4 ). The quadratic equation of RSM model can be given as below: The predicted maximum of the metal removal by RSM was 95.2%, where X 1 = 0.7 g/L, X 2 = 6.62, X 3 = 39 °C, X 4 = 1.1 mg/L. Based on statistical data analysis, the value of R 2 and adjusted R 2 achieved 0.9656 and 0.9318, respectively. However, the mean percentage error was 7.23. Under the optimized conditions, the removal of As(III) found experimentally was 94.3 ± 0.2%, which was in accordance with the predicted RSM model.
The hybrid ANN-GA based model. A neural network topology was designed for current biosorption system in order to refine the RSM optimized parameters and to represent the nonlinearities of the fitness function in a much better way by using ANN toolbox of MATLAB software (Version 7.0, Mathworks, Inc., MA, USA). The model was built with four input (viz. biomass dose, temperature, pH, and initial As(III) concentration) and one output parameter i.e. percentage As(III) removal by using error back-propagation method. CCD data was partitioned into the training and test set for keeping the network away from over-training and over-parameterization. In order to find out the best correlation between experimental and predicted results, the neural network was trained initially with minimum number of neurons (started with five) and further, step by step the number of neurons was varied up to a limit. The network training was also done for different sets of specialized functions such as random initialization, learning rate. The generalization ability of network was determined by assorting the weights generating least test set MSE. The data transfer commands for connecting input to hidden layer and hidden to output layer were utilized as tansig and purelin, respectively. Some key statements of the algorithm as already described in our other publications 36 were: %creating ANN: net = newff (minmax[input matrix]) 1,10 , {'tansig' , 'purelin'}, 'traingdx' , 'mse'); %training of ANN: net = train(net, input vector, target vector). %setting the maximum number of training epochs: net.trainParam.epochs = 10,000. %setting the performance of network in terms of 'mse': net.performFcn = 'mse' and %setting the 'mse': net.trainParam.goal = 0.001. %simulating the output function: Y = sim(net, input vector of testing dataset). The model was fully trained after 1393 epochs and the R 2 value between the model-predicted and desired As(III) removal with respect to entire data including training and test data was 0.9997 (Fig. 5a,b). The least MSE values for training (E trn ) and test data set (E tst ) were 0.023 and 0.031 respectively, which were achieved at the network topology 4-10-1 (i.e. hidden layer with 10 neurons) while learning rate was 0.5. The average prediction error (%) for training and test data set were 2.462 and 3.161, respectively. The lower values of R 2 , MSE and average prediction error (%) depicted that the ANN-based model had exhibited good approximation and generalization characteristics. Further, results are also comparable about ANN based prediction accuracy over regression model from the Table 2 confirms the better efficacy of ANN as compared to RSM as stated by others 26 .
Finally, genetic algorithm (GA) was implemented to optimize the input space of neural network model with objective of maximization of As(III) removal and it can be exhibited by following expression: Maximize Here, f denotes ANN based objective function; Y is maximum As(III) removal; W represents equivalent weight vectors; the L-dimensional decision vector, x, defines the input parameters (L = 4); and x L l and x U l indicates for the lower and upper limits on x l . All four process parameters, during GA optimization process, were confined with the followings constraints as described in experimental data listed in Table 2: The GA terms in the MATLAB software for the optimization of As(III) removal were fed with following functions: population type: double vector; original population size: 100; elite count: 20; cross over probability: 0.8; crossover function: @crossoversinglepoint; selection function: @selectionroulette; migration direction: forward; mutation function: @mutationgaussian; total generations: 100. For estimation of fitness score of a chromosome (candidate solution) in a random population, following mathematical expression was utilized: where εj = fitness score of jth candidate solution and ŷ j pred = As(III) removal as predicted by model in response to candidate solution.
GA based optimization process was iterated several times by changing the different input space parameters for best global solution. These reiterations at different GA input parameters verified that the entire searching space was surveyed thoroughly to obtain an optimum solution. Reoccurrence of similar optimal solutions for most of the input conditions confirmed that it is a global optimal solution. Finally, GA had resulted the optimum values of biomass dose, metal concentration, pH and temperature, which exhibited maximum As(III) removal. The best fit plot (Fig. 5c) generated during the iterations of GA over generations indicates that resulted solutions were gradually converged towards optimal value. At the ANN-GA optimized condition, As(III) removal was 99.2% at the condition of biomass dose 0.72 g/L, pH 7.31, temperature 42 °C, and initial As(III) concentration 1.1 mg/L (Fig. 5d). This result was authenticated by performing independent experimentation at the set of GA-specified optimum conditions. The percentage As(III) removal accomplished at the above experimental condition was 98.8 ± 0.12% which was in close accordance with the hybrid ANN-GA based prediction.
Results obtained in the present study was quite encouraging as compared to the previous studies done for biosorption of As(III) using different bio sorbents as shown in the Table 3.

Isotherm, kinetic and thermodynamic studies. Sorption isotherms. Sorption isotherm illustrates
the equilibrium between free and sorbed As(III) over the surface-liquid interface at a given temperature correlating the distribution of As(III) concentration at sorbate-sorbent interface and within aqueous phase with a certain adsorption pattern. The analysis of sorption pattern was performed by using Langmuir and Freundlich isotherm models at initial As(III) concentration varying in range of 0.1 mg/L to 1.1 mg/L while maintaining other parameters at their optimum values. Comparative higher values of correlation coefficients in results clearly indicated that Langmuir isotherm fitted better with the current sorption system (Fig. 6a, Table 4) and it showed that the sorption followed monolayer pattern with sorption energy uniformly distributed over biosorbent sur-  Kinetic studies. The studies of sorption kinetics suggest about the rate of solute elimination as well as the residence time of the sorbate at the solid-solution interface. Thus, sorption kinetic studies are pre-requisite for designing a continuous sorption system. These studies were carried out at optimum parameters and sampling was done at regular time interval to measure concentration variation profile of arsenic retained in the system as a function of time. In order to define the kinetic behavior of sorption system, two popular models viz. pseudo-first order and pseudo-second order pattern models were followed. Kinetic parameters and correlation coefficient for IIFB and FFB, shown in Table 4, indicates that As(III) sorption over FFB and IIFB follows the pseudo-second order kinetics model (Fig. 6b), which signifies that sorption process is directly proportional to As(III) con- www.nature.com/scientificreports/ centration and the square of the number of free sites over sorbent surface. It was also evident from better fit of pseudo-second order model with experimental data that the rate limiting step in current sorption system was chemisorptive valence force, produced during the exchange or sharing of electrons between sorbent and sorbate, coordination or via complexation 37 .
Thermodynamic studies. Thermodynamics of the current biosorption was studied along with varying temperature and constant optimal values of other parameters. Van't Hoff plots (lnKc vs. 1/T, Fig. 6c) were drown and values of correlated thermodynamic parameters were examined using equations from Table 1. The negative magnitude of ∆G at all the temperatures reflected the spontaneity of sorption process (Table 4). However, positive enthalpy change (ΔH) depicted that the process was endothermic in nature. Furthermore, change in entropy (ΔS) with positive sign signified about increasing the degree of freedom of metal ions during sorption process this might be owing to the randomness of the sorption at solid-liquid interface 38 .

Removal of As(III) by IIFB discs in packed bed bioreactor.
On the basis of performance in batch mode, IIFB was chosen as sorbet for packed bed bioreactor experiments and the ability of IIFB discs to act as a bio-filter was evaluated.
The bioreactor experiments were conducting in two sets: first with varying bed heights i.e. 5, 15, 25 cm and maintaining flow rate constant at 1.66 mL/min whereas second set of experiments with varying flow rate i.e. 1.66, 4.98, 8.30 mL/min and the bed height maintained at 25 cm. Breakthrough curves, Ct/C 0 vs. time, for both the sets of experiments and calculated parameters for biosorption in bioreactor are presented in Fig. 7a and Table 5 respectively. It has been observed from the results that when bed height was raised from 5 to 25 cm at constant flow rate of 1.66 mL/min, breakthrough time (t b ) increase from 510 to 1350 min and volume treated up to break through point (V b ) increase from 846.6 to 2241.0 mL It is explained from the fact that at lesser bed heights axial dispersion phenomena predominated in the mass transfer, resulting into reduction in diffusion of the metal ions i.e., metal ions will not have sufficient time to diffuse within whole sorbent 39 . Thus, increase in bed height minimizes this limiting factor i.e. provides enough time to metal ions to diffuse in to biosorbents mass leading to increase in breakthrough volume of treated water. Whereas results for second set of experiments depicts that t b decrease from 1350 to 250 min and V b decrease from 2241.0 mL to 2075.0 mL when flow rate was increased from As in a successful biosorption process, reusability of sorbent is an essential requirement which greatly reduces the process cost and opens the possibility of recovery of sorbate. In present study, the reusability of IIFB discs was evaluated up to four successive adsorption-desorption cycles using 10% NaOH solution as eluting agent. The results have been shown in Fig. 7b and calculated parameters for these adsorption-desorption cycles are summarized in Table 5. Results shows that breakthrough time, uptake capacity, elution time, and percent elution efficiency decrease with each consecutive cycle and sifts from 1350 to 1010 min, 0.82 mg/g to 0.62 mg/g,  IIFB disc as sorption based bio-filter. Normal filter consists of a filtration matrix, which separates impurities from water by using a fine physical barrier, a chemical process or a biological process. Here, in case of IIFB based filter, a filtration matrix is present in form of packed biomass which successfully blocks the passage of inlet As(III) through a physico-chemical biosorption process (Fig. 7c). The concentration of As(III) present in inlet water stream decreases gradually through the IIFB column since these ions get bounded with the active binding sites present over filtration matrix. A suitable bed height of the column allows complete entrapment with no As(III) present in the outlet stream. Figure 7d schematically represents the filter exhaustion profile at different times and corresponding sections (x, y and z) of breakthrough curve. Section x denotes complete removal, y shows incomplete removal i.e. intermediate stage of exhaustion whereas section z represents no removal and complete exhaustion of the filter bed where recharging required.
SEM EDX and FTIR analysis. The interwoven fibers within a luffa disc and immobilized fungal biomass over it are clearly visible as in Fig. 8a,b, respectively. SEM images have recorded at the accelerating voltage condition of 5.0kVand beam current 1.0 nA. SEM photograph of interwoven fibers in a luffa sponge was shown in Fig. 8c. Figure 8d-f represents the typical energy dispersive pattern of raw, modified and arsenic loaded modified biomass along with its SEM micrograph. Iron signal in case of modified biomass (CFB, Fig. 8e) confirms the iron impregnation on to the biomass surface whereas arsenic peak present in Fig. 8f denotes the biosorbed As(III) onto the surface of biomass. FTIR spectra for of FFB and IIFB were obtained for the range of 4000-400 cm −1 , prior after As(III) biosorption, using IR spectrophotometer (Perkin Elmer FTIR-1600) in order to analyze the groups present over the surface of biomass, changes after chemical treatments and groups involve in As(III) sorption. IR spectra shown in Fig. 9 represents different absorption bands, the wide band around 3200-3400 cm -1 represents -NH stretching of the protein as well as -OH groups of the glucose and the acetamido group of chitin fraction 38,40 . Absorption peak at 2949.80 cm −1 designate the -CH stretching. The absorption peak at 1651.32 cm −1 can be a characteristic to amide bonds of protein and the peak at 1550.62 cm −1 specifies for primary and secondary amines and amides (N-H bending). The absorbance band around 1404.51 cm −1 attributed to N-H stretching, O-H deformation and -CH 2 scissoring 41 . Peak at 1371.76 cm −1 is attributed to sulfamide bonds (S=O). Peaks around 1035.65 cm −1 and 1080.38 cm −1 represent -CN stretching vibration of chitin and protein fraction. Chemo-tailoring and ironimpregnation, sharpens the -NH stretching and polymeric association band (3200-3400 cm -1 ), which would be owing to emergence of moderately pure amino sugars (D-glucosamines), produced by partial hydrolysis of chitin 42,43 . Along with this, a new strong peak appears at lower frequency around 527 cm −1 after iron impregnation, which signifies the iron oxide skeleton 44 . After As(III) adsorption, following siftings in peaks were observed; peak of -NH group sifted from 1651.12 to 1626.99 cm −1 , peak of O-H deformation shifted from 1404.51 to 1446.37 cm −1 and peak of iron oxide skeleton shifted from 527.90 to 555.70 cm −1 . Which indicate that hydroxyl and amino groups participates in As(III) biosorption. Moreover, the sharpening of band around 3200-3400 cm −1 further confirms the involvement of-OH groups in As(III) biosorption.

Conclusion
The main aim of this work was effective removal of As(III) from water with the help of a sorption based bio-filtration system. The objective was fulfilled through the development of a biosorbent material by immobilization of dead fungal biomass over luffa sponge. The efficiency of the biosorbent was further improved by chemo-tailoring Table 5. Sorption data for As(III) sorption in up-flow packed-bed bioreactor containing IIFB discs as matrix.

Bed height (cm)
Flow rate (mL/min) t b (min) t e (min) V b (mL) V e (mL) q tot (mg) q (mg/g) EBCT (min) www.nature.com/scientificreports/ and iron impregnation. The finally developed biosorbent (i.e. IIFB) was found to be much efficient in batch experiments with 99.2% As(III) removal. The biosorbent, proved to be an excellent filtration biomaterial in a continuous system within a packed-bed bioreactor, because it showed complete removal of As(III) up to 1350 min (≈ 22.5 h) at a bed height of 25.0 cm with a flow rate of 1.66 mL/min and treated 2241.0 mL of water successfully.
The system could be termed as a biofilter as it contained microbial biomass attached to a supporting matrix. The  www.nature.com/scientificreports/ overall process involved can be explained with the help of Fig. 10. Though, the biofilter used does not possess metabolic activity but it has filtration ability due to its excellent sorptive properties, it would be better to refer this system as a sorption based fungal biofilter. The involved non-electrostatic legend exchange sorption mechanism for As(III) sorption onto the hydroxyl and amino groups of biosorbent was proposed and established also through pHzpc, FTIR, SEM-EDX analyses and aqueous chemistry of arsenic. Moreover, the sorption characteristics were explored and it was found that sorption followed monolayer pattern in an endothermic way with a spontaneous and pseudo-second order kinetic path. The Biofilter designed could also play significant role in removal of chances of toxicity of other heavy metals in the water as well as As(III) which might be present in Municipal Solid Waste Compost (MSWC) prepared from landfill sites.

Data availability
All data generated or analysed during this study are included in this published article [and its supplementary information files].