Green Synthesis of Silver Nanoparticles Using Spent Coffee Ground Extracts: Process Modelling and Optimization

Large amounts of spent coffee grounds (SCGs) are produced annually worldwide. SCGs contain high levels of phenolics and other bioactive compounds that make them a potential source of reducing and stabilizing agents for the synthesis of metal nanoparticles. This study investigates the use of SCG extracts as a green strategy to produce silver nanoparticles (AgNPs). SCG extracts were obtained using aqueous ethanol as the solvent and then contacted with a silver nitrate solution under the selected conditions. A central composite design coupled with response surface methodology was used to evaluate the effects of solvent composition (C = 30–70% v/v), silver-to-phenolic ratio (R = 3–7 mol/mol), temperature (T = 25–55 °C) and pH (10–12) on the production of AgNPs. Characterization of AgNPs by DLS, TEM and XRD techniques showed that they were highly crystalline with a narrow size distribution. Under optimal reaction conditions, AgNPs with an average size of about 10 nm and a zeta potential of −30.5 to −20.7 mV were obtained. Overall, the results of this study indicate that SCGs are a promising material for the green synthesis of small-sized and stable AgNPs.


Introduction
Silver nanoparticles (AgNPs) continue to attract attention from researchers due to their unique antibacterial, cytotoxic and catalytic properties, which make them ideal candidates for a variety of medical and non-medical applications [1][2][3]. This has stimulated interest in the development of new methods of synthesis in which toxic chemicals are replaced by natural compounds. Most of the proposed methods are based on the use of bioactive compounds from plants [4], with a special recent focus on the possibility of combining the principles of green chemistry with those of circular economy [5]. In this regard, the synthesis of AgNPs using agro-industrial or food wastes as a source of reducing and stabilizing agents could represent a valuable opportunity. Examples of this approach are the production of AgNPs using extracts from rice husks [6], bilberry and red currant wastes [7], tamarind shells [8] and mandarin peels [9].
Plant-derived materials contain several bioactive compounds, such as terpenoids, polyphenols, alkaloids, sugars, phenolic acids and proteins, that can reduce metal ions, allowing the formation of nanoparticles [10]. Some of these compounds also behave as capping agents, binding to the surface of nanoparticles and controlling their size and shape [11]. The degree of stabilization depends on the molecular features of the capping agent, such as the type, number and relative position of the functional groups interacting with the nanoparticles [12,13]. The synthesis of metal nanoparticles using plant materials has several advantages over traditional chemical and physical methods, including, among others, the absence of toxic compounds, the simplicity of the production process and the use of mild operating conditions [4,10,14]. Furthermore, studies on AgNPs suggest that the

Synthesis of AgNPs
AgNPs were synthesized in 20-mL glass vials thermostated at the selected temperature and magnetically stirred at 400 rpm. The vials were loaded with appropriate amounts of 0.1 M silver nitrate solution and SCG extract. The pH of the reaction solution was adjusted to the desired value by addition of NaOH. The formation of AgNPs was monitored spectrophotometrically by measuring the intensity of the surface plasmon resonance (SPR) band of silver at about 430 nm. All experiments were made at least in duplicate. The reaction temperature was varied between 10 and 70 • C, the pH between 9 and 13, and the silver-to-phenolic ratio, defined as the ratio of Ag + concentration to total phenolics, between 1 and 9 mol/mol.

Analytical Methods
Total phenolics were determined by the Folin-Ciocalteu method according to the procedure reported in [26]. The results were expressed as gallic acid equivalents (GAE) using a calibration curve obtained with gallic acid standards.
An X'Pert PRO diffractometer (Philips, Eindhoven, The Netherlands) was used for XRD measurements. The instrument was operated at 40 kV and 30 mA with Cu Kα radiation (λ = 1.5406 Å). The 2θ angle was varied from 20 • to 80 • . The step size was 0.04 • and the counting time was 20 s per step.
Nanoparticle size and zeta potential were measured using a Litesizer™ 500 instrument (Anton Paar, Graz, Austria).
Transmission electron microscopy (TEM) images were obtained with a Zeiss EM10 instrument (Carl Zeiss, Thornwood, NY, USA) operated at 60 kV. Samples were prepared by placing a drop of the nanoparticle solution onto a standard carbon-coated copper grid. Images were analyzed by the ImageJ software (ImageJ, National Institutes of Health, Bethesda, MD, USA).

Experimental Design
An experimental approach based on a Central Composite Design (CCD) [27] was used to investigate the effects of solvent composition (C), namely, the volume percentage of ethanol in the solvent, silver-to-phenolic ratio (R), temperature (T) and pH on the production of AgNPs. In all experiments, the reaction time was fixed to 5 h. The CCD consisted of a full two-level factorial design (2 4 points), eight axial points at a distance ±α from the central point and six replicates of the central point. The value of α was taken as (2 4 ) 1/4 = 2 to ensure the orthogonality and rotatability of the design.
The variation range of each factor was determined based on the results of preliminary experiments and on previous studies. Then, the factor levels to be explored were evaluated. They are reported in Table 1 in actual and coded values. According to the DOE (Design of Experiments) terminology, actual levels denote the values of factors in dimensional units, while coded levels indicate their corresponding dimensionless values. Actual (X i ) and coded (x i ) levels are linked by the following relation: where X i,0 is the value of the i-th factor at the center-point level and ∆X i is the step change value for that factor. The intensity of the SPR band at λ max was taken as the response variable. Overall, the CCD consisted of 30 runs (Table 2), which were performed in random order to minimize the effects of uncontrolled factors.
The statistical analysis of the results was performed using the Design-Expert ® software (version 7.0.0, Stat-Ease, Inc., Minneapolis, MN, USA).

Production of SCG Extracts
The moisture content of SCGs was 60.5 ± 1.8 wt% and was reduced to 6.4 ± 0.1 wt% after drying. The phenolic content was 30.25 ± 0.43 mg GAE/g, a value that is in line with those reported in previous studies [21,22,25,28], confirming that SCGs are an important source of phenolic compounds.
Preliminary experiments were made to investigate the effects of solvent composition on the extraction of phenolic compounds. The results shown in Figure 1 indicate that the highest extraction yields were achieved for an ethanol concentration in the solvent close to 50% v/v. However, since SCGs contain a variety of phenolic compounds with different Ag + reducing abilities, the solvent composition that maximizes the overall extraction efficiency can differ from that for the optimal production of AgNPs. In addition, co-extracted nonphenolic compounds from SCGs could also contribute to the formation and/or stabilization Preliminary experiments were made to investigate the effects of solvent composition on the extraction of phenolic compounds. The results shown in Figure 1 indicate that the highest extraction yields were achieved for an ethanol concentration in the solvent close to 50% v/v. However, since SCGs contain a variety of phenolic compounds with different Ag + reducing abilities, the solvent composition that maximizes the overall extraction efficiency can differ from that for the optimal production of AgNPs. In addition, co-extracted non-phenolic compounds from SCGs could also contribute to the formation and/or stabilization of AgNPs. For this reason, solvent composition was included in the list of factors and its central point value was set to 50% v/v.

Spectrophotometric Characterization of AgNPs
In Figure 2, some UV-Vis spectra of the solution containing silver nitrate and SCG extracts at different reaction times are displayed. The spectrum was dominated by a strong SPR band at about 430 nm, indicating the formation of AgNPs. The reaction was accompanied by a color change of the solution from pale yellow to deep brown and was completed within about 5 h. As is known, the intensity and sharpness of the SPR band can be related to the amount of nanoparticles produced and their size distribution, respectively. In particular, the sharper the peak, the narrower the size distribution [29].

Spectrophotometric Characterization of AgNPs
In Figure 2, some UV-Vis spectra of the solution containing silver nitrate and SCG extracts at different reaction times are displayed. The spectrum was dominated by a strong SPR band at about 430 nm, indicating the formation of AgNPs. The reaction was accompanied by a color change of the solution from pale yellow to deep brown and was completed within about 5 h. As is known, the intensity and sharpness of the SPR band can be related to the amount of nanoparticles produced and their size distribution, respectively. In particular, the sharper the peak, the narrower the size distribution [29].
source of phenolic compounds.
Preliminary experiments were made to investigate the effects of solvent composition on the extraction of phenolic compounds. The results shown in Figure 1 indicate that the highest extraction yields were achieved for an ethanol concentration in the solvent close to 50% v/v. However, since SCGs contain a variety of phenolic compounds with different Ag + reducing abilities, the solvent composition that maximizes the overall extraction efficiency can differ from that for the optimal production of AgNPs. In addition, co-extracted non-phenolic compounds from SCGs could also contribute to the formation and/or stabilization of AgNPs. For this reason, solvent composition was included in the list of factors and its central point value was set to 50% v/v.

Spectrophotometric Characterization of AgNPs
In Figure 2, some UV-Vis spectra of the solution containing silver nitrate and SCG extracts at different reaction times are displayed. The spectrum was dominated by a strong SPR band at about 430 nm, indicating the formation of AgNPs. The reaction was accompanied by a color change of the solution from pale yellow to deep brown and was completed within about 5 h. As is known, the intensity and sharpness of the SPR band can be related to the amount of nanoparticles produced and their size distribution, respectively. In particular, the sharper the peak, the narrower the size distribution [29].

Model Fitting
Different mathematical models, including the linear, the two-factor interaction, the quadratic and the cubic models, were tested for their ability to fit the experimental design data. The best results were obtained with the quadratic model: where y is the intensity of the SPR band at λ max and x i are the coded independent variables. In addition to the intercept (a 0 ), Equation (2) contains 4 linear (a i ), 4 quadratic (a ii ), and 6 interaction (a ij ) coefficients, for a total of 15 parameters. The full quadratic model was reduced by stepwise regression to keep only the statistically significant terms, while preserving the hierarchy of higher order terms. By this procedure, the following reduced model was obtained: The nine parameters were estimated by the least-square methods, giving the results listed in Table 3. The R-squared and adjusted R-squared values were 0.9637 and 0.9499, respectively. As can be seen from the ANOVA results shown in Table 4, the model was statistically significant (p < 0.0001) and that the lack-of-fit was not significant (p = 0.1972). Studentized model residuals were randomly scattered between -3 and +3 (Figure 3), indicating that the ANOVA assumptions were met. Table 3. Estimated coefficients for the reduced model described by Equation (2) with the associated standard errors (SE) and 95%-confidence intervals (CI).   Figure 4 shows the Pareto chart for the model coefficients. From this diagram, the following points can be made:

Analysis of Influencing Factors
(a) Two of the four investigated factors, namely, the ethanol concentration in the aque-  Figure 4 shows the Pareto chart for the model coefficients. From this diagram, the following points can be made:

Analysis of Influencing Factors
(a) Two of the four investigated factors, namely, the ethanol concentration in the aqueous solvent (C) and the pH of the reaction medium, affected the production of AgNPs through both a linear and a quadratic term; (b) Concerning the linear terms, the silver-to-phenolic ratio (R) gave a negative contribution on the response variable, while the contributions of the remaining factors were positive and increased in in the order: C < T < pH; (c) There were two positive interactions: between the silver-to-phenolic ratio (R) and the temperature (T), and between the silver-to-phenolic ratio (R) and the pH, indicating that the R factor had a more pronounced effect on the production of AgNPs at higher temperature and pH.  Figure 4 shows the Pareto chart for the model coefficients. From this diagr following points can be made:

Analysis of Influencing Factors
(a) Two of the four investigated factors, namely, the ethanol concentration in th ous solvent (C) and the pH of the reaction medium, affected the production of through both a linear and a quadratic term; (b) Concerning the linear terms, the silver-to-phenolic ratio (R) gave a negative bution on the response variable, while the contributions of the remaining were positive and increased in in the order: C < T < pH; (c) There were two positive interactions: between the silver-to-phenolic ratio (R) temperature (T), and between the silver-to-phenolic ratio (R) and the pH, ind that the R factor had a more pronounced effect on the production of AgNPs a temperature and pH. For a better appreciation of the effects of the investigated factors on the pro of AgNPs, an analysis of perturbation and response surface plots was performed.
Perturbation plots allow a comparison of the individual effects of factors on sponse variable at a specific point in the design space. Each factor was changed full factorial range (−1, 1), while setting the other factors to their central point val As is evident from Figure 5, the response variable varied non-monotonically with  For a better appreciation of the effects of the investigated factors on the production of AgNPs, an analysis of perturbation and response surface plots was performed.
Perturbation plots allow a comparison of the individual effects of factors on the response variable at a specific point in the design space. Each factor was changed over its full factorial range (−1, 1), while setting the other factors to their central point values (0). As is evident from Figure 5, the response variable varied non-monotonically with solvent composition and pH, which is consistent with the presence of quadratic terms for these factors in the model equation. Concerning the silver-to-phenolic ratio and the temperature, a negative linear dependence and a positive linear dependence were, respectively, observed. From the slopes of these lines, the higher sensitivity of the response variable to temperature can also be deduced.
Response surface and contour plots were generated to visualize the combined effects of influencing factors. Some representative plots, obtained by setting two of the four factors constant at their central point values, are shown in Figures 6 and 7. The curvature of the response surfaces reflects the quadratic effects of factors and their shape also provides an indication of the interaction between factors. composition and pH, which is consistent with the presence of quadratic terms for these factors in the model equation. Concerning the silver-to-phenolic ratio and the temperature, a negative linear dependence and a positive linear dependence were, respectively, observed. From the slopes of these lines, the higher sensitivity of the response variable to temperature can also be deduced.  ture, a negative linear dependence and a positive linear dependence were, respectively, observed. From the slopes of these lines, the higher sensitivity of the response variable to temperature can also be deduced.

Process Optimization
An examination of the model structure (Equation (2)) and its 3D and 2D graphical representations (Figures 6 and 7) clearly indicate that the production of AgNPs can be optimized by appropriate selection of process conditions. The search for the optimum was performed over the whole investigated domain (-α ≤ xi ≤ α) by maximizing the response variable. For this purpose, the gradient descent method was used with different randomly selected starting point. The following result was obtained: C = 49.2% v/v; R = 8 mol/mol; T = 65 °C; pH = 12.2. The predicted response was 3.836, with a 95% confidence interval of 3.620-4.053. The model was validated by performing a new experiment under these optimum conditions, which gave: yexp = 3.909 ± 0.08. This value differs by about 1.8% from the predicted one and falls within the 95% confidence interval, further supporting the reliability of the developed model. Figure 8 shows the XRD pattern of the synthesized AgNPs. In the 2θ range of 20-80°, there were four sharp peaks at 38.1°, 44.2°, 64.5° and 77.5°, corresponding, respectively, to the planes (111), (200), (220) and (311) of the FCC structure of silver. The peak at 38.1° was the most intense, indicating that the preferred orientation was along the (111) plane.

Characterization of AgNPs
The average nanoparticle size obtained from dynamic light scattering (DLS) measurements ranged from about 23.7 to 39.9 nm, depending on the reaction conditions, while the zeta potential varied between −30.5 and −20.7 mV.
Some TEM images of the produced nanoparticles are displayed in Figure 9. These images reveal that the AgNPs were spherical. The average nanoparticle size derived from the analysis of TEM images ranged from about 8.2 to 10.4 nm. These values are smaller than those from DLS, which can be attributed to the fact that DLS measures the hydrodynamic diameter of nanoparticles rather than the physical one [30]. Furthermore, the capping layer around the nanoparticles is transparent to electrons and therefore does not contribute to the size obtained from TEM measurements [31].

Process Optimization
An examination of the model structure (Equation (2)) and its 3D and 2D graphical representations (Figures 6 and 7) clearly indicate that the production of AgNPs can be optimized by appropriate selection of process conditions. The search for the optimum was performed over the whole investigated domain (-α ≤ x i ≤ α) by maximizing the response variable. For this purpose, the gradient descent method was used with different randomly selected starting point. The following result was obtained: C = 49.2% v/v; R = 8 mol/mol; T = 65 • C; pH = 12.2. The predicted response was 3.836, with a 95% confidence interval of 3.620-4.053. The model was validated by performing a new experiment under these optimum conditions, which gave: y exp = 3.909 ± 0.08. This value differs by about 1.8% from the predicted one and falls within the 95% confidence interval, further supporting the reliability of the developed model.  The average nanoparticle size obtained from dynamic light scattering (DLS) measurements ranged from about 23.7 to 39.9 nm, depending on the reaction conditions, while the zeta potential varied between −30.5 and −20.7 mV.

Characterization of AgNPs
Some TEM images of the produced nanoparticles are displayed in Figure 9. These images reveal that the AgNPs were spherical. The average nanoparticle size derived from the analysis of TEM images ranged from about 8.2 to 10.4 nm. These values are smaller than those from DLS, which can be attributed to the fact that DLS measures the hydrodynamic diameter of nanoparticles rather than the physical one [30]. Furthermore, the capping layer around the nanoparticles is transparent to electrons and therefore does not contribute to the size obtained from TEM measurements [31].

Discussion
This study was designed to investigate the use of SCGs as a source of reducing and stabilizing agents for the synthesis of AgNPs. The results obtained indicate that SCGs are a suitable material for this purpose and that the production process can be optimized by operating at 65 °C and pH 12.2, using aqueous ethanol at 49.2% as the extraction solvent and a silver-to-phenolic ratio of 8 mol/mol.
To provide an interpretation of these results, the characteristics of SCG extracts should first be considered. SCGs are the solid residue left after brewing of coffee beans. The latter are obtained by subjecting the seeds of coffee cherries, known as green coffee beans, to roasting. Green coffee is mainly composed of carbohydrates (55-65.5%), lipids (10-18%), N-containing compounds (11-15%), purine alkaloids (0.8-4%), chlorogenic acids (6.7-9.2%) and minerals (3-5.4%) [32]. In addition to chlorogenic acids, other bioactive compounds, such as diterpenes, caffeine and trigonelline, are also present [33]. When the beans are roasted, the characteristic organoleptic properties of coffee are developed. The variables that most affect the quality of the final product are the treatment temperature (typically, 200-260 °C) and duration (typically, 5-15 min) [34]. During roasting, a series of chemical reactions occur, like pyrolysis and the Maillard reaction, that modify the composition of coffee beans [35,36]. In particular, some chlorogenic acids and other coffee components are degraded while new volatile and non-volatile compounds are formed.

Discussion
This study was designed to investigate the use of SCGs as a source of reducing and stabilizing agents for the synthesis of AgNPs. The results obtained indicate that SCGs are a suitable material for this purpose and that the production process can be optimized by operating at 65 • C and pH 12.2, using aqueous ethanol at 49.2% as the extraction solvent and a silver-to-phenolic ratio of 8 mol/mol.
To provide an interpretation of these results, the characteristics of SCG extracts should first be considered. SCGs are the solid residue left after brewing of coffee beans. The latter are obtained by subjecting the seeds of coffee cherries, known as green coffee beans, to roasting. Green coffee is mainly composed of carbohydrates (55-65.5%), lipids (10-18%), N-containing compounds (11-15%), purine alkaloids (0.8-4%), chlorogenic acids (6.7-9.2%) and minerals (3-5.4%) [32]. In addition to chlorogenic acids, other bioactive compounds, such as diterpenes, caffeine and trigonelline, are also present [33]. When the beans are roasted, the characteristic organoleptic properties of coffee are developed. The variables that most affect the quality of the final product are the treatment temperature (typically, 200-260 • C) and duration (typically, 5-15 min) [34]. During roasting, a series of chemical reactions occur, like pyrolysis and the Maillard reaction, that modify the composition of coffee beans [35,36]. In particular, some chlorogenic acids and other coffee components are degraded while new volatile and non-volatile compounds are formed.
SCGs contain several compounds with reducing activity, such as chlorogenic, gallic, caffeic, ellagic, ferulic, p-coumaric and p-hydroxybenzoic acids, esters of caffeic and ferulic acids with quinic acid, caffeine, rutin, quercetin and trigonelline [37]. In this study, ethanolwater mixtures at different ratios were used to recover bioactive compounds from SCGs. As is known, in a solid-liquid extraction process, the properties of the solvent strongly affect the overall extraction yield as well as the relative amounts of extracted compounds [38]. The results presented here showed that changes in the solvent composition affected both the production of AgNPs and the amount of phenolic compounds extracted from SCGs. Furthermore, the optimal solvent composition for nanoparticle production (C = 49.2% v/v) was very close to that of the maximum yield of phenolics extraction, which suggests that phenolic compounds are the main bioactive components involved in the synthesis of AgNPs.
Regarding the mechanism of nanoparticle formation, it should be considered that the reducing ability of phenolic compounds is related to the presence in their molecules of one or more hydroxyl groups [6,39]. In the case of AgNPs, silver ions are reduced to metals by transfer of electrons provided by the phenolic compounds [40].
For the generic phenolic compound (PC), the following electron transfer process, involving the oxidation of PC to its quinone form (PCQ) and the reduction of silver ions to metallic silver, can be assumed: n Ag + + n e − → n Ag 0 , Figure 10 shows the oxidation reactions for three of the most common phenolic acids in SCGs (chlorogenic acid, caffeic acid and p-coumaric acid). The positive effect of increasing pH on the production of AgNPs can be explained by the fact that H + ions are released during the oxidation of phenolic compounds (Equation (4)), which means that the nucleation of nanoparticles is favored under alkaline conditions. Nevertheless, at high pH, the production of Ag2O can also occur [43]: together with a weakening of the interactions of surface-capped molecules with the nanoparticle surface [39]. Both phenomena could lead to a progressive reduction in the positive effect of pH and be responsible for the occurrence of an optimum pH (in this case, pH 12.2) for the synthesis of AgNPs. The analysis conducted also revealed a positive influence of temperature on the synthesis of AgNPs, which can be primarily related to its positive effect on the reaction kinetics. However, due to the complexity of the reaction system and the presence in the reaction medium of many different compounds with reducing and/or stabilizing abilities, it is difficult to provide an adequate explanation to the observed behavior. The detected interaction between temperature and silver-to-phenolic ratio could be a reflection of this complexity. For example, temperature has a different impact on the reduction of silver ions by phenolic compounds and the formation of the coating layer to which these compounds also contribute [44,45]. Although the results of the present study do not allow us to iden- Combining reactions (4) and (5) gives the following overall reaction: According to the classical nucleation theory [41], the produced silver atoms (Ag 0 ) can aggregate to form metal clusters (or embryonic nuclei): Nuclei with a radius smaller than the critical one (r < r*) will dissolve in the solution, while those of greater radius (r ≥ r*) will grow up to form AgNPs. Capping agents can interact with the surface of nanoparticles and form a coating layer which prevents aggregation affecting, at the same time, the final size of the nanoparticles. In the case of phenolic acids, the build-up of this layer is determined by the binding of their carboxylic moieties to the nanoparticle surface followed by the formation of intermolecular hydrogen bonds between hydroxyl groups of surface-capped molecules [39]. However, other functional groups of phenolic compounds could also play a role in stabilization. This is the case, for example, of catechol groups of some phenolic acids, which can easily adsorb on metal surfaces [42].
The positive effect of increasing pH on the production of AgNPs can be explained by the fact that H + ions are released during the oxidation of phenolic compounds (Equation (4)), which means that the nucleation of nanoparticles is favored under alkaline conditions. Nevertheless, at high pH, the production of Ag 2 O can also occur [43]: together with a weakening of the interactions of surface-capped molecules with the nanoparticle surface [39]. Both phenomena could lead to a progressive reduction in the positive effect of pH and be responsible for the occurrence of an optimum pH (in this case, pH 12.2) for the synthesis of AgNPs. The analysis conducted also revealed a positive influence of temperature on the synthesis of AgNPs, which can be primarily related to its positive effect on the reaction kinetics. However, due to the complexity of the reaction system and the presence in the reaction medium of many different compounds with reducing and/or stabilizing abilities, it is difficult to provide an adequate explanation to the observed behavior. The detected interaction between temperature and silver-to-phenolic ratio could be a reflection of this complexity. For example, temperature has a different impact on the reduction of silver ions by phenolic compounds and the formation of the coating layer to which these compounds also contribute [44,45]. Although the results of the present study do not allow us to identify their contributions, the methodology used can be adapted to include more response variables, such as the particle size and morphology, that could help elucidate the underlying mechanisms.

Conclusions
The results of this study indicate that small-sized and stable AgNPs can be obtained using SCGs as a source of reducing and stabilizing agents. The production process was investigated by a rigorous approach based on the response surface methodology, which provided the following set of optimal operating conditions: C = 49.2% v/v; R = 8 mol/mol; T = 65 • C and pH = 12.2.
The proposed process is very simple to perform and easily scalable. Furthermore, the use of aqueous ethanol as solvent and the exploitation of a plant waste make it environmentally friendly and compliant with the circular economy principles.