Substrate

A mixture of VW, FWCW, CP, SCJ, and water was used as the substrate. The selection of VW and FWCW was based on their availability as carbon sources and the presence of microorganisms with high potential for hydrogen production through DF, as reported by Cárdenas et al. (2020). The FWCW, obtained from the traditional processing method (fermentation for 24 hours), and coffee pulp (CP) were collected at Los Naranjos farm, located in the municipality of Pitalito (Huila) at 1,641 meters above sea level (masl) with an average temperature of 19 °C. The vegetable waste (VW) was obtained from the Pitalito marketplace, containing residues of lettuce, chard, cabbage, parsley, celery, and broccoli, residues in equal proportions, which were not suitable for human consumption. Meanwhile, sugar cane juice (SCJ) was obtained from a panela processing plant located in the municipality of Palestina, Huila. Prior to the experimental stage, preliminary fermentation trials (data not included in this document) were conducted to determine the proportions of the raw materials. From the trial with the highest hydrogen production, 2 L of residual sludge were collected and used as untreated inoculum, which was added to each treatment. The solid waste was crushed using a commercial blender (Oster® BLSTTDG-NBG) to obtain a homogeneous sample (Figure 1). Only the pH of the substrate was recorded (pH_i); no initial physicochemical characterization was performed.

Figure 1: Substrate and raw materials: A. Wastewater from the first coffee wash (FWCW); B. Coffee pulp (CP); C. Vegetable waste (VW); D. Sugar cane juice (SCJ); and E. Mixed Substrate (FWCW, CP, VW, and SCJ).

Bioreactor

It consisted of a 50 L high-density polyethylene container with a useful volume for 35 L substrate, which was placed inside another container with a capacity of 200 L and an opening at the top. The space between the two containers was filled with water, which was heated by three electrical resistors of 800 W (127 V), until reaching a temperature of 35 °C in the bioreactor. Water and substrate temperature were recorded with six sensors; four were DS18B20 and two PT100 thermocouples. The stirring system consisted of oblique blades driven by an electric motor (XD-153) at 1,600 rpm (115 V). The pressure in the bioreactor was measured with a Rockage® analogue pressure gauge (0-413.68 kPa). An industrial pressure sensor (0-1.2 MPa) allowed the operation of a 2 w-160-15 solenoid valve (Air control, 110 V), releasing the gas when a value of 68.94 kPa is reached. For the control, an Arduino Uno R3® board was implemented, programmed in C++ language. The pressure and temperature were shown on an LCD display, screen 1602 (16x2) (Figure 2). A gas meter was coupled to the bioreactor (Humcar® G1.6, minimum flow 0.016 m³ h^-1 and maximum pressure of 50 kPa), recording the total volume of gas generated. The bioreactor was installed at the Chemistry of Basic Sciences laboratory of the Universidad Surcolombiana (USCO) at Pitalito, where the tests were carried out.

Figure 2: Control and automation system.

Experimental design

Batch fermentations were made under a completely randomized experimental design to evaluate the effect of the proportions (% v/v) of the raw materials on the substrate. Three treatments were set with different amounts of VW, FWCW, CP, SCJ, and water, with three replicates per treatment. The temperature, agitation, acidification time (t_a) and operating pH (pH_o) in the fermentations (Table 1) were measured according to the invention patent 31671 granted to Universidad Nacional de Colombia by the Superintendencia de Industria y Comercio, conditions favorable for hydrogen production. All fermentations began with the grinding of solid residues (VW and CP) to a particle size of <2 cm. Subsequently, FWCP, SCJ, and treated water suitable for human consumption were added in the proportions established by the experimental design. The substrate was transferred to the bioreactor and left to rest for 72 hours, without agitation and at room temperature, to promote lactic acidification. After this period, the pH (pH_o) was adjusted to 6.5 using sodium carbonate (Na₂CO₃) to initiate hydrogen production.

Table 1: Substrate proportion and operation conditions.

The pH was monitored with a HI 98107 Hanna Instruments® pH-meter, with 0.1 resolution and ±0.1 accuracy. The response variables were hydrogen production rate for the day of maximum production (HP, L H₂ h^-1 d_max ^-1), cumulative hydrogen production (CHP, L H₂), maximum hydrogen content in the gas (MCH, %H₂), substrate yield (Y_S, L H₂ L_Substrate ^-1), average maximum hydrogen production rate (R_max-avg, L H₂ h^-1), and specific hydrogen production rate per biomass (SYPB, L H₂ h^-1 L^-1).

Kinetic and statistical analysis

The cumulative production of H₂ was adjusted to the modified Gompertz model (Equation 1). The model was applied using Visual Studio Code (version 1.87.2) and Python programming language (version 3.12.2), where H is the cumulative hydrogen production (mL), λ the lag phase time (h), P the potential hydrogen production (mL), R_max the maximum hydrogen production rate (mL h^-1), t corresponds to the elapsed time in hours, and e is 2.718281828.

The results of H and R_max were subjected to a non-parametric Wilcoxon signed-rank test (α=0.05) to determine differences between treatments using VS Code and the Python programming language.

Chromatographic analysis

To determine the production of H₂, gas samples were taken every 16 h using 1 L Tedlar bags. Concentrations of hydrogen (H₂), methane (CH₄), carbon monoxide (CO), carbon dioxide (CO₂), and nitrogen (N₂) were identified and quantified using gas chromatography with a MicroGC 3000 Agilent equipped with a terminal conductivity detector (TCD). This GC-TCD was coupled to the Molsieve and PLOTU columns, using argon and helium as carrier gases, respectively. The temperature for the injector and column was set at 60 °C, the pressure was 206.8 kPa, and injection flow was 0.83 mL s^-1. The analyses were conducted at the Energy Sciences laboratory at Universidad Nacional de Colombia Medellín Headquarters.

Gas Composition

The highest H₂ content was recorded for treatment L1 (43.99±3.89%), with a maximum value of 48.34% observed in trial L1-R2 (treatment 1, repeat 2). Al-Haddad et al. (2023), using glucose as the substrate, reported H₂ concentrations of 28.70% with untreated inoculum, 30.70% with thermal pretreatment at 115 °C for 20 min, and 27.40% H₂ with acid-treated inoculum. In this study, without substrate pretreatment, significantly higher H₂ values were obtained. The highest CH₄ concentration was 1.04% (L1-R2), a favorable outcome indicating that no significant H₂ consumption occurred for methane formation. The mean value for L1 was 0.93±0.35%, the highest among all treatments. The highest average CO content was observed in treatment L1 (9.85±3.93%), with a maximum value of 17.20% in trial L1-R2 (Figure 3).

Figure 3: Gas composition across three treatments (L1, L2, L3), each with three repeats (R1, R2, R3), assessed on their respective sampling days (D).

The formation of CO is attributed to homoacetogenic bacteria, which can grow using H₂ and CO₂ as energy sources (Mehi et al. 2024). The highest CO₂ content in this study was 33.59±12.16% for treatment L1. However, considering the remarkable results in H₂ production, in this work, CO could be more associated with metabolic pathways that do not consume H₂.

The highest N₂ mean was obtained in treatment L3 (33.93±7.89%), with a maximum value of 50.01% in trial L3-R3. Rojas-Sossa et al. (2017) studied the production of H₂ using coffee wastewater and indicated that the Comamonadaceae family was the most abundant group of proteobacteria. This genus performs the denitrification and decomposition of organic acids through the enzyme nitrate reductase, whose final product is N₂. Some researchers report that H₂, CH₄, CO₂, and N₂ have concentrations of 18.08, 0.20, 8.96, and 51.30%, respectively, using fruits and vegetables as substrates (Moreno Cárdenas et al. 2013). Mixed cultures from environmental sources may contain H₂-consuming microorganisms, such as methanogenic bacteria, which produce nitrate and iron, propionate, lactate and caproate; in these, H₂ can be consumed as reducing equivalent (NADH₂; potential H₂) or as molecular H₂ (Saady 2013). These results demonstrate that untreated substrates maximize H₂ production, minimizing unwanted byproducts such as methane while maintaining low levels of CO, suggesting an efficient and sustainable alternative for clean energy generation.

Bio-Hydrogen production and yield

Hydrogen production varied between 1.37±1.37 and 1.87±0.38 L H₂ h^-1 d_max ^-1, the MCH ranged between 34.37±9.45 and 43.99±3.89%, and the CHP between 43.63±17.39 and 70.03±2.65 L (Table 2). The maximum H₂ yield was found in trial L1-R2 with 2.30 L H₂ h^-1 d_max ^-1. Additionally, a cumulative H₂ production of 25.94 L H₂, with an H₂ content of 35.85%, has been reported after one day of acidification at a temperature of 30 °C, pH of 6.5, and a chemical oxygen demand (COD) of 60 g O₂ L^-1 of coffee mucilage and organic waste (Cárdenas et al. 2020). Likewise, an average H₂ production rate of 1,398.3 NmL L_substrate ^-1 d^-1 has been found, and an average concentration of 39% of H₂ using coffee mucilage and swine manure at a pH of 5.5 and a temperature of 55 °C (Hernández et al. 2014). García-Depraect and León-Becerril (2023) investigated the use of a specialized biocatalyst to produce hydrogen, lactate, and butyrate, reporting 0.2 NL H₂ L^-1, low butyrate production (3.3 g L^-1), and rapid decreases in lactate, from coffee industry wastewater (20% v/v from the fermentation stage; 80% v/v from the demucilaginization stage). The authors suggest that co-fermentation with other substrates, such as fruit and vegetable waste, could improve process efficiency.

Table 2: Production yield indicators.

The best yield obtained in the present study (Y_S) was 2.00±0.08 L H₂ L_substrate ^-1 for L1 under non-sterile conditions and without inoculation. Researchers have reported 1.29 L H₂ L_substrate ^-1 and a yield of 1.65 mol H₂ mol_hexose ^-1 with coffee mucilage combined with organic residues, no inoculation, and an organic load of 60 g O₂ L^-1 (COD), and a pH of 6.5 (Cárdenas et al. 2020). Previous studies report 1.29 mol H₂ mol_hexose ^-1 from wastewater generated during beverage manufacturing, subjected to heat treatment at 90 °C for 20 minutes (Jung et al. 2010). A yield of 49.2 mL H₂ g^-1 COD^-1 has been found using coffee pulp as a substrate, with inoculation (Miñón-Fuentes and Aguilar-Juárez 2019). Yields of 1.90 L H₂ L_substrate ^-1 have been achieved with urban organic waste in a bioreactor with pulsed stirring, without inoculation (Cano Quintero and Moreno-Cárdenas 2019). In addition, Hernández et al. (2014) reported that using coffee mucilage with inoculum and an organic load of 12.1 kg COD m^-3 d^-1, the yield was 2.5 mol H₂ mol_glucose ^-1.

During hydrogen production, various organic acids and alcohols have been identified, including lactic acid (pH=3.2-4.5), butyric acid (pH=4.7-5.0), acetic acid (pH=4.5), valeric acid (pH=6.0), propionic acid (pH=6.0), butanol (pH=4.7-4.9), and ethanol. Their presence varies depending on the microbial community in non-sterilized substrates (Villa Montoya et al. 2020a). This confirms that, in non-sterilized substrates, the microbial community influences the variability of compounds and metabolic pathways for hydrogen production. It has been reported that at pH=5.5, volatile fatty acids (VFAs) such as acetic, propionic, and isocaproic are accumulated, which could influence hydrogen production (Tiegam Tagne et al. 2024). Although this study did not evaluate VFA production, it is possible that favorable VFA formation pathways occurred, contributing to hydrogen production (Table 3).

Table 3: Temperature and pH values in the trials.

Figure 4 shows that the maximum production was not recorded on the same day for all the trials. Previous studies have demonstrated that H2 production is influenced by microbial diversity and that processes may be affected by operating conditions; hence, bacteria can adapt to dynamic conditions and alter the speed and efficiency of H2 production (Mehi et al. 2024). The results show that production lasted between three and five days, reaching its maximum between the first and second day, while on days four and five it was marginal, according to the batch operation. The use of native communities has been found to favor hydrogen production pathways. A yield of 596.3 mL H₂ L-1 and 25 g L-1 of lactic acid were reported from native communities of organic substrates, as they are well adapted to the substrates, efficiently converting lactic acid into hydrogen (Villanueva-Galindo et al. 2024).

Figure 4: Hydrogen production across three treatments (L1, L2, L3), each with three repeats (R1, R2, R3).

Modified Gompertz model applied to cumulative hydrogen production (CPH)

The kinetics of product formation (H₂) using the modified Gompertz model showed that the highest volume of H₂ was found in trial L2-R1 with 75,556.69 mL (Figure 5). The lag phase did not have the same duration in all the trials, since after the acidification time, the reaction rate of the base was not the same in all of them. The multiple correlation coefficient in relation to the Gompertz logistic model was 0.99 for all the tests (Table 4). The maximum H₂ production rate (R_max) was observed in trial L1-R2 with 3,638.34 mL H₂ h^-1, with lag phase time of 28 h. The best R_max-avg and the best SYPB were obtained in L1, with 2.70±0.82 L H₂ h^-1 and 0.08±0.02 L H₂ h^-1 L^-1, respectively (Table 4). However, all the parameters shown by the Gompertz model must be analyzed simultaneously since high values in the production rate and yield do not always imply high production, such was the case of trial L3-R3 that obtained outstanding values of R_max-avg and SYPB but showed the second lowest production (H_max).

Figure 5: Accumulated hydrogen production fitted to the modified Gompertz model across three treatments (L1, L2, L3), each with three repeats (R1, R2, R3).

Table 4: Parameters of the Gompertz modified model.

Villa Montoya et al. (2020a) report H_max values of 244 mL, R_max of 11.40 mL h^-1, λ of 17.10 h at pH of 5.5 and temperature of 30 °C, using pulp and coffee husk as a substrate with hydrothermal pretreatment at 150 °C. For additional studies using pulp and coffee wastewater pretreated at 180 °C for 15 minutes, the H_max results were 8 mL, R_max of 0.80 mL h^-1, and λ of 14.70 h (Villa Montoya et al. 2020b). Meanwhile, using a synthetic substrate, an R_max of 59.6 mL h^-1, H_max of 758.70 mL, and λ of 27.30 h were recorded, thus establishing that high organic loads cause inhibition and affect production (Laothanachareon et al. 2014). Similar results are reported by Moreno Cárdenas and Zapata Zapata (2019) for H₂ production using fruit and vegetable waste in co-digestion with fresh coffee mucilage; they found that H₂ production decreased with organic loads greater than 70,000 mg of O₂ L^-1. Additionally, Abreu et al. (2009) reported H_max of 137.20 mL, R_max of 1.70 mL h⁻¹, and λ of 13.70 h using hemicellulosic biopolymers (L-arabinose) at 30 g L⁻¹ and 37 °C, demonstrating the impact of substrate type and operational conditions on hydrogen production. Muri et al. (2016) documented H_max of 346 N mL, R_max of 91 N mL h⁻¹, λ of 6.50 h, and Y_S of 1.55 mol H₂ per mol glucose using glucose (5 g L^-1) at 37 °C and pH=6.4, noting that low pH suppresses hydrogenase activity, reducing H_max and R_max. This trend is consistent with the results achieved in this research, where the lowest H_max values were presented in treatment 3, which had the lowest proportion of water (18:68:0:0:14) with 14%, which implied a lower dilution and greater organic load.

Comparing medians: Statistical analysis for cumulative hydrogen production (CPH)

The results for the CPH variable indicate that the data are normal (P-value=0.10) according to the Shapiro-Wilk test and homogeneity of variance (P-value = 0.10), according to the Levene test. However, given the number of data available in the study, a non-parametric Wilcoxon test was preferred for comparing medians for the variables CHP and R_max (Figure 6).

Figure 6: Mean and interval for cumulative hydrogen production (CHP) variable in the treatments.

After applying the Wilcoxon test (Table 5), significant differences were observed between L1 and L3 in the variable CHP. In contrast, when comparing the medians between treatments L2 and L3 and between L1 and L2, no significant differences were found.

Table 5: Wilcoxon test results for cumulative hydrogen production (CHP) and maximum hydrogen production rate (Rmax).

Since there were no statistical differences in the variable CHP between the treatments L1 (composition of 18:25:5:5:47, in its order VW, FWCW, CP, SCJ and water) and L2 (composition of 18:48:0:0:35, in its order VW, FWCW and water), the substrate can be simplified to VW and FWCW, which opens a door for such residues to be used as main substrates without significantly affecting the H₂ production. Moreover, given that there are significant differences between the L1 treatment (with 47% water) and the L3 treatment (water was reduced to 14%), and that in the latter the production of H₂ was the lowest, there could have been an inhibitory effect due to organic overload. The rate of organic load is an essential factor in the DF, it represents the availability of substrate for H₂-producing microorganisms, allowing higher productions as long as the process is not inhibited by substrate overload (García-Depraect et al. 2021). High organic loads favor the production of undesirable by-products such as propionate and isocaproic acid negatively affecting the process (García-Depraect and León-Becerril 2023; Tiegam Tagne et al. 2024). Previous studies have reported a decrease in H₂ production due to organic overload when using different substrates, including mixtures of pig manure, coffee mucilage, and cocoa at concentrations between 40 and 50 g VS L^-1 (Rangel et al. 2020).

The Wilcoxon test results indicate no significant differences between the evaluated treatments for R_max. Although the limited number of replicates and the variability in the data suggest that these findings should be interpreted with some caution, the results support that, under the conditions evaluated, variations in substrate composition do not substantially affect R_max.