1. Introduction

Hydrogen (H₂) is a renewable energy source and a promising alternative to conventional fossil fuels because it is capable of eliminating most of the problems caused by widespread fossil fuel use. Several processes are known to produce H_2, including the electrolysis of water, thermocatalytic reformation of hydrogen-rich organic compounds, as well as geological and biological processes ^[1]. Biological production of H₂ using microorganisms via photo- and dark-fermentation is now an exciting area of technology development. This approach offers much promise because it requires a lower energy supply and provides other benefits compared with chemical-physical technologies to produce H₂ from a variety of renewable resources ^[2,3]. Specifically, via fermentation processes, H₂ can be generated directly from renewable substrates, such as sugars, biomass, or even organic residues, such as agricultural and food-industry waste and wastewater, in high concentrations ^[4].

Dark fermentative H₂ production is a ubiquitous phenomenon under anoxic or anaerobic conditions ^[5]. Many bacteria use the reduction of protons to H₂ via hydrogenases as a way to oxidize the reduced carriers during the fermentation process ^[6]. The theoretical yield of H₂ from glucose fermentation could be estimated from its known metabolic pathway, offering a maximum yield of 4 mol H₂/mol glucose if all of the substrates were converted to acetic acid as per the chemical reaction ^[7].

Dark fermentative H₂ production has been studied for a large group of pure bacterial cultures, including species of Enterobacter, Bacillus, Ethanoligenens, and Clostridium ^[2]. In taking advantage of their high-yielding H₂ production, many strains of Clostridium have been isolated and studied both as pure strains and in co-cultures for realizing efficient H₂ production ^{[8,9,10,11,12,13,14,15]}.

In the cow, the rumen is the largest stomach compartment, which harbors a complex community populated by microorganisms such as bacteria, archaea, protozoa, and fungi ^[16]. The rumen microbes can ferment plant fibers (e.g., hemicellulose and cellulose), starch, sugar, and protein to produce volatile fatty acids (e.g., acetate, propionate, butyrate, and lactate), microbial protein, and gases, such as CO₂, H₂, and CH₄ ^[17]. Many microorganisms produce H₂ in the rumen; however, methane bacterial populations also grow in the rumen, converting the CO₂ and H₂ into CH₄ ^[18]. Consequently, very little of the total rumen gases consists of H₂ Nevertheless, cow rumen fluid has been investigated and studied for its H₂ production from lignocellulose or cellulose under the inhibition of rumen methanogenesis by heat or an acid pre-treatment ^[18,19]. These studies, however, focused solely on the H₂ production of the rumen fluid enrichment culture as a mixed system composed of dominant Clostridium species (based on their 16S rDNA sequence and PCR-DGGE profiles analysis).

To specify the promising Clostridium strains for H₂ production, this study aimed to isolate, characterize, and identify the H₂-producing Clostridium spp. in cow rumen. Among the isolated Clostridium strains, three new ones (ST1, ST4, and ST5) were designated as C. beijerinckii ST1, C. bifermentans ST4, and C. butyricum ST5, based on their 16S rDNA gene sequence analysis and physiobiochemcal characteristics. Sucrose, glucose, lactose, xylose, and molasses were used as model carbon sources, to evaluate the H₂-production ability of these strains from different culture modes: three pure Clostridium cultures (ST1, ST4, and ST5) and four different co-cultures which consisted of (1) ST1 and ST4, (2) ST1 and ST5, (3) ST4 and ST5, and (4) mixing all three strains.

In the other hands, the food industry wastes such as cassava tuber wastes (cassava stumps) and a by-product of ethanol industry (distillers wet grains with soluble) are abundant, cheap, renewable, and rich nutrition of carbon source in tropical countries including Vietnam. Therefore, such wastes can be exciting and promising substrates for biological H₂ production by Clostridium sp. and they can give a significant benefit for bioenergy industrial in economic. However, until now, there has been no research on H₂ production from cassava stumps and distillers wet grains by Clostridium sp. yet. Hence, we use cassava stumps and distillers wet grains with soluble from cassava flour and ethanol manufacturing processes as the main substrate for H₂ production by the isolated Clostridium strains in this study.

Furthermore, response surface methodology using the Box-Behnken design and the software Design-Expert v7.1.5 was applied to optimize the operational conditions—temperature, time, and substrate concentration—to determine the most effective H₂ production by the three-strain co-culture.

---

2. Materials and methods

---

2.1. Isolation and identification of H₂-producing strains

One cow rumen (Quoc Oai, Hanoi, Vietnam) was sampled and pretreated by heating at 90 ℃ for 20 min to obtain the isolated strains. A peptone–yeast extract (PY) medium—10 g of glucose (Purity of 99%, Biobasic, Canada), 10 g of peptone (Biobasic, Canada); 10 g of yeast extract (Biobasic, Canada); 1 mg of resazurine (Purity of 99.9%, Sigma, USA), and 10 mL of a solution of salts ^[20]—was used to culture and screen the H₂-producing bacterial strains. The initial pH of the medium was adjusted to 6.5. The experiments were performed in 15 mL glass bottles using a 10 mL working volume. After inoculation, the headspace of the bottles was flushed with nitrogen gas for 15 min, to ensure an anaerobic environment in them. The bottles were kept in a constant temperature-controlled incubator at 37 ℃, with agitation at 120 rpm for 48 h. This process was replicated three times, and then inocula were spread onto PY-agar medium in an anaerobic culture box to obtain single colonies producing H₂. Then, these obtained strains were transferred onto fresh medium and cultured at 37 ℃ for 48 h.

The isolated strains that yielded high H₂ production were selected for identification by standard methods ^[20] and by 16S rDNA gene sequence analysis. Total genomic DNA was extracted from each strain by using the Magpure Bacteria DNA Kit (ANABIO Research and Development, Hanoi, Vietnam). The 16S rDNA gene sequence was amplified by conventional PCR using two primers, 27F (5'-AGAGTTTGATAMTGGCTCAG-3') and 1527R (5'-AAAGGAGGTGATCCAGCC-3'). PCR reactions were performed in a DNA thermal cycler, for which the reaction conditions were set as follows: 96 ℃ for 5 min, followed by 30 cycles of 94 ℃ for 30 s, 62 ℃ for 45 s, 72 ℃ for 90 s, with a final extension at 72 ℃ for 5 min. The PCR products were purified by an Anapure PCR Product Kit (ANABIO Research & Development, Hanoi, Vietnam) and sequenced by the 1st Base company (Singapore). The 16S rDNA sequence of these strains was analyzed with BioEdit v7.2 software (http://www.mbio.ncsu.edu/BioEdit/bioedit.html) and aligned against existing sequences in the GenBank database by using the BLAST program (https://blast.ncbi.nlm.nih.gov/Blast.cgi). A phylogenetic tree was constructed in CLUSTAL v1.8 based on the neighbor-joining method with 1000 bootstrap replicates.

---

2.2. Physiobiochemical characteristics

Pure isolated strains were tested for their morphology, physiology, and biochemistry properties. Although this approach is unlikely to be entirely accurate, it represents the first step taken in microbial classification research. The methods used for physiobiochemical characterization of isolated strain followed the standard protocol that has been conventionally used in bacterial systematic ^[20].

---

2.3. Enzymatic activity of the isolated strains

To evaluate the capacity of the isolated Clostridium strains for biohydrogen fermentation from different organic substrates, the enzymatic activities (amylase, protease and cellulase) were investigated via the diffusion method on agar plates. The Clostridium spp. were cultured separately on PY medium, which was added 10% (w/v) soluble starch (Biobasic, Canada), casein (Sigma, USA), and carboxymethyl cellulose (CMC) (Sigma USA), under conditions of 37 ℃, with a shaking rate of 200 rpm, for 24–48 h. Bacterial suspensions were centrifuged at 7000 rpm, discarding the biomass while keeping the supernatant. We used 1000-µL tips to perforate Petri dishes containing agar (with 1% soluble starch, casein, and CMC added). After dropping the bacterial suspension into the agar perforations, the agar Petri dishes were kept at 4 ℃ for 3–6 h, and then incubated at 37 ℃ for 24–48 h. Two petri dishes with 1% soluble starch and CMC were then stained with a Lugol-5% solution (Sigma, USA).

---

2.4. Pure cultures and co-cultures

The cultures were grown in 15 mL serum bottles sealed with a rubber stopper and an aluminum stopper containing 10 mL of PY medium including 10 g carbon source at 37 ℃ and pH 6.5 with 10% (v/v) inoculation, which released OD₆₀₀ of 0.1 for the primary medium and shaken at 220 rpm. The headspace of the serum bottle was sparged under an atmosphere of 99.999% N₂ to exclude oxygen and thereby create a stable anaerobic environment completely. To select the best suitable carbon source for fermentative H₂ production, each candidate carbon source such as glucose, sucrose, lactose, xylose, molasses, cassava stumps, and distillers wet grains with soluble was alternately tested at 10 g/L at final concentration.

---

2.5. Sampling and analyses

Growth was monitored by optical density (OD₆₀₀) with the sterile medium as the control. Hydrogen gas in the headspace was sampled with a gas-tight syringe (100 µL injection volume, Hamilton, USA) and determined by Shimadzu GC-8A gas chromatograph equipped with a thermal conductivity detector (GC/TCD) and two columns (Alltech, USA). Accumulative H₂ production was calculated by using the standard curve as described in the Figure S2 in the Supplementary data. Operating conditions were set: the carrier gas was nitrogen, under a pressure of 80 kPa; temperature of the column was fixed at 145 ℃, and the temperature of both the injector and detector was 150 ℃. The glucose concentration was determined with the dinitrosalicylic acid (DNS) colorimetric method as described by Miller ^[21]. All analyses were run in triplicate.

---

2.6. Experimental design for response surface methodology

To optimize the key parameters for enhancing the H₂ yield of the dark fermentative process, a 3^k Box-Behnken Design was employed using the software Design-Expert v7.1.5. For the statistical calculations, the relationship between the coded values and real values is described as follows:

${{\rm{X}}_{\rm{i}}} = \left( {{{\rm{A}}_{\rm{i}}} - {{\rm{A}}_{\rm{o}}}} \right)/\Delta {\rm{A}}$

(1)

where X_i codes the variable value, A_i is the real variable value, A_o is the value of A_i at the center-point; and ∆A is the step-change of the variable. Table S1 summarizes the levels of the variables and the experimental design containing the 17 trials; the response values were expressed as the mean of triplicates. The second-order polynomial coefficients were also calculated and analyzed in the Design Expert software (v7.1.5) statistical package. The H₂ yield was selected as the response variable, while the carbon source concentration (X₁), fermentation time (X₂), and fermentation temperature (X₃) were selected as the three independent variables. An analysis of variance (ANOVA) was carried out to validate the statistical results and the model equation.

---

3. Results and discussion

---

3.1. Isolation and biochemical characterization of strains showing their potentials

---

3.1.1. Isolation and identification of strains

Eight strains were isolated from the rumen of cow (Table S2, Supplementary Data). Among the isolates, the strains ST1, ST4, and ST5 respectively produced the highest H₂ production (mean ± SD) of 459.25 ± 5.31,568.05 ± 8.27, and 439.19 ± 8.42 mL/L medium, from the PY standard medium containing glucose as the carbon source. These strains were identified by standard biochemical analysis ^[20] for specific characteristics, as shown in Table S3 and Figure 1. The results indicated that the isolated strains ST1, ST4, and ST5 possessed general features of the Clostridium genus, such as Gram-positive staining, negative catalase functioning, spore formation, etc.

The colony of strain ST1 was milk-white in color, circular, opaque, and had a diameter of c. 1 mm after culturing on an agar plate at 37 ℃ for 48 h. Morphologically, cells of ST1 in the PYG culture appeared as straight rods with rounded ends, 0.5–1.7 × 1.7–8.0 μm (Figure 1A, right). The colony of strain ST4 was circular with irregular margins, flat or raised, lobate or scalloped, translucent or opaque, gray, shiny and smooth, and had a diameter of c. 0.5–4.0 mm after culturing on an agar plate at 37 ℃ for 48 h. Morphologically, cells of ST4 in the PYG culture appeared as straight rods, 0.6–1.9 × 1.6–11.0 μm (Figure 1B, right). The colony of strain ST5 was convex, translucent, gray-white, lobate or slightly scalloped, and had a diameter of c. 4–6 mm after cultivation on an agar plate at 37 ℃ for 48 hours. Its cells in the PYG culture appeared as straight rods with rounded ends, 1–1.7 × 1.7–8.0 μm, occurring singly or in pairs (Figure 1C, right).

For further confirmation, the 16S rDNA sequences of these strains were analyzed and aligned against existing sequences in the NCBI GenBank database using the BLAST program. The 16S rDNA gene sequence was directly amplified from the genomic DNA obtained from each strain. The results indicated a 100%, 99%, and 100% identification with Clostridium beijerinckii NCIMB 8052- CP000721 (NCBI No. MF136817), Clostridium bifermentans ATCC638 (NCBI No. MF125286), and Clostridium butyricum DSM10702 (NCBI No. MF125285) for the isolated strain ST1, ST4, and ST5, respectively (Figure 2). Thus, these three isolated strains were designated new strains of the species C. beijerinckii ST1, C. bifermentans ST4, and C. butyricum ST5, as determined by the 16S rDNA gene analysis and their physicochemical characteristics.

---

3.1.2. Enzymatic activity of isolated Clostridium strains

The three newly isolated strains C. beijerinckii ST1, C. bifermentans ST4, and C. butyricum ST5 produced protease (Figure S1A, Supplementary Data). A light ring surrounding the perforation appeared in the Petri dishes containing agar (with 1% casein added) after incubating at 37 ℃ for 24 h. This result indicated that casein (a protein in milk) was hydrolyzed by protease. To get a good visualization, the Lugol reagent was utilized to get a good visualization. The ring size of C. bifermentans ST4 indicated that its ability to decompose protease was superior to the other two strains.

The activity of amylase in the Clostridium isolated strains was also confirmed (Figure S1B, Supplementary Data). The part of the agar medium not hydrolyzed by amylase became blue-violet when stained with the Lugol reagent. The main component of the Lugol reagent is potassium iodide (KI). Thus, a reaction occurred between soluble starch (the substrate) and KI to create the characteristic blue-violet color. A light ring around the perforation revealed that three strains C. beijerinckii ST1, C. bifermentans ST4, and C. butyricum ST5 could all produce amylase. Their ring sizes indicated that the enzyme decomposition potential of these isolated strains was quite high (ring diameters of 10–15 mm).

As Figure S1C shows, the three isolated strains all produced cellulase. A light ring around the perforation appeared after staining with Lugol reagent. Comparing the ring sizes indicated that the ability to decompose cellulose was stronger in C. bifermentans ST4 than in the other two strains (Figure S1C, middle; Supplementary Data).

Based on the enzymatic activity of C. beijerinckii ST1, C. bifermentans ST4, and C. butyricum ST5 (Figure S1, Supplementary Data), the organic wastes and agriculture residues, are available for H₂ production via these three strains.

---

3.1.3. BioH₂ production from the model substrates

To evaluate the capacity of C. beijerinckii ST1, C. bifermentans ST4, and C. butyricum ST5, various substrates, including monosaccharides (glucose and xylose), disaccharide (sucrose and lactose), oligosaccharide (molasses) from sugar manufacturing, and complex matrix substrates from food industry wastes such as cassava stump and distillers wet grains with solubles (DWGS), were used as different carbon sources for H₂ production in a series of batch cultures that tested the capacity of pure cultures and co-cultures.

As Figure 3A shows, of the five substrates examined, H₂ production was observed by using pure Clostridium spp. cultures with four main types of carbon sources: glucose, sucrose, lactose, and xylose. However, the amount of H₂ production varied depending on the kind of substrate and the particular strains used. For glucose, the highest H₂ production (mean ± SD) of 732.15 ± 11.8 mL H₂/L (1.36 mol H₂/mol glucoseconsumed) was obtained using C. beijerinckii ST1 (Table S4, Supplementary Data), for which the obtained H₂ percentage was 51.7% and the OD₆₀₀ value was 1.794 ± 0.035. Glucose is also a suitable substrate for C. bifermentas ST4: this strain had an H₂ production of 675.37 ± 8.32 mL H₂/L or H₂ yield of 2.47 mol H₂/mol glucoseconsumed and an OD₆₀₀ value of 1.489 ± 0.045 (Table S3, Supplementary Data). The H₂ production obtained using C. butyricum ST5, with glucose as substrate, was lower, at 548.4 ± 3.74 mL H₂/L or H₂ yield of 0.7 mol H₂/mol glucoseconsumed (Table S4, Supplementary Data). Glucose, however, is one of the most suitable carbon sources because it is directly involved in the digestion process, which facilitates the growth of bacteria and, by extension, their H₂ yield. These results suggest that glucose is indeed a suitable carbon source for H₂ production, which agrees with prior research ^[22]. For example, Lin et al. reported H₂ yields of 2.81 and 1.8 mol H₂/mol glucoseconsumed when using C. beijerinckii L9 and C. butyricum ATCC19398, respectively ^[22]. In comparison with the previous studies (Table 1), C. beijerinckii ST1 and C. bifermentas ST4 indicate relatively strong potential for H₂ production.

Sucrose is also considered a suitable substrate for H₂ production. Previous research had shown that H₂ was relatively high when sucrose served as a carbon source. Chen et al. reported an H₂ yield of 2.78 mol H₂/mol of sucrose from C. butyricum CGS5 ^[37]. In our study, the highest H₂ production of 725.8 ± 5.6 mL/L was obtained using C. bifermentans ST4, accounting for 55.3% of the total gas production, with an OD₆₀₀ value of 1.694 ± 0.014. Next one was C. beijerinckii ST1, which obtained 654 ± 8.12 mL/L and had an OD₆₀₀ value of 1.634 ± 0.057. The H₂ production with C. butyricum ST5 (606.4 ± 2.08 mL/L) was lower than that with either C. beijerinckii ST1 and C. bifermentans ST4. Although H₂ production from lactose was lower than that from glucose and sucrose, it remained a suitable substrate for C. bifermentans ST4, which had an H₂ production of 585.5 ± 4.9 mL/L on this substrate (Table S5, Supplementary Data). Figure 3A indicates that xylose also was a substrate for H₂ production by fermentation with C. beijerinckii ST1, C. bifermentans ST4, C. butyricum ST5; their corresponding H₂ production were 387.8 ± 5.3,417 ± 6, and 364.8 ± 1.7 mL/L (Table S5, Supplementary Data). Molasses was not a suitable substrate for H₂ production. Previous study reported that the molasses contains several phenolic compounds derived from sugarcane such as dehydrodiconiferylalcohol-9'-O-beta-D-glucopyranoside and isoorientin-7, 3'-O-dimethyl ether, which possesses antibacterial activity against bacteria ^[38]. As a consequence, the highest H₂ yield achieved was only 33.8 ± 5.3 mL/L, for C. beijerinckii ST1 (Figure 3A, Table S5, Supplementary Data).

---

3.2. The use of strains in co-cultures with different substrates

---

3.2.1. H₂ production from the pure substrates

Based on the pure culture results, sucrose, glucose, and lactose as the carbon sources were chosen to investigate H₂ production by co-cultures at 37 ℃ for 48 h. Figure 3B shows that the highest H₂ production was achieved by the C. butyricum ST5 and C. bifermentans ST4 co-culture using glucose, for which the H₂ percentage was 54.7% and the H₂ production was 793.8 ± 8.1 mL/L (Table S5, Supplementary Data). This value was 1.17- to 1.44-fold that of either pure culture and greater than the highest H₂ production obtained with C. beijerinckii ST1 (732.2 ± 11.8 mL/L) (Table S5, Supplementary Data). In addition, the OD₆₀₀ value also increased from 1.021 ± 0.023 to 1.965 ± 0.056. Collectively, these results confirm that glucose is a suitable substrate for co-cultures of C. butyricum ST5 and C. bifermentans ST4. However, the hydrogen yield of the C. butyricum ST5 and C. beijerinckii ST1 co-culture was < 18% that of C. beijerinckii ST1, while the OD value decreased from 1.794 ± 0.035 to 1.632 ± 0.031. A similar result was obtained for the C. beijerinckii ST1 and C. bifermentans ST4 co-culture, which points to resource competition among microorganisms as a possible explanation.

For sucrose, the highest H₂ production was obtained for the co-culture of C. butyricum ST5 and C. beijerinckii ST1, which amounted to 699.8 ± 8.8 mL/L, and accounted for 55.1% of the total volume of biogas produced. This value is 1.15- to 1.07-fold greater than what their pure cultures achieved, with an increase in OD₆₀₀ from 1.094 ± 0.034 to 1.765 ± 0.054. However, this particular Clostridium co-culture still had a lower H₂ production than that of C. bifermentans ST4 (725.8 ± 5.6 mL H₂/L, OD₆₀₀ = 1.694 ± 0.014). An H₂ production of 676.4 ± 5.24 mL/L and an OD₆₀₀ value of 1.774 ± 0.043 were obtained for the co-culture of C. butyricum ST5 and C. bifermentans ST4. Although this OD₆₀₀ value is higher, the H₂ production is lower; this suggests that, when using sucrose as a substrate, microorganisms would likely grow better in co-cultures of ST5 and ST4, though in the decomposition process acetic, butyric, and ethanol byproducts were generated. The results for the C. beijerinckii ST1 and C. bifermentans ST4 co-culture also indicated that both H₂ production and OD₆₀₀ values decreased, which may be explained by competition between the Clostridium strains.

Besides using co-cultures of two Clostridium species, many studies have reported on the mixing of Clostridium with other species to produce H₂. For example, Ding et al. obtained a maximum H₂ yield of 3.47 mol H₂/mol glucose_consumed by using a co-culture of C. butyricum and immobilized Rhodopseudomonas faecalis RLD-53 ^[39]. Geng et al. obtained an H₂ yield of 1387 mL/L when they combined C. thermopalmarium and C. thermocellum ^[40]. This is consistent with that of Liu et al., who obtained an H₂ yield of 1232 mL/L with co-cultures of C. thermocellum JN4 and C. thermosaccharolyticum GD17 ^[41]. Finally, from a co-culture of Bacillus themoamylovorans I and C. beijerinckii L9, Chang et al. obtained an H₂ yield of 1145 mL/L ^[42].

Figure 3C depicted the growth and H₂ production vitality when the three isolated strains were evenly mixed (1:1:1) to investigate their H₂ production at 37 ℃ within 48 h. In this experiment, the maximum H₂ production of 1120.0 ± 14 mL H₂/L (Figure 3C, Table S5, Supplementary Data) was obtained when using sucrose as a substrate, for which the percentage of H₂ gas was 61.8% of the total biogas produced, and the OD value was 2.046 ± 0.027. This result clearly shows that sucrose is the most suitable carbon source for both bacterial growth and H₂ production from co-culture mixing C. beijerinckii ST1, C. bifermentans ST4, and C. butyricum ST5. A plausible explanation is that, when the organisms are present in combination, genes encoding enzymes hydrolyzing sucrose to glucose and fructose are strongly expressed, so that sucrose is readily cleaved into monosaccharide for metabolism and subsequent hydrogen generation. Particularly, it can be explained that sucrose supports vigorous growth of the saccharolytic Clostridium species. Also, co-culture three Clostridium strains can active a cluster of sucrose catabolic genes of C. beijerinkii, comprising all genes required for the transport, hydrolysis and subsequent phosphorylation to cleave sucrose to glucose and fructose via a glycosidic linkage. Molecular analysis indicated that C. beijerinkii contains four genes: scrARBK, encoding a sucrose-specific transport protein; ScrA, a regulator of the LacI–GalR family (ScrR); the sucrose-6-P hydrolase (ScrB); and a fructokinase (ScrK), respectively. This mechanism was proved in the previous study ^[43]. Therefore, it could be a reason why the co-culture of the three strains achieved the highest H₂ production (Figure 3C) from sucrose, which was much higher than that by any pure culture or co-culture from glucose (Figure 3A, B).

---

3.2.2. H₂ production from the food industry wastes

As the unlimited source of natural nutrients, the food industry wastes including cassava stumps (cassava tuber wastes) and distillers wet grains with soluble (DWGS—by-product of ethanol industry) was used as the main substrates to investigate the activity of H₂ production by C. beijerinckii ST1, C. bifermentans ST4, C. butyricum ST5. Figure 4 shows the H₂ yield of the Clostridium species in different culture modes from the food industry wastes, including cassava stumps or DWGS. For the cassava stumps, the highest H₂ production of 895.2 ± 5.6 mL/L was given by C. butyricum ST5, accounting for 59.2% of the total volume of biogas, and was 1.3-fold that from C. beijerinckii NBRC 109359 (658.3 ± 14.4 mL/L) in this study (Figure 4, Table S5, Supplementary Data). Similarly, H₂ production of 805.8 ± 11.5 mL/L and 787.5 ± 9.2 mL/L were obtained from C. beijerinckii ST1 and C. bifermentas ST4, respectively (Figure 4, Table S5, Supplementary Data). Using the cassava stumps, the H₂ production from co-cultures of two or three Clostridium species decreased from 1.7 to 3.6 times compared to that of the pure culture of C. butyricum ST5. H₂ production respectively was 248.5 ± 6.8 mL/L, 529.6 ± 8.7 mL/L, and 635.9 ± 7.5 mL/L for the co-cultures of C. beijerinckii ST1 and C. bifermentans ST4, C. butyricum ST5 and C.beijerinckii ST4, and C. beijerinckii ST1, C. bifermentans ST4, and C. butyricum ST5 (Figure 4, Table S5, Supplementary Data). This result may be explained by Clostridium species competing with each other for resources, which reduce the strains' ability to convert starch to H₂. As an intriguing carbon source, there are many studies on H₂ production from starchy substances. In a previous report ^[44], H₂ yields of 12.52–9.9 mmol/g starch were obtained by C. butyricum CGS2. H₂ yields of 240,224.4, and 165.2 mL H₂/g of cassava starch were reported for combined cultures of mesophilic organisms taken from three locations in Thailand ^[45]. Moreover, through the combination of dark- and photo-fermentation, Su et al. obtained H₂ production of 979.9–2541.1 mL/L when using starchy substance concentrations of 10–25 g/L ^[46].

Regarding DWGS as the main substrate (Figure 4 and Table 5S, Supplementary data), a higher H₂ production (765.1 ± 8.6 mL/L) was archived from C. bifermentans ST4 compared to that of C. beijerinckii ST1 (689.8 ± 11.5 mL/L) or C. butyricum ST5 (637.0 ± 8.6 mL/L). In addition, the H₂ production of these strains indicated a higher H₂ production compared with C. beijerinckii NBRC 109359 (458.8 ± 7.26 mL/L), that was used as a Clostridium strain reference. Interestingly, the highest H₂ production of 809.3 ± 7.9 mL/L was obtained by combining C. butyricum ST5 and C. bifermentans ST4. In contrast, the co-culture of C. beijerinckii ST1 and C. bifermentans ST4 gave an H₂ production of 139.5 ± 2.5 mL/L. Perhaps C. beijerinckii ST1 and C. bifermentans ST4 were better able to cleave organic compounds into acetate, butyric acid, etc. The H₂ production of 658.4 ± 9.4 mL/L obtained with the co-culture of C. butyricum ST5 and C. beijerinckii ST1 was similar to that from their pure cultures, a result suggesting that these strains are neither inhibited nor interchangeable in hydrogen decomposition. However, in the mix of the three isolated strains C. beijerinckii ST1, C. bifermentans ST4 and C. butyricum ST5, the H₂ production was 959 ± 8.8 mL/L, and the H₂ percentage was 54.5%, which was much higher than that by any pure culture or co-culture of two strains. This result indicated an advantage of combining three trains for H₂ conversion from DWGS as the main substrate.

---

3.3. Optimization of the key fermentation conditions for H₂ production by mixing three newly isolated Clostridium strains

Temperature is considered one of the most important factors affecting the activities underpinning fermentative H₂ production by H₂-producing bacteria ^[40,47]. Based on the characterization of different mesophilic and thermophilic H₂-producing bacteria, the optimal temperature for the fermentative H₂ production is expected to vary. For example, the temperature applied to fermentative hydrogen production by different mesophilic H₂-producing bacteria Clostridium isolates occurs between 30 ℃ and 40 ℃ ^{[28,48,49,50]}. The optimum temperature for efficient H₂ production by C. butyricum TM-9A was deemed to be 37 ℃ ^[51]. For fermentative H₂ production by C. butyricum EB6, Chong et al. optimized H₂ production via response surface methodology, and determined the optimal temperature as 36 ℃ ^[52]. In addition, the activities of key enzymes related to fermentative H₂ production (such as the hydrogenases) depend considerably upon on temperature. Hence, the optimization of fermentative temperature for the efficient H₂ production is needed.

Nevertheless, substrate concentration is also a critical factor which directly affects the distribution of metabolic products during fermentation ^[53]. We know that over an appropriate range, increasing the substrate concentration could increase the ability of fermentative H₂ production in H₂-producing bacteria to produce H₂, but much higher substrate concentrations could decrease this activity ^[54,55]. Furthermore, addressing the effect of substrate concentration on fermentative H₂ production must also be carried out adequately, since this is a critical parameter in determining the economic and technical feasibility of the process ^[56]. Many studies, therefore, have reported the influence of substrate concentration on fermentative H₂ production ^{[8,57,58,59,60]}. Understanding the influence of these key factors—temperature and substrate concentration—on bio-H₂ production will help to optimize the operating fermentative H₂ production by a co-culture that mixes the three newly isolated Clostridium strains studied here.

In our optimization experiments, the variables X₁ (sucrose concentration), X₂ (fermentation time), and X₃ (fermentation temperature) were taken according to the design in Table S1, and their function was Y_H2. The results of these experiments revealed that the highest H₂ yield (1.12 L/L) was obtained under the following specific conditions: a sucrose concentration of 10 g/L, with a fermentation time of 48 h, under a fermentation temperature of 37.5 ℃ (Table S1). Based on such conditions, it was possible to establish a regression equation describing the relationship between the function Y_H2 and the variables X₁, X₂, and X₃, as follows:

$\begin{array}{l} {{\rm{Y}}_{\rm{H}}}{\rm{ = 1}}{\rm{.10 + 0}}{\rm{.072}}{{\rm{X}}_{\rm{1}}}{\rm{ + 0}}{\rm{.066}}{{\rm{X}}_{\rm{2}}}{\rm{ - 0}}{\rm{.10}}{{\rm{X}}_{\rm{3}}}{\rm{ - 0}}{\rm{.010}}{{\rm{X}}_{\rm{1}}}{{\rm{X}}_{\rm{2}}}{\rm{ - }}\\ {\rm{0}}{\rm{.030}}{{\rm{X}}_{\rm{1}}}{{\rm{X}}_{\rm{3}}}{\rm{ - 0}}{\rm{.057}}{{\rm{X}}_{\rm{2}}}{{\rm{X}}_{\rm{3}}}{\rm{ - 0}}{\rm{.14}}{{\rm{X}}_{\rm{1}}}^{\rm{2}}{\rm{ - 0}}{\rm{.20}}{{\rm{X}}_{\rm{2}}}^{\rm{2}}{\rm{ - 0}}{\rm{.31}}{{\rm{X}}_{\rm{3}}}^{\rm{2}} \end{array}$

(2)

The ANOVA validated the statistical results and the significance of the fitting model equation for the experimental data (Table 2). The model's F-value of 293.95 implied a significant model fit, with the probability of error as noise at < 0.0001, which indicated only a 0.01% probability that the model fit the data by chance. To check the significance of each variable, as well as the interaction strength between each independent variable, the P-values were used as a tool. As seen in Table 2, the interaction between sucrose concentration (X₁) and fermentation time (X₂) had a low significance (P > 0.05), which indicates that these two variables did little to change each other's influence on the response variables. The low F-value of 1.07 for the lack of fit implied the latter was not significant relative to the pure error, with 45.07% chance it occurred strictly due to noise. In sum, that lack of fit is insignificant means that the model is robust. The multiple correlation coefficient (R² = 0.9974), the coefficient of determination (adjusted R²), and the predicted coefficient (predicted R²) were also evaluated. The adjusted R² = 0.994 indicated that 99.4% of the variation in the response variable "Y_H2" was explained by the fitted model. These results indicate good agreement between the experimental and predicted values. Hence, this regression model is very reliable as an accurate representation of the experimental data on hydrogen production as reported in this study.

To investigate the effects of sucrose concentration, temperature, and fermentation time on H₂ production, the three-dimensional response surface and two-dimensional contour plots were constructed to show the relationship between Y_H2 as a function of its predictor variables (i.e., X₁, X₂, and X₃) (Figure 5). The maximum predicted value is defined by the surface confined in the smallest elliptical contours, which indicates a perfect interaction between the independent variables ^[61]. Figure 5 shows that the response surface and contour plots indicated that the highest H₂ production (1.13 L/L) was obtained when the sucrose concentration was 11.63 g/L, the fermentation time was 51.13 h, and the fermentation temperature was 36.09 ℃.

To confirm the applicability of the above-constructed fermentative model, a confirmation trial was conducted under the optimal conditions at 36.1 ℃ in medium containing 11.63 g/L sucrose for a culture time of 51.1 h. As a result, the maximum H₂ production was found to be 1.129 L/L. Thus, the difference between the calculated yield according to the model we designed and the empirical yield obtained from a trial of this model was within the allowable range. Hence, this model is deemed useful for further applications.

---

4. Conclusions

Three novel H₂-producing Clostridium strains were isolated and screened from the rumen of cow in Vietnam and identified as C. beijerinckii ST1, C. bifermentans ST4, and C. butyricum ST5. These strains showed a capacity for utilizing a broad range of substrates, including sucrose, glucose, lactose, xylose, and molasses, in addition to food industry wastes such as cassava stumps and distiller wet grains with soluble, for efficient H₂ production in both pure cultures and co-cultures. The highest H₂ production was achieved from a mixed culture consisting of the three strains when compared with the other culture modes (of pure cultures and co-cultures with two strains). Response surface methodology with the Box-Behnken design was used to successfully optimize the operational conditions, including temperature, time, and substrate concentration, for the high-level production of H₂ in this mixed co-culture of the three newly isolated Clostridium trains. Statistical analysis revealed that all three variables influenced its H₂ production significantly. A mathematical model was established, with a confidence level of 99%, with P < 0.0001. The optimized condition for maximized H₂ production was a sucrose concentration of 11.63 g/L, a fermentation time of 51.13 h, and a fermentation temperature of 36.09 ℃ that produced 1.13 ± 0.015 L of H₂ per L of medium.

---

Supplementary material

See supplementary material for Box-Behnken experimental design with 3 independent variables and the corresponding experimental results (Table S1), for H₂ production from 8 isolated strains (Table S2), for physiobiochemical characteristics of Clostridium isolated strains (Table S3), for enzymatic activity of isolated Clostridium strains (Figure S1), for growth and H₂ production from isolated Clostridium strains (Table S4), for a summary of H₂ production by single and co-culture isolated Clostridium strains from various substrate (Table S5), and standard curve for H₂ production analysis (Figure S2).

---

Acknowledgments

The authors thank VNU University of Science for facility support. The Vietnam National University funded this work under grant QG.16.03.

---

Conflict of interest

The authors declared no potential conflicts of interest respecting to the research, authorship, and pulication of this paper.

---