A Review of Biohydrogen Productions from Lignocellulosic Precursor via Dark Fermentation: Perspective on Hydrolysate Composition and Electron-Equivalent Balance

Abstract

This paper reviews the current technological development of bio-hydrogen (BioH₂) generation, focusing on using lignocellulosic feedstock via dark fermentation (DF). Using the collected reference reports as the training data set, supervised machine learning via the constructed artificial neuron networks (ANNs) imbedded with feed backward propagation and one cross-out validation approach was deployed to establish correlations between the carbon sources (glucose and xylose) together with the inhibitors (acetate and other inhibitors, such as furfural and aromatic compounds), hydrogen yield (HY), and hydrogen evolution rate (HER) from reported works. Through the statistical analysis, the concentrations variations of glucose (F-value = 0.0027) and acetate (F-value = 0.0028) were found to be statistically significant among the investigated parameters to HY and HER. Manipulating the ratio of glucose to acetate at an optimal range (approximate in 14:1) will effectively improve the BioH₂ generation (HY and HER) regardless of microbial strains inoculated. Comparative studies were also carried out on the evolutions of electron equivalent balances using lignocellulosic biomass as substrates for BioH₂ production across different reported works. The larger electron sinks in the acetate is found to be appreciably related to the higher HY and HER. To maintain a relative higher level of the BioH₂ production, the biosynthesis needs to be kept over 30% in batch cultivation, while the biosynthesis can be kept at a low level (2%) in the continuous operation among the investigated reports. Among available solutions for the enhancement of BioH₂ production, the selection of microbial strains with higher capacity in hydrogen productions is still one of the most phenomenal approaches in enhancing BioH₂ production. Other process intensifications using continuous operation compounded with synergistic chemical additions could deliver additional enhancement for BioH₂ productions during dark fermentation.

1. Introduction

Although the idea of hydrogen economy was firstly proposed in the early 1970s, it was not until the dawn of the 21st century that industry and academia realize its pivotal potential in shaping a future sustainable society [1,2,3,4]. The status quo of existing gaps in the hydrogen economy lies in the cost-effective and reliable hydrogen generation, storage, and utilization, which are the main blocks in the chain of hydrogen economy [5,6,7]. Among them, the cost-effective and reliable hydrogen generation is one of the most challenging and difficult barriers to be overcome [8,9,10,11]. At present, hydrogen is predominantly produced via a thermal route on the industrial scale. The thermal-catalytic conversion processes are: steam reforming (SR), partial oxidation (POX), and auto-thermal reforming (ATR) [12]. Although the steam reforming of methane (SRM) is currently the most cost-effective approach in industrial scale hydrogen generation, its significant drawbacks of large green-house gas emission and heavy reliance on fossil derived material will significantly deteriorate its future prospects [13]. Apart from using fossil-based material as feedstock, the biomass, which is the world’s fourth largest energy resource and contains about 5–7 wt. % of hydrogen element on the dry basis, is emerging as a significant contributor in hydrogen generation [14,15,16]. In addition to deploying thermal techniques to convert biomass into energy, the biochemical approach via anaerobic and aerobic processes gradually steps in and becomes attractive in both academia and industry [17,18,19]. The produced hydrogen, so-called biohydrogen (BioH₂), has begun to provide supplementary subsidize for the entire hydrogen productions due to its inherent advantages such as insensitivity to the moisture of raw materials, mild operational conditions (low temperature and atmospheric pressure), less theoretical energy intensity, and suitability in decentralized energy productions [20,21,22]. BioH₂ production can be achieved via two-type of biological metabolisms, namely photo-dependent (photo-fermentation-PF and bio-photolysis-BP) and photo-independent (dark fermentation-DF) processes [23]. In addition, these processes are apt to be coupled either sequentially or simultaneously on the basis of energy and material cascade utilization and process intensification [24]. The growing trend of publications collected from Scopus (Figure 1) witnessed the endeavors that worldwide scholars have made for the R&D of BioH₂ production using biomass as substrate.

The discernible dominant fraction of publications (Figure 1) comes from BioH₂ productions via DF due to its inherent appealing features such as independence to light, easiness in operation, higher hydrogen conversion, flexibility in cultivation, and simultaneous utilization of organic substrate [25,26]. The reports on the combination of DF, PF, or ME maintain at the numbers of 30–50 and the core challenges in combining the photo-dependent, photo-independent, and microbial electrolysis techniques into an integrated BioH₂ process still awaiting being tackled [27,28]. In regard to the utilization of biomass, although the lignocellulosic biomass is notorious for its recalcitrant nature to the pretreatment (i.e., acid hydrolysis, alkaline delignification, and steam explosion) and the liberation of inhibitory chemicals to the subsequent process, inevitably weakening its fermentation performance in BioH₂ production, its abundant availability and allusive economic impetus make the BioH₂ production from lignocellulosic biomass as one of the most attractive renewable fuel production processes [29,30]. Consequently, the emerging consolidated bioprocess (CBP) successfully integrating the pretreatment and the fermentation to desirable metabolic products has gained substantial research attention in the last five years [18,31].

To narrow the scope of discussions and align with the consensus on the advantage of DF, this review will mainly focus on the BioH₂ productions via DF using biomass, especially lignocellulose as the feedstock. In this paper, instead of making simple BioH₂ production comparisons across literature reports, the collected data (such as HY, HER, and compositions matrix in the hydrolysate) were used to construct the data matrix for supervised machine learning using the developed artificial neuron networks (ANNs) via the feed backward propagation with one cross out validation approach for more insightful and quantitative correlations between the key operational parameters and performance parameters of BioH₂ productions. In addition, when electrons are transported from the substrate to electron acceptors, the reductive and oxidative (redox) reactions occur, which drives the entire biological metabolism. Thus, the electron equivalent balances can be established to represent these redox reactions so as to obtain an insightful understanding of the metabolism during dark fermentation. The reports of the application of supervised machine learning via constructed artificial neuron networks (ANNs) via the feed backward propagation and one cross out validation approach in statistically analyzing the biohydrogen performance using lignocellulosic feedstock and rigorous analysis of electron equivalent balances during dark fermentation, to the best of our knowledge, has never been published before. These are the main objectives in this review paper.

2. Approaches and Techniques

In this work, artificial neuron networks (ANNs) were established in Python 2.7 (Python Software Foundation). For supervised learning, there are no consensuses in regarding to how many hidden layers should be utilized [32]. In this work, we employed the most widely used feed backward three layers to train data. During data training, the simplified one cross-out method was employed for cross validation. The simplified procedure can be summarized as following: (1) Set observation i (1−n) from the data set being out and train the ANNs using the remaining data. (2) Computes the (mean square error) MSE and MARR (mean relative residual relation) as the following:

$$MSE\%=\frac{1}{N_{sam}}{\displaystyle \sum }_{j=1}^{N_{sam}}{\left(r|i^{\mathrm{sam}}-r_i^{\mathrm{cal}}\right)}^2\times 100\%$$

(1)

$$MARR\%=\frac{1}{N_{sam}}{\displaystyle \sum }_{j=1}^{{\mathrm{N}}_{\exp }}\left(\frac{\left|r_i^{\mathrm{sam}}-r_i^{cal}\right|}{r_i^{\mathrm{sam}}}\right)\times 100\%$$

(2)

where N_sam is the number of sample data set, $r_i^{\mathrm{sam}}$ and $r_i^{\mathrm{cal}}$ are actual and calculated values, respectively. (3) Repeat step 1 until i = 1…n. (4) Average all responses from ANNs. (5) Repeat steps 1–4, twice. Once allowable accuracy reaches over 95%, the established ANNs will generate prediction via designed matrix such as Box–Behnken design (BBD) and the central composite design (CCD). Finally, the established hybrid ANNs-response surface methodology (ANNs–RSM) with BBD design was evaluated using analysis of variance (ANOVA). The commercial Design Expert^®, Version 11 software package (Stat-Ease, Inc., 1300 Godward St NE, Suite 6400, Minneapolis, MN 55413 USA) was used for ANOVA.

3. Dark Fermentation

The hydrogen generation is achieved under the anaerobic condition through microbial oxidation of organic substrate and neutralization between the proton (H⁺) and electron (e⁻). Because of different combinations of metallic active catalytic sites in the metal clusters of the hydrogenase, there are three types of hydrogenases involving the hydrogen reaction, which are (FeFe), (NiFe), and (Fe) type hydrogenases [33,34,35]. While (Fe) type hydrogenases only exist in some specific methanogens, the (FeFe) type, catalyzing the formation of H₂, and (NiFe) type, catalyzing the consumption of H₂, are widely found in the anaerobic microbes [36,37,38].

Although the matrix of fermentation broth generated from different pretreatment approaches varies greatly [38], sugars in these different pre-treated lignocellulosic fermentation broths will be utilized via different metabolic pathways and these pathways will be eventually funneled into existing hydrogen generation metabolic pathways shown in Figure 2.

Although alternative pathways such as the Entner–Doudoroff pathway exist, the formation of pyruvate via glycolysis is one of the most important and widely adopted pathways by the bacteria [40]. The delicate balances among different metabolic pathways are well maintained by intertwining the cell growth, substrate decomposition, and energy harvesting through a serious of coupled reductive and oxidative biological reactions (Figure 2). From a thermodynamic perspective, the theoretical hydrogen production per glucose reaches a 12 (mol·mol⁻¹) H₂/glucose ceiling value assuming glucose being completely consumed to produce hydrogen, as depicted in Equation (3):

$${\mathrm{C}}_6{\mathrm{H}}_{12}{\mathrm{O}}_6+6{\mathrm{H}}_2\mathrm{O}\stackrel{}{\to }6{\mathrm{CO}}_2+12{\mathrm{H}}_2$$

(3)

Although Equation (3) shows a large potential hydrogen yield, much energy has to be spent to convert substrate into biomass to maintain the metabolism within the microbes, which compromises the maximum theoretical hydrogen generation. Three main indicative metabolic products, namely acetate, propionate, and butyrate are directly related to the hydrogen generation process. From a stoichiometric perspective using one mole of glucose, the following metabolites (acetate, butyrate, and propionate) will be produced [41]:

$${\mathrm{C}}_6{\mathrm{H}}_{12}{\mathrm{O}}_6+2{\mathrm{H}}_2\mathrm{O}\stackrel{\mathrm{Hydrogenase}}{\to }2{\mathrm{CH}}_3\mathrm{COOH}+2{\mathrm{CO}}_2+4{\mathrm{H}}_2$$

(4)

$${\mathrm{C}}_6{\mathrm{H}}_{12}{\mathrm{O}}_6\stackrel{\mathrm{Hydrogenase}}{\to }{\mathrm{C}}_2{\mathrm{H}}_5\mathrm{COOH}+2{\mathrm{CO}}_2+3{\mathrm{H}}_2$$

(5)

$${\mathrm{C}}_6{\mathrm{H}}_{12}{\mathrm{O}}_6+2{\mathrm{H}}_2\mathrm{O}\stackrel{\mathrm{Hydrogenase}}{\to }{\mathrm{C}}_3{\mathrm{H}}_7\mathrm{COOH}+2{\mathrm{CO}}_2+2{\mathrm{H}}_2$$

(6)

Broadly speaking, the phases of metabolites are divided into two categories: gaseous and liquid products. In regard to the liquid metabolites, the route of decomposition of glucose into acetate will lead to the largest hydrogen generation (Equation (4)) and this provides the guidance for the enhancement of BioH₂ generation by adjusting or skewing metabolic pathways towards the formation of acetate during the DF [42]. In addition, other metabolites such as lactate, valeric acid, and ethanol (though their formations do not directly lead to hydrogen formation), are critical because of their participation in the redox reactions forming either the reduced nicotinamide adenine dinucleotide (NADH) or oxidized nicotinamide adenine dinucleotide (NAD+) and their subsequent ratio (NAD+/NADH) is a typical indicator of the activities of metabolism and health of the microbes [43]. In regard to gaseous metabolites, H₂ is accompanied by the formation of CO₂. Therefore, to better and comprehensive gauge the entire dark fermentation, the measurement of the key liquid metabolites (acetic acid, butyric acid, propionic acid, lactate, valeric acid, and ethanol) and gaseous metabolites (H₂ and CO₂) are necessary.

4. Pretreatment of Lignocellulosic Biomass and the Matrix of Substrate

One of the most challenging aspects of renewable hydrogen generation technology, such as water electrolysis, to compete with current techniques based on fossil fuel, such as methane steam reforming (MSR), lies in its unfriendly cost. Exploring BioH₂ production using renewable feedstock, such as biomass, presents a rosy prospect if the production of the feedstock is not competing with the production of food and arable land [44].

Table 1. Summary of different types of biomass as feedstock.

Types of Feedstock	Substances	Examples	Advantages	Limitations
First generation	Food crops	Converting sugars including: glucose, xylose, lactose. ucrose, maltose, etc. Starch, molasses, etc.	Easy to be utilized by the cultivated microbes Commercially available such as E10 fuel. Centralized energy generation and utilization.	Competing with food source and arable lands High cost
Second generation	Lignocellulosic resources (hemicellulose and cellulose)	Agricultural waste including: corn stover, wheat bran, wheat straw, rice straw, sweet sorghum potato steam peels, cassava stillage, sugarcane bagasse, and beer lees, grass, etc. Paper pulping: pulping wastewater. Forestry by-products: saw dust, wood chip	Great abundance Available for solid fermentation Cost-effectiveness (not including pretreatment cost) Decentralized energy generation and utilization.	Inert to most microbes for BioH2 productions. High cost of necessary pretreatment Detoxication process required, such as neutralization by adding acid/base
Third generation	Sucrose-based, Starch based, Microalgae.	Sugar cane and molasses. Wheat, corn, barley, sweet sorghum, and sweet potato, Microalgae	Rapid growth rate, Cultivability without soil, Simultaneously green-house gas capture, such as CO2. Fast turn-around cycle (1–10 days), High contents of carbohydrate and lipids, less harsh pretreatment.	Early stage of the study Potential inhibitors to the subsequent fermentation

summarizes different categories of bio-feedstock for renewable energy generation. There are three types of feedstock categories from biomass that can be used for bioenergy generation. Comparing with plant-based feedstock (second generation feedstock), microalgae are efficient photosynthetic organisms, have high growth rates and they do not need arable land and fresh water, since they can grow in brackish water, seawater and even wastewater [45]. In addition, the cell wall compositions mainly the cellulose, hemicellulose, and absence of lignin, requires much less harsh pretreatment comparing that with processing the lignocellulosic feedstock [46]. In order to achieve integrated utilization of microalgae compositions, such as proteins, lipids etc., mature solvent extraction process is often employed to selectively separate the high-valued compounds. However, this separation process is often wrestling with the limitations of harvesting, high-cost, non-volatile features, subsequently restricted back extraction, and unknown environmental impacts, which prevented its development at industrial scale [47]. Apart from fractionation, cultivation of microalgae and harvesting especially on the large scale is also facing thorny challenges, i.e., relative high cost, risk of contamination, and relatively low biomass productivities. The common economical approaches of algae cultivation include: open ponds (open system), tubular, modified photobioreactors (PBRs), flat panel PBRs (closed system), etc. [48]. The common harvesting microalgae includes: physical, chemical, biological, magnetic, and combined process, of which each one possesses its inherent pro and cons [49]. The likelihood of dark fermentative biohydrogen from microalgal culture lies with the fact that hydrogen is generated by heterotrophic satellite colonies thriving anaerobically within the microalgal biomass. Although, this processes are more related to anaerobic fermentative bacteria (e.g., Escherichia coli, Clostridium spp., Thermococcales spp, Rhodobacter spp., and Rhodopseudomonas spp.), certain microalgal species (e.g., Chlamydomonas sp., Chlorella sp., and Scenedesmus sp.) have been reported to convert organic substrate into hydrogen anaerobically [50]. Due to many appealing advantages, especially cost-effectiveness and availability of readily available industrial processes, efforts upon using the second generation feedstock have been made extensively both in industry and academia. Second-generation feedstock, which come from non-food crops and agricultural waste, indeed provides even bigger potential due to its large availabilities and some existing technologies that are readily mature to process the lignocellulosic precursors. In this work, to narrow the scope of discussions, we will further focus on the BioH₂ production from lignocellulosic biomass as the feedstock.

One of the main disadvantages of lignocellulosic biomass is the required pretreatment to achieve fermentative substrates.

Table 2. Biohydrogen generation from lignocellulosic biomass.

Microbes †	Biomass ⸸	Composition after Pretreatment/g·L−1				HY	HER	Reference
		Glucose	Xylose	Acetate	Inh ‡	mol·mol−1	mmol·L−1·h−1
TT-W16	CS	2.1	9.0	1.5	0.5	2.24	4.9	[51]
AS	CS	5.5	27	1.0 !	0.1 !	2.84	0.2	[52]
CB-AS1.209	CS	3.5	1	0.5 !	0.5	1.09	1.4	[53]
MM	WS	2 !	4.2 !	1.0 !	0.6 !	1.01	0.3	[54]
AS	WS	1.2	1.1	0.7	0.5	2.81	0.9	[55]
CB-CGS5	RS	0.1	9.2	0.3	2.6	0.76	0.6	[56]
CB	SB	11	11	2.5	0.1	1.73	1.6	[57]
CS*	SS	31	13	2.1	0.1	2.6	2.1	[58]
MM	SS	2 !	6 !	1.1 !	0.1 !	2.1	0.4	[59]
CS*	WS	7	3	0.4	0.1	3.1	9.7	[60]
CS-8903	SG	1.8	1.5	1.1	0.1	1.6	0.2	[61]
TT-MJ1	SB	1.5	12	2.5	0.1	2.2	1.1	[62]
STDEV		8.53	7.36	0.81	0.71	0.78	2.77

^† TT-W16: Thermoanaerobacterium thermosaccharolyticum W16, AS: activated sludge, CB: Clostridium butyricum AS1.209, MM: mixed microflora, CB-CGS5: Clostridium butyricum CGS5, CB: Clostridium butyricum, CS*: Caldicellulosiruptor saccharolyticus, CS-8903: Caldicellulosiruptor saccharolyticus DSM 8903, TT-MJ1: Thermoanaerobacterium thermosaccharolyticum MJ1. ^⸸ CS: corn stover, WS: wheat straw, RS: rice straw, SB: sugarcane bagasse, SS: sweet sorghum, SG: switchgrass. ^‡ Inh refers to the inhibitors other than acetate, such as furfural compounds, aromatic compounds. ^! Estimate. STDEV: standard deviation.

summarizes the results of BioH₂ production from different lignocellulosic biomass hydrolysates pretreated by different approaches, using different microbes. The matrix of hydrolysate (the substrates such as the concentrations of glucose, xylose, and the inhibitors such as acetate, furfural, and aromatic compounds) plays the most critical role in determining the effectiveness of BioH₂ production once the microbial strains are chosen. In order to provide insightful understanding between the matrix of hydrolysate and BioH₂ production (HY and HER), we employed a recently developed novel correlation algorithm using the established artificial neuron networks (ANNs) combined with Box–Behnken design (BBD) design [32]. The schematic diagram of the established ANNs using training data from the references in Table 2 is depicted in Figure 3.

Because of its inherent features of fast converging and less computational time, together with equipped statistical analysis, this hybrid ANNs–RSM technique has presented very appealing application future. In this work, the references from Table 2 (in total 12 data sets) were utilized for data training. Four inputs (Glu, Xyl, Ace, and Inh), two outputs (HY and HER), together with three hidden layers (10 nodes in each layer) were used for the establishment of ANNs shown in Figure 3. Before training in ANNs, all training data were normalized using the maximum reported value from the references. For example, in the column ‘Glucose’ in Table 2, the maximum reported value for glucose is 31 (g·L⁻¹), therefore, this value was used to normalize the entire column. The detailed established ANNs–RSM using BBD design can be found in Supplementary Table S1.

Among the investigated four variables (Glu, Xyl, Ace, and Inh), the binary combinations of Glu/Ace and Ace/Inh were found to be significant to HY and HER (Supplementary Table S2). Using the established ANNs–RSM approach, the quadratic correlations between the four investigated parameters and HY or HER can be constructed with statistical confidence over 95% as the following:

$$Y=b_0+{\displaystyle \sum }_{i=1}^4{b}_i{X}_i+{\displaystyle \sum }_{i=1}^4{b}_{ii}{X}_i^2+{\displaystyle \sum }_{i<j,j=2}^4{b}_i{X}_i{X}_j$$

(7)

where Y is the predicted response (HY or HER), b_i is the linear coefficient, b₀ is the intercept coefficient, b_ii is the quadratic coefficient, b_ij is the interaction term and x_i and x_j are the coded values of variables. The obtained coefficients of the established model are summarized in Supplementary Table S3. Within investigated range from the literature reports in Table 2, the 3D plots of the effects of the combined parameters towards the HY and HER are shown in Figure 4. There are optimal conditions for the optimum values of HY and HER. Figure 4a suggests that the maximum HY will be around 3 H₂ mol per mol of the substrate, if the glucose and acetate levels are maintained at 14 and 1.3 (g·L⁻¹) respectively. Similarly, the HY will be around 3 (H₂ mol·mol⁻¹ substrate) if acetate and Inh are carefully managed at 1.4 and 1.6 (g·L⁻¹), respectively. In regard to HER, to achieve the maximum HER, the glucose vs acetate and acetate vs. Inh needed to be around 14 vs. 1.3 (g·L⁻¹) and 1.3 vs. 1.1 (g·L⁻¹), respectively. The validation results from one cross out approach is shown in Table S4. The calculated MSE and MARR suggest a good training outcome.

Although the numbers of reports and the training data sets were very limited, it does provide obvious statistical correlations between the investigated four variables (Glu, Xyl, Ace, and Inh) and two outputs (HY and HER). This also clearly suggests that the careful manipulation of the compositions, especially the ratio of glucose to acetate and acetate to inhibitors (Glu/Ace/Inh optimal around 14:1.3:1) in the pretreated broth from lignocellulose biomass, are statistically vital in optimizing the HY and HER prior to DF. These results provide very useful guidance for future optimization of BioH₂ production, let alone if great improvements could also be made from microbial strains via strain selection and genetic engineering.

5. Electron-Equivalent Balances in BioH₂ Productions

In biological conversions, a substrate such as glucose is both the source of energy and the material used for the synthesis of cells and the products. Therefore, there is a balance between the fraction of total electrons present in the substrate being transferred to the electron acceptor by metabolic reactions to generate energy (e_e) in the form of Adenosine triphosphate (ATPs) and the remaining fraction of total electrons being transferred to the biomass synthesis (e_s). The sum of e_e and e_s is 1.0 as seen in the following [39]:

$${\mathrm{e}}_{\mathrm{e}}+{\mathrm{e}}_{\mathrm{s}}=1$$

(8)

The oxidative and reductive reactions are two half-reactions together representing the net reactions. The fraction of electrons used for energy generation (e_e) and biomass synthesis (e_s) need to be known in balancing microbial reactions and the following equations will hold [41]:

$$R_e=R_a+R_d$$

(9)

$$R_s=R_c+R_d$$

(10)

where $R_e$ and $R_s$ refer to the net-generation of energy and biomass synthesis, respectively. $R_a$, $R_d$, and $R_c$ refer to the electron acceptor, donor and cell synthesis (electron acceptor) half-reactions, respectively, which leads to the fractional electron utilization expression as the following:

$$R_{overall}={\mathrm{e}}_{\mathrm{e}}\left(R_a+R_d\right)+{\mathrm{e}}_{\mathrm{s}}\left(R_c+R_d\right)$$

(11)

By coupling Equations (9) and (11):

$$R_{overall}={\mathrm{e}}_{\mathrm{e}}{R}_a+{\mathrm{e}}_{\mathrm{s}}{R}_c+R_d$$

(12)

We will get the electron-equivalent (e⁻ eq) balance as the following:

$${\mathrm{e}}^{-}\left(glucos{e}_i\right)={\mathrm{e}}^{-}\left(glucos{e}_f\right)+{\mathrm{e}}^{-}\left(SMP\right)+{\mathrm{e}}^{-}\left(biomass\right)+{\mathrm{e}}^{-}\left(H_2\right)$$

(13)

where ${\mathrm{e}}^{-}\left(glucos{e}_i\right)$ is the e⁻ eq balance of the initial glucose, ${\mathrm{e}}^{-}\left(glucos{e}_f\right)$ is the e⁻ eq balance of the residue glucose, ${\mathrm{e}}^{-}\left(SMP\right)$ is the e⁻ eq balance of soluble metabolic product, and ${\mathrm{e}}^{-}\left(H_2\right)$ is the e⁻ eq balance of the hydrogen formation. The empirical formula for bacteria cells is often assumed to be C₅H₇O₂N, which slightly varies with the type of microbes and cultivation media. On the basis of one mole, the quantities of electrons in the e⁻ eq balance with its corresponding compounds are summarized as the following: $C_5{H}_7{O}_{2}N=20{\mathrm{e}}^{-}$ eq, acetic acid $=8{\mathrm{e}}^{-}$ eq, butyric acid $=20{\mathrm{e}}^{-}$ eq, propionate $=14{\mathrm{e}}^{-}$ eq lactate = 12e⁻ eq and H₂$=2{\mathrm{e}}^{-}$ Equation By incorporating the above quantities, the established stoichiometric equations of the electron donor-acceptor half-reactions for different compounds are expanded as the following [63]:

Glucose:

$$\left(1/24\right){\mathrm{C}}_6{\mathrm{H}}_{12}{\mathrm{O}}_6+\left(1/4\right){\mathrm{H}}_2\mathrm{O}\stackrel{}{\to }\left(1/4\right){\mathrm{CO}}_2+{\mathrm{H}}^{+}+{\mathrm{e}}^{-}$$

(14)

Acetic acid:

$$\left(1/8\right){\mathrm{CO}}_2+\left(1/8\right){\mathrm{HCO}}_3^{-}+{\mathrm{H}}^{+}+{\mathrm{e}}^{-}\stackrel{}{\to }\left(1/8\right){\mathrm{CH}}_3{\mathrm{COO}}_{}^{-}+\left(3/8\right){\mathrm{H}}_2\mathrm{O}$$

(15)

Butyric acid:

$$\left(1/5\right){\mathrm{CO}}_2+\left(1/20\right){\mathrm{HCO}}_3^{-}+{\mathrm{H}}^{+}+{\mathrm{e}}^{-}\stackrel{}{\to }\left(1/20\right){\mathrm{CH}}_3{\mathrm{CH}}_2{\mathrm{CH}}_2{\mathrm{COO}}_{}^{-}+\left(7/20\right){\mathrm{H}}_2\mathrm{O}$$

(16)

Propionic acid:

$$\left(1/7\right){\mathrm{CO}}_2+\left(1/14\right){\mathrm{HCO}}_3^{-}+{\mathrm{H}}^{+}+{\mathrm{e}}^{-}\stackrel{}{\to }\left(1/14\right){\mathrm{CH}}_3{\mathrm{CH}}_2{\mathrm{COO}}_{}^{-}+\left(5/14\right){\mathrm{H}}_2\mathrm{O}$$

(17)

Biomass:

$$\left(1/5\right){\mathrm{CO}}_2+\left(1/20\right){\mathrm{HCO}}_3^{-}+\left(1/20\right){\mathrm{NH}}_4^{+}+{\mathrm{H}}^{+}+{\mathrm{e}}^{-}\stackrel{}{\to }\left(1/20\right){\mathrm{C}}_5{\mathrm{H}}_7{\mathrm{O}}_2\mathrm{N}+\left(9/20\right){\mathrm{H}}_2\mathrm{O}$$

(18)

Lactate:

$$\left(1/12\right){\mathrm{CO}}_2+\left(1/12\right){\mathrm{HCO}}_3^{-}+{\mathrm{H}}^{+}+{\mathrm{e}}^{-}\to \left(1/12\right){\mathrm{C}}_3{\mathrm{H}}_5{\mathrm{O}}_3^{-}+\left(2/12\right){\mathrm{H}}_2\mathrm{O}$$

(19)

Hydrogen formation:

$${\mathrm{H}}^{+}+{\mathrm{e}}^{-}\stackrel{}{\to }\left(1/2\right){\mathrm{H}}_2$$

(20)

With the detailed characterization of fermentation broth of the above liquid and gaseous metabolites on time on stream (TOS), the dynamic variations of complicated metabolisms are quantitatively gauged and monitored during the DF. Due to the limited numbers of reports with detailed characterizations of the distributions of electron sinks during DF, we only summarize some recent research works with available data of detailed metabolism during DF in

Table 3. Stoichiometry variation by the effects of using different substrates.

Substrate	Biomass /%	Acetate /%	Butyrate /%	Propionate /%	HY mol·mol−1	HER mmol·L−1.h−1	Microbes	Operation /−	Reference
Cornstover/	22.6	22.1	31.5	15.5	0.9	0.2	CB	Bat	[39]
Cornstover+NP Ni	21.9	21.9	31.6	15.3	1.2	0.3	CB	Bat	[39]
Organic wastewater	2	56.5	24.4	9.1	1.3	0.5	AS	Cont	[64]
Sugarcane	17.3	22.1	47.7	7.1	0.9	2.3	CL	Bat	[65]
Sugarcane+Fe2+	13.5	23.3	47.5	2.1	1.3	2.2	CL	Bat	[65]
Pistia stratiotes	25.3	25.9	19.2	2.2	2.6	1.7	AS	Bat	[66]
Glucose	1	37.8	3.4	17.9	2.7	9	AS	Bat	[67]
Xylose	5.6	18.7	-	-	1.1	0.2	KO	Bat	[68]
Sucrose	9.1	47.6	41.2	3.5	0.7	10	AS	Cont	[69]
SW-CWP	1.2	11.8	39.9	27.3 *	1.4	2.9	AS	Cont	[70]
Glucose	6.8	15	10	3.5	1.0	2	CT	Cont	[71]
Glycerol	8.3	6	42.5	1.8 #	0.6	2.7	CP	Bat	[72]
Glucose	1	47.3	−	−	3.5	41	TC	Bat	[73]
Lac/Ace	7.2	17.2	68.8	−	0.4	6.6	AS	Cont	[74]
TV	5.8	16.1	54.4	11.9 #	1.9	7	PT	Cont	[75]

* lactate. ^# succinate. NP Ni: nanoparticle Ni, SW-SWP: synthetic wastewater containing cheese whey powder, Lac: lactate, Ace: acetate, TV: Tequila vinasse. CB: Clostridium butyricum, CL: Clostridium spp, AS: activated sludge, KO: Klebsiella oxytoca, CT: Clostridium tyrobutyricum, CP: Clostridium pasteurianum, TC: Thermobrachium celere, PT: PTA-124566. Bat: batch, Cont: continuous.

. In regard to comparisons of BioH₂ performances across different reports, the challenges such as failure or omission of complete mass balance during DF still remain. The HY versus HER among the literature reports are summarized and compared in Figure 5.

The BioH₂ performances over the averaged value were annotated for easiness of discussion (Figure 5). There are three different regions of the BioH₂ yield (HY) versus rates (HER). In our recent work, the performances of BioH₂ yield (HY) versus rates (HER) from DF among different literatures were rigorously compared [2]. It was found that the averaged value of HY = 1 mol·mol⁻¹ for HER = 0.3 (mol.L⁻¹·h⁻¹) generally agrees well with most of the reported values with the exception of two reports [67,69], of which HER values reach 6 and 10 (mol.L⁻¹·h⁻¹), respectively. For these two relative higher HER in the region (I), they both employed a mixed culture of microbes isolated from activated sludge and the substrates used for cultivation were sucrose and lactate, respectively. The appreciable difference of reported works in the region (II) lies in the obvious increased HY though HER also increase slightly compared with the averaged value in the region (I). Among these three reports, using glucose as substrate by mixed culture presents the best performance. There is one exceptional high value for both HY and HER (3.5 mol·mol⁻¹ and 40 mol.L⁻¹·h⁻¹, respectively) using glucose as substrate with a pure culture of Thermobrachium celere (TC) strain (region III). Possible reasons not only derive from the inherent metabolic features of microbe during DF but can also be potentially attributed to the types of operations such as continuous operation at the optimized conditions.

The distributions of electron sinks in the major metabolites across different works using different microbial strains are summarized and shown in Figure 6a (HY) and Figure 6b (HER). For the convenience of comparisons, three top high values by considering both HY and HER (for HY using a red square and HER using a red circle) were highlighted. To yield higher HY and HER, the relatively larger electron sinks in acetate are generally observed except using Klebsiella oxytoca (KO) microbial strains that lead to a low HER possibly due to the direct employment of xylose that mainly relies on the formate cleavage metabolic pathway [68]. This agrees with the pyruvate metabolism during fermentation on the stoichiometric basis. In regards to the biomass synthesis, the correlations between the electron sinks and BioH₂ performances present discrepancy. Since the microbial cell is the host for reactions including hydrogen generation, the metabolism for BioH₂ generation needed to be compromised with biosynthesis metabolism. This agrees well with the literature reports obtained from DF using sugarcane hydrolysate as substrate in batch operation, with the electron sinks for acetate (36%) and biomass (35%) being similar to each other. On the other hand, larger HY and HER are also observed in some reports where electron sinks in the biomass only takes about 1–2% of the entire liquid metabolite (Figure 6a,b). This discrepancy most likely stems from the approaches of operations between batch and continuous operations (Chemostat), of which the batch operation will experience a continuous biomass increases after inoculation while the Chemostat will maintain a dynamic steady balance for biomass after start-up. Although the numbers of literature reports for electron-equivalent balance comparison are limited, it potentially suggests that the continuous operations tend to produce better BioH₂ performances if all other conditions are maintained the same. In addition, the distributions of electron-equivalent balances also indirectly reflect the activities of different genus of bacteria if a mixed culture is employed during the fermentation. Depending on the substrate used during fermentation, the activities of different genus in the microbial ecology vary accordingly. The hexose especially glucose is one of the most easily utilized carbon sources during fermentation across all different types of microbes [76]. As for pentose, when acetate is produced as a by-product, xylose can be converted to H₂. While a wide range of bacteria has the capability of utilizing hexose to produce H₂, only a few pentose-fermenting microorganisms have been identified [68,77]. For example, it has been found that the Caldicellulosiruptor saccharolyticus could utilize the substrate (xylose and glucose) simultaneously for cells growth and H₂ production and the consumption efficiency was higher than xylose as a sole carbon source [78,79]. Apart from conventional hexose such as glucose, other substrates such as lactate, acetate can be also utilized as the carbon source for BioH₂ generation by the species such as Clostridium beijerinckii, Clostridium butyricum, and Clostridium tyrobutyricum and these strains are found to be associated with the activities of butyrate metabolism from pyruvate [80,81,82,83]. To sum up, the level of acetate in metabolites is surely an indicator of relative higher BioH₂ yield (HY) and generation rate (HER). In batch operation, the electron sinks in biomass synthesis also needs to be maintained at a relatively higher level (similar level to that in acetate sink). The continuous operations tend to yield the larger HY and HER albeit the electron sinks in biomass synthesis is maintained at a low level.

6. Results Comparison among Different Process Intensifications

The ultimate goal in BioH₂ production research is to enhance the hydrogen production so as to be as cost-effective as possible. For process intensification, the approaches such as pure culture, continuous operation, consolidated saccharification and fermentation, chemical additions, are optimization avenues. In this work, for convenience of discussions, the comparisons were divided into the following five categories, namely pure culture (PC) [73,84,85,86,87,88,89,90,91,92,93,94], chemical additions (CA) [2,39,76,95,96,97], continuous operation with mixed culture (COMC) [98,99,100,101,102], batch operation with mixed culture (BOMC) [103,104,105,106,107,108] and consolidated saccharification and fermentation (CSF) [18,31,61,109,110,111], respectively. For quantitative comparison of hydrogen yield across different literatures, the necessary Carbon-molar-based mass balance for DF is pivotal and necessary. Because of the omission of mass balances during DF in some reported works, we only adopted those literature reports with complete mass balance and rate expressions. Within each category, all the collected BioH₂ (HY and HER) were averaged and corresponding standard deviations were calculated at the following:

$$\overline{x}=\sqrt{\frac{1}{N-1}{\displaystyle \sum }_{i=1}^N{\left(x_i-\overline{x}\right)}^2}$$

(21)

where N represents the numbers of references, $\overline{x}$ represents the averaged values, x_i refers to the specific one reported value. The comparison of results across different literatures is depicted in Figure 7. Apparently, one of the most prominent factors that determine the performance for BioH₂ production is the type of microbial strain. The selection of microbial strains with exceptional higher capacity in hydrogen productions is the most appreciable approach in enhancing BioH₂ production. Clearly, with the isolation of pure strain potentially compounded with genetic engineering, the recombined bacteria with relative stable plasmid that possesses appealing features of larger hydrogen production will be an attractive approach for future BioH₂ production research and development. On the other hand, more efforts should be focused on the issues of improving the stability of genetic expression, biosecurity, and the cost-effectiveness of cultivation of genetic engineered bacteria. Apart from microbial strains, chemical additions such as using synergistic factor like biochars and nanoparticles, are found to effectively improve BioH₂ production. Besides to chemical addition, continuous operation with mixed culture is also found to be effective in enhancing BioH₂ production. Obviously, the batch operation and simultaneous saccharification and fermentation, especially using pretreated lignocellulosic biomass as fermentation broth still present quite low ceiling values for BioH₂ productions. The effective enhancement of BioH₂ production not only lies in manipulating the optimal compositions in the pretreated solutions (maximizing the reducible sugars especially the glucose and minimizing toxic compounds especially the acetate) prior to the fermentation, but also relies on the high performance fermentation operations. From the above comparisons, a more realistic enhanced process approach might be incorporating all the appealing features from different types of process intensification approaches discussed above into an integrated bioprocess for hydrogen fermentation.

The inherent limitations i.e., finite reserves which are located in geopolitical sensitive region, volatile prices, not renewable, greenhouse gas emissions, etc. that limit the application of fossil fuels for sustainable development meeting future energy and chemical needs [112,113,114]. Currently, more and more countries have realized the importance of shifting the paradigm of the development of human society from the primitive concept of “take, make, and dispose of” to “reuse and recovery” of resources to achieve “healthy environment-healthy human being” and “socio-economic prosperity” [115]. From an energy cascade utilization and material recycling and reused perspective, the utilization of second-generation biomass feedstock (lignocellulosic biomass) might need to follow a more delicate pretreatment process to fractionate three main compositions (cellulose, hemicellulose, and lignin) first, then followed by the subsequent conversion [116,117,118]. Due to inherent resilient features of lignocellulose, the pretreatment process is one of the core processing step for lignocellulose conversion. The recent works deploying different types of pretreatment are summarized and commented (merits and drawbacks) in

Table 4. Merits and drawbacks of different pretreatment for lignocellulosic biomass.

Types of Pretreatment	Merits	Drawbacks	References
Physical Millings	Effective downsizing the particle size and degree of polymerization Relative less environmental impact and chemical consumptions	High power consumption Limited enzymatic digestibility. High CAPEX and OPEX	[119]
Microwave	Simple operation Fast heating rate Low energy consumption Specific destructive to crystallization of cellulose	Cost unfriendly Difficult to scale up Ineffective delignification Formation of inhibitors	[120]
Chemical Acid	Fast liberation of reducible sugars Mild temperature Cost effectiveness Low concentration Maturation	Generation of inhibitors Recovery of acidic catalyst Detoxification process Environmental impact	[116]
Alkaline	Reasonable low CAPEX Commercial available Industrial mature Highly delignification	Black liquor recovery Scale dependent Silicate colloids	[118]
Ionic liquid	Environmental compatibility Low operational pressure No release of toxic or explosive gases Thermal stability	High temperature Long residence time Cost unfriendly Poor delignification Recovery of ionic liquid	[121]
Organosolv	Low temperature and pressure Low cellulose lost Formation of less destructive lignin Effective delignification	High lignin dependent High CAPEX Solvent recovery Inhibitor generation	[122]
Ozonolysis	Low inhibitors generation Ambient operational condition	Flammability of ozone High CAPEX for ozone generation onsite	[123]
Steam explosion	Less toxic inhibitors No need for size reduction Higher yield for glucose and hemicellulose Short residence time	High CAPEX Hard to scale up Immature industrial scale Less destructive to hemicellulose Poor delignification	[124]
Biological Enzymatic	Selective delignification Less inhibitors Simple downstream process No need for chemical recovery	Long residence time High cost for enzyme production Low downstream yield Instability and deactivation of enzyme	[125]
Fungal	High efficiency and downstream yield Minimum inhibitors generation Low energy consumption No need for chemical recovery Short downstream process	Commercial immature Long treatment time Compromise by carbohydrate lost Low downstream yield Instability during cultivation Hard to industrial scale up	[126]
Microbial consortium	Relative shorter treatment time Increased enzymatic accessibility Enhanced productivity	Operational condition sensitive Sensitive to lignin content Complicated intermediates during cultivation	[127]

.

The merits and drawbacks of different pretreatment for lignocellulosic biomass are shown in Table 4, which summarizes both merits and drawbacks of some widely deployed pretreatment approaches (physical, chemical, and biological) for lignocellulosic biomass fractionation. Although the pretreatment processes for lignocellulosic biomass utilizations have re-emerged into scope of research interests both in academia and industry during last two decades, alkaline delignification process with the aims of production of cellulose based paper pulp still is the undebatable mature and profitable process for commercial lignocellulosic biomass fractionation. The key technical hurdles for these commercial processes to overcome lie in the cost-effective retrofitting the existing downstream process that is able to integrated-utilizations of other compositions (decomposed hemicellulose and lignin) other than cellulose alone. In the meanwhile, breakthrough for alkaline recovery other than incineration of black liquor after delignification needed to be made so as to avoid bypassing the utilization of lignin and hemicellulose in black liquor by incineration (alkaline recovery during paper pulping) to realize the material recycling and energy cascade utilization. Apart from alkaline pretreatment, the dilute acid pretreatment had been found as one of the most mature and reliable process for lignocellulosic fractionation followed by subsequent utilization [128]. By considering maturity and reliability, the efforts paid upon these realms might potentially lead to a big engineering breakthrough.

The cost estimation of any developed process can be performed by studying the cost of the different components of the overall process. In the case of biomass pretreatment, these components include the cost of the biomass, equipment (fixing the CAPEX), chemicals, energy consumption, and operational expenses (depreciations, labors, plant overhead, etc.) during the process. The current status quo of economic analysis or cost estimation of biorefinery using lignocellulosic biomass mainly depends on the laboratory scale (some were conducted on pilot scale) experimental data and overall mass balances analysis to roughly estimate the associated CAPEX and OPEX costs [129]. Among the costs for different pretreatment process, the conventional acid hydrolysis are still among the most competitive and widely industrial adopted process for subsequent biorefinery such as ethanol and butanol productions [130]. For example, Kazi FK et al. performed the economic analysis of four different pretreatment processes namely, dilute acid, two-stage dilute acid, ammonia fiber expansion (AFEX), and hot water pretreatment. The result indicates that the highest ethanol yield of 289 (L·t⁻¹) with lowest gasoline cost of 1.36 (US $·L⁻¹) of gasoline equivalent for dilute acid method [131]. According to André et al., total annual cost (TAC) of diluted acid hydrolysis process developed from national renewable energy laboratory (NREL) is still one of most competitive process achieving at 225 M$, while liquid hot water (LHW) and AFEX processes are 328 and 359 M$, respectively [132]. This might be the reasons that the numbers of available reported works focusing on using acid hydrolysis for pretreatment are still quite high. In this work, a potential process for the integrated utilization of biomass combining the conventional mature thermal process with DF to produce BioH₂ is proposed and depicted in Figure 8. The selection of biomass compounds for conversion into hydrogen depends heavily on the inherent physical and chemical features of three main compositions (hemicellulose, lignin, and cellulose). Taking lignin (mainly available from industrial black liquor) for example, containing large amount of aromatic compounds (resin like materials with relative higher heating value), it might be an appealing precursor for hydrogen generation via conventional thermal approaches such as gasification and steam reforming [133,134].

The hydrolysis of polysaccharide (cellulose and hemicellulose) also needs proper manipulation prior to the fermentation process. Therefore, the additional buffering or adjustable units that are able to adjust the optimal ratio of the reducible sugars and the main inhibitors are necessary prior to fermentation process. In addition, the careful selection of process intensification approach is needed as the matrix of fermentation broth from polysaccharide hydrolysate varies significantly. The necessity of using the mixed culture that is able to utilize different types of sugars (hexose and pentose) via different metabolic pathways needs to be carefully scrutinized in order to enhance BioH₂ production. Furthermore, the effective incorporation of approaches such as long term continuous operation and chemical additions (biochar and nanoparticles) will be helpful in further boosting BioH₂ production if the genetic stability and recovering the chemical additions could be cost-effectively improved. However, the challenges of this proposed integrated process still lies in: (a) the cost-effective fractionation of the three main compositions from biomass, (b) how to optimize the feed compositions in the fermentation broth, as assessed from ANNs study that the glucose/acetate/inhibitor needs to be maintained to a certain preferable range if natural microbial strains are inoculated, (c) the availability of large amount of water onsite, (d) the cost-effective CO₂ separation and utilization, (e) the recycling of the hydrolysate and cost-effective pH swing process with nearly zero emission, and (f) a comprehensive life circle analysis (LCA) for the total process.

7. Conclusions

The collected reference reports using lignocellulosic biomass for BioH₂ production were deployed for supervised machine learning via the constructed artificial neuron networks (ANNs) using feed backward propagation together with one cross-out validation approach to construct statistical and quantitative correlations between inputs of data matrix (including carbon sources i.e., glucose, xylose, and inhibitors i.e., acetate, furfural, and aromatic compounds) and outputs (including hydrogen yield (HY) and hydrogen evolution rate (HER)). The statistical analysis indicates that the optimal compositions of glucose (around 14 g·L⁻¹) and acetate (around 1.3 g·L⁻¹) will yield optimal HY (3 H₂ mol·mol⁻¹ substrate and maximum HER, which are independent to the microbial strains and approaches of operations among the investigated literature reports. The comparative studies also indicate that the larger electron sinks in the acetate is found to be appreciably related to relatively higher HY and HER. Apart from the selection of microbial strains with exceptional higher capacity in hydrogen productions, which is still potentially substantial approach for the improvement of BioH₂ production, other process intensifications using continuous operation compounded with synergistic chemical additions might also enhance the BioH₂ productions during dark fermentation. The integrated utilization of lignocellulosic biomass for the hydrogen production relies on the seamless coupling of conventional mature thermal hydrogen generation process with the highly effective and process enhanced biological hydrogen production route.