Oils derived from oleaginous microbes, such as algae, yeast and bacteria (so-called single-cell oils (SCOs)) have long held promise as a biofuel precursor due to the high lipid content (>20%), high growth rate and the ability of these microbes to be cultivated on feedstocks or in areas that do not compete for terrestrial food production (1). Historically, the availability of microbial lipids has been quite limited, but technology developments in food and feed applications may begin to benefit lipid production for biofuels as well. Corbion produces an algae oil high in oleic acid for culinary use, and an algae-based omega-3 fatty-acid-enriched product for fish feed. Qualitas Health, DSM, Evonik and Veramaris also produce omega-3 enriched oils for food and feed applications, though to our knowledge only Qualitas Health is producing the algae photoautotrophically. While the target omega-3 fraction is not ideal for fuel production due to long carbon chains and numerous double bonds that would increase hydrogen consumption during hydroprocessing, the process to enrich omega-3 oils likely generates a byproduct of lighter and more saturated fatty acids that may be more suitable for fuel production than the omega-3 fraction. Additionally, although many oil majors and startup companies have pivoted away from algal biofuels, a partnership between Synthetic Genomics and ExxonMobil to improve algal biology for fuel production is still active. The increasing availability of cellulosic sugars may also pave the way for larger scale cultivation of heterotrophic oleaginous microbes.

While both the growth and characterisation of oleaginous microbes (1–13) and the conversion of vegetable oils to fuels (13–27) have been extensively studied, there has been considerably less research on converting SCOs to fuels, and especially on converting SCOs to so-called ‘drop-in’ hydrocarbon fuels that are compatible with existing fuel infrastructure (28–30). However, production of hydrocarbon fuels from SCOs is a growing area of research and hydroprocessing of these oils requires some considerations that are unique to the differing composition of SCOs. Enough recent studies have been published that a summary of findings and perspective on research directions is warranted, especially with respect to the differences between terrestrial oils and SCOs. In particular, lipid recovery, lipid cleanup and catalyst selection all appear to play a key role in hydroprocessing performance, and there are clear needs for deeper research in each of these areas.

In most cases, SCOs are produced intracellularly, and thus necessitate separation from the other cell components prior to upgrading. Additionally, the microbial biomass is typically recovered with a very high water content (≥80%) relative to terrestrial biomass. Thus, while vegetable oils can be effectively extracted by pressing, solvent extraction or a combination thereof, these operations are usually not effective when applied directly to microbial biomass. Furthermore, it is energy intensive to dehydrate microbial biomass and therefore strategies that can extract oil in high yields from wet biomass are required. The first step is commonly to rupture the microbial cell wall which allows access to the lipids.

Various cell wall lysis methods have been applied for this application, such as high-pressure homogenisation (HPH), bead milling, ultrasound, pulsed electric field (PEF), osmotic shock, microwave, subcritical water hydrolysis, enzymatic hydrolysis, autolysis and chemical hydrolysis (31). From the perspective of energy consumption, pretreatment methods such as HPH, subcritical water hydrolysis, enzymatic and chemical hydrolysis are more attractive due to the relatively lower energy requirements. Moreover, all these methods can be scaled up to industrial application (32). Researchers at the National Renewable Energy Laboratory (NREL) have shown that dilute acid pretreatment is an effective, low cost, energy efficient and scalable means to recover lipids from wet microbial biomass (31–34).

It is also necessary to consider the compatibility of the cell lysis method with lipid extraction. For example, hexane is a commonly employed extraction solvent, but does not always maximise extraction yields. In many cases, modification of solvent polarity (for example employing a mix of hexane with ethanol or isopropanol) is necessary for high lipid recovery (35–39). However, incorporation of a polar solvent also often leads to more prevalent co-extraction of other polar materials, such as polar lipids, pigments, proteins and sterols, that require more severe cleanup of the lipid phase. In particular, extraction methods that incorporate chlorinated solvents, such as the well-known Folch and Bligh-Dyer extraction protocols, tend to extract significant amounts of polar lipids and other polar materials (36, 38). Combinations of hexane with alcohols, such as ethanol and isopropanol, have the potential to more selectively extract hydrophobic lipids while maintaining high extraction yields of these components (31, 39, 40). The optimal balance is thus a function of lipid speciation, content and speciation of other potential co-extractives, the lipid cleanup strategy and the hydroprocessing approach.

SCOs that are considered for biofuel production are mainly recovered from oleaginous algae, fungi (yeast and filamentous) and bacteria, with the number of known oleaginous species decreasing by family in the order listed. These microbes may be cultivated autotrophically (using only CO₂ as a carbon source), mixotrophically (using a mixture of CO₂ and organic carbon sources) or heterotrophically (using organic carbon sources, such as glucose and acetate). Several recent reviews have compiled oil content and broad composition of oleaginous microbes and their production of oils with an emphasis on wastes and lignocellulosic carbon sources (7–9, 11). A sampling of these compositions is reproduced in Table I, with comparison to common vegetable oils. It is important to note that analytical methods vary greatly across the references cited in Table I. The lipid content data, especially, should be taken with a grain of salt.

Both SCOs and terrestrial plant oils typically contain high proportions of palmitic (16:0), palmitoleic (16:1), stearic (18:0), oleic (18:1) and linoleic (18:2) acids. On the surface, the similar composition of both oils appears to make SCOs an attractive feedstock for catalytic upgrading. However, SCOs can contain a broader distribution of lipid classes than terrestrial plant oils, including higher proportions of polar lipids, free fatty acids (FFA), sterols, terpenes, carotenoids and chlorophyll, as well as unidentified compounds, that can be co-extracted with the desired monoacylglycerides (MAG), diacylglycerides (DAG), triacylglycerides (TAG) and FFA. Examples of these other lipid classes are shown in Figure 1. The proportions of each vary greatly depending on the organism, growth mode, growth condition, growth stage and postharvest processing conditions, including biomass and extracted oil storage and solvent extraction conditions (36, 60). For example, glycolipids and phospholipids can account for 17–90% of the total lipids in autotrophically-grown algae (61), compared to ~2% in soybean oil (62), and lipase enzymes present in the cells can convert acylglycerides to FFA during lipid or biomass storage (60).

Fig. 1.

TAGs and FFAs are usually considered as the favoured precursors for biodiesel or hydrocarbon-based biofuels such as renewable diesel or renewable jet fuel. Other components are often detrimental in catalytic hydroprocessing as they either increase hydrogen consumption due to a high degree of unsaturation (for example terpenes, carotenoids and chlorophyll), contain heteroatoms that can poison metal or acid catalysts (for example polar lipids and chlorophyll) or contribute to poor cold flow properties (for example sterol-derived cycloparaffins). For example, polar lipids mainly consist of phospholipids, glycolipids, lipoproteins and sulfolipids. A phospholipid molecule consists of a polar phosphorus-containing moiety (such as phosphate, phosphatidylethanolamine or phosphatidylcholine), whereas a glycolipid contains a polar carbohydrate moiety in place of the phosphor group. The sidechains of polar lipids (sugars, proteins or phosphorous-containing molecules) may also include other nitrogen, sulfur or phosphorus-containing moieties (21).

On the other hand, hydrotreating of some of these contaminants can be beneficial in that the high degree of branching (for example in terpenes and the phytol chain of chlorophyll) can potentially improve cold flow properties and increase the value of the renewable diesel fuel. Additionally, when a large proportion of the lipid fraction comprises nominally undesirable molecules (such as polar lipids), removal of the entire molecule in a cleanup strategy would result in low fuel yields. Thus, while some cleanup of the oils is likely necessary, the most desirable cleanup strategy is not universal across SCOs, and tradeoffs between overall yields and catalyst longevity must be considered.

Cleanup of SCOs targets many of the same impurities that are removed in refining of vegetable oils for human consumption, and thus cleanup approaches can be broadly categorised into the same four operations: bleaching to remove pigments and polar compounds, degumming to precipitate phospholipids, deodorisation to remove FFAs and distillation to remove additional FFAs and other volatile matter. When refining SCOs for hydrotreating, some modifications may be necessary. For example, FFAs are one of the preferred feeds for hydrotreating so deodorisation and distillation will not always be needed. However, in some cases it may be advantageous to hydrolyse the entire lipid stream to FFAs, distil these FFAs away from impurities, and route the distillate to hydrotreating. In these cases, bleaching and degumming may not be needed. Researchers have explored to some extent each of these steps except deodorisation. These techniques and their advantages and disadvantages are summarised in Table II.

Crude extracted oils from autotrophically-grown microbes are often high in chlorophyll content. The removal of chlorophyll is important in both hydroprocessing and biodiesel production as chlorophyll contamination can deactivate catalysts and degrade fuel quality (60, 63, 64). For example, the porphyrin head of the chlorophyll molecule is highly unsaturated and contains the catalyst-poisoning heteroatoms nitrogen and magnesium. The two most common approaches to remove chlorophyll are adsorption on a solid material and dealkylation via acid treatment.

Bleaching earths, activated carbon and silica are common adsorbents for oil refining. Adsorption of chlorophyll is initiated by ion exchange of the Mg²⁺ centre of the porphyrin head with a proton. The metal-free porphyrin is then protonated and adsorbed (65). Adsorbents are commonly pretreated with mineral acids to increase the number of exchangeable protons (66). Ultrasound-assisted adsorption has also been demonstrated to increase the rate of chlorophyll removal from oils (67). It is worth noting that the loadings of polar adsorbents to fully remove chlorophyll from SCOs may be much higher than typically used for edible oil bleaching due to the higher levels of chlorophyll (and other adsorbing contaminants). For instance, Chen et al. employed bleaching earth to adsorb chlorophyll from microalgal biodiesel at a mass loading of 16 wt% (60), whereas a typical edible oil bleaching process uses a mass loading of 1.5 wt% (68). Increasing adsorbent loading, while effective in reducing chlorophyll content, decreases yields due to co-adsorption of lipids. This is particularly true of bleaching earths, which can retain up to a third of their mass in oil (69), though even higher retentions have been claimed (70). Similarly, Santillan-Jimenez et al. removed chlorophyll from crude extracted algae oil via adsorption on K10 montmorillonite clay and activated carbon, recovering only 58 wt% and 46 wt% of the original extract mass, respectively (71). Gas chromatography-mass spectrometry (GCMS) analysis of the recovered materials indicated a high content of fatty acids, hydrocarbons and phytol isomers, while ultraviolet-visible (UV-vis) spectroscopic and inductively coupled plasma (ICP) analyses revealed complete chlorophyll, phosphorus and magnesium removal. Notably, the relative abundance of hydrocarbons in the activated carbon-bleached material was lower than in the K10-bleached material, likely due to the greater hydrophobicity of the activated carbon retaining more of the hydrocarbons. This observation is also consistent with the lower total recovered yield from the carbon adsorbent. Thus, adsorbents with higher polarity may be more selective for removal of the target compounds.

Phosphoric acid-catalysed dealkylation was originally developed by Diosady to refine Canadian canola oil, which has anomalously high chlorophyll content for a terrestrial oil due to weather conditions in the region (70). The acid bleaching approach employs anhydrous acids, such as phosphoric acid or sulfuric acid to cleave the phytol side chain from the porphyrin structure, the latter of which is not soluble in oil. Thus, this precipitable form of pheophorbide can be easily separated from the oil by filtration while the hydrophobic phytol is preserved. Retaining the phytol in the oil confers the additional benefit of improved biofuel yield and performance due to a high degree of branching. Dong et al. employed a modified version of Diosady’s method to remove 99.7% of chlorophyll from algae oil (63). In a direct comparison of adsorptive and dealkylative approaches, Kruger et al. found that phosphoric-acid-catalysed dealkylation was more effective than adsorption on silica for the removal of chlorophyll from algae oil (72). In the comparison, adsorption on silica gel removed 85% of the oil nitrogen content and still showed a detectable chlorophyll signal in UV-vis analysis, while acid bleaching removed 92% of the oil nitrogen and did not show a detectable chlorophyll UV-vis signal. Moreover, phosphoric acid bleaching is compatible with conventional vegetable oil refining practice, since phosphoric acid is routinely added for oil degumming. However, for the same reason and also because the acid must be washed out of the oil by adding water, phosphoric acid bleaching may suffer from yield losses and poor bleaching performance in oils with high levels of phospholipids.

Apart from adsorption and acid treatment, other novel methods of chlorophyll removal have been reported. Sathish developed a multi-step process to remove chlorophyll from microalgal crude oil, involving sequential aqueous acid, base and acid treatments to lyse cells, saponify lipids and precipitate chlorophyll, respectively (73). Though the extraction process was not optimised, roughly 20% of the lipids remained with the biomass solids and another 20% with the precipitated chlorophyll. However, the purified lipids showed no chlorophyll signal in UV-vis analysis. The process is advantageous in that it does not require anhydrous conditions, but the sequential use of concentrated acids and bases is likely to be expensive and generate highly saline wastewater. On the other hand, Li et al. removed chlorophyll from intact microalgae by saponification with sodium hydroxide (74).

Analogous to chlorophyll, SCOs can have higher phospholipid content than most terrestrial oils, and removal of these polar lipids is necessary to prevent catalyst fouling in downstream processes. Further complexity arises as a large portion of phospholipids in SCOs are non-hydratable, namely phosphatidic acid and phosphatidylethanolamine (75). These species complex with metal cations and cannot be removed by conventional water degumming processes that precipitate hydratable phospholipids. Other degumming methods have been developed to remove these non-hydratable phospholipids (76).

Acid degumming is the most common approach to removing non-hydratable phospholipids. An acid is added to decompose phospholipid salts to improve their hydratability. Phosphoric acid is routinely used in algae oil degumming given the co-benefit of chlorophyll reduction (32, 54, 60). A drawback of acid degumming, however, is that some decomposed phosphatic acid will remain in the oil phase (76). Subsequent addition of a small amount of diluted base – enough to neutralise phosphatic acid while avoiding saponification – has been shown to be effective in further reducing phosphorus content of acid-degummed oils (77). Other methods which target both hydratable and non-hydratable phospholipids, such as membrane or enzymatic degumming, are in nascent stages of development. While promising, the cost of these techniques are prohibitively expensive for SCO degumming in their current state (76).

Given the potentially high content of microbial polar lipids, a remarkable amount of microbial lipid may be lost in the degumming process (60). Nevertheless, in some SCOs degumming may be appropriate, and examples of both dilute acid and solvent degumming exist in the patent literature (78–80). For SCO streams containing large fractions of polar lipids, hydrolysis to a FFA stream is likely preferable for fuel production (34).

As an alternative to removing impurities from the lipid stream, approaches have also been developed to remove only the desired components from the lipid stream and leave everything else. Within this realm, hydrolysis and methanolysis are the primary approaches applied, generating FFA and FAME, respectively, both of which are distillable.

Hydrolysis can employ acids, bases or high temperature water to cleave fatty acid and other esters. Alkaline hydrolysis (i.e. saponification) is advantageous in that it requires mild temperatures and short reaction times (typically 80°C and 1 h, respectively). However, the stoichiometric consumption of alkali likely makes this approach too expensive for a fuel production scenario. Acid hydrolysis is favourable in that the acid is catalytic and mild temperatures are also typically employed (100°C or less), but reaction times are typically 8–24 h. Acid hydrolysis was the basis of the industrial Twitchell process, which was subsequently superseded by steam splitting. Steam splitting via the Colgate-Emery process is the current industrial standard, employing temperatures of 250–330°C and reaction times of 2–3 h (81). Higher temperatures can significantly shorten the reaction time (for example to ~10 min), but also significantly increase the operating pressure (for example to ~2500 psig) (82). Steam splitting is effective with both acylglycerides and with phospholipids (82). Lawal et al. demonstrated the effectiveness of hydrolysing algal lipids to FFAs and distilling the hydrolysed FFAs away from impurities (83). An alternative approach involves engineering microbes to produce and excrete FFAs directly into the growth medium, which could simplify the cell lysis and lipid extraction operations. This approach was recently demonstrated in the cyanobacteria Synechocystis sp. PCC6803 (45).

Alternatively, microbial lipids can be converted to methyl esters, which facilitates distillation without vacuum. The (trans)esterification to methyl esters is widely used on a commercial scale to produce biodiesel, and occurs under similarly mild conditions as saponification. Murzin and coworkers demonstrated the effectiveness of this cleanup approach for removing impurities in algae oil (84–86). However, this approach suffers from two significant drawbacks, namely that it requires a dry feedstock and that it consumes methanol stoichiometrically. Drying of microbial biomass is much more energy intensive than drying terrestrial crops, and may be economically untenable for fuel production. Process engineering may make it feasible to recycle the methanol (or valorise it, for example in a methanol-to-gasoline-type process) though this has not been explored to our knowledge.

Several types of SCOs have been upgraded to hydrocarbon fuels with and without prior cleanup. Among these, algae oils are by far the most common. Hydroprocessing approaches and cleanup approaches for use on algae oils have varied significantly, though many employed techniques to specifically remove chlorophyll. Hydroprocessing of yeast oils, which do not contain chlorophyll but may contain sterols, has also been reported by a few researchers. Among other microbes, to our knowledge only lipids from the methanotrophic bacterium Methylomicrobium buryatense have been upgraded to hydrocarbons. A summary of SCO hydroprocessing literature is provided in Table III.

Lawal and coworkers explored deoxygenation of algae oils from Nannochloropsis salina and Chlorella vulgaris over platinum, rhodium and nickel-molybdenum-based catalysts (83, 87, 88). The oils were obtained from Valicor, which employed a thermochemical pretreatment of the algal biomass to render the lipids extractable (83, 106). Notably, this pretreatment reduced the amount of phosphorus and iron in the Chlorella oil by nearly 100%, while sulfur was only reduced by 30%. For the N. salina oil, however, the sulfur and phosphorus content remained relatively high at 2033 ppm and 246 ppm, respectively. Additionally, the N. salina oil contained only 47% material that could be identified as acylglycerides or FFAs (83, 87). Analysis showed that the oil had up to 0.8% chlorophyll, 0.5% carotenoids, 5% sterols, 1–5% mannitol and a large fraction of unsaponifiable matter that was not identified (83). While most reaction conditions explored showed significant catalyst deactivation within 7 h time on stream (TOS), Zhou and Lawal were able to find a set of conditions for both a NiMo/Al₂O₃ and a Pt/Al₂O₃ catalyst that did not show significant deactivation within this timeframe, and obtained hydrocarbon yields above 60 wt% (87, 88). The optimal conditions for the platinum and nickel-molybdenum catalyst were significantly different, underscoring the need for reaction engineering studies for each combination of catalyst and oil.

Crocker and co-workers explored deoxygenation of algae oils from Scenedesmus acutus grown on power plant flue gas as a carbon source using nickel-based catalysts (71, 89–91). While the focus of these experiments was generally on model compounds, all of the catalysts employed with algae oils (Ni/Al₂O₃, Ni-Al layered double hydroxide and Ni-Cu/Al₂O₃) deactivated faster with the algae oils (within 4 h TOS) than with vegetable oils or model compounds such as triolein, despite the fact that the oil concentration was only 1.3 wt% in a dodecane solvent. The fast deactivation may have been in part due to the low temperature employed (260°C). In particular, the Ni-Al layered double hydroxide catalyst showed significantly less deactivation at 300°C. The faster deactivation with algae oils is also notable in light of the cleanup procedures employed: the lipids were purified by column chromatography using either K10 montmorillonite clay or activated carbon with silica gel, which was effective at removing chlorophyll and other pigments, magnesium and phosphorus. K10 montmorillonite appeared to be a more effective adsorbent than activated carbon, resulting in a greater fraction of products in the diesel range. The authors hypothesised that several unquantified components in the algae oil, including FFAs, polar lipids, sterols or fatty amides could be contributing to the deactivation (90). However, experiments with mixtures of model TAGs and FFAs suggested FFAs were not the culprit (90). A more detailed analysis of impurities also suggested that the oil cleanup quantitatively removed phosphorus and magnesium, though a small amount of nitrogen remained (71). Santillan-Jimenez et al. (71) also noted that the residual extraction solvent (namely chloroform) was likely the source of chloride detected on the catalyst in post-reaction analysis (71), underscoring the observation that the extraction protocol can impact downstream processes through routes other than influencing the types of lipids extracted.

Lercher and coworkers explored nickel-based catalysts for algae oil deoxygenation under similar conditions to those above (though also in batch mode), using commercial algae oil of an unspecified species (92–94). In these experiments, support acidity (HBEA zeolite or zirconia) was shown in mechanistic studies with model compounds to play a key role in the reaction kinetics, and the nickel particle size was also key to catalyst activity. The oil was not analysed for impurities (only fatty acid composition), but did not deactivate the catalysts for at least 72 h (for Ni/ZrO₂) (93) or 120 h (for Ni/HBEA) (92) TOS. Notably, some isoalkanes, which improve the cold flow properties of the diesel blendstock, were observed in the products due to the support acidity.

Nguyen et al. explored both nickel and molybdenum-based catalysts in deoxygenation of Chlorella algae oil extracted by supercritical hexane (95, 96). In these studies, the catalysts were rapidly deactivated (within 1 h) and produced less than 10% yield to hydrocarbons. In contrast, FFAs were the major product, with acylglycerides featured prominently as well. The authors noted that the crude extracted lipids comprised only around 50% fatty acids and 3–4% sterols, with the remainder unidentified (95). While low conversion of the FFA and acylglycerides was observed, no conversion of the sterols or steryl esters was found (96). Nguyen et al. also noted that while molybdenum nitride catalysts had much lower initial activity than nickel-based catalysts, the molybdenum nitride catalysts did not show as strong a deactivation as the nickel catalysts did (95). The low activity of the molybdenum nitride catalysts represents a significant challenge nonetheless. However, recovery of the lipids as methyl esters and purification over a bleaching clay produced a feedstock that could be readily converted over a Ni-HY zeolite catalyst (84–86). Interestingly, a Pd/C catalyst did not perform as well as the Ni-HY catalyst in deoxygenation of the purified methyl esters, possibly due to the lower acidity of the carbon support.

Researchers at NREL demonstrated both deoxygenation and hydroisomerisation of Scenedesmus algae oil recovered after acid pretreatment and hexane extraction (72, 97). Deoxygenation employed a Pd/C catalyst, while hydroisomerisation employed a Pt/SAPO-11 catalyst. Kruger et al. found that while relatively severe temperature and pressure (450°C and 1300 psig) and dilution of the algae oil in a hexane solvent were required for complete conversion and stable performance of the Pd/C deoxygenation catalyst, bleaching the oils with either silica gel or with phosphoric acid did not improve the hydroisomerisation performance, at least through 10 h TOS (72). The deoxygenation step also resulted in significant denitrogenation as well. The severe conditions required for deoxygenation led to significant cracking and lower yields to diesel range products, but hydroisomerisation produced renewable diesel blendstocks with cloud points below –10°C (72). The latter step is important to demonstrate given the non-zero levels of residual nitrogen in the deoxygenated feedstock that could poison acidic isomerisation catalysts.

Robota et al. also employed a Pd/C catalyst for deoxygenation of a neat algae oil, and then hydroisomerised the resulting alkanes with a Pt/USY zeolite catalyst (98). Despite the higher concentration of oil, less severe conditions were required for nearly complete conversion of the oil to alkanes than was observed by Kruger et al. (72), though a second pass and a polishing step to remove residual oxygenates were used to produce a clean alkane stream for hydroisomerisation. With the large-pore Pt/USY zeolite catalyst, isomerisation and cracking were competitive, resulting in a 40% or more mass loss to naphtha-range hydrocarbons. Though cloud points of the resulting products were not measured, solvent dewaxing yielded a product that remained liquid at –20°C.

Soni et al. (99) evaluated cobalt/natural clay catalysts for deoxygenation of a commercial algae oil, and achieved hydrocarbon yields around 85% after 8 h in a batch reactor at 580 psig hydrogen. This oil contained nearly 70% docosahexaenoic acid in its FFA profile, suggesting that the product alkanes would require some degree of cracking to serve as a diesel blendstock.

Fu et al. (100) demonstrated hydrogen-free decarboxylation of Chlorella oil at 330–370°C for 2 h in a batch reactor. The heating value and carbon content of the organic phase both increased and the organic product contained significantly isomerised compounds, though conversion, yield, oil characterisation and catalyst characterisation were not reported.

Kandel et al. (101) reported the conversion of an algae oil from Solix Biofuels into hydrocarbons using a nanoparticle iron catalyst supported on mesoporous silica. After 6 h at 290°C and 440 psig hydrogen, 67% of the algae oil had been converted to both saturated and unsaturated alcohols and hydrocarbons. The authors noted that the presence of unsaturated compounds suggests hydrogen may have been limiting.

Tang et al. (102) evaluated deoxygenation of a Scenedesmus algae oil in a batch reactor over a NiMo/Al₂O₃ catalyst. As the reaction temperature increased from 290°C to 350°C (and as catalyst loading increased from 0 wt% to 50 wt% of the oil), the product spectrum shifted from acids, esters and alcohols to paraffins, olefins, naphthenes and aromatics, with a small amount of isoparaffins also present. The authors used multiple solvent combinations to extract the lipids. Dichloromethane:methanol was the highest yielding solvent on a weight basis, but the primary components were sterols and tocopherols rather than acylglycerides and FFAs. Other solvents were lower yielding, but the extracts were also not analysed.

Poddar et al. deoxygenated Nannochloropsis lipids over a phosphorus-promoted CoMo/Al₂O₃ catalyst, achieving nearly complete conversion to alkanes and nearly 80% yield to liquid-phase products in 6 h at 375°C and 50 bar hydrogen (103). The authors conducted a detailed kinetic analysis but did not report detailed product analysis or post-reaction catalyst characterisation.

Wang et al. explored deoxygenation of a FFA stream derived from Dunaliella lipids (104). The lipids were hydrolysed in 250°C to release FFAs, which self-separated from the aqueous phase and were decarboxylated in fed-batch mode over a Pd/C catalyst. The catalyst maintained steady operation over 5 h of reaction, even with a hydrogen-poor headspace in the reactor.

Finally, Schulz et al. (105) deoxygenated Synechocystis (cyanobacteria) lipids over a Pd/C catalyst and isomerised the resulting alkanes over a Pt/CaY zeolite catalyst. The lipids were unique in that they were recovered in the form of excreted FFAs (mainly C12), adsorbed on and eluted from a resin. Once recovered, the lipids were also subjected to a number of cleanup protocols, including recrystallisation, activated carbon treatment, saponification and preparatory-scale liquid chromatography. These protocols had varying effectiveness in reducing sulfur and phosphorus contaminants in the oils (especially sulfur), which were reflected in varying performance of the Pd/C deoxygenation catalyst across FFA streams, giving 33–88% alkane yield depending on the stream. The two most promising streams gave relatively low conversion, but high isomerisation selectivity over the Pt/CaY catalyst.

Aside from algae lipids, yeast lipids have also been evaluated for deoxygenation and hydroisomerisation. Sànchez i Nogué et al. (10) employed the same conditions as those used by Kruger et al. (72) and Knoshaug et al. (97) to deoxgygenate lipids from Rhodosporidium toruloides yeast. The yeast lipids produced a renewable diesel blendstock of similar quality as that from the algae oils. Process optimisation to determine if the yeast lipids could be satisfactorily converted under less severe conditions due to the lack of impurities such as chlorophyll was not conducted. The lipids were found to contain a small amount of steryl esters and a significant amount of polar lipids as well.

Chuck and coworkers have also evaluated conversion of yeast lipids to fuels, though not through conventional hydroprocessing approaches (107, 108). In these studies, the yeast lipids were either catalytically cracked (107) or metathesised with ethylene using a Hoveyda-Grubbs catalyst (108). Although the Pd/C cracking catalyst and reaction temperatures were similar to those used for deoxygenation by other researchers, the gas atmosphere was quite different. Interestingly, however, the catalytic cracking experiments were performed with a yeast lipid from Metschkownia pulcherrima that included 9 wt% sterols, and the authors inferred that the sterols can act as hydrogen donors as they undergo aromatisation at 350–400°C in an argon atmosphere, which improved the yield to linear alkanes from the triglyceride fraction.

Among other oleaginous microbes, the only report of hydroprocessing to our knowledge is that of Dong et al. (34), who converted polar lipids from Methylomicrobium buryatense to linear alkanes over a Pd/SiO₂ catalyst. The bacterial biomass was hydrolysed by a sequential alkaline-acid method that both ruptured the cell wall (rendering the lipids extractable) and cleaved the polar head groups off the predominantly polar lipids, producing a clean FFA stream for deoxygenation. The high degree of polar head group cleavage was confirmed by the presence of only 4 ppm phosphorus in the oil, though the oil contained sulfur, nitrogen, sodium and halogens in higher amounts. With a relatively high catalyst loading and deoxygenation in batch mode, full conversion to linear alkanes was achieved. The authors also noted that a preliminary catalyst screening with nickel- and copper-based catalysts confirmed the hypothesis that these catalysts would be rapidly deactivated in the presence of phosphorus-containing feeds, while noble metals performed better.

Several interesting themes emerge in lipid composition, cleanup and hydroprocessing of microbially-derived oils. First, the studies reporting the longest and most stable catalytic performance have used refined algae oils, for example from Phycal or Solix. Although the details of the refining process are not readily available, these processes appear to sufficiently clean the oils of catalyst-deactivating impurities. In contrast, lipid streams extracted by researchers directly prior to catalytic processing have tended to deactivate catalysts more rapidly or perform more poorly, even when diluted by a factor of 100 into an inert solvent. Perhaps this is unsurprising given the differing descriptions of the oils’ appearances, for example bright orange transparent liquid vs. dark brown semisolid. The situation is additionally confounded by the variety of techniques and solvents used to extract the lipids, which may co-extract different undesirable components. Hexane is the simplest solvent employed, but can result in low extraction yields and form emulsions with wet biomass. Alcohols, such as methanol and ethanol can be added to increase the extraction yields and minimise emulsions, but can also increase co-extraction of impurities. Halogenated solvents can serve the same purpose as alcohols, but further confound the process by introducing additional heteroatoms (typically chloride from chloroform or dichloromethane) that can remain in the extract in trace amounts even after solvent removal.

Second, studies using noble metal catalysts have tended to report better performance than those using base metal catalysts. This is frequently true in other catalysis applications as well, wherein base metals such as nickel, molybdenum and copper are more prone to coking, oxidation by feedstock oxygen and poisoning by feedstock heteroatom impurities. Indeed, in the examples where microbial oils have been thoroughly characterised for heteroatom impurities, it is tempting to infer that supported noble metals may even be able to tolerate some degree of phosphorus, nitrogen and sulfur in the feedstock. However, times on stream in the literature reviewed here, less than 200 h TOS, have typically been too short to make a meaningful assessment. Nevertheless, while noble metals are frequently avoided due to the high cost, there may be environmental advantages (at least for metals such as ruthenium and platinum) (109), and some of the cost can be recouped by regeneration and recycling of the spent catalysts (110). It is worth noting that industrial processes producing green diesel fuels from vegetable oils, such as Honeywell-UOP’s Ecofining, Neste’s NExBTL, Axens’ Vegan and Renewable Energy Group’s Bio-Synfining processes employ nickel-molybdenum or cobalt-molybdenum sulfide catalysts similar to those used in petroleum refining (111). These catalysts tend to favour hydrodeoxygenation over decarbonylation or decarboxylation, which preserves carbon in the product, but also consumes more hydrogen in the process. At least one evaluation of the tradeoffs between carbon yield and hydrogen consumption concluded that decarboxylation would be economically favourable, provided that hydrogen consumption through subsequent methanation and reverse water-gas shift reactions is not significant (112). On the other hand, the deployment of large-scale wind and solar power has at times resulted in excess power production and low-cost electricity that can be used for water electrolysis to produce inexpensive hydrogen. Nickel-molybdenum catalysts are also frequently employed for hydrodesulfurisation and hydrodenitrogenation, which may indicate a higher level of robustness to heteroatoms than other types of catalysts. In particular, nickel-molybdenum catalysts typically require a co-fed source of sulfur, often hydrogen sulfide or dimethyl disulfide, to mitigate conversion of the active sulfide phase to a less-active oxide. Concomitantly, these catalysts tend to leach sulfur into the product (21), which may inhibit the ability of these catalysts to meet increasingly stringent sulfur limits in fuels. It is unknown whether the sulfur-containing species in microbial oils can satisfy this requirement of nickel-molybdenum catalysts.

Third, lipid stream cleanup by a number of approaches appears promising, but all of the techniques have some drawbacks and only a few can be reasonably expected to apply universally across microbial lipids. Saponification of the lipid stream to FFA followed by distillation is perhaps the most robust as it should be effective even with lipid feeds containing high levels of polar lipids, sterols, chlorophyll and metals. However, the distillation step is energy intensive compared to most of the other bleaching methods and the cost of the alkali is high. Methyl esterification and distillation should be similarly effective, but the methanol would need to be recovered for high material and economic efficiency, or else recovered as methane after hydrotreating. Degumming (precipitation of phospholipids) and phosphoric acid bleaching can work well for lipid streams that are low in polar lipids, but would suffer from significant yield losses otherwise, and the phosphoric acid must be thoroughly washed out of the cleaned lipid stream to avoid contamination from the bleaching agent. In lipid streams with high polar lipid content, hydrolysis to FFAs may prove more economical. Adsorption on a polar material such as silica, bleaching clay or activated carbon appear effective for some target impurities (for example chlorophyll and some metals), but similar to degumming, may result in unacceptable yield losses for lipid streams that contain large amounts of polar lipids. Additionally, post-adsorption oils have still tended to deactivate the catalysts. Interestingly, even when chlorophyll is removed to below detection limit, some nitrogen frequently remains in the oils, likely in the form of hydrophobic protein. There is thus a clear need for both fundamental science and process development on this topic.

Looking forward, converting microbial lipids to fuels is a promising approach to displace conventional fossil fuels, especially for diesel and jet fuels which share the same carbon number range as the microbial lipids. While previous research has focused primarily on algae lipids, the advent of third generation cellulosic sugars may spur further development of heterotrophic oleaginous microbes as well. Finally, to best allow for economically-viable biofuels production from SCOs, it is critical to design an integrated process, in which certain complementary units can be intensified to maximise the yield of high quality products for a given step. For example, the lipid hydrolysis step can be integrated with the biomass pretreatment or cell wall rupture step to simultaneously rupture the cell walls for bulk oil extraction and liberate FFA from polar lipids. In addition, the selectivity of the extraction solvent should be carefully considered to increase extraction of fatty acids while reducing the co-extraction of impurities. With these considerations, SCOs are poised to make a significant impact in sustainable fuel production.

Hydrogen is the first element of the Periodic Table and the most abundant element in the Universe. Currently, we are witnessing the emergence of hydrogen gas as an increasingly powerful energy vector for storing and delivering electricity (1). Firstly, besides having very low physical density, hydrogen has the highest gravimetric energy density of any known non-nuclear fuel (for example three times higher than gasoline and 150 times higher than a state-of-the-art lithium-ion battery), which sets the stage for hydrogen as an efficient energy-storage solution (2). Secondly, hydrogen is an environmentally benign fuel, since only energy and water are the end products of the reaction between hydrogen and oxygen, giving access to zero-emission electricity production when used in fuel cells. These benefits have made hydrogen a priority area within the “climate and resource frontrunners” of the European Green Deal.

Currently, most hydrogen (ca. 95%) is produced through energy-demanding steam reforming reaction of water with natural fossil fuels. Unfortunately, this translates into undesired generation of greenhouse gases. As an alternative, WE (Equation (i)) can be employed for hydrogen production, providing a completely carbon-neutral solution when working on renewable electricity from wind, solar, wave, tide, biomass or geothermal, thus eliminating carbon dioxide emissions.

(i)

Importantly, the resultant high-purity hydrogen can be stored in a compressed or liquefied form or in chemical compounds (metal hydrides and liquid organic hydrogen carriers), and then delivered as needed for fuelling small- and large-scale applications (3).

Although the era of WE began more than two centuries ago, its worldwide implementation is still limited to hydrogen production up to the megawatt range using liquid alkaline electrolysis (4). WE is a kinetically controlled process characterised by slow charge transfer and insufficient chemical reaction rates. A large overpotential, defined as the difference between the required and thermodynamic value of the WE voltage (E⁰ = 1.23 V vs. reference hydrogen electrode (RHE)), needs to be applied to drive this reaction. Therefore, catalysts, primarily based on pgms, are used to accelerate WE by reducing the value of the applied overpotential to conduct the cathodic HER and the anodic OER, which are the two half reactions of the WE (5).

The best-performing catalysts for WE are platinum for HER and iridium oxide/ruthenium oxide for OER, featuring topmost activity and long-term stability in both acidic and alkaline electrolytes. Unfortunately, these pgms are scarce and expensive, having the status of critical raw materials in the European Union (EU). Iridium and ruthenium are scarce even when compared to platinum because both are byproducts of platinum mining. Hence, the use of pgms has been recognised as one of the major bottlenecks for hydrogen production by WE on terawatt scale, and, therefore, the development of pgm-free catalysts is an economically sound strategy towards technological and industrial expansion of WE.

On the electrolyser side, the commercially available technologies for hydrogen generation are based on: (a) alkaline electrolysers (AELs); (b) solid oxide electrolysers (SOELs); and (c) (semi)solid polymer electrolysers (SPELs) (4, 6–9). The AEL technology has been commercially available for >50 years. It is characterised by long lifetime and low cost. Recent developments of AEL deal with the employment of high temperatures and pressures, targeting improvement of efficiency (7). The AEL electrodes are typically composed of abundant nickel with a catalytic coating. Shortcomings of the AEL are the use of corrosive liquid electrolytes (for example potassium hydroxide), low purity of the resultant hydrogen due to high cross-permeation through cathode and anode separator, and long start-up times not suitable for dynamic operation. The currently developing SOEL (also known as steam electrolysis) omits the need for corrosive liquid electrolyte, but operates at temperatures above 700°C, making the whole process energy demanding. Another issue is the instability and degradation of the electrode materials at the high operating temperatures. WE technology based on SPEL is well-suited to intermittent supply applications at scale (dynamic operation), offering high current densities compared to AEL, high-purity hydrogen at pressures up to 30 bar and efficiency of ca. 70%. Nevertheless, a number of obstacles, such as high capital costs, prevent the widespread use of SPEL (9).

Advantageously, SPELs operate at low temperatures (50–80°C), and the thin humidified polymer membranes feature high conductivity and low resistance, resulting in minimal current losses and low cross-permeation of hydrogen. There are three SPEL technologies: (a) proton-exchange membrane (PEM); (b) anion-exchange membrane (AEM); and (c) bipolar membrane (BPM).

PEM electrolysers are the present state-of-the-art SPEL technology. Relatively compact PEMWE systems are available from several small, medium and large companies, such as ITM Power, NEL, Giner Labs, Enapter and Siemens. Despite the fact that PEMWE has high technology readiness level (TRL), it is still mainly used for demonstration projects, largely subsidised by local or national governmental funding organisations. The main reasons for the limited commercial adoption are high capital expenditures (CAPEX) and operational expenditures (OPEX), making the produced hydrogen economically non-competitive compared to the traditional steam-reforming process.

More specifically, due to the extremely acidic protonic matrix of the ionomer and membrane (for example Nafion^TM, Aquivion^®, Celtec^®), the technology relies on scarce platinum/carbon and iridium oxide/ruthenium oxide catalysts, which exhibit high intrinsic activity and durability (10). Notably, only these materials can withstand the acidic PEMWE operation conditions over a 20 year electrolyser lifetime. Additionally, highly acidic and oxidative operation conditions dictate that the porous transport layers (PTLs) have to be made from corrosion-resistant but rather expensive titanium, which, combined with pgm catalysts, represents ca. two thirds of the cost of the PEMWE stack (several individual PEM cells stacked together to achieve higher hydrogen production).

The alternative AEMWE approach is rapidly growing from TRL2–3 to the level of TRL6–7 (11–13). AEM combines the advantages of AEL and SPEL, eliminating the need of corrosive liquid electrolyte and leveraging efficient membrane-based WE (14). Until recently, the main bottleneck in AEMWE was the lack of highly hydroxide conductive, stable and durable AEMs as compared to PEMs (15, 16). Fortunately, several promising AEMs have recently emerged on the market (17), such as those from Tokuyama, Fumatech, Ionomr Innovations, W7energy^®, Ecolectro, Dioxide Materials^TM and AGC. The non-aggressive AEM environment at pH ≈ 13 and the possibility to produce AEMs without the use of fluorine (regulated in several countries) makes AEM production more environmentally friendly and safer for workers and nearby inhabitants (11).

Importantly, AEMWE may occur efficiently on pgm-free catalysts based on inexpensive metals (such as iron, cobalt or nickel) that can also withstand prolonged operation in alkaline membrane environment. Therefore, the development of an electrolyser based on emerging AEMs and pgm-free catalysts should enable a significant reduction in both CAPEX and OPEX of electrolysis (18). Interesting work on AEMWE within the EU Horizon 2020 ANIONE project additionally illustrates these points very well. Notably, a further CAPEX reduction can be achieved by replacing the expensive titanium PTLs by simple stainless steel, which is stable under AEMWE operation. Nevertheless, the detailed technoeconomic analysis of AEMWE with pgm-free catalysts is complicated, and the projected cost of the as-produced hydrogen strongly depends on the selected model and input parameters (19, 20).

A third, significantly less explored, avenue for SPEL is the BPM (21). Here, a PEM and an AEM are connected in series, thus allowing the use of non-iridium or ruthenium catalysts on the anode side. Notably, BPM provides a thermodynamic advantage due to the pH gradient across the membrane, which makes it possible to conduct the electrolysis under applied potential below the standard potential of WE, thus requiring less current input. BPM technology is still in its infancy, and more research is required to improve its design, performance, stability and cost.

Our research is focused on the development of pgm-free catalysts for alkaline HER and OER that could be further translated into real-life application of AEMWE. Various classes of pgm-free materials, such as alloys, (oxy)hydroxides, borides, carbides, nitrides, phosphides, chalcogenides, oxides, spinels, perovskites, metal-organic frameworks (MOFs) and metal-free carbon-based compounds have been investigated as catalysts for alkaline WE. In this review, we will highlight several titled prospective pgm-free catalysts that have been studied in our laboratories.

In the simple scheme of a AEMWE cell (Figure 1), the OER and HER catalyst layers are directly coated on the respective sides of the AEM membrane, thus forming a membrane electrode assembly (MEA). Alternatively, the catalysts can be incorporated into the respective gas-diffusion layer (GDL), thus forming a gas-diffusion electrode (GDE). Using porous GDLs is preferential compared to planar ones, since the porosity accelerates the mass transfer of the reactant or product over the catalyst layer. The main role of the AEM is to ensure hydroxide transfer from the cathodic side of the cell to the anodic side. Normally, AEM electrolysers use a water feed with the addition of HCO₃^–/CO₃^2– or dilute potassium hydroxide electrolytes to achieve better performance.

Fig. 1.

AEMWE allows for the direct generation of hydrogen and oxygen through the following electrode reactions (Equations (ii)–(iv)):

(ii)

(iii)

(iv)

As shown in Figure 2(a), HER starts with electrochemical hydrogen adsorption according to Equation (v) (Volmer reaction), where H* designates a hydrogen atom chemically adsorbed on an active site of the electrode surface (M).

(v)

Fig. 2.

The initial adsorption can be followed by either electrochemical desorption according to Equation (vi), (Heyrovsky reaction), or chemical desorption in Equation (vii) (Tafel reaction). The exact mechanism and rate-determining step of the HER over certain catalysts can be established by analysis of the respective Tafel slope data (22).

(vi)

(vii)

The proposed alkaline OER (Figure 2(b)) commences with the electrochemical adsorption of hydroxide anion on the electrode active site (Equation (viii)), followed by the electrochemical formation of O- and OOH-bound intermediates (Equations (ix) and (x)). Finally, O₂ molecule is generated as a result of the last single-electron charge-transfer step according to Equation (xi).

(viii)

(ix)

(x)

(xi)

Notably, a scaling relation (23–26) between binding energies of the reaction intermediates results in a minimum theoretical excess potential, the so-called ‘overpotential wall’ of 0.37 V (27). This means that one would need to apply a minimum potential of 1.23 V + 0.37 V (i.e., 1.6 V vs. RHE) to conduct the OER reaction, which has found experimental confirmation (28). Breaking the OER scaling relation and thus reducing or eliminating the ‘overpotential wall’ still remains a challenge (29).

Table I summarises parameters typically used to present laboratory half-cell data for the evaluation of the catalyst performance and comparison between different catalysts with similar mass loading per geometric area. A number of these parameters will be used in the subsequent discussion. Readers interested in best practice for investigating novel HER and OER catalysts are referred to (30), while (31) provides useful guidance on analysing and presenting the catalytic data. Moreover, some critical aspects of electrochemical data evaluation are reported elsewhere (22, 32–34).

OER, as a kinetically slow 4e^– transfer reaction, governs the overall efficiency of WE. Since the largest overpotential stems from OER, catalyst development for this half reaction will offer the largest efficiency gains (26).

Interestingly, initial reports employing transition metal borides, carbides, nitrides, phosphides and chalcogenides (in general, ‘Xides’) as catalysts for alkaline OER putatively attributed the observed catalytic activity to the pristine materials. However, a careful consideration of following reasons indicated that all the aforementioned Xides should undergo oxidation during the OER into the respective oxides or (oxy)hydroxides (35):

Detailed characterisation studies, including diffraction, spectroscopy and microscopy analyses of the catalysts before and after alkaline OER, provided strong support for in situ oxidation. Accordingly, all transition metal Xides could be considered as precatalysts or ‘active’ supports for alkaline OER. The degree of oxidation could be different and largely controlled by the chemical nature of the Xide precatalysts. Moreover, the size and microstructure of the catalysts play an important role. If the catalyst nanoparticles (NPs) are ultrasmall or the active catalyst surface layer is very thin, the compounds will rapidly undergo full oxidation. In contrast, if the catalyst NPs or films are reasonably large or thick, then partial oxidation of the surface occurs, forming distinct core@shell NPs and nanoheterostructured films, preserving the bulk Xide structure underneath. In other words, in the case of poor catalysts, the oxidation of the bulk compound occurs forming non-conductive oxides or hydroxides, while for the good catalysts the oxide or hydroxide layer passivates the catalyst surface, preserving the bulk structure of the original compound.

The surface oxidation yielding the real catalyst is an unavoidable but crucial phenomenon to obtain highly active and stable OER catalysts. In the vast majority of studies, the resultant in situ formed catalysts show remarkably higher OER performance than their respective metal oxide or hydroxide counterparts. Full understanding of this performance enhancement is still lacking. Generally, there are several factors that should be pointed out: (a) the development of high surface area due to the formation of amorphous-like (oxy)hydroxide surface; (b) the high electrical conductivity of the non-oxide Xide precatalyst underneath the real catalyst; and (c) synergetic catalyst−precatalyst electronic interactions within the as-formed nanoheterointerface (35–37). These factors give rise to a larger number of catalytically active sites and faster charge transfer kinetics over the anode/electrolyte interface, thus beneficially boosting the OER performance.

Currently, there is a limited knowledge of the mechanism of surface transformation during OER because mainly ex situ studies are performed before and after OER. In situ surface-sensitive studies (such as spectroscopy, diffraction, imaging, electrochemical imaging, cyclic voltammetry, nanoimpacts and combination thereof), despite being extremely challenging, are crucial for the rational development of the OER precatalysts (38–42). Such knowledge will be essential to achieve control over the formation of the real catalyst, thus guiding future efforts towards active, stable and economic catalysts for alkaline OER (Figure 3).

Fig. 3.

To summarise, the structure of the bulk phase or core of NPs of a prominent OER catalyst is preserved during the reaction. The surface is, undoubtedly, oxidised due to the extreme oxidation potential applied and the presence of multiple active species, such as hydroxide radicals. The formed oxide layer passivates the surface, thus preventing further oxidation of the bulk of the precatalyst. The same oxide layer is an active catalyst. Thus, the formation of the oxide layer should not be avoided but rather judiciously directed to enhance stability and activity of the catalysts. A combination of computational and detailed in situ studies is a promising strategy in this research direction.

Thorough and comprehensive studies conducted by research groups around the globe have narrowed down the range of industrially-relevant pgm-free HER/OER catalysts for AEMWE to: (a) nickel, nickel alloys and intermetallic compounds as the most active and stable for the HER side of the MEA (43–51); and (b) base metal oxides, (oxy)hydroxides, spinels and perovskites for the OER side of the MEA (52–58).

In general, the catalysts for alkaline HER are similar to those used in AEM fuel cells (AEMFC) (59). Carbon-supported nickel, nickel-molybdenum, nickel-copper and other nickel-based catalysts have been synthesised and investigated (43–51). As a lighter analogue of pgms (especially being chemically similar to palladium), nickel plays the main role in HER, while the addition of other elements increases the cathode durability. Nickel-molybdenum catalysts have been intensively studied for both hydrogen oxidation reaction and HER under realistic AEMFC and AEMWE conditions, respectively (45, 48, 50). The nickel-molybdenum alloys enriched with nickel up to ≈87 at% were found to be not only active in alkaline HER, but also less prone to poisoning by functional groups of ionomers, which is a well-known shortcoming of pgm catalysts in AEMFC and AEMWE (45, 48). The properties of carbon supports, such as the surface area, level of graphitisation, bulk and surface chemical composition and hydrophobic or hydrophilic properties, significantly affect the AEMWE performance and should be carefully tuned as a function of the type, composition and weight loading of the HER catalysts (60).

Considering the OER catalysts, nickel (oxy)hydroxides or oxides with iron impurities in alkaline electrolyte, as well as nickel-iron (oxy)hydroxides, have become benchmark catalysts for AEMWE (61, 62). Although the exact nature of the active catalytic sites in such systems is not well understood (63), the nickel-iron-based materials for alkaline OER deserve a more careful study in the future, especially with respect to their low durability issues (64, 65).

Furthermore, a general direction in the development of catalysts for alkaline OER is to improve the electronic conductivity of already established active oxide catalysts, since the conductivity of the oxides is typically very low. Such an improvement is critical to reducing the applied overpotential during real AEMWE operation, where loading of OER catalysts can often be as high as 4–12 mg cm^–2 (53–55). Unfortunately, the oxidative operation conditions during OER at high potential of ≈1.9–2 V forbid the utilisation of traditional highly conductive carbon supports, which would simply oxidise and degrade during prolonged AEMWE. At the same time, using carbon as a supporting material for transition metal oxide catalysts is a common laboratory practice in short half-cell experiments for the initial assessment of intrinsic catalytic activity of the materials and performance losses due to insufficient electronic conductivity (52). For instance, highly graphitic carbon nanotubes or high surface area graphenes are widely used during the screening and selection of promising alkaline OER materials (12, 52). An alternative strategy is based on employing non-stoichiometric mixed oxides, perovskites, delafossites or spinels (58). Although these materials possess high conductivity (> 0.1 Ω^–1 m^–1) (58), specific synthetic methods should be developed to produce such conductive oxides on a large scale.

Notably, the commercial metal oxide products made on the multi-tonne scale are mainly silica, zeolites and titania. The protocols used to manufacture these materials are well-matured and based on the sol-gel, precipitation, hydrothermal, flame or spray pyrolysis. Nevertheless, these approaches have a limited ability to control physicochemical properties required for OER catalysts, including phase purity (required for high electronic conductivity), the surface area (required for higher density of active sites) and the primary particle size (required for the uniform distribution in the electrode structure).

For the past eight years Pajarito Powder redesigned and modified their proprietary manufacturing platform VariPore^TM for upscaling mixed oxide catalysts production. The schematic of the method is shown on Figure 4. The main idea of this method is to design material by a bottom-up approach with the ability to control all required properties of the final catalyst at every stage of the process. For example, selecting particle and pore forming agents (P&PAs) with different morphology will afford a final catalyst with an inverted structure of the respective P&PAs. The chemical nature of the precursors will allow selection of manufacturing conditions to obtain materials under low energy demanding regime, substantially decreasing the cost of final materials.

Fig. 4.

The method was successfully used for the preparation of phase-pure CuCo₂O₄ spinel catalyst for OER (Figure 5(a)). The P&PAs used in this design derived from high surface area spherical silica particles, which prevented NPs from agglomeration. The phase purity of the spinel was achieved by thermal treatment at just 550°C, allowing to preserve the unique spherical shape of agglomerates (Figure 5(b)). The latter is important for manufacturing dense electrodes for OER, allowing maximal catalyst utilisation on the GDEs. At the moment, this method is established as a robust, flexible and modular manufacturing platform for making different classes of HER/OER catalysts and practiced for commercial production at the kilogram scale per batch (54).

Fig. 5.

Among the large variety of pgm-free catalysts that are currently extensively explored, the interest of our groups have been focused on transition metal phosphides and borides, which belong to the larger family of non-oxide catalysts referred to as Xides. In the next two sections, we focus specifically on these two groups of Xides, highlighting some of our results along with important findings from other research groups. For a more general overview of pgm-free non-oxide electrocatalysts, the readers are referred to several recent reviews relevant to this topic (36, 66, 67).

6.1 Crystal and Electronic Structure

6.2 Hydrogen Evolution Reaction Catalysis

6.3 Alkaline Oxygen Evolution Reaction Catalysis

7.1 Crystal and Electronic Structure

7.2 Hydrogen Evolution Reaction Catalysis

7.3 Alkaline Oxygen Evolution Reaction Catalysis

Electrocatalysis is a rapidly evolving research area, and there are plenty of pgm-free catalysts reported in the literature, including nickel alloys, oxides, TMPs and TMBs, as detailed here. Recent findings have shown that, while HER most likely takes place on the surface of the pristine Xide catalyst, the in situ formation of oxide or oxohydroxide NPs as the true catalyst is clearly observed during OER as a result of surface oxidation of the Xides. Even though pgm-free nickel alloys, TMPs and TMBs do not outperform platinum, their promising HER activity can be of practical importance due to significantly lower costs. In contrast, many iron-, cobalt- and nickel-based catalysts outperform the standard rutile-type iridium/ruthenium oxides in OER activity, but little is known about the durability of these pgm-free materials. This is mainly due to the fact that the vast majority of the literature reports the results of laboratory half-cell measurements and the durability is only tested for a few weeks at most.

With respect to the synthesis, it is difficult to formulate the exact key requirements (for example chemical composition; promoters; crystallographic, electronic and surface structures; physical properties; morphology; fine microstructure; specific surface area; particle size distribution and supporting material) that should be fulfilled for the HER/OER catalysts to be economic, active, stable and durable for AEMWE. This challenge is mainly explained by the ever increasing complexity of the reported catalysts, their dynamic reconstruction during reactions and large scattering of the resultant half-cell testing data. The way to solve this shortcoming is to foster the international community working on WE to follow well-defined characterisation and testing protocols. Recent efforts in this direction (30, 31) provide hope that the aforementioned requirements will be established in the near-term, also involving the help of data mining and computation. It would be also interesting to see how Xides will compete and, more importantly, participate in the rapidly developing area of mass-efficient atomically dispersed catalysts for HER/OER (104, 105).

Considering the discovery of new, more efficient catalysts, solid-state chemistry offers hundreds of ternary and multinary phosphides, borides and other Xides, in addition to the relatively limited range of binary materials that have been studied thus far as potential HER/OER electrocatalysts. Such materials have unique crystal and electronic structures and may exhibit quite different catalytic properties as compared to their binary counterparts. Tuning the transition-metal d-orbital filling, local metal and non-metal (boron, carbon, nitrogen, silicon, phosphorus, sulfur, selenium, tellurium) coordination environments, and the density of states at the Fermi level is important to optimising the HER/OER catalytic performance. The large structural and compositional variety is a challenge as the synthesis and property characterisation protocols are tedious. Application of computational and machine learning approaches can substantially accelerate identification of the most promising candidates. Coupling the computational screening, which accounts for surface dynamics, together with experimental research should result in emergent catalysts with improved performance.

Assembling even a single AEMWE cell is a tedious process that requires expertise not commonly found in academic research laboratories. For instance, to the best of our knowledge, no full AEM electrolyser has been tested with promising TMP/TMB cathodes and anodes. Nevertheless, real AEMWE testing is quite important to demonstrate the applied potential of Xides as catalysts. Here, the early involvement of industry will be highly beneficial, allowing the real future prospects of the reported pgm-free catalytic systems to be understood.

Another issue is the lack of benchmarked AEMWE components (membrane, ionomer, GDL and PTL) and standard HER/OER catalysts accepted by the electrolyser community. This state of matters makes comparison of performances not only difficult, but in many cases meaningless. Substantial input from academia, national laboratories and industry into standardisation of materials and test protocols is needed. This work has been initiated in the international US–EU collaboration under the HydroGEN consortium and has already resulted in the harmonisation of testing protocols for PEMWE, which will be followed by expansion to the promising AEMWE field. Establishing such protocols will allow the electrolyser community to work at the device level to engineer AEMWE electrolysers with high performance and durability (16, 106).

The direct transformation of CO₂ to methanol requires the addition of three equivalents of hydrogen to generate methanol and water from CO₂. This can be performed at moderate temperatures and pressures (220–250°C and 10–30 bar) over heterogeneous metal oxide catalysts, particularly copper and zinc oxides with alumina (CZA, Cu/ZnO/Al₂O₃), which are based on catalysts that date back to the 1930s (8, 9). This process is now being carried out commercially by Carbon Recycling International, producing 4000 tonnes a year of sustainable methanol from Icelandic geological CO₂ and geothermal energy (10).

For this direct conversion, heterogeneous copper-based catalysts are the most extensively studied with catalyst performance thought to be generally dependent on the structure of the copper surface and also possibly the interface between the copper and the other transition metal components, most commonly zinc, zirconium or their oxides (11). Further weight is also given to the degree of dispersion of the copper within the catalyst structure, with increasing copper dispersion and thus copper surface area correlating directly with increased methanol yield (12).

Computational studies have suggested that CO₂ reactions on the stepped and close-packed Cu(211) representative surface gives the primary low-temperature reaction pathway via both formic acid and formaldehyde, the route via CO being an alternative and potentially competitive route (13). This CO, generated by the reverse water gas shift reaction (RWGS) can be problematic, particularly at higher temperatures, where excess carbon monoxide is generated from CO₂ and hydrogen (see Figure 2).

Fig. 2

Both the CO and the byproduct water can limit the selectivity to and yield of the desired methanol product, especially where carbon monoxide side production increases and the RWGS reaction dominates (14). While the water that is inevitably generated while the reaction proceeds can sinter and degrade the catalyst alone, high CO and H₂ concentrations from the RWGS reaction have also been shown to overly reduce these copper surfaces, making sintering happen more easily (15, 16). Additionally, the excess CO production typically promotes hydrocarbon and also higher-alcohol generation via well-understood Fischer-Tropsch chemistry, further increasing hydrogen consumption and complicating product purification (17, 18). However, as discussed in the later section of this article, the production of higher alcohols and other products directly from CO₂ by taking advantage of the greater propensity for CO to form new C–C bonds on these catalyst surfaces might be a way to generate CO₂ fuels more effectively than individual production of methanol and then carrying out subsequent dimethyl ether (DME), methanol to gasoline (MTG) or methanol to olefins (MTO) processes, thereby avoiding multiple individual processes (19).

One method used to avoid CO production and to generally promote milder reaction conditions is the use of precious metal catalysts such as palladium or platinum instead of, or in addition to, the traditional copper as these help to reduce the required temperature for the reaction and therefore increases reaction selectivity as the RWGS reaction is less favourable (20, 21). In addition, these catalysts can promote the hydrogenation of any CO to methanol, further reducing the CO:CO₂ ratio within the reaction mixture (22). However, these catalysts typically show poor CO₂ conversion, thought to be due to the lower strength of the bonding between the CO₂ and metal surface (23). Yet another approach is to use copper encapsulated in metal organic framework (MOF) catalysts which further promote the exclusion of water from the catalyst surface, increasing the catalyst turnover number. The frameworks can also increase the effective surface area of the active copper species by limiting the growth of the metal surface, a result that can also be achieved by using a hydrotalcite-like compound as the catalyst precursor (24).

For the indirect production of methanol, the RWGS reaction can alternatively be harnessed to generate CO as an intermediate that can then be used to create a sustainable synthesis gas (syngas). This syngas can then be used for methanol production separately. Using CO in this way has the benefit of very high selectivity (over 99.5%) and high yield, facilitated by the absence of water byproduction (which in this case would be removed during CO formation) and the fact that CO is a more reactive starting material than CO₂ (25). In some cases, when producing methanol from CO, the product methanol solution could even be used for further product generation without additional purification or drying (26). It is due to these advantages that general industrial (non-sustainable) methanol production typically uses this syngas route, and it supplies the overwhelming majority of the global methanol demand, which was 83 million tonnes in 2019 (27). The routes for creating the CO starting material from CO₂ are numerous, including classic hydrogenation (RWGS), disproportionation reactions with biochar (where elemental carbon reacts with carbon dioxide to form two equivalents of CO), electrochemical and even plasmolytic routes (28–31).

Compared with methanol, producing the four-carbon chain butanol from CO₂ is a far more challenging synthesis. However, butanol is a valuable potential ‘drop-in’ replacement for petrol as a liquid transport fuel and is compatible with existing fuel infrastructures. It can even be used alone as fuel for unmodified vehicles, with an octane number of 96 (32). It is hydrophobic enough to prevent water and salt corrosion in modern engines, which is a major drawback when using high ethanol content in road vehicle fuel in much of the world (33). Butanol in standard petrol engines has also been shown to have similar or even superior fuel economy than the petrol it replaces, despite having approximately 11% less energy density. This is due to its nature as a single-component fuel, rather than the wildly diverse mixture of compounds that make up fossil fuels, with the entirety of the fuel burning at the optimum rate. Unlike both methanol and ethanol, butanol can also be blended with aviation fuels in limited amounts and its corresponding diester, butyl butyrate, shows good compatibility with aviation kerosene (34). This may be crucial for decarbonisation of the aviation industry as major industrial nations move towards net zero CO₂ emissions in the coming decades.

While there are biological routes for the creation of sustainable butanol from carbon dioxide, particularly acetone–butanol–ethanol (ABE) fermentation, separation of low-concentration butanol from water is challenging due to the low volatility of the butanol and the fact that high concentrations of butanol are toxic to microorganisms, limiting the extent of the fermentation for butanol synthesis (35). One method to avoid these issues is to instead dimerise easily manufactured bioethanol via borrowed hydrogen or Guerbet chemistry (36) (see Figure 3).

Fig. 3

As with the biological routes above, most potential methods to synthesise butanol from CO₂ will similarly involve a multistep process. For example, the Guerbet route could also be appropriate for a CO₂ to fuels approach, by first carrying out the conversion of CO₂ to ethanol and then using the ethanol as a feedstock for the generation of butanol or other higher alcohols. This can be carried out using conventional transition metal catalysis, with reduction being combined with methylation of the absorbed and partially reduced CO₂ to generate the C₂ alcohol. The chemical stability of ethanol, a quirk of its structure, allows this process to be carried out under surprisingly mild conditions and with high selectivity. Catalysts that have shown good activity have included palladium-copper nanoparticles and cobalt-alumina catalysts, both have been able to produce ethanol at over 90% selectivity under 200°C, with the palladium-copper nanoparticles achieving production of over 100 mmol ethanol per gram catalyst per hour (37). Interestingly however, the cobalt-alumina catalysts, while less active for ethanol production, showed trace production of butanol directly, suggesting that a direct conversion of CO₂ to butanol may be possible with the development of the right catalyst and conditions (38).

CO₂-to-ethanol has also been carried out using both atomically divided copper and traditional Cu/ZnO/Al₂O₃ catalysts, used for methanol synthesis, with electrochemical or plasma assistance. The former has recently been carried out with a Faradaic efficiency of over 90% at −0.7V, although both routes showed very low turnover rates (39, 40). Once again, the fact that butanol has also recently been generated directly from CO₂ by copper electrocatalysis in low yield is worthy of note: in this case, it is thought that the acetaldehyde intermediate has undergone in situ condensation, as found in Guerbet chemistry (41).

An alternative method to convert CO₂ to butanol, using Grignard chemistry, has been demonstrated by the authors (Figure 4). First methylmagnesium bromide is synthesised, which can be generated by the reaction of methyl bromide, produced from CO₂-derived methanol, with magnesium. This then can be reacted with dilute gas phase CO₂ at room temperature and atmospheric pressure to generate acetic acid, thus incorporating the CO₂ capture process itself into the fuel generation. This acetic acid was then dimerised via Claisen condensation and reduced using copper on zinc oxide and hydrogen to form a mixture of alcohols including butanol. In principle the magnesium halide byproduct could then be recycled by high efficiency electrolysis, also allowing for the creation of further methyl halide and an overall electrosynthesis-by-proxy route (42). The magnesium can be regenerated in existing electrolysis processes and returned to the process, an example of stoichiometric metal looping. As the electrolysis process is routinely used to produce magnesium from sea water, additional magnesium halide in the process will increase the efficiency of the metal production.

Fig. 4

As a nascent part of the CO₂-to-fuels research field, other routes could yet be discovered. Some of these routes could include a selective MTO process to produce either butenes or ethylene from CO₂-derived methanol. After dimerisation (if ethylene is used) the resulting butenes can be simply hydrated to produce (primarily) 2-butanol, which is the main industrial manufacturing method for 2-butanol, and also the preferred butanol isomer for transport fuels (43). Finally, butanol can be synthesised directly from sustainable synthesis gas by Cr/ZnO catalysts, although as might be anticipated, selectivity and yield to butanol by this route has so far been low (44).

Synthetic transport fuels will be in increasing demand as fossil-based fuels are phased out. The initial focus in this paper has been on methanol, however internal combustion engines need to be modified because of the corrosivity of high concentration methanol fuels (these are generally limited to below 18% of the fuel as a consequence) as discussed previously. Drop-in hydrocarbons such as synthetic diesel and kerosene have been developed, however these are expensive approaches and require considerable quantities of dihydrogen as a major byproduct of the reduction process is water. Recent developments have focused on DME, formed by the condensation of CO₂-derived methanol (45). Indeed, Volvo (46) and Ford (47) trucks in North America have developed DME vehicles that have been deployed in commercial environments. DME has a lower energy density than diesel but, unlike lower alcohols, is a direct drop-in fuel at 100% concentration. As it contains no C–C bonds it has much lower (approaching zero) particulate emissions than diesel (5) and as it is not fossil-derived it has no SO_x emissions. DME (CH₃OCH₃) has less hydrogen than diesel (C₁₈H₃₈) and so becomes a more economical proposition as less dihydrogen needs to be produced in order to reduce the CO₂. One major challenge will be to produce DME in a single-step process from CO₂ and dihydrogen, or preferably water. Photocatalysis may offer a viable route to achieve this sustainably.

In many countries, governments are proposing a transition to electric vehicles (EVs) as a means to defossilise road transport. However, while there are advantages such as zero tailpipe greenhouse gas emissions, there are also many problems. For a true picture of environmental impact, comparative ‘well to wheel’ and ‘weather to wheel’ technologies should be compared and the latter should be based solely on low carbon energy as a changing grid mix will lead to different carbon intensities (48) with the actual emissions being transferred to point source emitters; power stations. Furthermore, we will need fuels that are compatible with existing internal combustion (IC) engines for the foreseeable future if we wish to avoid the risk of creating a social transport underclass; restricting the use of older spark injection (SI) and compression injection (CI) vehicles.

Ford in Aachen, Germany announced the world’s first original equipment manufacturer (OEM)-built DME passenger vehicle, a Ford Mondeo, at the DME Sustainable Mobility Workshop at Landesvertretung NRW in Berlin in 2019 (49). Ford has claimed that the well-to-wheel CO₂ emissions for a DME powered CI engine could be as low as 5 g km⁻¹, compared to a conventional diesel fuel value of 116 g km⁻¹ (4). A technoeconomic analysis (TEA) of DME derived from CO₂ has shown that DME can be produced at a 740 tonnes per day scale with a minimum selling price of €2193 per tonne, which compares well with fossil-derived DME of around US$3000 per tonne (50).

Conventional DME synthesis relies on the formation of methanol from natural gas or syngas, the latter being an exothermic process as shown in Equation (iii). Direct hydrogenation (Equations (iv) and (v)) can also be achieved catalytically as discussed previously, with the enthalpy of reaction being less than that of CO.

(iii)

(iv)

(v)

Acid catalysed dehydration in the condensation of two molecules of methanol gives DME in an exothermic reaction (Equation (vi)). In order for this be considered a sustainable process, the methanol should be produced from captured CO₂ or from biogenic methanol. Both of these approaches have been applied to the commercial synthesis of DME by Oberon Fuels (51). The provenance of the supply has led to the DME gaining Renewable Fuel Standard approval from the US Environmental Protection Agency.

(vi)

Recently, considerable effort has been directed to the direct synthesis of DME from captured CO₂ and hydrogen (Equation (vii)). The advantages include a single process operation and a highly exothermic reaction. However, there is still an issue that of the six equivalents of hydrogen consumed, half end up in the water byproduct.

(vii)

It has been shown that DME can be produced by the direct reduction of CO₂, however a more exothermic process is the direct hydrogenation of CO in the form of syngas in a 1:1 stoichiometry (Equation (viii)) (52, 53).

(viii)

A benefit of this process is that all the hydrogen is retained in the DME product, however the byproduct is one equivalent of CO₂. There is of course the opportunity to separate the CO₂ and feed it into a second reactor to produce more DME using the processes described in the following discussions. This offers an opportunity to utilise waste gases from iron and steel industries which contain high concentrations of CO and CO₂ as well as hydrogen in the furnace off-gas (54).

Bifunctional catalysts, such as Cu-In-Zr-O with SAPO-34 zeolite (CIZO-SAPO) have been reported (55) for the direct conversion of CO₂ to DME at 250°C and 30 bar. While the selectivity to DME is good (64%) the conversion of CO₂ is disappointing at 4.2%.

Polierer et al. (56) have reported a direct synthesis of DME from CO₂ using a mixed catalyst system based on Cu/ZnO/ZrO₂ produced by continuous precipitation. The catalyst combines methanol synthesis with subsequent dehydration using CO/CO₂ mixtures at 230°C and 50 bar. DME synthesis was enhanced when CO₂-rich gas feeds were used.

The use of a hybrid Cu/ZnO/ZrO₂-ferrite (CZZ(C)-FER) catalyst to produce DME directly from CO₂ has been reported (57). A mechanism was proposed in which hydrogen is adsorbed onto the copper atom and CO₂ onto the Zn-Zr-FER surface. Formation of formate on the surface is achieved by hydrogen transfer which then undergoes dehydration through further hydrogen transfer to give the surface bound methoxide. Two methoxides then combine to give DME, completing the dehydration step.

Modak et al. (58) have recently reviewed the catalytic reduction of CO₂ to give amongst other things methanol and DME and have included the use of hydrides and carbon instead of hydrogen as the reducing agent. An optimised reactor has been used for the direct synthesis of DME from CO₂ (59) using CuO/ZnO/Al₂O₃ and γ-Al₂O₃ as the catalyst system.

In common with many reports, Kornas et al. (60) have reported the Cu-ZrO₂ system as an active catalyst for DME direct synthesis from CO₂, modified using a heteropolyacid, montmorillonite K10 as the acidic motif.

The energy density (MJ l⁻¹) of DME is only 54% that of diesel, and 60% of its specific energy (MJ kg⁻¹). However, it also contains significantly less carbon and so when it burns in air or oxygen it follows the following stoichiometric reaction, Equation (ix):

(ix)

By contrast, the combustion of diesel can be generalised as follows, Equation (x):

(x)

It has been shown that despite the lower energy density and specific energy, in a conventional compression engine, DME demonstrates lower well-to-wheel emissions when compared to fossil-oil derived diesel (61).

DME is a gas under ambient conditions so needs to be stored under slightly pressurised conditions (5.1 bar at 20°C), but considerably lower than the pressures required for hydrogen. In the vehicle, the DME must be pressurised to 12–30 bar to ensure the material is in the liquid phase (62). Oxymethylene ethers (OME_x ) may ultimately be a better option as those with more than three carbons (OME₁ and higher) are liquids that can be stored in conventional tanks. However, the emissions reductions decrease as the value of x increases as while the density becomes similar to diesel the energy density does not increase proportionately. Furthermore, as x increases so do the production costs.

A one-pot synthesis of OME₁ by hydrogenation of CO₂ in methanol using a 3% ruthenium catalyst over the high silica zeolite BEA has been reported (63) to give good selectivity at 150°C (Equation (xi)). A mechanistic study also suggests that the reaction proceeds via a bound formate species.

(xi)