Semi-biological approaches to solar-to-chemical conversion

This review presents a comprehensive summary of the recent development in semi-artificial photosynthesis, a biological-material hybrid approach to solar-to-chemical conversion that provides new concepts to shape a sustainable future fuelled by solar energy. We begin with a brief introduction to natural and artificial photosynthesis, followed by a discussion of the motivation and rationale behind semi-artificial photosynthesis. Then, we summarise how various enzymes can be combined with synthetic materials for light-driven water oxidation, H 2 evolution, CO 2 reduction, and chemical synthesis more broadly. In the following section, we discuss the strategies to incorporate microorganisms in photocatalytic and (photo)electrochemical systems to produce fuels and chemicals with renewable sources. Finally, we outline emerging analytical techniques to study the bio-material hybrid systems and propose unexplored research opportunities in the field of semi-artificial photosynthesis.


Introduction
The consequences of anthropogenic carbon emissions call for innovative strategies to develop renewable energy technologies. Photovoltaics (PV) is the leading technology for solar energy conversion, 1,2 but it shows disadvantages in intermittency and long-distance electricity transmission. Artificial photosynthesis is a process that converts solar energy into fuels and it thereby circumvents these drawbacks by storing solar energy in chemical bonds using synthetic light absorbers and catalysts. 3,4 While synthetic catalysts still encounter challenges in solar-to-chemical conversion, nature provides evolutionarily-optimised biocatalysts. This review summarises how this biological machinery can be leveraged to catalyse light-driven reactions by an emerging technology termed ''semi-artificial photosynthesis'', in which biocatalysts in the form of enzymes or microorganisms are incorporated with particles or structurally-crafted electrodes to produce fuels or chemicals.

Natural photosynthesis
Photosynthesis occurs in photoautotrophs such as cyanobacteria, algae and higher plants, which harvest solar energy to produce biomass and O 2 from CO 2

and H 2 O. Photosynthesis is accomplished
Department of Chemistry, University of Cambridge, Lensfield Road, Cambridge CB2 1EW, UK. E-mail: reisner@ch.cam.ac.uk

Xin Fang
Xin Fang obtained his PhD degree from the University of Cambridge under the supervision of Prof. Erwin Reisner. His doctoral research focused on establishing efficient solar-to-chemical conversion avenues with biocatalysts in the form of enzymes and whole cells. His research broadly traverses the fields of biocatalysis, materials science, electrochemistry, and photocatalysis, with the aim to shift the paradigm of artificial photosynthesis by rationally integrating biocatalysts with synthetic components. Before that, he studied at Fudan University and Sichuan University for a Master and Bachelor's degree in polymer science, respectively.

Shafeer Kalathil
Shafeer Kalathil obtained his PhD in Chemical Engineering at Yeungnam University under the supervision of Prof. Moo Hwan Cho, where he worked on microbial fuel cells and microbial cell factories. He was a JSPS Fellow at the University of Tokyo (with Prof. Kazuhito Hashimoto) and a Postdoctoral Fellow at KAUST (with Prof. Pascal Saikaly). He joined the group of Prof. Erwin Reisner (University of Cambridge) as a Marie Curie Fellow to work on the microbial electrosynthesis of value-added products from waste-carbon. His current research focuses on the integration of light-harvesting materials with microbes to produce industrially relevant chemicals from carbon dioxide. by two phases of reactions: the light reaction uses solar energy to generate the energy carrier adenosine triphosphate (ATP) and the reduced nicotinamide adenine dinucleotide phosphate (NADPH) as the reducing agent; the dark reaction then uses the ATP and NADPH to reduce atmospheric CO 2 into carbohydrates through the Calvin cycle. 5 1.1.1 Light reactions. The light reaction is carried out in the thylakoid membrane by the photosynthetic apparatus, including photosystem II (PSII) and photosystem I (PSI), cytochrome b 6 f (cyt b 6 f ), the plastoquinone (PQ) pool, plastocyanin (PC), ferredoxin (Fd), ferredoxin-NADP + reductase (FNR) and ATP synthase (ATPase) (Fig. 1a). 6 The two light absorbing photosystems both contain antenna complexes and reaction centres. The antenna complex comprises hundreds of pigments (chlorophylls and carotenoids), which capture photons and funnel the light energy to the reaction centre at a high quantum efficiency of 495%. The reaction centre is a transmembrane protein-pigment complex, where electrons are excited and transferred to the downstream chemical synthesis pathway. The reaction centre chlorophylls in PSII and PSI are referred to as P680 and P700, respectively. The two photosystems function in concert to efficiently transfer electrons from H 2 O to NADP + through a ''Z-scheme'' (Fig. 1b). The electron transfer in the thylakoid membrane is initiated by PSII, which absorbs photons and extracts electrons from H 2 O to reduce its terminal electron acceptor, plastoquinone B (Q B ). The fully reduced plastohydroquinone (Q B H 2 ) dissociates from the reaction centre and transfers its electrons to cyt b 6 f, which forwards the electrons to PSI via a small copper-containing protein PC. At PSI, the arriving electrons are energised by a second excitation and finally delivered to FNR to reduce NADP + to NADPH (Fig. 1a). 5 Simultaneously, the water oxidation and the electron transport induce proton translocation from the stroma to the lumen, which generates a proton gradient and chemiosmosis driving the synthesis of ATP by ATPase.
1.1.2 Dark reactions. The NADPH and ATP produced by the light reaction are then used to reduce CO 2 to carbohydrates in the Calvin cycle. During the process, CO 2 and H 2 O are combined with the carbon acceptor ribulose-1,5-bisphosphate to yield two molecules of 3-phosphoglycerate. This intermediate is then reduced to glyceraldehyde-3-phosphate with ATP and NADPH. The cycle closes with the regeneration of ribulose-1,5bisphosphate, the initial CO 2 acceptor. 6 The Calvin cycle accounts for more than 90% of carbon assimilation on earth. Central to this reaction is ribulose-1,5-bisphosphate carboxylase/oxygenase (Rubisco) that covalently fixes CO 2 to a carbon skeleton. 7,8 Rubisco is abundant in nature and operates using atmospheric CO 2 concentrations (B415 ppm). However, it also displays slow catalysis and low selectivity. The turnover frequency (TOF) of Rubisco is typically less than 10 s À1 , rendering the photosynthetic CO 2 fixation inefficient under optimal conditions. 8 It also confuses the substrate (CO 2 ) and the product (O 2 ) of photosynthesis, Fig. 1 Schematic illustration of the light reaction in natural photosynthesis. (a) Electron and proton transfer pathways in the thylakoid membrane. At the start of the photosynthetic chain is PSII that oxidises H 2 O and releases O 2 and protons upon light absorption. The electrons are then delivered to PSI via a plastoquinone (PQ) pool, cyt b 6 f and plastocyanin (PC). Electrons are photoexcited for a second time at PSI to reduce NADP + to NADPH via a ferredoxin (Fd) and a ferredoxin-NADP + reductase (FNR). The water oxidation and electron transport also induce proton translocation from the stroma to the lumen, which generates a proton gradient and chemiosmosis driving the synthesis of ATP by ATP synthase (ATPase). Protein data bank ID: PSII (4ub6), cyt b 6 f (4h44), PC (1bxu), PSI (5oy0), Fd-FNR (2yvj), ATPase (6fkf). (b) Energy level diagram of the Z-scheme electron transfer in the thylakoid membrane. The redox carriers are placed at their midpoint redox potentials at pH 7.0. which offsets the CO 2 uptake and saddles oxygenic phototrophs with energy-dissipating photorespiration. 9 Photorespiration can consume 25% of carbon photosynthetically fixed by C 3 plants and this carbon loss is more severe at higher temperature. Some tropical plants, namely C 4 plants, overcome the disadvantageous photorespiration by deploying an additional C 4 pathway to increase the local concentration of CO 2 . In these plants, phosphoenolpyruvate is first carboxylated into a four-carbon compound oxaloacetate, which undergoes further transformations and decarboxylation to release CO 2 . The released CO 2 is then scavenged by Rubisco and drawn into the Calvin cycle with minimal photorespiration. 6 1.1.3 Photosynthetic efficiency. Photosynthetic organisms have evolved highly efficient light-harvesting systems with a quantum efficiency of more than 90%. Nevertheless, the overall photosynthetic efficiency is low: the highest efficiency measured under field conditions is 2.9% for C 3 and 4.2% for C 4 plants and typical efficiencies are 0.2-1% for crop plants. 10,11 Several reasons are responsible for the low photosynthetic efficiency: 10,12 (1) As the pigments in photosystems and antennae are primarily sensitive to visible light (l = 400-700 nm), photosystems can only intercept 48.7% of the incident solar energy; (2) light reflection and transmission due to weak absorption of green light (l = 500-600 nm) cause at least 4.9% of energy loss; 13 (3) 6.6% of energy will be dissipated as heat during the transfer in antenna complexes, before reaching the red lightabsorbing chlorophylls in the reaction centres; (4) energy expenditure associated with carbohydrate synthesis accounts for 24.6% and 28.7% for C 3 and C 4 plants, respectively; (5) oxygenation and ensuing photorespiration cost extra 6.1% of energy in C 3 plants; (6) respiratory metabolism for maintenance and growth consumes a minimum of 1.9% (C 3 plants) and 2.5% (C 4 plants) of the entire energy input from solar irradiation. Summing all of these energy losses, the maximal theoretical solar-to-biomass conversion efficiencies are 4.6% for C 3 plants and 6.0% for C 4 plants (30 1C). 10, 12 1.1.4 Photobiological H 2 production and N 2 fixation. Cyanobacteria and microalgae contain thylakoid membranes to carry out photosynthetic reactions, 14 and they can produce H 2 in anaerobic conditions by partially diverting photo-generated electrons from PSI towards H 2 -producing hydrogenases (H 2 ases). [15][16][17] However, inhibition of H 2 ase by evolved O 2 renders this process transient and economically unfeasible. In addition, electron transport from PSI to H 2 ase requires diffusional electron carriers such as ferredoxins, which impose a kinetic limit on the overall process. 18 Some cyanobacteria can also fix atmospheric N 2 using solar energy when nitrogen-containing substrates (NH 3 , NO 3 À , etc.) are not available. 19,20 However, nitrogenase (N 2 ase), the enzyme responsible for N 2 fixation, is sensitive to oxygen. 21 Therefore, cyanobacteria have developed strategies to separate the oxygenic photosynthesis and N 2 fixation spatially (in different cells) or temporally (during the night) to protect the N 2 ase from irreversible damage. 20

Artificial photosynthesis
Artificial photosynthesis does not reproduce the exact reactions occurring in photoautotrophs, but exploits light absorbers and catalysts to produce fuels and chemicals with earth-abundant feedstock chemicals such as H 2 O and CO 2 . A prototype reaction is light-driven water splitting, which produces H 2 as a suitable energy carrier to overcome the intermittency of solar irradiation. 22,23 Solar water splitting has been under intense investigation over several decades, but it still faces challenges arising from solar light harvesting, catalyst stability, interfacial charge transfer and the water oxidation reaction. 24 The fourelectron water oxidation to oxygen involves multiple bond rearrangements and concerted proton release, and thus causes both a thermodynamic barrier and kinetic sluggishness that confront the existing catalytic chemistry. 25 Thermodynamics determines the suitability of light absorbers from an energetic perspective (Fig. 2), as illustrated by the water splitting reaction, where the energy band positions should straddle the electrochemical potentials of the half reactions, i.e., hydrogen evolution reaction (HER) and oxygen evolution reaction (OER) (Fig. 3a). The band gap of the light absorbers should not be too wide to enable the visible light (400-800 nm, band gap: 1.53-3.10 eV) utilisation. 26 The thermodynamic redox potentials of light absorbers should also be properly aligned with respect to their energy band positions to prevent themselves from photoinduced corrosion. 27 Several semiconductors whose electronic structures suffice in theory for overall water splitting with visible light have been under investigation, such as C 3 N 4 , CdS and Y 2 Ti 2 O 5 S 2 . 22 An alternative strategy is to align the band position of two light absorbers and catalyse HER and OER with the aid of electron mediators (Fig. 3b), 22,28 which resembles the photosynthetic electron transfer (''Z-scheme'') in the thylakoid membrane (Fig. 1b). This two-step excitation system can expand the light absorption by harvesting lower-energy photons, 22,29 and can bring together materials that can only drive either HER or OER alone. [30][31][32] Such tandem systems enable larger driving forces for water splitting, and permit higher theoretical solar-tohydrogen conversion efficiencies than the single light-absorber systems. 33 Yet two light absorbers also multiply chances of charge recombination and pose challenges to balance the electron transfer kinetics of the half reactions. 3,22 The implementation of artificial photosynthesis is often envisioned in two forms: first, light absorbers are loaded with electrocatalysts and then suspended in a photoreactor (Fig. 3c). 34 Second, solar fuel synthesis is performed in a photoelectrochemical (PEC) cell, where catalysts are immobilised on photoelectrodes and photoinduced electrons flow across the external circuit (Fig. 3d). 35 From a techno-economical viewpoint, the photocatalytic system may be more advantageous, as PEC systems may be required to achieve up to B25% solar-to-hydrogen (STH) conversion efficiency to rival with petrol in energy prices, primarily due to higher investment in installations, whereas a STH efficiency of 5-10% would possibly suffice to render photocatalytic reactors costcompetitive. 4,34 However, the highest STH conversion efficiency of particle-based systems is around 1%, 28 whereas that of electrodebased systems can typically achieve more than 10%. 36,37 PEC systems also have several merits that are appealing for fundamental research and practical applications: (1) immobilisation of photocatalysts on electrodes enables in-depth studies of half-reactions individually without sacrificial reagents and allows for in situ probing of photoredox chemistry through spectroscopic methodologies; [38][39][40] (2) from an application's perspective, the electrode-based configuration enables separation of H 2 and O 2 in a two-compartment configuration, 41 and can be transformed into a continuous fuel-production system with a flowing electrolyte solution. 42 On the other side, drawbacks emerge: (1) the preparation of photoelectrodes requires conductivity and high film-forming capability of semiconductors; (2) the operation of a PEC cell can generate pH gradients and cause mass transfer limitations that account for substantial potential loss and subject electrodes to increasingly corrosive environments; 43,44 (3) the internal resistance often necessitates external bias voltages to drive the electron flow from a photoanode to a photocathode.
Sustainable transformation of CO 2 to chemicals and fuels would provide a means to close the loop of the anthropogenic carbon cycle and offer a viable solution for carbon capture and utilisation. 45 The linear CO 2 molecule is chemically stable, and thus activating the CQO bond for the ensuing endothermic reactions may incur a substantial energy penalty, which is reflected in the negative potential (À1.9 V vs. SHE, pH 7.0) for the oneelectron reduction to CO 2 À . [46][47][48] Although such a high energy barrier can be overcome by large electrochemical potentials or high temperature, it results in poor energetic efficiencies for subsequent fuel synthesis. The activation of CO 2 is followed by a stepwise proton and/or electron transfer, which gives rise to furcate reaction pathways towards miscellaneous products. 45,49 In addition, due to the low solubility of CO 2 in water (0.033 mol L À1 at 25 1C, 100 kPa), 46 the reduction of CO 2 usually competes with H 2 evolution from H 2 O, which further reduces the selectivity and the conversion efficiency. 48 Current photocatalytic CO 2 reduction systems can produce CO and formate, but still struggle to produce multicarbon chemicals such as ethylene and ethanol selectively, which challenges the economic exploitation of this approach. 50 Artificial photosynthesis has recently been extended to drive reactions such as N 2 reduction. Today, industrial N 2 reduction  119 For molecular dyes, the lowest unoccupied molecular orbital (LUMO) and the highest occupied molecular orbital (HOMO) are used as conceptual equivalents of conduction band edge and valence band edge in semiconductors. Data source: DPP, 147 RuP, 147 Eosin Y (EY), 292 TiO 2 , 293 g-C 3 N 4 , 294 carbon dot (CD), 283,295 CdS, 296 BiVO 4 , 157 p-Si. 162 The redox potentials CO 2 /CO (catalysed by CODH), 186   Energy level diagrams of light-driven water splitting at pH 7.0: single-step excitation (a) and two-step excitation (b). Dye and Dye*: the light absorber in the ground and excited state, respectively. Med.: redox mediator. (c) Representation of a two-step photocatalytic water-splitting system with semiconductor particles and HER and OER co-catalysts. The HER and OER photocatalysts are interfaced with a solid-state electron mediator. (d) Representation of a PEC tandem water-splitting system. relies on the energy-demanding Haber-Bosch process that requires H 2 , high temperature (300-500 1C) and high pressure (150-200 atm). 51,52 The Haber-Bosch process also entails considerable CO 2 emission during the production of the inlet H 2 from natural gas (1.87 kg CO 2 per 1 kg NH 3 ). 51 The intense energy consumption and environmental stress arising from the Haber-Bosch process have spurred explorations in sustainable routes for N 2 -to-NH 3 transformation. Photocatalytic N 2 reduction uses solar light to replace fossil fuels as the energy source and water instead of natural gas as the hydrogen source and operates at room temperature and atmospheric pressure. [53][54][55][56] The prospect of this approach largely depends on the photocatalysts that generate photoelectrons to reduce the stable and inert NRN bond. However, the existing photocatalytic N 2 reduction systems still display low solar-to-chemical conversion efficiencies (typically below 0.02%), 51,55,57 and current studies encounter technical difficulties in reliable NH 3 quantification and elimination of probable experimental artefacts. 58 2 Semi-artificial photosynthesis Artificial photosynthesis establishes solar-to-fuel pathways to address the global energy challenge, but the bottlenecks are key-step reactions during the fuel-forming process, such as water oxidation and selective CO 2 activation. Biology is highly capable of tackling these synthetic challenges through the naturally-refined biocatalytic machinery. Enzymes comprise only a handful of earth-abundant metal atoms as catalytic centres and cofactors, which are embedded in polypeptide chains (Fig. 4a). 46 Enzymes perfectly orchestrate the electron and proton transfer, reactant delivery, bond rearrangement and product removal at the active site, thereby reducing the energy threshold and accelerating the reaction kinetics. Enzymes widely employ steric hindrance and electrostatic or hydrogen bond interactions to stabilise selected intermediates and transition states, and therefore provide an efficient reaction path towards a single product. 59 The catalytic prowess of enzymes has inspired synthetic endeavours to mimic their active sites and ligand environments, and to exploit the biological strategies to stabilise intermediates and control selectivity. [60][61][62] Although synthetic mimics are making progress, they have not yet reproduced the performance of their natural models in terms of catalytic rate, selectivity and electrochemical overpotentials under benign aqueous conditions.
Microorganisms can produce complex multicarbon compounds from simple feedstocks through their intracellular pathways (Fig. 4b). They can also maintain metabolic homeostasis at varying environmental and nutritional conditions. The enzymatic machinery embedded in its well-controlled and confined native environment benefits from the innate regulation mechanisms and can be repaired and regenerated when necessary. On the other side, microorganisms rank their physiological needs to survive ahead of synthetic efficiency to produce target chemicals. Carbon and electron fluxes in the metabolism are partially directed towards biomass synthesis for cell growth and maintenance, which reduces the pathway efficiency in chemical synthesis. Moreover, rerouting biochemical pathways towards desired products encounters resistance from the intracellular regulation, which renders such alteration problematic. 63,64 Artificial photosynthesis, however, permits more flexibility in system design and modification. Taking light harvesting as an example: most chlorophylls in nature have a minimum absorption to green light (l = 500-600 nm), 13 which partially accounts for the low photosynthetic efficiency, whereas in artificial photosynthesis, broadband absorption can be readily achieved with a variety of semiconductors and molecular dyes. 11 Furthermore, artificial systems allow for coupling fuelforming reduction reactions with useful oxidative chemical transformations. Several studies have already demonstrated that electrons could be extracted from biomass, organic compounds and even plastics to produce H 2 with versatile light-absorbing systems. [65][66][67][68][69] Artificial photosynthesis systems reduce the dissipation of energy and electrons along the pathway, enabling high solar-to-fuel efficiencies routinely surpassing their natural counterparts. [70][71][72][73] Artificial photosynthesis is also empowered by an array of analytical techniques such as electrochemistry, spectroscopy and operando methodologies, 40,45,74,75 which are still less commonly employed for quantitative studies in biological research. The well-defined features of synthetic materials permit systematic investigations to understand reaction mechanisms and establish structure-function relationships for system design and optimisation. 76 Semi-artificial photosynthesis provides a hybrid approach to solar-to-chemical conversion, by integrating the biocatalytic machinery (enzymes and microbes) with synthetic materials (dyes, electrocatalysts, semiconductors, electrodes, mediators). 77 The photosynthetic biohybrid systems seek to outsource tasks to the components that can perform them best and thus combine strengths of both while bypassing limitations of each. In such hybrid systems, enzymes and microbes function often as catalysts to drive endergonic and complex chemical reactions, whereas synthetic materials are commonly scaffolds to immobilise biocatalysts and functional components that carry out light absorption, charge transfer, chemical transformations and product separation. Several synthetic materials such as polymers and metal organic frameworks can also provide additional protection to fragile enzymes for more sustained operation against environmental stressors. [78][79][80] According to the form of biocatalyst, semi-artificial photosynthesis falls into two categories: the enzymatic and the microbial system. Each of these systems can be further distinguished as either a colloidal suspension or a PEC cell (Fig. 5).

Enzymatic hybrid systems
An enzymatic hybrid system employs separated electroactive or photoactive enzymes as catalysts or light absorbers, to function in concert with synthetic components in driving reactions selectively at high rates and yields ( Fig. 5a and b). Amongst a myriad of enzymes available in nature, only a handful of them are of interest in synthetic reactions starting from the simplest feedstocks (e.g., H 2 O, CO 2 and N 2 ). Translation of their catalytic ability in vivo into advantages in vitro is confronted with several challenges. Whereas synthetic catalysts are either integral parts of electrodes or discrete entities (particles or molecules), these redox enzymes are proteins with molecular weights of 10 4 -10 6 g mol À1 and areal footprints of 50-400 nm 2 . Their active sites (catalytic centres) are embedded within the protein matrices, which prevent indiscriminate reactions with substrate analogues. These enzymes have evolved intraprotein electron relays such as haems, quinones, and [Fe-S] clusters that can carry electrons between active sites and redox partners. 81 The electron relays define directional electron transfer pathways within the protein and allow electrochemistry to activate, monitor and modulate the enzymatic redox chemistry. 82 The current knowledge of the structure, functionality and catalytic mechanism within the enzymes discussed in this section has already been reviewed previously and is therefore not discussed here. 59,[83][84][85][86] To approach maximal catalytic rate, enzymes are expected to accelerate the chemical reactions towards diffusion controlled kinetics. 87 As such, enzymes should engage with electrodes in an ''electroactive'' orientation to enable rapid electron transfer kinetics between the distal relay centre and the electrode surface. In this regard, the enzyme-electrode interface must be rationally engineered with respect to topology, porosity and surface chemistry (hydrophilicity, surface charge, functional moieties, etc.). From a kinetic perspective, redox enzymes  145 Under irradiation, the dye is excited and donates electrons to the conduction band of TiO 2 . The photoelectrons are then delivered to the catalytic centre of H 2 ase for proton reduction via intraprotein electron relays. The oxidised dye is regenerated by extracting electrons from a sacrificial electron donor (SED). (b) An electrode system with a three-dimensional indium tin oxide (ITO) electrode and PSII. 102,120 PSII is wired by the porous ITO scaffold and photocatalyses water oxidation, aided by an applied potential. (c) A colloidal system with light absorbing CdS nanoparticles deposited on the CO 2 -reducing bacterium M. thermoacetica. 226 The excited CdS transfers photoelectrons to the bacterium, which generates reducing equivalents to produce acetate from CO 2 via the Wood-Ljungdahl pathway. The holes in CdS are quenched by the SED. (d) A hybrid tandem system with TiO 2 nanowires as the photoanode and p-Si nanowires integrated with acetogenic bacteria S. ovata as the photocathode. 248 When irradiated, the TiO 2 electrode oxidises water while p-Si nanowires generate photoelectrons that are delivered to the interfacing S. ovata via extracellular electron transfer pathways. S. ovata uses these electrons to reduce CO 2 into acetate through the Wood-Ljungdahl pathway.
reduce the activation energy and thus, minimise the electrochemical overpotential needed to drive a reaction. 59,88 Therefore, enzymes can often catalyse redox reactions very close to their thermodynamic potentials and reduce the energy penalty arising from large overpotentials. The low overpotentials can simplify system design by judiciously pairing with light absorbers and electron mediators according to their energy levels, and carry out reactions that were otherwise not possible with conventional electrocatalysts.

Water oxidation
Water is the ''ideal'' terminal electron source for the downstream fuel-forming reactions in artificial photosynthesis, but only few molecular catalysts with earth-abundant elements can handle this multielectron/proton process in an efficient fashion, which is further complicated by the combination with light absorbers. PSII, however, assigns the tasks of light harvesting and water oxidation to antenna proteins and the oxygen evolution centre (OEC), respectively, and orchestrates photochemistry in concert with catalysis in a stepwise manner via efficient intraprotein electron relays. 89 However, the intrinsic fragility of PSII under light irradiation limits the longevity of its catalytic activity (t 1/2 o 30 min) and thereby strangles the continuous downstream synthesis with isolated PSII. 90 Antenna complexes in PSII harvest solar light and funnel the light energy to P680, where charge separation occurs within a few picoseconds. 91 This is followed by an electron transfer towards a pheophytin molecule (Phe D1 ) and later to plastoquinone A (Q A ) (Fig. 6a). The resulting highly oxidising P D1 + in P680 can extract an electron from a redox-active tyrosine Y Z (Tyr Z ) and then initiate water oxidation at the Mn 4 CaO 5 cluster. 91 On the stroma side, Q A is tightly bound to the D2 protein and acts as a single-electron acceptor, whereas Q B in the D1 protein can accept two electrons and be fully protonated. 92 The charge transfer from Q A to Q B is aided by a non-haem iron midway between them. 93 The formed Q B H 2 departs from the reaction centre and is replenished by the plastoquinone pool in the thylakoid membrane. PSII is a biological model OER photocatalyst that inspires synthetic endeavours to mimic its core components, 61,94-96 and stimulates mechanistic investigations to understand its functionality. [97][98][99] Wiring PSII to an electrode allows the resulting biohybrid electrode to supply photoelectrons from water, and permits protein film-photoelectrochemistry (PF-PEC) to benchmark enzymatic activity in vitro, dissect photoinduced electron transfer pathways, and repurpose biogenic electrons to drive endergonic reactions. 90,100 Attempts to interface PSII with an electrode surface started several decades ago, 101 when isolated PSII was deposited on a Pt electrode and generated a photocurrent of a few microamperes. However, the minuscule amount of PSII loaded on the electrode renders in-depth studies and proof-of-concept demonstrations unfeasible. Underlying PSII-PEC is the electrode that provides a physical scaffold to immobilise enzymes and an artificial electron acceptor for the biogenic electrons. PSII's photochemistry and the ensuing current output can be greatly influenced by the material, morphology and physical property of the electrode scaffold, which dictates the strength of protein binding, the capacity of protein loading, the accessibility to intraprotein electron relays, the depth of light penetration, and the transport of reactants and products therein. 102 The directionality of the intraprotein electron flow makes the interfacial electron transfer highly dependent on the protein orientation. Q B , the terminal electron acceptor in PSII, can only undertake outward electron transfer if the stromal side of PSII is in close proximity to an electrode surface (Fig. 6b). The electron transfer via intrinsic plastoquinones, namely, direct electron transfer (DET) can be registered as an anodic photocurrent by chronoamperometry under irradiation, and verified in control experiments by the removal of the Mn 4 CaO 5 cluster or addition of 3-(3,4-dichlorophenyl)-1,1-dimethylurea (DCMU) that inhibits the Q B site and interrupts the electron flow from Q A . 103,104 DET often results in a relatively low photocurrent because PSII loading may be low and a significant portion of the enzymes may be in an unfavourable orientation or remote from the electrode surface (Fig. 6c). This problem can be mitigated by using diffusional mediators such as 2,6-dichloro-1,4-benzoquinone (DCBQ, E m = 0.32 V vs. SHE, pH 7.0) to shuttle electrons from the Q B site to the electrode (Fig. 6c). 103 This mediated electron transfer (MET), in biological and artificial systems, comes at an expense in both energetics and kinetics, as extra energy is needed to drive the redox turnover of electron mediators and the diffusion of mediators is likely to govern the overall rate of electron transfer.
From both fundamental and practical viewpoints, an electrode with a large number of proteins being wired in an electroactive orientation is desired to produce a high photocurrent that can afford reliable analysis and proof-of-concept demonstrations (Fig. 7a). The making of such electrodes enables new possibilities to create solarto-fuel pathways unattainable in biology. The chronology of the development of PSII electrode design will be outlined below. 100 In the nascent stage of PSII-PEC, PSII with a polyhistidine tag (His-tag) at its stromal side was immobilised on flat Au electrodes by a self-assembled monolayer (Fig. 7b). [105][106][107][108] The interaction between electrode-anchored nickel-nitrilotriacetic acid and His-tags permitted site-selective binding of PSII on Au electrodes. However, the flatness of the electrode surface limited the protein loading typically below 10 pmol cm À2 and thus resulted in submicroampere photocurrents. 108 Electrodes using a redox polymer matrix as a non-diffusional mediator to electrically wire a high amount of proteins enable redox centres on the polymer backbone to access intraprotein electron relays, regardless of their orientations and distances (Fig. 7c). 109 A polymer with the Os 3+ /Os 2+ redox couple (E 1/2 = 0.39 V vs. SHE) was employed as both an immobilisation matrix and an electron mediator to PSII, which gave rise to a photocurrent density up to 45 mA cm À2 . 110 The second generation of electrodes are mesoporous films made with metal oxides such as Fe 2 O 3 , TiO 2 and ITO. 104,[111][112][113][114][115] A mesoporous ITO (meso-ITO) film with a thickness of 2-10 mm and a pore size up to 100 nm can be made with a high degree of tunability (Fig. 7d). 104,111,116,117 The meso-ITO afforded a PSII loading of 19 pmol cm À2 , which resulted in a DET photocurrent of 1.6 mA cm À2 and a mediated photocurrent of 22 mA cm À2 . 104 The surface chemistry of ITO can be modified to covalently bind PSII in favour of an electroactive orientation. 111 The meso-ITO film can also be deposited onto the disk of a rotating ring disk electrode to detect the O 2 production on the ring and study the oxygenic photoreactivity in PSII. 118 The state-of-the-art third generation electrodes for PSII-PEC feature a hierarchical inverse opal structure with both macroporosity and mesoporosity (Fig. 7e). 102,[119][120][121][122][123][124][125][126][127] This electrode architecture has interconnected macropores that provide a large surface area accessible by protein diffusion. The skeleton is made of a mesoporous structure consisting of (semi)conducting and hydrophilic nanoparticles (e.g., ITO or TiO 2 ) where proteins are adsorbed. The inverse opal electrodes can be fabricated via a co-assembly method using polystyrene beads as the structural template and nanoparticles as the electrode material. Such method makes the electrode structure easily variable to fit biocatalysts with different dimensions. 128,129 The hierarchical structure also benefits both mass transport and light transmission, and permits high PSII loading in the range of 30-1000 pmol cm À2 (depending on the film thickness). [119][120][121][122] A 40 mm thick inverse opal-ITO (IO-ITO) electrode with PSII attained a high photocurrent of 17 mA cm À2 for DET and 930 mA cm À2 in the presence of a DCBQ mediator. 120 Os-based polymers can further be incorporated in the IO-ITO scaffold to improve the electrical wiring between PSII and the electrode to eliminate the need for diffusional additives. 121 Compared to flat electrodes, 110 the IO-ITO electrode with the polymer matrix achieved a higher photocurrent of 381 mA cm À2 due to higher PSII (336 pmol cm À2 ) loading. 121 The high photocurrent allowed for reliable O 2 quantification to calculate the Faraday efficiency, which was not possible with flat electrodes.
PSII immobilisation on porous ITO electrodes also allowed PF-PEC to dissect unexpected electron transfer pathways at the enzyme-electrode interface. 100 DET from Q A to the ITO electrode was observed when inhibiting PSII with DCBQ, showing the possibility to release electrons upstream of Q B . 104,130 Furthermore, analysing cathodic current contributions at low potentials allowed identifying a competing pathway stemming from photoinduced O 2 reduction, presenting a short-circuit to the natural water oxidation process. 118,130 Although PSII is the only enzyme that operates in nature for water oxidation to O 2 , it is not the only enzyme that can perform this reaction in vitro. Laccase, a blue multicopper oxidase that couples the oxidation of organic substrates with the reduction of O 2 to H 2 O, has been shown to catalyse the reverse water oxidation reaction when immobilised on an electrode (E 4 1.2 V vs. SHE, pH 7.4). 131 In an attempt to use this enzyme for light-driven water oxidation, laccases were adsorbed on a semiconducting In 2 S 3 electrode (band gap: 2.0 eV) as a visible-light absorber. 132 3.2 Hydrogen production H 2 ases are metalloenzymes that can catalyse the interconversion between H 2 and H + /e À with a TOF of 41000 s À1 . 84,133 The HER activity of H 2 ases per active site is therefore comparable with that of the benchmark Pt catalyst. 134 136 These [Fe-S] clusters are spaced 10-14 Å apart, which permits sequential electron tunnelling towards the active sites at a rate (10 7 s À1 ) faster than catalysis. 137,138 These intraprotein electron relays can couple the catalytic turnover at the active site with the electron exchange through the electrode, and thus enable protein film electrochemistry (PFE) to probe the H 2 -H + interconversion by deciphering resulting voltammograms and chronoamperograms. 135 136 [NiFeSe]-H 2 ases are a subclass of [NiFe]-H 2 ases with a selenocysteine residue coordinated to the nickel at the active site instead of a cysteine. 138 [NiFeSe]-H 2 ases are kinetically more biased than other [NiFe]-H 2 ases towards H 2 evolution without substantial product inhibition ( Fig. 8i and j). The presence of selenium provides additional protection to the nickel centre from oxidative attacks, and hence confers better oxygen tolerance. 138,140,141 [NiFeSe]-H 2 ase is therefore a common model biocatalyst and has been rationally coupled with various light absorbers for photocatalytic HER (Fig. 8). Early attempts employed dyes to generate photoelectrons, which were carried to H 2 ases by diffusional electron mediators such as methyl viologen (MV), in the presence of a sacrificial electron donor (SED). 142,143 A more recent study shows that an organic dye Eosin Y could directly transfer photoelectrons to [NiFeSe]-H 2 ases without any electron mediators (Fig. 8a). 144 The homogeneous system produced 0.5 mmol h À1 H 2 for up to 15 h, corresponding to a TOF H 2 ase of 13.9 s À1 . Notably, this Eosin Y-H 2 ase system also demonstrated some tolerance towards oxygen: more than 80% of photoactivity was retained with 5% O 2 and the system remained photoactive even at aerobic conditions (21% O 2 ) for 3 h. The system performance was limited by the interfacial electron transfer between dye molecules and H 2 ases.
To this end, semiconductors were used as light absorbers with immobilised H 2 ases. Photogenerated electrons can thereby stream directly from the conduction band towards the enzyme's distal relay centre (Fig. 8b). Co-adsorbing [NiFeSe]-H 2 ases on RuP-TiO 2 particles through the putative interaction between the TiO 2 surface and side-chain carboxylates near the distal [Fe-S] cluster yielded 3.56 mmol H 2 in the first hour of irradiation and a benchmark TOF H 2 ase of 50 s À1 . 145,146 Replacing RuP with a metal-free diketopyrrolopyrrole (DPP) dye to sensitise TiO 2 also attained similar HER activity (Fig. 8b). 147 Likewise, [FeFe]-H 2 ases were immobilised on CdTe nanoparticles or CdS nanorods that had their surfaces modified with 3-mercaptopropionic acid. 148,149 The negatively-charged surface induced [FeFe]-H 2 ases to interface with the semiconductors at their positive patch close to the distal [Fe-S] cluster. Transient absorption spectroscopy elucidated that the electron transfer occurred between the conduction band of CdS and the distal [Fe-S] cluster in [FeFe]-H 2 ase, regardless of the enzyme activity, at a rate (B100 ns) comparable with that of the charge recombination in CdS. 150 The rate of electron transfer is several orders of magnitude slower than that in synthetic systems (10 À3 -1 ns), 151 probably due to the longer tunnelling distance (410 Å) between the semiconductor surface and the distal [Fe-S] cluster. Nevertheless, the interfacial electron transfer is fast enough to outpace the incident photon flux, rendering the light absorption by CdS performancelimiting. 150 The H 2 evolution rate of the above systems declined over a few hours, presumably due to dye degradation or an unstable material-enzyme interface. Thus, robust carbon-based light absorbers such as carbon nitride (CN x ) and carbon dots were utilised to improve the long-term stability (Fig. 8c). [152][153][154] The CN x -H 2 ase hybrids remained photoactive for 48 h, totalling 2.5 mmol of H 2 production and a turnover number (TON) of 450 000. 152 But the weak interaction between CN x and H 2 ases rendered the TOF H 2 ase much lower (1.5 s À1 , in the first 4 h) than that of RuP-TiO 2 -H 2 ase hybrids. This problem can be partially resolved by blending CN x with TiO 2 that has a high affinity for H 2 ases. 153 The ternary hybrids exhibited better photocatalytic activity with a higher TOF H 2 ase of 7.8 s À1 within the first 4 h of irradiation and a longer H 2 evolution span of 72 h. The photostability of H 2 ases was even better than their synthetic analogue, a molecular DuBois-type Ni bis(diphosphine) catalyst, that decomposed within 3 h of irradiation. 152 [NiFeSe]-H 2 ases adsorbed on amine-capped carbon dots attained similar activity: a TOF H 2 ase of 1.1 s À1 , H 2 production of 2.6 mmol and a total TON of 52 000 in 48 h. 154 A comparison with carboxylic acid-capped carbon dots echoes with the previous finding that [NiFeSe]-H 2 ases prefer to electrostatically interact with positive surfaces, and underlines the importance of semiconductor surface chemistry that controls the electronic communication with enzymes. 148,149,155 This journal is © The Royal Society of Chemistry 2020 Chem. Soc. Rev., 2020, 49, 4926--4952 | 4935 The aforementioned systems employ SEDs such as triethanolamine (TEOA) or ethylenediaminetetraacetic acid (EDTA) to expediently bypass the kinetic difficulty of water oxidation. To enable water to supply electrons for reductive reactions, synthetic chemistry and biology utilise different approaches. Artificial photosynthesis can carry out water oxidation separately by a photoanode in a PEC system. BiVO 4 has a band structure well-suited for solar water oxidation, but it is insufficient to reduce protons ( Fig. 2 and  8d). 156,157 A photocathode hosting HER catalysts is thus required to further energise the electrons that are withdrawn from water by the BiVO 4 photoanode. The electrochemical window of ITO (À0.6 to +2 V vs. SHE, pH 7.0) limits its application at negative potentials needed to drive reducing reactions. 158 In view of this, ITO can be substituted with semiconductive TiO 2 to fabricate porous electrodes that are applicable under cathodic conditions. [159][160][161] To this end, [NiFeSe]-H 2 ases were interfaced with p-type silicon that was stabilised by a mesoporous TiO 2 layer. 162,163 The TiO 2 layer is necessary to protect the underlying p-Si from unwanted surface passivation, 163,164 but also to assist in binding [NiFeSe]-H 2 ases. An ensuing problem is that enzymes having a large footprint show a low loading capacity on the mesoporous TiO 2 layer. 163,165 As such, a hierarchical IO-TiO 2 structure was built atop the p-Si to increase the H 2 ase loading (up to 120 pmol cm À2 ) (Fig. 8e). 166 A 4 nm TiO 2 interlayer was deposited on the surface of p-Si by atomic layer deposition to protect the electrode from the formation of an insulating SiO 2 layer, followed by the construction of the IO-TiO 2 scaffold. The p-Si|IO-TiO 2 |H 2 ase electrode gave a photocurrent onset potential of 0.0 V vs. SHE, pH 6.0, which is more positive than that of a water oxidising BiVO 4 -TiCo photoanode (À0.1 V vs. SHE, pH 6.0). Thus the BiVO 4 -TiCo||p-Si|IO-TiO 2 |H 2 ase tandem PEC system afforded overall solar water splitting without an external voltage, resulting in a stoichiometric production of H 2 (0.47 mmol, Z F = 98%) and O 2 (0.20 mmol, Z F = 84%) after 5 h of irradiation. 166 [FeFe]-H 2 ases have also been interfaced with black Si electrodes with submicron porosity, but they are prone to desorb from the electrode surface due to weak interfacial interactions. 167 Instead of using a p-type semiconductor, the H 2 ase-loaded IO-TiO 2 scaffold was also interfaced with an encapsulated lead halide perovskite solar cell that generated a photovoltage of 0.9 V. In a tandem configuration with a BiVO 4 -TiCo photoanode, the resulting system produced H 2 coupled to O 2 evolution with a STH efficiency of 1.1%. 168 To streamline the electron transfer in biological H 2 production, photosystems and H 2 ases could be directly wired via molecular wires, redox polymers, protein subunits or external circuits to eliminate the diffusion-governing steps. 18,120,166,[169][170][171] For example, PSI has been connected with [FeFe]-H 2 ases by 1,6-hexanedithiol to allow photogenerated electrons to tunnel between the two proteins (Fig. 8g). 18,169,172 Alternatively, PSII and H 2 ases were immobilised on IO-ITO electrodes and compartmentalised in a PEC cell (Fig. 8h). 120 The onset potential of an IO-ITO|H 2 ase cathode (À0.4 V vs. SHE, pH 6.5) remained close to the thermodynamic value of proton reduction, whereas electrons from P680 were attenuated during the intraprotein electron transfer in PSII (Fig. 2).
The energy is further offset by the use of electron mediators such as DCBQ that aid in the electron transfer from PSII to the electrodes. As such, a minimum bias voltage of 0.6 V was required to drive the photoelectrons from PSII towards H 2 ase. At an applied voltage of 0.9 V, this biohybrid PEC cell (IO-ITO|PSII||IO-ITO|H 2 ase) generated 0.52 mmol O 2 (Z F = 104%) and 0.96 mmol H 2 (Z F = 98%). 120 Note that the system performance with respect to photocurrent, onset potential and longevity was limited by the IO-ITO|PSII photoanode. To reduce the bias voltage in the future, one could possibly extract electrons from the upstream electron relays such as Q A or pheophytin (Fig. 2). 104,130 An easier way is to substitute for an electron mediator with a more negative redox potential than DCBQ whilst also introducing a second light absorber to reduce the bias voltage. This has been achieved by the use of 3,5-di-tert-butyl-1,2-benzoquinone (DTBoQ, E m = 0.29 V vs. SHE) as the electron mediator, which shifted the photocurrent onset by 0.1 V earlier than DCBQ. When paired with a p-Si|IO-TiO 2 |H 2 ase photocathode, the required voltage was reduced to 0.24 V. At 0.4 V, the tandem PEC cell produced 0.125 mmol H 2 (Z F = 91%) within 3 h. 166 At the anodic side, a second light absorber can be judiciously selected to imitate the function of PSI or to complement the light absorption of PSII. A working example is DPP that has a negative excited-state redox potential (À1.15 V vs. SHE) and strong light absorption between 400 and 550 nm (l max = 496 nm). 147 DPP and PSII were loaded in an IO-TiO 2 scaffold as the light absorbers and electrically connected by an Os-based redox polymer (P Os ) hydrogel. 119 Thus, a vectorial electron transfer from water to the TiO 2 electrode was established to prime electrons with sufficient reducing power for HER. Moreover, DPP conferred supplementary green-light absorption on this PSII-containing photoanode, thereby enabling a panchromatic sensitivity to visible light. By directly wiring the IO-TiO 2 -DPP|P Os -PSII photoanode with an IO-ITO|H 2 ase cathode, autonomous solar water splitting can be carried out using wired enzymes without a bias voltage. 119 After 1 h of irradiation at zero bias, 0.015 mmol H 2 was detected with a Faraday efficiency of 76%. H 2 ases at the cathode were not driven to their full capacity in catalysing proton reduction, due to the photodegradation of PSII in vitro. An initial applied bias STH conversion efficiency of 0.14% was obtained at a voltage of 0.3 V during the first hour. 119 PSI, the native second light absorber in the Z-scheme, has also been rewired with PSII and H 2 ase by redox polymers in an attempt to reconstruct a biological H 2 production pathway in vitro. 173

Carbon dioxide reduction
From a viewpoint of carbon products, the biomass produced by the Calvin cycle is not ideal to substitute for petrochemicals as feedstocks for the chemical industry. 174 However, biology has evolved more than one pathway for CO 2 fixation, 175 and the reductive acetyl-coenzyme A (acetyl-coA) pathway (Wood-Ljungdahl pathway) is an energy-conserving CO 2 -fixing pathway, through which inorganic carbon is assimilated into cellular metabolism in the form of formate and CO by carbon monoxide dehydrogenase (CODH) and formate dehydrogenase (FDH). [176][177][178] Both [NiFe]-CODH and metal-dependent FDH can catalyse the interconversion between CO 2 and their two-electron reduced carbon products, CO and formate, respectively. They control the binding of CO 2 and the transfer of electrons and protons to partition intermediates on the reduction pathway towards CO or its hydrated form, formate. 46 [NiFe]-CODH is an O 2 -sensitive metalloenzyme with a [NiFe]active site from anaerobes such as Moorella thermoacetica and Carboxydothermus hydrogenoformans (Fig. 8k). It efficiently catalyses CO oxidation with a TOF up to 40 000 s À1 and CO 2 reduction with a TOF up to 45 s À1 , at a nearly thermodynamic potential (À0.5 V vs. SHE, pH 6.7) (Fig. 8l). 46,179,180 CO 2 is bound to the [NiFe]-cluster and undergoes a bifunctional attack by the electrophilic iron centre and the nucleophilic nickel centre, which enables a twoelectron pathway for CO 2 activation. 181,182 Such two-electron reduction is also applied by [Mo]-or [W]-FDH, a metalloenzyme reversibly interconverting CO 2 and formate with minimal overpotentials (À0.39 V vs. SHE, pH 7.0) (Fig. 8m and n). 183,184 It has been determined that [W]-FDH isolated from Syntrophobacter fumaroxidans exhibited a TOF up to 3400 s À1 for formate oxidation and 280 s À1 for CO 2 reduction (with MV as the redox partner). 183 Both [W]-and [Mo]-FDH can achieve B100% Faraday efficiency for electrocatalytic CO 2 reduction with formate as the sole product at modest overpotentials. 183,184 These enzymes provide an efficient two-electron CO 2 reduction pathway that bypasses the thermodynamically-uphill formation of CO 2 À . 185 In view of the close thermodynamic redox potentials for CO 2 /CO (À0.52 V vs. SHE, pH 7.0), proton reduction (À0.42 V vs. SHE, pH 7.0) and CO 2 /HCOO À (À0.39 V vs. SHE, pH 7.0) (Fig. 2), similar strategies can be employed to drive CO 2 reduction with solar energy.
[NiFe]-CODH was adsorbed on RuP-TiO 2 nanoparticles for light-driven CO 2 reduction to CO in the presence of an SED (Fig. 8b). During 4 h of irradiation, the RuP-TiO 2 -CODH hybrids produced B5 mmol CO, corresponding to an average TOF CODH of 0.15 s À1 . 186 The relatively low turnover rate compared with H 2 ase hybrids was partly due to a smaller driving force for CO 2 reduction than proton reduction, as the conduction band edge of TiO 2 (À0.52 V vs. SHE, pH 6.0) nears the thermodynamic potential for CO 2 reduction to CO. Factors such as interaction with TiO 2 and enzyme orientation also affect performance. The enzyme activity began to decrease after 3 h, and the low activity might be partly due to inefficient dye regeneration. 186,187 CODH has also been integrated in a PEC system comprising a P1 dye-sensitised p-type NiO cathode (Fig. 8f). 188 The excited P1 dye receives an electron from the valence band of the NiO electrode, followed by electron injection into an adjacent CODH, via intraprotein electron relay centres, down to its [NiFe]-active site, where CO 2 was reduced to CO (Fig. 8f).
Light-driven CO 2 -to-formate conversion has been previously realised by FDH-based photoredox systems, where electrons from SEDs are energised by photosensitisers and delivered to FDH via electron mediators, 78,[189][190][191] or by an FDH-based PEC system, where photoelectrons from the semiconductor photocathode p-InP were mediated to FDH by the MV/MV 2+ redox couple. 192 The diffusional mediator can be eliminated by immobilising FDH on RuP-TiO 2 nanoparticles (Fig. 8b). 193 FDH showed a TOF of 11 s À1 in the first 6 h of photocatalysis and yielded 2.6 mmol formate. 193 Quartz crystal microbalance and infrared spectroscopy revealed that TiO 2 binds FDH strongly through both electrostatic interaction and chemisorption. FDH can also be wired with an electrode to work in tandem with a photoanode. 122,194,195 A recent study coupled an FDH-functionalised IO-TiO 2 cathode with an IO-TiO 2 -DPP|P Os -PSII tandem photoanode to drive CO 2 -to-formate reaction with the aid of a small bias voltage. 122 After 1 h of irradiation with a bias of 0.3 V, 0.046 mmol of formate (Z F = 70%) was detected as the only product in the cathodic chamber of the PEC cell. The photocurrent decayed by half after 8 min, arising from the photodamage of PSII in the anode. Self-driven CO 2 reduction was recently achieved by a PV-PEC tandem system with a perovskite solar cell, a BiVO 4 -FeOOH photoanode and a TiN nanoshell|FDH cathode. 194 The resulting PV-PEC tandem cell yielded formate at a rate of 1.06 mmol h À1 with a Faraday efficiency of 83% in 8 h, corresponding to a solar-to-fuel efficiency of 0.08%. The formate production could last for three days, and the decrease in activity was ascribed to the degradation of the perovskite solar cell outside the PEC cell. FDH can also be loaded on a TiO 2 -deposited CuFeO 2 -CuO photocathode to couple with a BiVO 4 -FeOOH photoanode for autonomous light-driven CO 2 reduction. 196 This bias-free PEC cell allowed CO 2 -to-formate conversion at a rate of 0.098 mmol h À1 with a Faraday efficiency of 33.5% in 8 h.
The product of CO 2 reduction can be extended from formate to methanol via an enzyme cascade. [197][198][199] A prototypical PEC system employed CoP i -modified a-Fe 2 O 3 and BiFeO 3 as the photoanode and photocathode, respectively, and contained a mixture of enzymes, i.e., FDH, formaldehyde dehydrogenase and alcohol dehydrogenase, in the cathodic chamber. 198 Electrons extracted from water are energised by photoelectrodes and vectored through enzyme cascades to reduce CO 2 to methanol via NADH and diffusional electron mediators. Instead of being directly wired on the photocathode, enzymes in this system were dispersed in the catholyte, and depended on NADH, the reducing agent, to carry out redox reactions. As such, despite the well-aligned band positions, an external voltage (0.8 V) was needed to efficiently regenerate NADH and produce methanol (1.31 mM in 6 h). 198 An enzyme cascade was also immobilised in a silica matrix to work with an ''artificial thylakoid'' in a colloidal system. 199 The ''artificial thylakoid'' constituted a microporous protamine-TiO 2 hollow sphere (microcapsule) with CdS nanoparticles deposited on its luminal surface. The TiO 2 microcapsule received electrons from the excited CdS and exported electrons to regenerate NADH that furnished the enzyme cascade with reducing equivalents. In this way, photocatalytic oxidation and enzymatic CO 2 reduction were decoupled so that the enzymes were protected from photoinduced reactive oxygen species (ROS). This biohybrid system has demonstrated a high rate and yield in renewing NADH and thus generated 85 mM methanol from CO 2 in 2 h. 199

Chemical synthesis
With enzymes as biocatalysts, photogenerated electrons from light absorbers can realise a broad spectrum of reactions of synthetic interest, such as N 2 reduction, reduction of CQC bonds, formation of C-C bonds and hydroxylation of C-H bonds etc. Several recent studies are briefly highlighted here.
Biology converts N 2 into metabolically tractable NH 3 using N 2 ases under physiological conditions. N 2 ase is a two-component system comprising a  3 and one H 2 are generated from one N 2 and eight protons and electrons (Fig. 8o). 86 To leverage N 2 ases for light-driven N 2 fixation, the isolated [MoFe] proteins were adsorbed on a CdS nanorod to form a biohybrid complex, where light energy replaces ATP hydrolysis to drive the enzymatic turnover in N 2 . 200 The photoexcited CdS nanorods generated electrons with low reducing potential (À0.74 V vs. SHE, pH 7.0), which were then forwarded to the interfacing [MoFe] proteins for N 2 reduction. The CdS-[FeMo] protein hybrid attained an NH 3 production rate of 315 nmol mg N 2 ase À1 min À1 and an average TOF N 2 ase of 1.25 s À1 . The NH 3 production lasted up to 5 h with a TON of 1.1 Â 10 4 mol NH 3 mol N 2 ase

À1
. 200 The reduction of CQC bonds is exemplified by the conversion of fumarate to succinate, which can be catalysed by the flavoenzyme fumarate reductase (FccA) (Fig. 8p and q). 201 The light-driven fumarate reduction can be either performed in a colloidal system or within a PEC cell, using RuP-sensitised TiO 2 nanoparticles, and carbon dots or a W-BiVO 4 -CoP i electrode, respectively (Fig. 8b-d). 154,202 The FccA adsorbed on RuP-TiO 2 exhibited a TON of 5800 over 4 h of irradiation, corresponding to a TOF FccA of 0.4 s À1 , 202 while that on amine capped carbon dots achieved a similar TOF FccA of 0.47 s À1 (averaged in the first 2 h of irradiation) with improved photostability up to 24 h. 154 Photoinduced enzyme degradation was likely to account for the slowdown of productivity. In a PEC system with a W-BiVO 4 -CoP i photoanode, the turnover rate of the immobilised FccA dropped down to B0.01 s À1 . 202 Another example is the use of old yellow enzymes (NADPH dehydrogenases) for stereoselective CQC reduction. Light absorbers including Au-TiO 2 or N-doped carbon dots were employed to produce photoelectrons that drive the regeneration of reducing equivalents such as NAD(P)H and its synthetic substitute, for trans-hydrogenation of conjugated CQC bonds. 203,204 With molecular xanthene dyes, photoelectrons could be directly transferred to the prosthetic flavin moiety in the old yellow enzyme, eliminating the need of using costly nicotinamide cofactors. 205 Such light-driven enzymatic catalysis yielded enantioselective products (enantiomeric excess 4 90%) with conversion yields up to 80-90%. 204,205 Photochemical C-C bond formation has recently been realised in a CdS-2-oxoglutarate:ferredoxin oxidoreductase (OGOR) hybrid system. 206 The OGOR catalyses the amalgamation of CO 2 and succinate into 2-oxoglutarate and is part of the reductive tricarboxylic acid cycle responsible for CO 2 fixation in many autotrophic microorganisms. The irradiated CdS generates reducing equivalents to drive the catalytic turnover at OGOR which involves large substrates, significant conformational changes during catalysis, and eventual formation of C-C bonds. 206 Functionalisation of C-H bonds provides a straightforward and atom-economical access towards a plethora of organic products, 207 where enzymes such as cyt P450 show their synthetic advantages. The hallmark reaction of cyt P450 is the hydroxylation of C-H bonds, during which the delivery of electrons from the NADH reductase is synchronised with the activation of oxygen at the haem centre. The disadvantage of activity dependence on NADH cofactors and redox partners can be overcome by a semi-biological solution. A Ru(II)-diimine dye was covalently attached to a mutant haem domain of a cyt P450, which enabled excited dyes to inject electrons into the haem centre to catalyse the hydroxylation of lauric acid at an initial rate of 2.1 s À1 with a TON of more than 900 in 2 h, without the aid of NADH and reductase. 208

Microbial hybrid systems
Microbial cells include an entire system of biochemical pathways and can produce metabolites of synthetic interest with high specificity at physiological conditions. Microorganisms utilise numerous enzyme cascades to maintain the intracellular metabolism. Different metabolic pathways are spatially organised to divert metabolites or enzymes that can react promiscuously, to maintain selectivity, concentrate reactants to drive unfavourable reactions, and protect enzymes or unstable intermediates from harmful cytoplasmic contents. 209 Moreover, these biosynthetic pathways are under dynamic regulation to keep cellular functionality in tune with physiological needs at different conditions. These features allow microbes to synthesise complex products from the simplest and stable feedstocks (e.g., H 2 O, CO 2 , N 2 , etc.) and render microbial catalysis resilient to environmental variations. Appealing to synthetic chemistry is their ability to produce multicarbon products from CO 2 via inherent carbon assimilation pathways such as the Wood-Ljungdahl pathway and the Calvin cycle. 175,178 Such synthetic complexity is challenging to achieve with individual enzymes and a fixed reaction stoichiometry. Microbial catalysis can be further empowered by synthetic biology that employs genetic tools to access and engineer the intracellular metabolism, and create new pathways beyond the native metabolic pattern. 210 Compared with enzymatic counterparts, the microbial hybrid systems are less explored, but have provoked increasing interest in solar-to-chemical conversion, due to the possibility to synthesise complex chemicals, achieve better stability and prospects for scalability. 211,212 In the following discussions, we will summarise the recent achievements that take advantage of the microbial metabolism for chemical synthesis in solar energy conversion with photocatalysis and (photo)electrochemistry (Table 1).
Whereas enzymes catalyse biochemical reactions with specific substrates, products and stoichiometry, microorganisms are more versatile due to a set of diverse metabolic pathways involving a multitude of enzymes. Moreover, the microbial hybrid systems are being developed to establish an efficient conversion strategy of solar energy to desired multicarbon chemicals, a catalytic prospect particularly suitable for microorganisms. 213 Thus, we summarise progress for microbial hybrid systems from a different perspective: instead of organising this section by individual reactions enabled by such systems, we put the emphasis on key questions imperative for research in this field: how microorganism and synthetic materials are integrated into hybrid structures, how non-photosynthetic microbes are empowered with light-absorbing capabilities, and how fluxes of reducing equivalents are guided into intracellular metabolic pathways.

Microbial photocatalysis in colloidal systems
The discussion begins with the microbial hybrids operating in colloidal systems (Fig. 5c). We consider the HER as a model reaction in the context of solar-to-fuel conversion. Photosynthetic H 2 production can transiently take place in nature by microalgae and cyanobacteria under anaerobiosis, during which electron flux on the photosynthetic chain is shifted away from its normal path towards H 2 ase. 17,214 However, such process is O 2 -sensitive and competes with NADPH-dependent biological processes. Although the O 2 -sensitivity can be partially alleviated by protective encapsulation, 215,216 the rate and yield of H 2 evolution are intrinsically limited by the light reaction, which saturates at fairly low light intensities (10-20% of solar light). 214 Whereas these drawbacks can be partially overcome by streamlining the solar-to-H 2 pathway with isolated H 2 ases and light absorbers, laborious purification and fragility of the enzymes diminish the practicality of this approach.
An alternative are microbial hybrids where non-photosynthetic live cells are employed as biocatalysts, and the task of light harvesting is outsourced to synthetic light absorbers. Several early studies illustrate this strategy: H 2 ase-(and N 2 ase-) containing bacteria such as Clostridium butyricum, Rhodopseudomonas capsulata, Rhodospirillum rubrum were coupled with semiconductor nanoparticles, specifically, TiO 2 and Bi 2 O 3 via MV as the electron mediator (Fig. 9a). In the presence of SEDs, photogenerated electrons were delivered into cells where H 2 evolved. [217][218][219] The low growth rate of these bacteria limits the availability of biocatalysts. As such, Escherichia coli (E. coli), a well-established bacterial chassis for synthetic biology, 220 was engineered to heterologously express genes encoding [FeFe]-H 2 ases from Clostridium acetobutylicum. 221 The recombinant E. coli was mixed with TiO 2 nanoparticles and MV, and such hybrid system produced 216 mmol H 2 during 15 h of irradiation.
To enable visible light absorption, CdS was employed, and instead of inducing heterologous expression of [FeFe]-H 2 ases, E. coli can produce H 2 in anaerobic conditions with endogenously synthesised [NiFe]-H 2 ases. Moreover, CdS nanoparticles were precipitated on the bacterial surface to directly supply photoelectrons without diffusional mediators (Fig. 9b). 222 It has been found that photogenerated electrons interacted with the intracellular metabolism and guided more reducing equivalents along the pathway for H 2 production. To extend its aerobic uses, the CdS-E. coli hybrid was further encapsulated with biomimetic silica to protect the bacterium from oxygen. 216 Under aerobic conditions, the hybrid system continuously produced H 2 within 96 h of irradiation.
The catalytic ability of microorganisms is better exemplified in CO 2 reduction, where they can produce complex organic products not readily accessible through synthetic methods. 211 One example is CO 2 reduction to CH 4 , which involves eight electron transfers that can facilely diverge to form a wide range of products. 223 In the realm of biology, this reaction is handled by methanogenic archaea as part of their energy-conserving metabolism. 224 In light of this, Methanosarcina barkeri (M. barkeri), an anaerobic methanogen, was employed as the catalytic machinery for CO 2 -to-CH 4 conversion. 225 Like the aforementioned CdS-E. coli hybrids, 216,222 CdS nanoparticles were deposited on the surface of the M. barkeri as the light absorber. Photoelectrons produced by the excited CdS were delivered to the intracellular pathways via membrane-bound cytochromes or H 2 ases and were used by the bacteria for methanogenesis (Fig. 9b). The CdS-M. barkeri hybrid produced 13.7 mmol CH 4 after three-day irradiation with a quantum efficiency of 0.34%. 225 Isotopic labelling experiments confirmed that the CH 4 was derived from CO 2 reduction. The CH 4 evolution halted after three days, likely due to the depletion of the SED cysteine.
CdS nanoparticles can also be interfaced with Moorella thermoacetica (M. thermoacetica), an acetogenic and electrotrophic bacterium. M. thermoacetica could directly use photogenerated electrons from the excited CdS to reduce CO 2 into acetate via the intracellular Wood-Ljungdahl pathway, while holes produced by CdS were quenched by cysteine (Fig. 9b). 226 A peak quantum yield of 2.44% was recorded at low-intensity simulated sunlight, which is comparable to the year-long averages (B0.2-1.6%) for plants and algae. 227 The electron transfer pathways between CdS and M. thermoacetica was further elucidated by transient absorption spectroscopy: photoelectrons were transported via membrane-bound electron acceptors such as cytochromes, ferredoxin, flavoproteins and menaquinones at the initial stage (3 h) of photosynthesis; at longer duration (24 h), a membrane-bound H 2 ase-mediated pathway dominated, which correlated with increased rates of acetate generation and thus a higher quantum efficiency. 228 A latest proteomic study revealed that the photoexcited CdS induced changes to the metabolic pattern of M. thermoacetica: a host of enzymes associated with the tricarboxylic acid cycle and glycolysis were up-regulated. Both pathways together with the Wood-Ljungdahl pathway constitute the energy-conservation scheme of the CdS-M. thermoacetica hybrid system. 229 To make the hybrid system more sustainable, the added cysteine, the SED to CdS, could be regenerated photocatalytically by TiO 2 nanoparticles loaded with Mn(II) phthalocyanine co-catalysts, which ultimately extracted electrons from water and delivered them to the CdS-M. thermoacetica hybrids via the cysteine/cystine redox couple. 230 The light absorbers can be replaced with those that are more biocompatible and accessible, for example, Au nanoclusters. 231 Au nanoclusters could be ingested by M. thermoacetica and directly deliver photoexcited electrons with longer lifetime to the Wood-Ljungdahl pathway via cytoplasmic redox mediators (Fig. 9c). The intracellular interplay between photochemistry and metabolism bypassed kinetic and energetic drawbacks stemming from transmembrane diffusion. The Au nanoclusters also could reduce ROS formed during irradiation, thus improving the cellular viability in this hybrid system and increasing acetate production with a higher quantum yield of B2.9%. 231 A similar protocol is applicable to various microorganisms with different physiological functionality. For example, CdS nanoparticles were precipitated on the photosynthetic bacterium Rhodopseudomonas palustris for CO 2 fixation. 232 Under visible light, the excited CdS generated additional reducing equivalents for the innate Calvin cycle, which promoted the production of biomass and valuable multicarbon compounds such as carotenoids and poly-b-hydroxybutyrate (PHB), as well as the photosynthetic efficiency. When CdS was hybridised with Thiobacillus denitrificans (T. denitrificans), an autotrophic denitrifier, the resulting CdS-T. denitrificans hybrid could use the photoelectrons for NO 3 À reduction to N 2 O. 233 In light of its cytotoxicity, CdS has been substituted with organic semiconductors, i.e., perylene diimide derivative (PDI) and poly(fluorene-co-phenylene) (PFP) as the photosensitiser. 234 PDI and PFP could be immobilised on the surface of M. thermoacetica through electrostatic and hydrophobic interactions, which enabled efficient electron transport across the cellular membrane. Such microbial hybrid system registered a quantum efficiency of 1.6%.
Light-driven fumarate reduction to succinate has been achieved by an enzymatic hybrid system with flavoenzymes (FccA), 154,202 but such reaction can also be accomplished by a dye-sensitised whole cell containing flavocytochromes. 235 This hybrid system comprised of photosensitisers (Eosin Y or proflavine), Shewanella oneidensis MR-1 (S. oneidensis), an electron mediator (MV) and a SED. Another model reaction is C-H bond functionalisation. Cyt P450 was employed to catalyse the hydroxylation of lauric acid with a covalently attached Ru(II) light absorber that was meant to deliver photoelectrons to the haem domain without the need of redox partners. 208 The same strategy also applies in a microbial hybrid system: the cyt P450 was expressed in an engineered E. coli bacterial chassis and Eosin Y was absorbed into the cytoplasm and directly bound to the haem domain of the enzyme. 236 Under visible light irradiation and with an SED, photoexcited Eosin Y can continuously supplement electrons to sustain the catalytic turnover of the cyt P450 more than 18 h (average TOF P450 : 2.5 Â 10 À4 s À1 ). 236 A comparison with the previous P450-Ru(II) dye hybrids illustrates the disparity that epitomises general differences between enzymatic and microbial systems. First, the P450-Ru(II) dye hybrids need post-translational modification to link the Ru(II) dye molecule to the specific cysteine residue, and such modification is specific to a particular type of cyt P450 and may affect the activity when applied to different cyt P450 variants. 208 In contrast, the whole cell system has demonstrated success with both bacterial and human cyt P450 in producing high value chemicals such as drugs and steroids. 236 Lastly, the enzyme hybrids showed an initial TOF P450 (2.1 s À1 ) several orders of magnitude higher than the whole cell system, but the activity of the former began to languish within 2 h (average TOF P450 : 1.25 Â 10 À3 s À1 ), while productivity of the latter can be compensated by the prolonged longevity (418 h) and the increased quantity of biocatalysts. 208,236 Besides bacteria, eukaryotic cells such as yeasts are also favoured as a platform cell factory for biosynthesis of fuels and chemicals, as their well-studied biology primes genetic and analytical tools to unravel the mechanisms governing electron transport and metabolic flux in biohybrid systems. 64 The intracellular redox reactions are dependent on cofactors, namely NADH or NADPH, as hydride sources and their generation is strongly intertwined with the central carbon metabolism. 237 The expense of NAD(P)H precludes a stoichiometric supply to sustain the biosynthesis, and an effective cofactor regeneration was achieved photocatalytically by hybridising the yeast strain Saccharomyces cerevisiae (S. cerevisiae) with semiconducting InP nanoparticles (Fig. 9a). 238 S. cerevisiae was genetically engineered to suppress the cytosolic synthesis of NADPH and to enhance the carbon flux towards shikimic acid production. The NADPH was regenerated by the inward electron transport by irradiated InP nanoparticles bound on the cell surface. A maximum light-to-chemical conversion efficiency of 1.58% was attained within 12 h of irradiation. 238

Electrolysis coupled to homogeneous microbial synthesis
Another biohybrid system integrated planktonic M. barkeri with a PEC cell system consisting of a water oxidising TiO 2 photoanode and a p-InP-Pt photocathode, to couple artificial H 2 evolution with biological methanogenesis. The unassisted water splitting under irradiation generated H 2 in the cathodic chamber, which was consumed by the methanogens therein as reducing equivalents to reduce CO 2 to CH 4 (Fig. 9d). This hybrid system produced 68.8 nmol CH 4 over three days, corresponding to a Faraday efficiency of 74%. 239 Similar strategies can be extended to other strains of bacteria to obtain CO 2 -reducing products that are more feasible for storage and transport. 240,241 For example, in a water electrolyser inoculated with genetically-engineered Ralstonia eutropha, a water oxidising CoP i anode and a Co-P alloy cathode worked together at an applied voltage of 2.0 V to furnish the bacteria with H 2 . 241 The bacteria transformed CO 2 and H 2 following a desired route into an array of fuels and chemical products such as poly(3hydroxybutyrate), isopropanol, isobutanol and 3-methyl-1-butanol (Fig. 9d). The energy input can be readily outsourced to a photovoltaic device with a routinely attained efficiency of 18%, 2 and the resulting system could anticipate a solar-to-chemical conversion efficiency of 47%. 241,242 Such a modular platform has furthermore been leveraged beyond CO 2 reduction, towards more complex ambient NH 3 synthesis from N 2 and H 2 O with the H 2 -uptake diazotrophic microorganism Xanthobacter autotrophicus. 243 In the above hybrid systems, H 2 acted as the vector of reducing power to connect microbial metabolism with artificial electrolysis. However, the low solubility of H 2 (0.79 mM, 25 1C) and inefficient H 2 delivery from the electrode limit the rate of throughput. A recent study shows that this bottleneck can be alleviated by introducing perfluorocarbon nanoemulsions as biocompatible H 2 carriers to increase the H 2 availability around cells, accelerate the H 2 transfer and thereby improve the productivity of CO 2 reduction. 244

Photoelectrosynthesis with immobilised microorganisms
Instead of suspending microbes in solution, they can also be immobilised on electrodes to catalyse redox reactions driven by light or electricity (Fig. 10a). Such systems are referred to as microbial (photo)electrosynthesis. Underlying this process is the electron exchange ability of microorganisms with the interfacing electrode (Fig. 10b-d). 245 This can either occur directly between the electrode and microorganism or be mediated by H 2 and diffusional redox couples. The direct electron transfer for microbial electrosynthesis has the advantage of bypassing the handicap of low solubility of H 2 and the diffusional limitations of electron shuttles. Besides, it avoids the potential loss from intermittent H 2 generation and the potential toxicity of some mediators (e.g., MV). 246 From a practical perspective, the advantages over a planktonic system largely resemble the benefits of heterogeneous catalysis over homogeneous systems such as easy catalyst separation and recycling. One can readily and flexibly apply the microbial electrode for photosynthesis, either by pairing with a photoelectrode in a PEC cell or connecting with a solar cell for PV-driven electrolysis. Of more interest to fundamental studies is that the hybrid electrode system permits electrochemical toolsets to access the intracellular redox chemistry via the current exchange at the biointerface.
Oxygenic photosynthetic microorganisms such as cyanobacteria have been immobilised on IO-ITO electrodes with tailored porosity to perform semi-artificial photosynthesis. 128,129 The cyanobacterium encloses the entire photosynthetic apparatus within the cell, rendering themselves more robust than isolated PSII to carry out light-driven water oxidation. The cyanobacterial biofilm produced much lower photocurrent densities than those of PSII, but with increased longevity (45 days). 128 Compared with the IO-ITO|PSII electrodes, the IO-ITO|biofilm systems exhibited different photocurrent profiles, which is derived from different electron transfer mechanisms at the interface. The cyanobacterium likely relies on intracellular redox mediators that are released under irradiation to carry electrons towards the electrode surfaces. 247 There are only few reports to directly interface nonphotosynthetic microorganisms with photoelectrodes. 248,249 The difficulty is in part due to the large footprint of cells, which requires appropriate electrode geometry to achieve a high loading density while maintaining light penetration. A notable example is an array of Si nanowires that served both as a lightabsorbing semiconductor and a physical scaffold for the acetogenic bacterium, Sporomusa ovata (S. ovata) (Fig. 5d). 248 The bacteria were integrated with the Si nanowires by taking photoexcited electrons from the conduction band to sustain their metabolism. The resulting biohybrid electrode was assembled with a photoanode made of water oxidising TiO 2 nanowires for autonomous photoreduction of CO 2 to acetate. The PEC cell produced a stable photocurrent of 0.3 mA cm À2 for more than 120 h, yielding an acetate titre of 20 mM with a Faraday efficiency of 86%. The overall solar-to-chemical conversion efficiency registered 0.38% during 200 h operation. Acetate could further be activated into acetyl-coA, a common biochemical intermediate, to access a variety of biosynthetic fine chemicals. The downstream synthesis could be performed by genetically-engineered E. coli that transformed acetate into n-butanol, polyhydroxybutyrate polymer, and isoprenoid compounds. 248 Recently, the acetate production was accelerated as a result of a larger bacterial population and a better bacterium-nanowire interface, which was realised by improving the local pH environment inside the electrode. 250 As such, the solar-to-chemical conversion efficiency was increased to 3.6% over seven days when coupling to a photovoltaic device.
Few light-absorbing semiconductors have been optimised to fit the dimension of microorganisms. Integration of microbes into the electrode scaffold also poses challenges on the chemical stability, biocompatibility, conductivity and surface chemistry of the semiconducting materials, not least the impact on light transmission. A simpler strategy is to immobilise microorganisms on electrodes with tailored morphologies while outsourcing light harvesting to photoanodes or external PV devices. This approach decouples light absorption from microbial metabolism, and thereby suppresses unwanted parasitic reactions and eliminates possible oxidative stresses stemming from photoinduced ROS. The simplified design also grants more flexibility in the making of electrode architectures to enhance the loading capacity and engineer the electrode surface to improve the interaction. Therefore, more efforts converge on interfacing microbial cells with tailored electrodes and improving the interfacial electron transfer and mass transport.
Such exploitation can capitalise on decades of research in microbial fuel cells, 251 and be guided by recent understandings of the interaction between synthetic materials and microorganisms. 252,253 As these hybrid electrodes can readily operate with light absorbing electrodes or devices, to broaden the scope of discussion, this section will continue with a survey of recent progress in microbial electrosynthesis, where sessile bacteria exchange electrons with electrodes and carry out synthetic reactions.
An early study of microbial electrosynthesis employed graphite electrodes to supply Geobacter sulfurreducens (G. sulfurreducens) with electrons for fumarate reduction. 254 This is reminiscent of a previously described electrode-bacterium hybrid system, where H 2 acted as the electron vector (Fig. 10b). Yet this scenario has been ruled out because the electrode barely produced H 2 at the applied potential (À0.3 V vs. SHE, pH 6.8). Instead, G. sulfurreducens was able to directly receive electrons from the electrode (Fig. 10d). 255 The same reaction could also be accomplished by S. oneidensis cultured on a graphite electrode or in a three-dimensional reduced graphene oxide (RGO) scaffold. 256,257 An envisioned application for microbial electrosynthesis is bioremediation, namely, decontaminating the wastewater possibly powered by solar energy. 258 For example, under applied negative potentials, G. sulfurreducens could precipitate soluble U(VI) ions on electrodes, which can be easily collected. 259 Bacteria such as Geobacter lovleyi and Anaeromyxobacter dehalogenans with the ability to respire with chlorinated compounds can also be exploited to dechlorinate contaminants (e.g., tetrachloroethene, 2-chlorophenol) in this manner. 260,261 More recently, an IO-ITO electrode scaffold was employed to culture a large population of G. sulfurreducens (Fig. 10f). 262 The resulting biohybrid electrode catalyses the reduction of soluble fumarate and heterogenous graphene oxide (GO), with electrons from an external power source or an irradiated dye-sensitised (RuP-TiO 2 ) or water oxidising BiVO 4 photoanode. 262 The microbe-modified IO-ITO electrode enabled high current densities and also served as a platform underpinning a variety of analytical methods to decipher the electrical interplay at the biointerface. 263 However, Geobacter and Shewanella strains lack CO 2 -fixation pathways. Autotrophic microorganisms that can source energy from inorganic chemical reactions (chemolithotrophs) or light (photoautotrophs) for carbon assimilation are more promising in this regard, as they can adapt to use an electrode as the electron donor to generate reducing power. 264 Acetogens such as S. ovata, Clostridium ljungdahlii, and M. thermoacetica have been shown able to directly receive electrons from electrodes and reduce CO 2 into acetate with Faraday efficiencies of 480%. Moreover, the electron uptake and CO 2 fixation can also be decoupled using two bacteria strains (Acetobacterium woodii and Desulfobacterium corrodens), which can separately receive electrons from the electrode and reduce CO 2 , respectively, and communicate by means of interspecies H 2 transfer (Fig. 10e). 265 As microbial electrosynthesis systems are anticipated to operate together with PV devices, the ultimate solar-to-chemical efficiency will to a large extent depend on the biochemical pathway that dictates the energetics and stoichiometry of the substrateproduct conversion. Whereas a sulphur-oxidising phototroph Prosthecochloris aestaurii can sustain its photosynthesis with electrons supplied from a cathode, 266 its application for photoelectrosynthesis is largely unfeasible because the Calvin cycle is rather inefficient for solar fuel production and genetic tools to expand their synthetic diversity are not yet developed for some less studied microorganisms. Compared with methanogens of which CH 4 is the only attainable product, acetogens enable a wide repertoire of chemicals such as ethanol, isopropanol, n-butanol and 2,3-butanediol, which can be further extended by metabolic engineering. 267,268 The alliance of microbial metabolism with photocatalysis and (photo)electrochemistry is formed by virtue of the translocation of reducing power from abiotic materials to biotic entities. This can be achieved by interfacing microorganisms with light absorbers/electrodes, where electrons are either directly transferred to microorganisms or shuttled inward via in situ generated electron carriers (e.g. H 2 ). Alternatively, electrons can be delivered to planktonic cells by diffusive redox mediators. Sessile microorganisms on electrodes can manage a high Faraday efficiency up to 90%. Yet the slow electron uptake (typically o 1 mA cm À2 ), stemming from the kinetic disparity between the electrochemical and biochemical processes, which is less amenable to improvement by interface engineering, diminishes the economic feasibility of this approach. 269 Thus a technically favoured strategy may be to decouple the extracellular electrochemistry and intracellular biochemistry in a modular fashion, as exemplified by a recent microbial electrosynthesis setup that combined PV, electrolysis and fermentation modules. 212 CO 2 and H 2 O was electrochemically reduced into H 2 and CO in a PV-powered electrolyser at a current density of 50 mA cm À2 . The abiotically generated syngas was then fed into a microbial fermenter where a mixed culture of Clostridium autoethanogenum and Clostridium kluyveri worked together to produce butanol and hexanol at an energetic efficiency of 78% and an overall Faraday efficiency of B100%. 212 As such, the highest attainable solar-to-chemical efficiency of the integrated system could be projected at 15%, assuming that the existing PV modules deliver a routine energy conversion efficiency of 20%. 212

Future perspectives
Semi-artificial photosynthesis opens a wide spectrum for new solar-to-chemical conversion designs by rationally integrating biocatalysts in the form of proteins and whole cells with synthetic components, which traverses the fields of enzymology, microbiology, materials science, electrochemistry, and photocatalysis. The ultimate goal is to catalyse the transformation of H 2 O, CO 2 , and N 2 into fuels, chemicals and fertilisers via a sustainable path and synthesise value-added fine chemicals with high yield and selectivity with renewable energy sources. Here we briefly propose several promising opportunities for the future development.
At the biotic side, more enzymes or bacteria relevant to fuel/ chemical production can be added into the inventory. The methodologies established with model biocatalysts on bespoke electrodes and semiconductor particles can be extrapolated to new biocatalysts to create new reaction pathways. For example, RuP-TiO 2 particles were used in early work to adsorb H 2 ases for photocatalytic HER. 145 The use of the same light absorber was later extended to CODH, FDH and FccA. 186,193,202 CdS nanorods developed for CdS-H 2 ase complexes have been employed to drive N 2 -to-NH 3 conversion with N 2 ases. 149,200 Likewise, CdS nanoparticles have been in situ deposited on the surface of E. coli and M. thermoacetica. 222,226 It is reasonable to envision that a similar strategy can be applied to other acetogens such as S. ovata and methanogens like M. barkeri.
Biocatalysts are widely-appreciated for their high product selectivity, yet such catalytic prowess is confined to naturallyoccurring reactions. One way to expand the reaction scope is orchestrating various catalysts, including biotic and synthetic, to drive cascade reactions towards more complex and valuable products. 270 For example, an enzyme cascade has been introduced into a tandem PEC cell to reduce CO 2 into methanol. 198 More opportunities are emerging in this respect to judiciously construct reaction pathways and sensibly direct electron and carbon flux towards desired products.
The second approach arises when advancing genetic techniques meet with enzymology and microbiology. With the derived methodologies, enzymes can be engineered with improved promiscuity and stability. PSII has been introduced with a polyhistidine tag near the stromal side in order to enhance the linkage with Ni-terminated moieties on an electrode surface. 108 Such modifications and single site mutations more broadly are available for other enzymes to improve the interaction with light-absorbing materials. Pathways in microorganisms can be altered to maximise the chemical production or heterologously introduced to unlock unnatural reactions. Metabolic engineering can allow heterotrophic E. coli to synthesise sugar from CO 2 , 271 and cyanobacteria to drive the reduction of CQC bonds, 272 or produce valuable aromatic compounds directly from CO 2 . 273 Engineered microbes with additional pathways can work with light absorbers to yield synthetically useful products beyond their natural metabolites. 238 Moreover, the genetic toolkit can alter the existing or enable new physiological functionalities in model microorganisms. For example, current production of electrogenic microorganisms can be enhanced by increasing metabolic fluxes, promoting electron shuttle secretion, encouraging biofilm formation, etc. 274 Heterologous electron conduits have been transplanted into E. coli to confer extracellular electron transfer capability. 274,275 S. oneidensis has been engineered to metabolise glucose as carbon and energy source. 276 Beyond this, these techniques also provide powerful means to generate insights into the poorly understood questions with regard to microbial metabolism strategies and electron uptake pathways, which, in turn, will facilitate the rational design of next-generation biohybrid systems.
In parallel, progress at the artificial side can be translated into strengths to stimulate the advancement of semi-artificial photosynthesis systems. Developments in materials science and synthetic chemistry will provide more opportunities in tuning a material's composition, morphology and property at various levels to improve the interaction with biocatalysts or adapt to different reaction conditions. A recent report demonstrated the application of a metal-organic framework to confine FDH within its precisely defined volume and porosity to stabilise enzymatic activity at non-physiological conditions. 277 A metal-organic framework was also employed to encapsulate anaerobic bacteria (M. thermoacetica) to protect them against oxidative stress arising from irradiation. 215 Such a strategy is feasibly applicable to other enzymes and bacteria. Another emerging possibility is to foster interaction with biocatalysts at a sub-entity level, instead of forming heterogeneous biointerfaces. This has been realised by fusing dye molecules with cyt P450 to replace native redox partners, 208 or anchoring Pt nanoparticles with PSI for light-driven H 2 evolution. 278 At a subcellular scale, Au nanoclusters or molecular dyes have been exploited to intervene pathways within cells with light irradiation as bioorthogonal opportunities for intracellular modulation. 231,236,279 Furthermore, the intracellular metabolism can be coupled with extracellular redox transformations to manufacture functional materials such as polymers and inorganic particles. [280][281][282] These materials may provide additional benefits to promote extracellular electron transfer or cytoprotection. 281 Advanced photochemistry provides another thrust in future development. Many photosynthetic biohybrid systems are limited by light absorbers that display non-ideal absorption and driving force to perform reactions, or suffer from degradation faster than the inactivation of biocatalysts. Potential solutions may arise from the iterative cycle of design, synthesis and characterisation in the field of artificial photosynthesis, which yields a variety of photocatalysts with fine-tuned properties and morphologies. Progress in this field can instantly contribute to the construction and optimisation of biohybrid systems. For example, carbon nitride and carbon dots emerged as an attractive visible-light absorber due to their well-fit band structure, low cost, and robustness. 66,283 Carbon nitride-H 2 ase and carbon dot-H 2 ase complexes have been shown effective in photocatalytic H 2 production. [152][153][154] As such, more applications can be anticipated by replacing H 2 ase with other enzymes, or substituting the toxic CdS to interface with acetogenic bacteria. Studies into these newly-derived systems will gain more knowledge of the biotic-abiotic interplay and generate guidelines for the assembly of better systems.
Finally, semi-artificial photosynthesis, in a broader context, also enables a suite of powerful analytical techniques to delve into fundamental questions relevant to physiological functionality, biocatalytic redox chemistry and biotic-abiotic interaction (Fig. 11). 284 Wiring enzymes or cells to an electrode allows electrochemical methods to probe their redox chemistry and physiological functionality under turnover and non-turnover conditions, through efficient interfacial electron transfer (Fig. 11a). 59,82,285,286 The simplicity of the electrochemical apparatus (typically in a threeelectrode electrochemical cell) renders it readily compatible with an extra light source to investigate the light-driven redox reactions of photoactive enzymes (Fig. 11b). 90 Besides stationary electrodes, rotating electrodes, namely rotating disk electrodes or rotating ring disk electrodes, provide an effective way to eliminate the limitations in mass transport that often obscures catalytic features. 287 The rotating ring disk electrode, with an additional ring working electrode surrounding the central disk, enables quantification of reaction products formed at the disk electrode (Fig. 11c). 118 Complementary to electrochemical methods are spectroscopic techniques, which allow for in situ detection of electronic and vibrational changes with relevance to enzymatic or cellular functionalities under electrochemical control and provide structural or mechanistic insights that were otherwise intractable with electrochemistry alone. For instance, ATR-IR was previously deployed to track molecular catalysts and reaction intermediates during catalytic cycles, 288 while it also assisted to probe the enzyme-electrode interaction in H 2 ase-TiO 2 , FDH-TiO 2 and PSII-ITO systems (Fig. 11d). 102,166,193 Their versatility renders these tools readily transferable to biohybrid assemblies, and offers a high degree of flexibility to work together with (photo)electrochemistry in operando studies (Fig. 11e).
Time-resolved pump-probe spectroscopy, namely transient absorption spectroscopy, is universally applied in artificial photosynthesis systems to look into kinetic aspects of charge separation and transfer, which take place on a timescale (10 À12 -10 À3 s) intractable by other methods. 289 With such temporal resolution, electron transfer at a semiconductorenzyme interface and the ensuing catalytic turnover can be deconvoluted, enabling unique strength in unravelling kinetic intricacies in photosynthetic biohybrid systems. This technique has been exploited to track the charge transfer in CdS nanorod-H 2 ase hybrids and revealed that photogenerated electrons transferred from CdS to the distal relay centre in H 2 ase at a rate comparable to that of their relaxation process. 150 In a more complex microbial hybrid system (CdS-sensitised M. thermoacetica), transient absorption spectroscopic analysis provided kinetic insights into the electrical interplay between photoexcited semiconductors and acetogenic bacteria, and contributed to the elucidation of pathways connecting extracellular photochemistry and intracellular metabolism. 228 Beyond electrochemistry and spectroscopy, quartz crystal microbalance (with dissipation) enables a sensitive method to measure the mass change on a piezoelectric quartz chip (Fig. 11f). 76 This allows quantification of enzyme loadings in an electrode scaffold, which is not easily accessible by electrochemical or spectroscopic means and can be further extrapolated to evaluate the nature of enzyme interaction with electrode materials. 166,193 More tools are continuously being added into the toolbox. Recently, electron paramagnetic resonance spectroscopy has been combined with protein film electrochemistry to study metalloenzymes (Cu-Zn superoxide dismutase) adsorbed on a mesoporous ITO electrode under precise potential control. 290 Nanoscale secondary ion mass spectrometry was applied to study the metabolic activity of G. sulfurreducens biofilms. 291 These emerging analytic techniques can provide fresh insights from different perspectives to produce a more comprehensive picture of biohybrid systems.
Semi-artificial photosynthesis synergises functional components from different disciplines. Advances in each field will yield new opportunities to further the development and implementation of biohybrid systems. Conclusions drawn from these studies will deepen the current understanding of the biology-materials interplay and form concrete steps to expedite the evolution of semi-artificial photosynthesis towards maturity.

Conflicts of interest
There are no conflicts to declare.

Acknowledgements
This work was supported by a CSC-Cambridge PhD scholarship (to X. F.), a Marie Sklodowska-Curie fellowship (EMES, 744317 (c) A rotating ring disk electrode with a biocatalyst-adsorbed disk electrode and a Pt ring electrode. The rotation of the apparatus generates a convective flow that hydrodynamically carries reaction products from the disk to the ring electrode for product analysis. (d) An ATR-IR setup with an electrode scaffold deposited on an optical waveguide crystal (e.g., a Si prism). The internal reflection generates an evanescent wave that penetrates into the electrode scaffold by typically B500 nm. (e) In situ spectroelectrochemistry that combines vibrational/electronic spectroscopy with biofilm electrochemistry. Spectroscopic signatures with mechanistic significance can be captured during the course of redox variation with accurate electrochemical control. (f) A typical piezoelectric chip (electrode materials deposited on a Au substrate) for quartz crystal microbalance analysis. Through the mass-frequency relationship, the mass variation stemming from the biocatalyst binding or desorption can be monitored, enabling quantitative insights into bio-electrode interaction and biocatalyst loading capacity. to S. K.), and an ERC Consolidator Grant ''MatEnSAP'' (682833 to E. R.). The authors thank Dr Annika Eisenschmidt, Ms Esther Edwardes Moore, Prof. Julea Butt and Prof. Michael de Volder for the helpful feedback.