## Main

The reduction of CO2 coupled to water oxidation offers an attractive approach for the production of carbon-neutral fuels through a closed redox cycle1,2. However, the thermodynamically demanding and kinetically sluggish water oxidation reaction, as well as the lack of commercial value for the evolved O2 product, pose a challenge to the technoeconomic viability of this process3,4,5. An alternative oxidation pathway that requires a lower thermodynamic driving force may enable the simultaneous generation of fuels and cost-effective synthesis of valorized chemicals, and so provide a higher opportunity for profitability than the traditional reduction of CO2 coupled to water oxidation3,5,6,7. In particular, alcohols, which are readily obtainable from biomass8, can be (photo)electrochemically oxidized in aqueous conditions9,10. Upgrading biomass through oxidation provides greener routes for aldehyde and ketone synthesis, which are of importance in the fine chemicals industry6,11. To date, the chemical synthesis of value-added products coupled to CO2 reduction has mostly been studied in electrolysers, with an applied voltage bias being required to drive the reactions12,13,14, or in colloidal photocatalytic systems using organic solvents15,16.

Photoelectrochemical (PEC) cells offer the possibility to employ sunlight to power electrochemical reactions17. Oxidative organic transformations were recently explored in PEC cells, and typically employed heterogeneous narrow bandgap semiconductors (such as BiVO4) as the photoanode9,18,19. The reported photoanodes have been coupled to H2-evolving cathodes and typically require an additional electrical bias due to an insufficient driving force upon photoexcitation10,18. Similarly, an applied bias voltage was provided by an external photovoltaic module to a substrate oxidation photoanode and a CO2-reducing cathode20. Furthermore, these semiconductors often suffer from photocorrosion and competing water oxidation in aqueous media, and thus perform best in organic solvents, which potentially limits the environmental compatibility of this approach19,21,22. A molecular PEC-only cell capable of bias-free clean substrate oxidation coupled with CO2 reduction has thus far not been reported.

The lack of stability and tunability of heterogeneous materials can be addressed by separating the light absorption, and charge separation and transportation properties23. Consequently, a dye-sensitized TiO2 photoanode benefits from the fine-tuning of a molecular chromophore, as well as from the stability and high dielectric permittivity of the TiO2 semiconductor, which possesses a conduction band (CB) with just sufficient potential for CO2 reduction when using an ideal reversible electrocatalyst23,24. However, reports of dye-sensitized photoanodes for alcohol oxidation are scarce, and typically feature a low faradaic efficiency (FE, < 50%) and the use of precious metals25,26,27. In addition, most synthetic catalysts for CO2 reduction require a substantial overpotential and cannot operate with the driving force provided by the CB electrons of TiO2.

In this study, we report a single-light absorber PEC cell for bias-free simultaneous substrate oxidation and CO2 reduction (Fig. 1a). The precious-metal-free cell comprises an organic dye- and catalyst-sensitized TiO2 photoanode for alcohol oxidation, and a biohybrid cathode for CO2 reduction. Two dyes based on a diketopyrrolopyrrole (DPP) core, which was previously explored for PEC solar fuel production28,29, were synthesized, and feature two distinct anchoring groups (carboxylic and phosphonic acid, DPP-CA and DPP-PA, respectively; Fig. 1b). The dyes were co-immobilized with an organic nitroxyl radical alcohol oxidation catalyst, which was functionalized with a silatrane anchor (STEMPO; TEMPO, 2,2,6,6-tetramethylpiperidine-N-oxyl; Fig. 1b), on a mesoporous TiO2 (mTiO2) scaffold6,10,14. Comparison with a ruthenium-based chromophore, RuP ([Ru(bpy)2((4,4′-(OH)2PO)2bpy)]Br2 (4,4′-(OH)2PO)2-bpy, 2,2′-bipyridine-4,4′-diyldiphosphonic acid); Fig. 1b), typically used as a benchmark in dye-sensitized photoanodes and colloids25,30, showcases the advantages of organic sensitizers. The resulting photoanode was subsequently wired to a semi-artificial CO2 reduction cathode with a W-containing formate dehydrogenase (FDH) enzyme on a mesoporous indium tin oxide (mITO) electrode capable of direct electron transfer (Fig. 1a). mITO provides a biocompatible electrode material for the wiring of fuel-forming enzymes31,32, whilst FDH serves as an ideal bioelectrocatalyst for CO2 reduction owing to its high activity and selectivity for formate production close to the thermodynamic potential32,33. Visible light irradiation of the two-electrode PEC cell photoinduces an electron transfer cascade (Fig. 1a), which starts from the excited dye, through the mTiO2 to the mITO and thence to the FDH, which ultimately reacts with CO2. Simultaneously, hole transfer from the oxidized dye to the neighboring STEMPO triggers alcohol oxidation. Our two-electrode PEC cell is capable of sustaining photocurrents over a 6 h experiment, and results in quantifiable formate and aldehyde (4-methylbenzaldehyde (4-MBAd) or 2,5-diformylfuran (DFF) from 4-methylbenzylalcohol (4-MBA) or 5-(hydroxymethyl)furfural (HMF), respectively) under bias-free conditions.

## Results and discussion

### Dye synthesis and characterization

Synthesis of both organic chromophores (Fig. 2a) started with the reaction of DPPBr34 with phenylboronic acid to afford DPPPh in a 77% yield. Condensation in the presence of piperidine with either cyanoacetic acid or diethyl cyanomethylphosphonate yielded DPP-CA (68%) or DPP-PAEt (75%), respectively. Deprotection of the phosphonate ester was achieved by reflux in chloroform solution in the presence of bromotrimethylsilane (TMSBr), followed by hydrolysis with methanol (MeOH) to afford DPP-PA in an 80% yield. The composition and purity of the dyes were confirmed by 1H, 13C and 31P NMR spectroscopy (Supplementary Figs. 16), high-resolution mass spectrometry, infrared spectroscopy and elemental analysis.

The ultraviolet–visible (UV–vis) absorption spectra of DPP-CA and DPP-PA were recorded in N,N-dimethylformamide (DMF) solution (Fig. 2b), and both chromophores featured a maximum absorption band at λmax = 498 nm (maximum molar absorption coefficient εmax = 25.7 mM−1 cm−1) and at λmax = 502 nm (εmax = 14.1 mM−1 cm−1), respectively. RuP features a weaker light absorption (εmax = 11.0 mM−1 cm−1) centered at shorter wavelengths (λmax = 457 nm). Fluorescence spectra recorded in diluted DMF solution allowed us to calculate a zero–zero energy transition (E0−0) of ~2.25 eV for both organic dyes (Supplementary Fig. 7). The strong coloration of a mTiO2 film (anatase nanoparticles, ~20 nm diameter, ~6 µm thick) on a glass substrate after immersion in a solution of the dyes, followed by a washing step, confirmed the binding of the molecules to the surface (Supplementary Fig. 8).

To gain insight into the electronic structures of DPP-CA and DPP-PA, geometry optimizations were performed by employing density functional theory (DFT) at the PBE0/6-311G(d,p) level (Supplementary Tables 1 and 2). Subsequent time-dependent DFT (TD-DFT) calculations at the CAM-B3LYP/6311+G(2d,p) level were conducted34, with the calculated UV–vis absorption spectra from TD-DFT calculations matching those experimentally recorded for DPP-CA and DPP-PA (Supplementary Fig. 9). For both chromophores, the peak at ~500 nm corresponds mainly to a highest occupied molecular orbital to lowest unoccupied molecular orbital transition (Supplementary Table 3), which originates in the DPP core and extends to the anchoring moiety, suggesting a charge-density directionality in the excited state (Fig. 2c and Supplementary Fig. 10).

The electrochemical properties of DPP-CA and DPP-PA were studied via cyclic voltammetry in DMF solution, and chemisorbed on a mITO electrode (particle size < 50 nm, film thickness ~3 µm)29,35,36 in acetonitrile (MeCN), with tetrabutylammonium tetrafluoroborate as a supporting electrolyte (Supplementary Figs. 11 and 12 and Supplementary Table 4). Combining the dye (D) oxidation potentials (E(D+/D)) and E0-0 values enabled estimations of the oxidation potential in the excited state (E(D+/D*)) for DPP-CA and DPP-PA at −0.95 and −0.90 V, respectively, versus the normal hydrogen electrode (NHE). These values are more negative than the CB of TiO2 (ECB(TiO2) = −0.57 V versus NHE at pH 7) (ref. 24), which allows exergonic electron injection. Also, the E(D+/D) of immobilized DPP-CA (+1.33 V versus NHE) and DPP-PA (+1.28 V versus NHE) is more positive than the onset for alcohol oxidation by STEMPO (Eonset = +0.75 V versus NHE at pH 7) (ref. 14), which allows for hole transfer to the catalyst. The reported values for RuP (E(D+/D*) = −0.78 V versus NHE and E(D+/D) = +1.37 V versus NHE)37 are also favorable for the desired charge-transfer events and confirm that all the dyes met the thermodynamic requirements for their incorporation in an alcohol-oxidizing photoanode.

### Photoanode assembly

The photoanodes were prepared by optimized sequential immobilization of the catalyst, as the kinetic bottleneck29,38,39, followed by saturation of the remaining exposed surface by the chromophore to provide high light absorption. First, immobilization of the catalyst was achieved by immersion of a mTiO2 film (anatase nanoparticles, ~20 nm diameter, ~6 µm thick) in a solution of STEMPO (see Methods for details), which resulted in the hydrolysis of the silatrane moiety and the formation of strong siloxane bonds with the TiO2 surface (ref. 14). The resulting mTiO2|STEMPO electrodes were then immersed in a solution of the dye in dichloromethane (DCM) (0.2 mM DPP-CA or DPP-PA) or MeOH (0.2 mM RuP) to afford the photoelectrodes denoted as mTiO2|STEMPO/dye (see Methods for details).

We carried out photoanode characterizations via PEC experiments at room temperature in a one-compartment three-electrode cell using a Pt counter electrode, a Ag/AgCl/KClsat reference electrode, and a functionalized mTiO2 film on fluorine-doped tin oxide (FTO)-coated glass as the working photoelectrode. Linear sweep voltammetry (LSV) and controlled potential photoelectrolysis (CPPE) experiments were performed under chopped light irradiation. Ultraviolet-filtered simulated solar light (100 mW cm−2, AM 1.5G, λ > 420 nm) was used for all the PEC measurements, thus avoiding direct excitation of the TiO2 semiconductor.

Different aqueous electrolytes with a range of buffers and pH values were first examined in the presence of 4-MBA as a model oxidation substrate using DPP-CA (Supplementary Fig. 13), DPP-PA (Supplementary Fig. 14) or RuP (Supplementary Fig. 15) as photoabsorbers. CPPE experiments were performed for 6 h at an applied potential (Eapp) of +0.4 V versus the reversible hydrogen electrode (RHE), a sufficiently positive potential for the efficient extraction of charges from the CB of TiO2 (see above). We monitored the oxidation of 4-MBA to 4-MBAd by high-performance liquid chromatography (HPLC) at the end of the CPPE experiments, and all the electrolyte solutions enabled high FEs of 80−100% and 60−90% for 4-MBAd formation when employing DPP-CA (Supplementary Table 5) and DPP-PA (Supplementary Table 6), respectively. No overoxidation product (that is, 4-methylbenzoic acid) was observed, in line with the preference of the TEMPO catalyst for alcohol substrates10. Selection of the optimized electrolyte was performed by comparison of the PEC performance and evaluation of the dye stability, observed by UV–vis spectroscopy of the assembled photoanodes at the end of the CPPE experiment (DPP-CA, Supplementary Fig. 16; DPP-PA, Supplementary Fig. 17; RuP, Supplementary Fig. 18).

Comparison of the dye performances during CPPE under their optimized conditions, that is, sodium borate (Na2B4O7) buffer at pH 8.0 for DPP-CA and DPP-PA and sodium acetate (NaOAc) buffer at pH 5.6 for RuP (Fig. 3a), revealed a higher activity (J = ~90 µA cm‒2) of the organic dyes compared with that of RuP (J = ~9 µA cm‒2, Supplementary Table 7). This can be partially explained by the superior light harvesting efficiency of the organic dye-based electrodes over RuP (Supplementary Fig. 19). We focused on DPP-CA for further experiments due to its higher photocurrent activity and FE compared with those of DPP-PA, its maximum light harvesting efficiency, which ensured non-limiting light absorption, and its versatility in different electrolytes and conditions.

The solvent for DPP-CA immobilization was investigated with a higher loading (evaluated by UV–vis absorption) and catalytic activity obtained with DCM (Supplementary Fig. 20 and Supplementary Table 8) compared with that of DMF and MeOH. We then quantified the initial chromophore loading for mTiO2|DPP-CA and mTiO2|STEMPO/DPP-CA electrodes via UV–vis absorption (see Methods for details), which revealed loadings of 117 ± 16 and 54 ± 5 nmol cm−2, respectively. The catalyst loading on mTiO2|STEMPO electrodes was calculated as 176 ± 26 nmol cm−2 by inductively coupled plasma optical emission spectrometry (see Methods for details), and resulted in a catalyst/dye ratio of 3.2. The unusually high catalyst loading compared with those of molecular catalysts on metal oxides (50–100 nmol cm−2) (refs. 35,40,41) can be attributed to the polymerization of hydrolytically unstable silatrane moieties on the surface, which may result in not all catalysts being electrochemically active14, and thus provide a conservative value when calculating turnover numbers (TONs; see Methods for details).

### Performance evaluation

The LSV of the optimized mTiO2|STEMPO/DPP-CA electrode (Supplementary Fig. 21) displayed photocurrents with an early onset potenial close to the CB of TiO2 (ECB(TiO2) = −0.16 V versus RHE)24. A TONSTEMPO of 54 and a TONDPP-CA of 351 were achieved by the fully assembled mTiO2|STEMPO/DPP-CA photoanode over a 6 h CPPE experiment at +0.4 V versus RHE, with a FE of 94%, which confirms the catalytic nature of the photoanode with respect to both molecular components (Fig. 3b and Supplementary Table 9).

We further conducted control experiments to confirm that the substrate oxidation originated from both STEMPO and DPP-CA (Fig. 3b and Supplementary Table 9). Dye-free mTiO2|STEMPO electrodes led to a negligible photocurrent response, whereas STEMPO-free mTiO2|DPP-CA electrodes displayed an initial photocurrent of 36 µA cm−2, which rapidly decayed during the CPPE experiment due to the absence of an efficient catalytic turnover, which resulted in chromophore decomposition. Immobilization of TEMPO-free 3‐aminopropylsilatrane (Sil), followed by immobilization of DPP-CA resulted in a lower dye loading compared with that of mTiO2|DPP-CA electrodes (Supplementary Fig. 22), as well as low, rapidly decreasing photocurrents during CPPE. This confirms that the high activity and stability of the fully assembled system stems from the incorporation of the TEMPO moiety, and hence rules out alternative catalytic activity sources, such as a lower DPP-CA aggregation and the alteration of the TiO2 semiconductor by the silatrane moiety42,43. Employing the mTiO2|DPP-CA system in the presence of solubilized TEMPO (1 mM) in the electrolyte led to 4-MBAd production with high FE (99%). However, limited photocurrents of ~20 µA cm−2 with a TONTEMPO < 1 were reached, in line with previous reports on TEMPO-mediated dye-sensitized systems44, which highlights the benefits of immobilized molecular catalysts on heterogeneous scaffolds14,45.

Finally, we recorded the incident photon-to-current efficiency (see Supplementary Methods for details) at +0.4 V versus RHE for the fully assembled system (Supplementary Fig. 23), which reached a value of 2% at 425 nm and reflected the light-harvesting efficiency spectrum of the corresponding mTiO2|STEMPO/DPP-CA electrodes.

The electrodes retained 50% of their activity after 12 h of irradiation during CPPE, with a sustained drop in activity up to 22 h (Fig. 3c). This compares favorably with organic chromophore-driven photoanodes, which have commonly only reached a stability of 1 h (refs. 26,46). At the end of the experiment, the photoanode produced 23 ± 2 µmol cm−2 of 4-MBAd, with a high FE of (87 ± 4)%, an absolute yield for the conversion of 4-MBA to 4-MBAd of (6.6 ± 0.3)% (Supplementary Table 10) and TONSTEMPO and TONDPP-CA values of 131 ± 22 and 853 ± 107, respectively. The maximum TONs obtained are substantially higher than those reported for organic chromophores coupled with Ru-based catalysts on water-oxidizing photoanodes, which are consistently below 30 (refs. 47,48). Attempts to revive the catalytic activity by immobilizing fresh DPP-CA on an electrode after the CPPE experiment did not result in notable photocurrents, which suggests STEMPO decomposition as the main cause of the activity loss (Supplementary Fig. 24)14.

### Substrate scope

We replaced the 4-MBA substrate with HMF, a compound of interest as it can be sourced from biomass and its oxidation products are of value as industrially relevant building blocks8. Photocurrents of 70 µA cm−2 were achieved in Na2B4O7 buffer (0.1 M, pH 8.0) under the previously optimized conditions (Fig. 3d). We also tested NaHCO3 buffer (0.05 M, pH 6.4) to investigate the compatibility with a FDH-based cathode (see below), which resulted in currents of 40 µA cm−2. The higher activity in the pH 8.0 solution is consistent with a higher TEMPO activity at more alkaline pH values, which resembles the conditions typically used in the literature for HMF’s electrochemical oxidation6,10. HMF oxidation proceeds through an initial alcohol oxidation, which can be followed by further oxidation reactions to form a wide range of products (Supplementary Fig. 25)10,49. HPLC analysis revealed a selectivity for DFF, a high-value monomer employed in the polymer industry8, with no overoxidation products being detected. This may be a consequence of the low overall conversion yield (Supplementary Table 11) and of the experimental conditions, which can impact product selectivity50,51. DFF was produced with a FE of 86−90% under both electrolyte conditions (Supplementary Table 11).

### Photoelectrochemical CO2 reduction and chemical synthesis

To construct the CO2-reducing cathode, we assembled a mITO scaffold (particle size 50 nm, film thickness ~8.5 µm), followed by immobilization of FDH by drop casting the enzyme onto the electrode surface (274 pmol cm−2) (refs. 32,52). We selected a W-containing FDH from Desulfovibrio vulgaris Hildenborough for its high catalytic activity and ability to reduce CO2 with a marginal overpotential on mITO electrodes32.

We evaluated individually the mITO|FDH cathode and mTiO2|STEMPO/DPP-CA photoanode in a three-electrode configuration in a CO2-saturated NaHCO3 buffered solution (0.05 M, pH 6.4), with 4-MBA present for the photoanode experiment. The voltametric responses on the LSV of both systems revealed current matching at about +0.06 V versus RHE (Fig. 4a). CPPE of the assembled photoanode and cathode at +0.06 V versus RHE showed sustained (photo)currents over the course of 6 h, which resulted in 4-MBAd and formate production with a FE of 113% and 90%, respectively (Supplementary Fig. 26 and Supplementary Table 12). The excessively high FE from the photoanode at this applied potential is attributed to an inefficient electron extraction from the CB of TiO2 to the external circuit, which resulted in unproductive leakage current pathways and, overall, an underestimation of the counted charges53. This was confirmed by applying a more positive potential during the CPPEs, which resulted in FEs ~100%, whereas bringing the applied potential closer to ECB(TiO2) delivered abnormally high FEs (Supplementary Fig. 27 and Supplementary Table 13).

We then constructed a semi-artificial PEC cell by connecting the two electrodes in a two-compartment cell, separated by a Nafion membrane, with the 4-MBA substrate only in the photoanode compartment. Irradiation of the photoanode led to a net photocurrent at zero applied voltage (Uapp = 0 V) (Fig. 4b) during the LSV measurement. CPPE experiments under these bias-free conditions resulted in a stable photocurrent of 30 µA cm−2 over the course of 6 h of irradiation (Fig. 4c). Analysis of the liquid products revealed the production of 3.17 ± 0.31µmol cm−2 and 2.16 ± 0.26 µmol cm−2 of 4-MBAd and formate, formed with FEs of (108 ± 18)% and (74 ± 17)%, respectively (Supplementary Table 14). The high FE for the 4-MBAd synthesis is attributed to the inefficient electron extraction explained above, which, alongside inaccuracies in the formate quantification due to the detection limit of the apparatus, accounts for the observed higher ratio of 4-MBAd to formate. A conservative calculation of the TON values, in which all the drop-cast enzymes and immobilized STEMPO molecules are assumed to be electrochemically active, leads to a TONFDH of 10,372 ± 1,272 and a TONSTEMPO of 18 ± 3, which highlights the high efficiency of biological systems in comparison with that of synthetic catalysts.

We performed control experiments by omitting the molecular components in both compartments (Supplementary Fig. 28) in an otherwise complete mTiO2|STEMPO/DPP-CA||mITO|FDH PEC assembly. Employing a bare mTiO2 electrode as the photoanode led to a negligible photocurrent, whereas the use of a bare mITO cathode resulted in lower photocurrents than those of the fully assembled cell, with no formate production (Supplementary Table 15). This unproductive photocurrent is attributed to electron accumulation in the metal oxide cathode and proton reduction, and has previously been observed for analogous systems54. Although sampling of the headspace revealed no H2 present, the low amounts of gas produced would have largely remained dissolved in the liquid phase55.

We investigated the two-electrode PEC cell for HMF oxidation in a NaHCO3 buffer (0.2 M, pH 7). As explored above (Fig. 3d), the use of more alkaline conditions would have benefitted the activity of the photoanode. However, the lower activity of FDH at much higher pH values and ionic strength33, as a result of the increase of bicarbonate concentration at higher pH values1, required a compromise when selecting the electrolyte. Separate three-electrode experiments for the individual electrodes revealed that both assemblies were able to sustain stable (photo)currents at the potential at which a current matching was observed (Supplementary Fig. 29 and Supplementary Table 16). Thus, a bias-free two-electrode cell was capable of delivering a photocurrent of 10 µA cm−2 over the course of 6 h of irradiation (Supplementary Fig. 30), which resulted in TONSTEMPO and TONFDH values of 9 ± 2 and 4,532 ± 945, respectively (Supplementary Table 17). Based on the bias-free photocurrents of the PEC cell, the solar-to-fuel efficiency is expected to be lower than that of the state-of-the art systems for classical CO2 and water splitting56,57,58. Nonetheless, the reported PEC system provides a proof-of-concept demonstration for the development of future PEC devices for the simultaneous chemical synthesis and CO2 reduction that employs a (bio)molecular hybrid approach based on dye-sensitized photoelectrodes. Further improvement of the coupled system through the incorporation of a bipolar membrane, which allows the conditions in each compartment to be optimized independently59, would increase the versatility and activity of this cell configuration. Advances in protein engineering and its local environment may further prolong the stability, which remains one of the biggest challenges in systems that employ isolated bioelectrocatalysts60,61. Alternatively, the replacement of the enzymatic catalyst by microorganisms or more robust synthetic catalysts62,63, and developments in dye-sensitized solar cell technology, may further enable the improvement of the performance of such a paired PEC platform.

To date, coupling of CO2-reducing FDH-based (photo)cathodes to water-oxidizing photoanodes has either required an applied bias20,54,64,65,66 or displayed lower reaction rates for CO2 reduction products than those reported in this study (Supplementary Tables 14 and 17)67,68,69. A PEC cell that requires two photoabsorbers and consists of a BiVO4 photoanode, and a TiO2-coated CuFeO2 and CuO mixed photoelectrode functionalized with FDH, resulted in currents of up to 150 µA cm−2, albeit with a modest FE of 34% for formate production under bias-free conditions67.

Employing solely a single light absorber requires visible light absorption and a sufficient thermodynamic driving force to transport electrons to the enzymatic catalyst. A BiVO4 photoanode for water oxidation was coupled to a dark FDH-based cathode, but the low driving force provided by the BiVO4 photoanode resulted in minimal photocurrents of 0.1 µA cm−2 under bias-free conditions68. Although a bare TiO2 photoanode ensures a sufficient driving force for CO2 reduction with FDH, the lack of visible-light-absorption capabilities and use of electron mediators results in a system with a low activity and FE (15%)69. The reported rate of formate production (Supplementary Tables 14 and 17) of our device also compares favorably with that of a previously reported PEC cell coupled to a photovoltaic module20, whilst employing a single light absorber and producing a sole oxidation product.

Thus, the activity and efficiency reported in this study are a direct result of the optimized molecular engineering of the photoanode and the careful selection of the cathode material and biocatalyst. Utilizing a dye-sensitized TiO2 photoanode allows for intense visible light absorption by the organic chromophore and sufficient driving force for CO2 reduction provided by the CB of the TiO2 semiconductor. The mITO cathode allows the electrons to be transferred efficiently from the hybrid photoanode to the adsorbed FDH enzyme through direct electron transfer. Overall, the fully assembled system showcases simultaneous CO2 reduction and substrate oxidation in a PEC-driven cell. The cell is capable of utilizing a single light absorber, without the need for an external bias and the use of precious metals, to achieve good photocurrents, FE, and stability.

## Conclusions

We report an unbiased PEC cell capable of simultaneous CO2 reduction and chemical synthesis. We first constructed and characterized a dye-sensitized TiO2 photoanode with a co-immobilized organic molecular chromophore and catalyst. We synthesized a series of organic chromophores (DPP dyes), which were then coupled to a silatrane-functionalized TEMPO catalyst for alcohol oxidation in aqueous conditions. The system delivered up to 100 µA cm−2 for the oxidation of 4-methylbenzyl alcohol to the corresponding aldehyde in sodium borate buffer (pH 8.0) over the course of 6 h without substantial chromophore degradation. Both organic chromophores outperformed a ruthenium-based photoabsorber and allowed for the construction of a precious-metal-free dye-sensitized photoanode for chemical synthesis in aqueous conditions. The electrode delivered photocurrents over the course of up to 22 h of irradiation, which resulted in TONSTEMPO and TONdye values of 131 and 853, respectively. We then wired the photoanode to a formate dehydrogenase-modified cathode in a two-electrode fashion, which resulted in unbiased CO2 reduction and 4-MBA oxidation over the course of 6 h irradiation, while maintaining high FE for both products. Expansion of the substrate scope to selective 5-(hydroxymethyl)furfural oxidation was achieved and resulted in photocurrents of 10 µA cm−2 with simultaneous CO2 reduction. This study shows that, through careful selection of the molecular components and PEC design, single light absorbers can provide sufficient driving force for CO2 reduction and chemical synthesis without the need for applied potentials and precious-metal components.

## Methods

### Materials

Chemicals for the syntheses and analytical measurements were of the highest available purity unless otherwise noted: 2,5-diformylfuran (97%, Sigma Aldrich), 4-methylbenzyl alcohol (99%, Alfa Aesar), 4-methylbenzyldehyde (97%, Sigma Aldrich), 5-(hydroxymethyl)furfural (98%, ACROS Organics), absolute ethanol (VWR Chemicals), dl-dithiotreithol (DTT, BioXtra >99.5%, Sigma Aldrich), Tris-HCl (tris(hydroxymethyl)aminomethane hydrochloride >99.0%, Sigma Aldrich), formate solution (1 g l−1, ≥99.0%, Sigma Aldrich), glacial acetic acid (Fisher Chemical). MeCN, DCM, MeOH, piperidine, tetrahydrofuran and triethylamine were distilled on calcium hydride (triethylamine, MeOH and DCM), potassium hydroxide (piperidine) or sodium (tetrahydrofuran) before use. Dry 1,4-dioxane (99.8%) and dry DMF (99.8%) were purchased from Sigma Aldrich.

Column chromatography was carried out with silica gel 60 (0.040–0.063 mm mesh) from Material Harvest. FTO-coated glass sheets (SnO2/F, 7 Ω sq−1 sheet resistance, 300 × 300 × 2 mm) and ITO nanopowder (diameter <50 nm; Brunauer–Emmett–Teller area = 27 m2 g–1; 90% In2O3, 10% SnO2) were obtained from Sigma Aldrich. The TiO2 paste (15–20 nm, Ti-nanoxide T/SP, 100% anatase) was purchased from Solaronix. Sodium hydrogen carbonate (>99.998% trace metal basis) was purchased from Puratronic. Ti foil (0.25 mm thick, 99.5% purity) was purchased from Alfa Aesar. All the aqueous solutions were prepared with ultrapure water (deionized water; Milli-Q, 18.2 MΩ cm).

### Methods for computational calculations

Computations were conducted using Gaussian 0970, with DFT calculations performed for the ground state and TD-DFT for the excited state. The calculations were adapted from literature protocols successfully employed for DPP chromophores71. First, a rough optimization of the ground-state geometry was performed at the B3LYP/STO3G level with an extra quadratic convergence threshold72. The ground-state geometry was further optimized at the PBE0/6-311G(d,p) level73 via a force-minimization process using a self-consistent field convergence threshold (10−10 a.u.), which included modelling of bulk solvent effects through the polarizable continuum model with the permittivity set to DMF74. Subsequently, frequency calculations were performed to confirm that the optimized geometry corresponded to an energy minimum. Finally, the first 25 low-lying excited states were determined using TD-DFT calculations and the CAM-B3LYP/6311+G(2d,p) level of theory75, with a tight self-consistent field convergence threshold (10−7 a.u.). All the plots of computed species were produced with ChemCraft (graphical software for visualization of quantum chemistry computations (https://www.chemcraftprog.com)). Protons were omitted for clarity. The coordinates of all the relevant species are given in Supplementary Tables 1 and 2.

### Electrode preparation

#### Preparation of mITO films for chromophore characterization

Electrodes were prepared as previously described29,35. In brief, a dispersion of a 20% w/w suspension of commercial ITO nanoparticles (particle size <50 nm) was prepared in a 5 M acetic acid solution in ethanol, and subsequently spin coated onto cleaned FTO-coated glass. The electrodes were then sintered in a Carbolite furnace, and had a geometrical area of approximately 1 × 2 cm2 and a thickness of 3 µm.

#### Preparation of mTiO2 films

Electrodes were prepared as previously described29,40. In brief, a mTiO2 scaffold was deposited by slot-coating commercial Ti–nanoxide pastes (15–20 nm particles, 100% anatase, Solaronix) over a defined area (approximately 0.5 × 0.5 cm2) on a cleaned FTO-coated glass. The electrodes were then sintered in a Carbolite furnace and had an approximate thickness of 6 µm.

#### Preparation of mITO films for CO2 reduction

The mITO electrodes for CO2 reduction were prepared following a previously reported procedure52. The FTO-coated glass (2 × 1 cm) was sonicated sequentially with isopropanol and ethanol for 30 min and then dried at 180 °C. A scotch tape ring was placed onto the substrate to define the geometrical surface area as 0.19 cm2. The 50 nm ITO nanoparticles (43 mg), synthesized by a previously reported procedure52, were dispersed in an acetic acid (57 µL) and ethanol (143 µL) mixture by sonication for >1 h at 45 − 55 °C in a sealed vial. 5 µL of the cooled ITO suspension was drop cast onto the pre-defined area and after 10 s, spin-coated at 1000 rpm for 30 s. The electrode was allowed to dry for approximately 45 min before the scotch tape was removed. Finally, the electrodes were heated at a rate of 4 °C min−1 from room temperature to 400 °C and annealed at this temperature for 1 h. The resulting electrodes had a geometrical surface area of 0.19 cm2 and an average thickness of 8.5 µm.

### Photoelectrode assembly

#### Immobilization of the organic catalyst

Immobilization of the catalyst was achieved by employing the optimized conditions previously reported to ensure fast catalysis kinetics14. mTiO2 electrodes were placed in a capped vial and immersed in a solution of STEMPO (2 mM) in MeCN (which contained 0.2% v/v acetic acid and 0.1% v/v water). The solution was heated to 70 °C for 6 h under a N2 overpressure. After cooling, the mTiO2|STEMPO electrodes were rinsed with MeCN, dried under air and stored in the fridge before use.

#### Immobilization of the chromophores

The chromophores were immobilized in the dark at room temperature. DPP-CA was immobilized by the immersion of mTiO2|STEMPO electrodes in a bath of DPP-CA (0.2 mM) in DCM (2.5 h), DMF (overnight) or MeOH (overnight). DPP-PA was immobilized by the immersion of mTiO2|STEMPO electrodes in a bath of DPP-PA (0.2 mM) in DCM overnight. Immobilization of RuP was achieved by the immersion of mTiO2|STEMPO electrodes in a bath of RuP (0.2 mM) in MeOH overnight. The electrodes were then rinsed with the immobilization solvent, followed by DCM and then dried in air.

### Cathode assembly

The mITO electrode was wired to a metal rod by attaching a copper wire to the electrode using copper tape. The connection was then sealed using successive layers of Teflon tape, parafilm and then a final layer of parafilm over the metal rod and the electrode. Finally, electrical tape was used to cover any bare FTO surface so that only the predefined area was open to the electrolyte solution. W-FDH from D. vulgaris Hildenborough was expressed, purified and characterized according to a published method33. An 80 mM solution of DTT was made up in Tris-HCl (20 mM, pH 7.6). FDH (1.3 µL, 52 pmol) was incubated with the DTT solution (2.5 µL) under inert conditions for 5 min. The DTT solution containing FDH was then drop cast onto the mITO electrode 6 min before cell assembly.

### Photoelectrochemistry

The sensitized mTiO2 electrodes were wired to a metal rod by attaching copper wire to the electrode using copper tape. The connection was then sealed using layers of Teflon tape and parafilm.

Electrochemical measurements were carried out using an Ivium CompactStat potentiostat in a one-compartment three-necked custom-made cell equipped with a flat borosilicate glass window. A three-electrode set-up was used with a Pt counter electrode, a Ag/AgCl/KClsat reference and a sensitized electrode as the working electrode. The cell was purged with N2 or CO2 for 15 min prior to the measurements. Unless noted, back illumination for PEC assemblies was used for all the experiments using a calibrated Newport Oriel solar light simulator (150 W, 100 mW cm−2 across the solar spectrum, AM 1.5G) fitted with a UQG Optics UV Filter (λ > 420 nm) and infrared water filter. LSVs were conducted at a scan rate of 5 mV s−1 with chopped light alternating between dark and light every 5 s, followed by the chronoamperometry measurement.

The following conversion factor was applied for electrochemical experiments76:

$${E}_{{\rm{RHE}}}={E}_{{\rm{Ag/AgCl}}}+({\rm{pH}}\times 0.059)+{E}^{0}_{{\rm{Ag/AgCl}}}$$
(1)

PEC measurements of the two-electrode cell were carried out in a two-compartment two-necked custom-made cell equipped with a flat quartz glass window and separated by a Nafion membrane. The electrolyte employed for both compartments was identical, apart from the presence of the alcohol substrate in the anodic compartment. A two-electrode set-up was used with a mTiO2|STEMPO/DPP-CA working electrode placed in the compartment that featured the flat quartz glass window, and a mITO|FDH counter electrode was placed in the other compartment. A temperature of 25 °C was maintained for all the experiments.

### Quantification of molecular components

The molecular surface coverage of DPP-CA was quantified in triplicate using UV–vis spectroscopy after desorption of the chromophore from mTiO2|DPP-CA or mTiO2|STEMPO/DPP-CA electrodes by sonicating in tetrabutylammonium hydroxide 30-hydrate (TBAOH, 0.1 M, DMF, 1 − 2 mL) for 5 min until the electrode turned colorless. Higher tetrabutylammonium hydroxide concentrations and longer sonication times were avoided to prevent dye decomposition. The absorption maximum at 498 nm was determined and fitted to a calibration curve (conducted in 0.1 M TBAOH in DMF) to determine the loading values.

The molecular surface coverage of STEMPO was determined in triplicate by inductively coupled plasma optical emission spectrometry based on Si determination. To minimize the background from the glass substrate, mTiO2 films were prepared on a Ti foil substrate (Ti|mTiO2). The Ti substrate was cleaned by successive sonication in ethanol and acetone for 15 min each, followed by drying at 70 °C in air before further use. The mTiO2 films were then deposited as described above, which afforded the Ti|mTiO2 substrates. This was followed by immobilization of the STEMPO catalyst, overnight digestion of the electrodes (~0.36 cm2) in aqueous HNO3 (70%, 143 µL) and subsequent dilution to 2% v/v with MilliQ H2O. A background value, obtained in triplicate after digestion of the bare Ti|mTiO2 electrodes, was subtracted from the obtained values for Ti|mTiO2|STEMPO electrodes. The suspected polymerization of the silatrane moiety means that inductively coupled plasma optical emission spectrometry provides an absolute loading of the molecule, but does not allow for the precise quantification of the photoelectrochemically active molecules on the electrode, which results in a conservative value when TON numbers are calculated.

The surface coverage of FDH is assumed to be equal to the amount of enzyme drop cast onto the electrode surface.

### Product analysis and quantification

#### Gas product analysis

Measurements for H2 production were performed by manual injection of 50 µL from the headspace of the PEC cell using a gas-tight syringe (Hamilton, GASTIGHT) into a Shimadzu Tracera GC2010 Plus with a barrier discharge ionization detector. The gas chromatograph was equipped with a ShinCarbon micro ST column (0.53 mm diameter) kept at 85 °C using a helium carrier gas.

#### Liquid product analysis and quantification

After the reaction, a 100 µL aliquot of the solution was taken from the PEC cell and diluted with 900 µL of Milli-Q water and then analysed via HPLC. Substrate conversions and yields were deduced using a Waters 1500-Series HPLC, with an UV–vis detector (Waters 2489) set at 254 nm and 280 nm for the 4-MBA and HMF oxidation, respectively. For the 4-MBA oxidation, samples and standards (15 µL) were injected directly onto a 100 × 3.0 mm column (Luna Omega 3 µm PS C18 100 Å, LC Column) purchased from Phenomenex. The mobile phase comprised a 1:1 MeCN/H2O mixture with a total flow rate of 0.5 ml min−1 at 40 °C. For the HMF oxidation, samples and standards (15 µL) were injected directly onto a 300 × 7.8 mm column (Rezex ROA-Organic Acid H+ (8%), LC Column) purchased from Phenomenex. The mobile phase comprised 2.5 mM sulfuric acid with a total flow rate of 0.5 ml min−1 at 60 °C.

For the analysis, 1 h of equilibration was conducted before the first sample injection. Initially, the starting materials and expected main products were analysed separately to identify their respective retention times. This was carried out for 4-MBA and its possible oxidation products, 4-MBAd and 4-methylbenzoic acid. Similarly, this process was carried out for HMF and its possible oxidation products, DFF, 2,5-furandicarboxylic acid, 5-formylfuran-2-carboxylic acid and 5-hydroxymethyl-2-furancarboxylic acid (Supplementary Fig. 25). Standard calibration curves of the main product compounds (4-MBAd and DFF) were then produced for product quantification.

Formate production was quantified using ion chromatography on a Metrohm 882 Compact IC Plus ion chromatograph with a conductivity detector. The eluent buffer was an aqueous solution of Na2CO3 (3.0 mM) and NaHCO3 (1.0 mM). The system was calibrated for each batch of eluent buffer with samples that contained 0.14, 0.28 and 0.42 mM formate in each individual electrolyte. Calibrations were done for each electrolyte batch and the experimental samples matched to the correct calibration batch. Samples were diluted twofold with H2O before injection into the ion chromatography.