MiniSOG as a biodegradable heterogeneous photocatalyst for coupled redox biotransformations

Abstract

Visible-light photocatalysis has emerged as a powerful strategy to drive chemical reactions under mild conditions, often in tandem with biocatalysts to achieve one-pot cascades. In this work, we demonstrate that the photocatalytic flavoprotein miniSOG, originally engineered as a genetically encoded singlet oxygen generator, can be repurposed as a heterogeneous photocatalyst. By immobilizing miniSOG as cross-linked enzyme aggregates (CLEAs), we enable controlled in situ production of hydrogen peroxide (H₂O₂) and NADH regeneration, facilitating both oxidative and reductive biotransformations.

---

Introduction

Heterogeneous photocatalysts offer significant advantages over homogeneous counterparts, such as simplified catalyst separation, improved catalyst recyclability, increased stability, and reduced environmental impact owing to their solid-state nature.^1–3 Recent studies have explored various heterogeneous photocatalysts, including titanium dioxide (TiO₂),⁴ graphitic carbon nitride (g-C₃N₄),⁵ conjugated porous polymers (CPPs),⁶ and metal–organic frameworks (MOFs),⁷ for applications such as water splitting,⁸ pollutant degradation,⁹ and organic synthesis.¹⁰ Despite these advances, biodegradable heterogeneous (organic) photocatalysts have been relatively underexplored.

Recent research highlighted flavin-based photocatalysts as particularly promising due to their versatility and biocompatibility.¹¹ In particular, flavin mononucleotide (FMN) stands out among organic photocatalysts because its oxidised form absorbs strongly at ≈450 nm and, upon blue-light excitation, funnels efficiently to a long-lived triplet that is both a potent oxidant ( vs. NHE) and—after single-electron reduction—a potent reductant (FMN_{_sq}/FMNH˙ ≈ −1.1 V). This bimodal redox behaviour lets FMN toggle between one- and two-electron pathways in neat water, providing a metal-free, recyclable alternative to Ru/Ir complexes;¹² by replacing the N5 atom with carbon (deazaflavins, dFl) and harvesting a second photon, chemists have pushed the excited-state potential beyond that of lithium metal [E(dFl*/dFl˙⁻) = −3.3 V vs. SCE] while retaining the parent cofactor's green credentials.¹³ That versatility translates into a surprisingly broad reaction portfolio: free FMN mediates aerobic C–H oxygenation of alkyl benzenes on oxide supports,¹⁴ drives visible-light [2+2] cycloadditions when tailored for energy-transfer catalysis,¹⁵ and, in its super-reducing deazaflavin form, cleanly dechlorinates electron-rich arenes under air.¹³ Additionally, FMN's ability to generate reactive oxygen species (ROS) upon light activation has been exploited in environmental applications, such as the degradation of micropollutants in water treatment processes.^16,17 Strategies for heterogenisation of flavins like embedding FMN within porous organic polymers,¹⁷ inorganic particles (SiO₂,¹⁶ ZrO₂/TiO₂¹⁴) or co-immobilization of free FMN and enzymes onto porous materials,¹⁸ have been developed to enhance its photostability, ensuring prolonged catalytic activity, even in chemoenzymatic systems.

Beyond small-molecule catalysis, FMN's role as a natural cofactor in flavoproteins has been harnessed in photoenzymatic reactions. Enzymes such as old yellow enzymes (OYEs) utilize FMN to achieve stereoselective transformations under light irradiation, expanding the toolkit for asymmetric synthesis.^19,20 Particularly interesting is the miniSOG; a small, 106 residue (14 kDa) FMN-containing protein initially designed for singlet oxygen generation in correlative light and electron microscopy.²¹ miniSOG has been shown to undergo both one- and two-electron reductions, demonstrating potential for diverse catalytic reactions.²² Engineered miniSOGs, such as the Q103V variant, have been reported to enhance photocatalytic performance by extending the lifetime of its triplet excited state, enabling effective photoreduction reactions involving metal complexes such as Pt(IV) substrates.²³ Notwithstanding these very recent developments, miniSOG still remains primarily an imaging tool for microbiological applications, with a limited scope in its use as a photocatalyst.²⁴

By leveraging the inherent biodegradability and the versatile photocatalytic capabilities of FMN, we envisioned that utilizing miniSOG's scaffold for heterogenization of the chromophore could significantly enhance its practical applicability, stability, and recyclability, thereby addressing critical limitations of homogeneous systems.

Results and discussion

Preparation of miniSOG WT and Q103V mutant

The proteins were prepared in house as previously reported (see Fig. S1 and S2), achieving a yield of ∼30 mg L⁻¹ of purified protein, in line with literature reported values.²⁵

Singlet oxygen generation

To evaluate whether miniSOG can deliver synthetically useful fluxes of singlet oxygen (¹O₂) in solution, we first carried out photooxygenations under homogeneous conditions. Classical ¹O₂ reporters were chosen (Scheme 1): (i) α-terpinene, whose conversion to ascaridole can be followed by ¹H NMR,²⁶ and (ii) triphenylphosphine (PPh₃), which oxidises to triphenylphosphine oxide and can be monitored by ³¹P NMR.²⁷ Reactions were performed in CDCl₃ containing 10% v/v D₂O potassium phosphate (KPi) buffer (pD 7.4, 50 mM) to extend the lifetime (τ_Δ) of ¹O₂, which is 20–25-fold longer in D₂O than in H₂O.^28,29 A fully aqueous test with the water-soluble substrate 2-furoic acid, a recognised benchmark for ¹O₂ oxidation, was run in parallel (iii).³⁰ Even under conditions designed to maximise ¹O₂ yield (0 °C, deuterated solvents, O₂-sparging, 450 nm LEDs, miniSOG = 5 mg mL⁻¹) no ascaridole, triphenylphosphine oxide or 5-hydroxy-2(5H)-furanone could be detected within the NMR limits of quantification. This outcome is consistent with the low solution quantum yields reported for native miniSOG (Φ_Δ ≈ 0.03–0.06)³¹ and with structural studies showing efficient quenching of its FMN triplet by the surrounding protein matrix.²⁵ Although engineered variants and cellular contexts can enhance phototoxicity,^32,33 our results indicate that unmodified miniSOG is ill-suited for preparative ¹O₂ chemistry in homogeneous solution without further protein or cofactor optimisation.

Scheme 1 Attempts at use of miniSOG for photoinduced singlet oxygen generation.

Hydrogen peroxide and speroxide radical production

We then redirected our efforts to type-I electron-transfer photochemistry (PS* to ³O₂, producing ROS), specifically the light-driven formation of hydrogen peroxide (H₂O₂). A continuous, micromolar-scale flux of in situ H₂O₂ is especially valuable for biocatalysis, because many peroxide-dependent enzymes (e.g., heme-thiolate peroxygenases) suffer rapid, irreversible heme degradation when exposed to the millimolar peroxide boluses typically added from a syringe, yet remain active if H₂O₂ is fed slowly and uniformly.³⁴ We previously reported on the efficiency and stability of a photobiocatalytic flow system which could deliver brominations via an internally generated, steady trickle of peroxide.³⁵ Similar “self-dosing” strategies have proved crucial for unspecific peroxygenase cascades in which photocatalysts, including flavin photosensitisers furnish only as much H₂O₂ as the enzyme can safely consume.^36–38

For the heterogenization of miniSOG, covalent attachment of wild-type (WT) miniSOG to commercial epoxy- and amino-activated beads was explored first, yet the preparations lost fluorescence and activity quickly under irradiation (see Fig. S3 and S4). The Q103V mutant, whose triplet lifetime (∼100 μs) indeed boosts H₂O₂ output in solution as previously reported, proved equally impractical: the FMN cofactor detached during metal-affinity chromatography and again when the protein solution was reduced to achieve a concentration of ∼5 mg mL⁻¹ for downstream experiments, suggesting a weakened chromophore interaction.

The weakened binding most likely arises because Gln103 promotes FMNs stability via H-bonding (Fig. 1(c)); its replacement with hydrophobic residues such as valine or leucine eliminates the non-covalent interaction, a drawback noted previously for Q103V/L in other contexts.^39,40

Fig. 1 CLEA preparation workflow that included (a) determination of ideal precipitating agent and (b) glutaraldehyde (cross-linking) percentage (second step). See Fig. S7. (c) Catalytic pocket of miniSOG, containing the FMN chromophore.

These setbacks steered us toward the cross-linked enzyme aggregate (CLEA) strategy, a carrier-free immobilisation concept in which the target enzyme is (i) precipitated from a concentrated solution so that the molecules cluster into nano- to micrometre aggregates, and (ii) the resulting protein clusters are covalently “locked” with a bifunctional reagent such as glutaraldehyde, which forms Schiff-base bridges between surface lysines.⁴¹

Immobilising miniSOG as CLEAs would offer a carrier-free, recyclable and biodegradable photocatalyst format that tightly entraps the flavin in the protein matrix, while enabling straightforward recovery.⁴² Despite the fact that CLEAs have been prepared for a variety of biocatalysts, no reports on the efficiency of this strategy for small proteins like miniSOG has been published to date. In particular, structural distortion (with potential loss of the FMN) or minimal efficiency of the cross-linking could be encountered due to the distribution of the six surface lysine residues in miniSOG (SI, Fig. S5). We started our CLEA optimization campaign with two critical parameters, precipitant selection, and cross-linker amount (Fig. 1(a) and (b)).

Precipitant selection

CLEA preparation begins with the precipitation of folded protein (as opposed to inducing unfolding which would also results in protein precipitation). Because miniSOG lacks a routine activity assay, we monitored denaturation indirectly: precipitation (200 μL, 5 mg mL⁻¹) was induced with 12 candidate agents at ≈90% of their customary concentration; after 5 min, precipitates were redissolved by adding 1800 μL of buffer (100 mM KPi pH 7.0) solution, centrifuged through a 10 kDa filter, and FMN fluorescence in the filtrate (λ_ex 450 nm/λ_em 520 nm) quantified. Loss of FMN correlates with unfolding. Ammonium sulfate (NH₄)₂SO₄—long regarded as the gentlest salting-out agent for enzymes⁴³—released virtually no FMN (≲baseline), whereas 1-propanol liberated ≈90% of the chromophore, making it the worst choice (Fig. 2(a)). PEG-400, methanol and ethanol displayed intermediate effects. Thus, ammonium sulfate (90% sat.) was adopted for all subsequent CLEA preparations.

Fig. 2 (a) Precipitant screen. (b) Reaction additive screen. (c) Glutaraldehyde dose optimisation. All data represent mean ± SD (n = 3). H₂O₂ quantification determined by HRP/ABTS assay using a calibration curve (Fig. S6).

Reaction-medium engineering

Having established that 90%-saturated ammonium sulfate cleanly precipitates holo-miniSOG without ejecting FMN, our next priority was to tailor the reaction environment in which these CLEAs operate. The composition of this reaction medium governs the photoredox fate of the trapped flavin—especially its propensity, under blue light, to divert electrons toward oxygen and form superoxide. Triethanolamine (TEOA), a tertiary amine commonly used in photocatalytic cycles,⁴⁴ boosted H₂O₂ production from 1.4 μM (KPi only) to 10 μM at 10 mM, but fell to 6 μM at 100 mM—consistent with donor-excess quenching of excited FMN (Fig. 2(b)). Because superoxide dismutase (SOD) catalyses O₂˙⁻ → H₂O₂ + O₂ near the diffusion limit (∼2 × 10⁹ M⁻¹ s⁻¹),⁴⁵ adding 1 U SOD in the system doubled the H₂O₂ yield to ≈23 μM, confirming superoxide as the primary ROS intermediate. These observations are in agreement with classical FMN photochemistry, where reduced flavins (by an electron donor, like EDTA or TEOA) react with O₂ to form H₂O₂ via superoxide.⁴⁶ Moreover, it has been proven that miniSOG not only generates O₂˙⁻, but it's actually one of the primary ROS that it produces, and, is able to diffuse out of the protein structure into the surrounding medium.^47,48

Cross-linker optimisation

Aggregates were stabilised with glutaraldehyde (GA), the most widely used bifunctional cross-linker in CLEA technology owing to its low cost and broad reactivity with lysine ε-amines.⁴⁹ Yet excessive GA can rigidify the protein matrix and throttle mass transfer, while too little fails to lock particles together.⁵⁰ We therefore varied GA between 0.5 and 3.5% v/v. Photocatalytic assays (Fig. 2(c)) revealed a clear optimum at 0.5% GA, giving 100% relative H₂O₂ output; higher loadings depressed activity, in line with literature reports that over-cross-linking introduces diffusional barriers,⁵¹ while concentrations below 0.5% v/v did not lead to appreciable CLEA formation. It is also noted that after crosslinking, the NaBH₄ reduction step is crucial, as without it, the CLEAs degrade due to hydrolysis of the imine bonds formed between glutaraldehyde and lysine residues, and the FMN chromophore remains inactive. Consequently, the unreduced, deep-orange material does not generate detectable H₂O₂ upon irradiation. Once reduced, the finalised CLEA can be lyophilised and stored at −20 °C for at least one month without loss of ROS-producing ability.

Characterization of miniSOG CLEAs

To confirm successful CLEA formation and understand their structural features, we characterized the aggregates using Fourier transform infrared spectroscopy (FT-IR), dynamic light scattering (DLS), and scanning electron microscopy (SEM) (Fig. 3). IR spectra of the CLEA (orange trace) compared to the pre-crosslinked precipitate (blue trace) revealed a persistent amide I band at ∼1650 cm⁻¹ and amide II at ∼1540 cm⁻¹, suggesting successful retention of protein secondary structure in the CLEA material.⁵² Furthermore, significant increases in aliphatic C–H signals (e.g. 2925–2850 cm⁻¹) from glutaraldehyde's methylene groups, confirm that the protein molecules are covalently interconnected via glutaraldehyde. These features match typical CLEA spectral fingerprints reported previously.^53,54

Fig. 3 Physicochemical characterisation of miniSOG CLEAs. (a) FTIR spectra of precipitated miniSOG (blue) and cross-linked aggregates (orange). (b) DLS size distribution showing aggregated particles (5–8 μm). (c), (d) SEM micrographs revealing type I/II porous, spherical aggregates typical of protein CLEAs. Scale bars: (c) 10 μm, (d) 2 μm.

DLS analysis of suspended CLEAs revealed a broad particle size distribution centred between 5–8 μm, far larger than the native enzyme, indicating aggregation of many protein units into each CLEA particle.

Such large sizes are common in CLEA preparations without dispersants or co-feeders, where inter-protein crosslinking leads to extensive agglomeration.^53,55

SEM micrographs further confirmed the aggregated morphology: the CLEAs appeared as rough, porous clusters composed of smaller, 4–8 μm subunits (Fig. 3(c) and (d)), in agreement with the DLS observations. According to the classification by Sheldon et al.,⁵⁶ these can be classified as composite type I/type II CLEAs, with spherical subunits partially fused into irregular, porous aggregates.

Together, the spectroscopic, hydrodynamic, and morphological evidence confirms successful conversion of miniSOG into a covalently crosslinked, insoluble aggregate.

Evaluation of the system

Having established the conditions in which the miniSOG-CLEA/TEOA/SOD gave the most constant sub-millimolar H₂O₂ flux (Fig. 2(b)), we explored whether the peroxide stream is compatible with a peroxide-sensitive heme enzyme—an unspecific peroxygenase (UPO)—and could drive a synthetically useful oxy-functionalisation (Fig. 4(a)). As model substrate we chose 1H-indole-7-carboxylic acid (1 mM), whose carboxylate renders the indole core water-soluble, avoiding the mass-transfer limits that affect the parent indole.⁵⁷ Recombinant AaeUPO (10 μM; heme-thiolate active site) was combined with 5 mg miniSOG-CLEA in 100 mM KPi (pH 7.0) containing 10 mM TEOA and 1 U SOD; the suspension (1 mL, 2 mL Eppendorf) was irradiated for 30 min with 450 nm (10 W, orbital shake 100 rpm) and then analysed.

Fig. 4 (a) Reaction scheme and conditions for the blue-light mediated oxidation of 1 (1 mM) to oxindole 2 using rAaeUPO and in situ produced H₂O₂ from miniSOG-CLEA. (b) Recyclability test of the miniSOG-CLEA photocatalyst.

LC–MS revealed ∼82% conversion of the indole after a single 30-min light period (Fig. 4(b) and Fig. S8), and ∼80% of the product peak corresponded to the expected 3-oxindole (m/z 177.0 [M + H]⁺, Fig. S9), the most common mono-oxygenation product for UPO-catalysed indole oxidations.⁵⁸ Minor signals for dimeric or over-oxidised species—cross-linked indoles routinely observed in UPO chemistry—were also detected. By contrast, dark controls (no hν) or reactions lacking the CLEA photocatalyst or the UPO protein gave ≤3% conversion, confirming that in situ peroxide formation is mandatory (Fig. S10). Furthermore, the regioisomer of 1, 1H-indole-6-carboxylic acid was also employed under otherwise identical conditions, and the same outcome was observed; after 1 h of irradiation, 100% conversion was achieved (Fig. S11).

Re-usability

After the reaction was complete the reaction mixture was centrifuged, washed and the miniSOG-CLEA pellet was resuspended by addition of 1 mL of fresh reaction mixture for four further 30-min cycles (identical conditions, Fig. 4(b)). Conversion declined gradually from ∼82% (cycle 1) to ∼52% (cycle 5), reflecting progressive but not catastrophic FMN bleaching (as opposed to leaching); miniSOG is known to inactivate under prolonged exposed to excess light via ROS promoted chromophore destruction,^25,59 yet the CLEA material clearly tolerates this effect. HPLC analyses of the clear supernatant after each reaction showed only minor traces of leached FMN chromophore, confirming that photobleaching is the main avenue for catalyst degradation over time.

These observations demonstrate that the miniSOG-CLEA platform supplies a UPO-compatible, self-metered H₂O₂ flux, enabling the successful oxidation of carboxy-indoles and providing a proof-of-concept for coupling miniSOG-CLEAs with other peroxide-sensitive oxidations.

Oxidative system benchmarking

To benchmark the oxidative performance of miniSOG-CLEA we compared its apparent quantum yield Φ_app and photocatalyst-normalised turnover frequency TOF with two representative heterogeneous photocatalysts previously reported for visible-light-driven UPO reactions (Table 1). MiniSOG-CLEA not only gives the highest peroxide productivity per gram of catalyst (Table 1, entry 1), but also exhibits the best photon economy in the set outperforming Au-decorated TiO₂ nanoparticles (NP) (Table 1, entry 2) and melon-type g-C₃N₄ (Table 1, entry 3). These results show that a biodegradable protein scaffold can already match or surpass state-of-the-art inorganic supports, while leaving room for further improvement through chromophore engineering or light-management strategies.

Table 1 Visible-light heterogeneous H₂O₂ generators for peroxygenase reactions^a

Entry	PS	Light source/time (h)	H2O2 (μmol)	Φapp (%)	TOF (μmol g^−1 h^−1)
a Detailed calculations for deriving Φapp and TOF values can be found in SI, Page S27.
1	MiniSOG-CLEA (this work)	10 W LED strip, 0.5 h	0.82	0.0012	330
2	Au–TiO2 NPs	150 W halogen, 6 h	0.96	0.0011	32
3	g-C3N4 powder	205 W halogen, 1 h	0.12	0.00028	24

---

NADH recycling

Beyond their utility as recyclable H₂O₂ photo-generators, we reasoned that miniSOG-CLEAs might also function as photocatalysts for NADH regeneration, an ever-present bottleneck in nicotinamide-dependent biocatalysis.⁶⁰ Our interest was sparked by a 2019 report from Kinastowska et al. showing that, under strongly alkaline conditions (pH ≥ 10) and high triethanolamine (TEOA) loadings (0.5–1 M), a range of heterogeneous photosensitisers photo-oxidise TEOA to glycolaldehyde (GlyA); GlyA then reduces NAD⁺ to enzymatically active 1,4-NADH even after the photocatalyst is removed and the lights are off.⁶¹

TEOA thus could serve the dual role of sacrificial electron donor and a precursor of the GlyA reductant. To verify that miniSOG-CLEAs indeed photo-generate the glycolaldehyde required for nicotinamide reduction, we quantified GlyA production by repeating the trapping protocol of Yuehui et al.,⁶² which consists in reacting GlyA with 3-methyl-2-benzothiazolinone hydrazone (MBTH) to generate its oxime–hydrazone adduct (MBTH-GlyA) for subsequent UV and mass detection by LC–MS (Fig. 5(a) and Fig. S12–S14). Under our adapted conditions (0.5 M TEOA, NH₃/NH₄⁺ buffer, pH 10.5, 30 min irradiation) the CLEA suspension produced ≈1.8 mM GlyA (Fig. 5(b)), consistent with the 3–5 mM range reported for semiconductor photocatalysts.⁶¹ Moreover, ¹H-NMR analysis of the illuminated reaction mixture (Fig. 5(b), D₂O, 0.5 M TEOA, pD 10.5) revealed the clear appearance of glycolaldehyde-specific resonance, a singlet at δ 9.56 ppm (1 H, –CHO), identical to a GlyA standard. These signals are absent in both the dark control and the “no-CLEA” control, confirming that they originate from the photo-induced oxidation of TEOA mediated by miniSOG-CLEA.

Fig. 5 Photochemical generation of glycolaldehyde (GlyA) and NADH using miniSOG-CLEA. (a) GlyA derivatisation with 3-methyl-2-benzothiazolinone hydrazone (MBTH) to give the UV-detectable adduct MBTH-GlyA 3. (b) GlyA formation: irradiation of miniSOG-CLEA (5 mg) in 0.5 M TEOA/400 mM NH₃/NH₄⁺ buffer (pH 10.5) for 30 min (450 nm) affords 1.8 mM GlyA, quantified as MBTH-GlyA. No GlyA is detected in dark or “no-CLEA” controls. ¹H-NMR spectra of highlighted reaction products. Peak at ∼8.5 can be attributed to miniSOG degradation. (c) The pre-irradiated mixture (pre-NAD⁺ addition) reduces NAD⁺ (0.5 mM) to enzymatically active NADH (0.2 mM) within 1 h in the dark, confirming GlyA-mediated cofactor regeneration.

An additional experiment confirmed that, after 30 min irradiation of miniSOG-CLEAs at pH 10.5, the cleared reaction supernatant reduced NAD⁺ to ∼40% NADH within 1 h in the dark (Fig. 5(c)). These results corroborate that FMN is indeed capable of acting as a photocatalyst for the production of GlyA.

Because this highly alkaline medium (optimal GlyA formation is recorded at pH ≈ 11.5) is too extreme for most redox enzymes, we sought a compromise that still supports appreciable GlyA formation yet allows enzymatic turnover. Phenylalanine dehydrogenase (PheDH) is an attractive test case: it catalyses the NADH-dependent reductive amination of phenyl-pyruvate in NH₃/NH₄⁺ buffer (ammonia source) at an optimum pH of 9.5, while retaining more than half of its activity at pH 10.5.^63,64 Adopting pH 10.5 in a TEOA/NH₃ system therefore (i) keeps GlyA production within one order of magnitude of its optimum and (ii) places PheDH within its activity tolerance threshold, enabling the photogenerated GlyA to reduce NAD⁺ in situ and drive PheDH turnover (Fig. S15).

When PheDH (7 U mL⁻¹, as measured in these optimised conditions) and phenyl-pyruvate (3) (2 mM) were added and illuminated (450 nm light), > 60% of the starting material 3 was consumed to form L-phenylalanine within 30 min (Fig. 6(a)), demonstrating an operational photo-enzymatic NADH-recycling loop compatible with alkaline biocatalysts (Fig. S16). Upon blue light illumination, the miniSOG-CLEA produces glycolaldehyde (GlyA), which reduces NAD⁺ to NADH; the latter drives phenylalanine dehydrogenase (PheDH) to convert phenyl-pyruvate 3 (2 mM) into L-phenylalanine 4 in NH₃/NH₄⁺ buffer (400 mM, pH 10.5, 30 min). In principle, the dialdehyde (glyoxal) that is formed after GlyA is oxidised can then be a NAD⁺ reductant on its own. Furthermore, as with the oxidative system, we tested the durability of the photocatalyst under these alkaline conditions; in each recycling run we tracked both phenyl-pyruvate (PhePyr) conversion and the amount of glycolaldehyde (GlyA) trapped as its MBTH adduct (Fig. 6(b)). The two profiles correlate closely: GlyA is generated at 1.7 mM in the first light period and falls steadily to ≈0.8 mM by the fifth, mirroring the drop in PhePyr conversion from 62% to 29%. The slight rebound in cycle 5 (GlyA ≈0.8 mM; conversion 29%) reflects the experimental scatter but underscores that conversion is limited by the GlyA (and hence NADH) pool rather than PheDH itself, whose specific activity remains high at pH 10.5/NH₃ buffer. Control experiments confirmed that product formation requires all three components: (i) irradiated CLEA, (ii) TEOA and (iii) PheDH—no product was detected when light was omitted, or the atmosphere was replaced with N₂.

Fig. 6 Coupling photogenerated glycolaldehyde to PheDH-catalysed reductive amination. (a) Reaction scheme. (b) Recycling study (five 30-min irradiation cycles; 5 mg CLEA, 0.5 M TEOA, 0.2 mM NAD⁺, in 1 mL). Blue bars: PhePyr converted (%) green bars: GlyA formed (mM, quantified as MBTH adduct). Product formation parallels GlyA availability. No product was detected in “no-light” or “no-CLEA” controls.

Finally, the used miniSOG-CLEA material was further characterised with FT-IR and SEM to determine its physicochemical condition after use. The former showed an identical spectrum, with the key amide I and II bands remaining present (Fig. S17). Furthermore, the SEM micrographs showed that the surface morphology remained unaltered after five consecutive reaction cycles (Fig. S18). This observation shows that the protein backbone of the material remains intact, and the inactivation of the catalyst is indeed attributed to the photobleaching of the FMN chromophore.

We also attempted a combi-CLEA in which PheDH was co-precipitated (Fig. S19) and cross-linked with miniSOG, following precedents for tandem CLEAs in oxidoreductase cascades.^65,66 Although the material formed readily, PheDH lost >95% activity after cross-linking, likely due to glutaraldehyde modification of its catalytic lysine, and no detectable product formed under illumination. These observations emphasise that a two-component set-up—photocatalytic miniSOG-CLEA plus freely diffusing PheDH—is currently the most effective arrangement for the GlyA/NADH relay.

To ensure that the low, steady-state H₂O₂ released by miniSOG-CLEA would not compromise the reductive amination step, we conducted a separate experiment using PheDH with a deliberately harsh peroxide load. The enzyme (7 U mL⁻¹, NH₃/NH₄⁺ buffer, pH 10.5) was pre-incubated for 60 min with 1 mM H₂O₂—over twenty-fold higher than the ≈40 μM generated photochemically in each cycle—then assayed in the standard PhePyr → L-Phe reaction. Residual activity was indistinguishable from an untreated control (<3% difference, n = 3), indicating that PheDH tolerates peroxide levels well above those present during the coupled process. Consequently, the addition of catalase or other peroxide-scavenging enzymes, often required in flavin-driven NADH systems, was unnecessary in our set-up.

Taken together, the data demonstrate a clear stoichiometric link between photogenerated GlyA, NADH availability and product formation, and establish miniSOG-CLEA as a recyclable, carrier-free photocatalyst that can power NADH-dependent reductive biocatalysis under alkaline conditions.

Reductive system benchmarking

To situate miniSOG-CLEA within the broader NADH-photoregeneration landscape we assembled a concise comparison of apparent quantum yields (Φ_app) and catalyst-normalised initial rates (TOF) against three representative heterogeneous systems (Table 2). miniSOG-CLEA reaches Φ_app ≈ 1 × 10⁻⁵ (0.001%) and TOF ≈ 80 μmol g⁻¹ h⁻¹, values near those of the conjugated organic polymer PHTT_DMP and bulk g-C₃N₄ reported by Kinastowska et al.⁶¹ (Table 2, entries 2 and 3) and substantially lower than the engineered visible-light TiO₂@g-C₃N₄ core–shell catalyst of Dong Yang et al. (Table 2, entry 4).⁶⁷ Notwithstanding its modest photon economy, miniSOG-CLEA offers advantages unique to a biodegradable protein scaffold—straightforward genetic tuning, facile removal from product streams, and dual oxidative/reductive functionality—which collectively justify further optimisation efforts.

Table 2 Benchmark comparison of four heterogeneous NADH-photoregeneration systems

Entry	PS	Illumination and time	Reactor volume/catalyst loading	NADH formed (μmol)	Φapp (%)	TOF (μmol g^−1 h^−1)
a Φapp and TOF were calculated based on experimental values. Detailed calculations can be seen at SI, Page S30.b Φapp value was provided directly by the authors of the study.c In addition to the composite catalyst, an electron mediator, [Cp*Rh(bpy)H2O]^2+, was also employed in order to facilitate electron transfer from TEOA to NAD^+.
1^a	MiniSOG-CLEA (this work)	450 nm LED strip, 30 min	1 mL, 5 mg CLEA	0.20	0.001	80
2^a	PHTT_DMP polymer^61	400–700 nm lamp, 2 h	5 mL, 10 mg polymer	0.50	0.002–0.004	50
3^a	g-C3N4^61	400–700 nm lamp, 2 h	5 mL, 10 mg g-C3N4	0.50	0.002	50
4^b^,^c	TiO2@g-C3N4 core–shell^67	405 nm LED, 1 h	4 mL, 5 mg composite	4.0	0.17	800

---

Mechanistic pathway

With these experiments we have now exploited both branches of the miniSOG (FMN) photoredox manifold (Scheme 2): (i) the reductive leg, in which the FMN triplet oxidises TEOA step-wise to glycolaldehyde (GlyA), and GlyA in turn delivers the two-electron hydride required to convert NAD⁺ into enzymatically competent NADH, and, (ii) the oxidative leg, in which photo-reduced FMN˙⁻ transfers an electron to O₂ to give O₂˙⁻/H₂O₂, fuelling haem peroxygenase catalysis. To our knowledge this is the first demonstration that an FMN photosensitiser can drive GlyA formation from TEOA and propagate that chemistry into a functional NADH-recycling loop (photochemical GlyA → NADH → PheDH turnover). Conventional photobiocatalytic systems usually harness only one half-reaction, the oxidation of a “sacrificial” donor whose products are discarded, while the complementary reduction channel remains unused.

Scheme 2 Dual photoredox pathways of miniSOG's FMN: oxidative branch (O₂ → H₂O₂) and reductive branch (TEOA → GlyA → NADH). The TEOA → GlyA → NADH branch operates only under alkaline conditions (this work: 0.5 M TEOA in NH₃/NH₄⁺ buffer, pH 10.5, 450 nm irradiation, 30 min). The oxidative branch (photo-reduced FMN → O₂˙⁻/H₂O₂) is run in neutral phosphate buffer as described in the text.

By coupling both directions we achieve dual functionality from a single, carrier-free miniSOG-CLEA catalyst: a controlled H₂O₂ source for oxidative enzymes and, under alkaline conditions, a renewable NADH generator for reductive biocatalysis. This dual exploitation of FMN photochemistry expands the toolbox of flavin-based photocatalysts and showcases a rare example where the oxidation products of the sacrificial donor are valorised rather than wasted.

Conclusions

This proof of concept showcases that miniSOG has the potential to be exploited as more than a mere cell deactivator/fluorescent tag for microbiology studies and expand its application scope into the realm of synthetic chemistry. We were able to successfully use miniSOG in conjunction with other proteins (UPO, SOD, PheDH), further highlighting its compatibility with protein cascades. The miniSOG-CLEA is a recyclable heterogeneous photocatalyst: it retains activity over multiple reuse cycles and activity loss is due to photobleaching rather than leaching (the FMN cofactor remains entrapped), and stored samples remain active. Unlike conventional inorganic photocatalysts (e.g., TiO₂), here a single bioderived solid enables two biocatalysis-compatible modes under visible light and in water: (i) at neutral pH it delivers a controlled H₂O₂ feed for oxidative enzymes, and (ii) under alkaline conditions it photo-oxidizes TEOA to glycolaldehyde, which drives NADH regeneration, a functionality uncommon for inorganic materials. While its ability to produce a relatively small amount of the desired species (H₂O₂, NADH equivalents), remains a barrier, this study opens the road for further improvements, for instance by genetic engineering (more photostable fluorescent proteins etc), or more advanced reaction set ups. Continuous flow would be a good candidate for such an endeavour, as preparative-scale CLEAs have already been produced for lipases, showing that the precipitation/cross-link cycle scales with the same equipment used for routine enzyme processing.⁶⁸ The resulting mechanically robust aggregates can be pumped as slurries or packed into transparent packed-bed columns, and illuminated.⁶⁹ Continuous-flow reactors have operated smoothly with such CLEA formulations at litre-per-hour throughputs, demonstrating seamless integration of CLEAs into modern flow-chemistry platforms.⁷⁰

Author contributions

F. P. and E. B. conceptualized the idea. F. P. supervised the project and acquired funding. E. B. performed the experiments and wrote the initial draft. Both authors discussed the results and reviewed the manuscript.

Conflicts of interest

There are no conflicts to declare.

Data availability

The data supporting this article have been included as part of the SI. See DOI: https://doi.org/10.1039/d5ma00861a.

Acknowledgements

This project was supported by the SNSF (200021_232199, F.P). We thank Prof. Frank Hollmann for providing the lyophilised AaeUPO enzyme and for useful scientific discussions.

Notes and references

1. J. Chen, J. Cen, X. Xu and X. Li, Catal. Sci. Technol., 2016, 6, 349–362.
2. A. Savateev and M. Antonietti, ACS Catal., 2018, 8, 9790–9808.
3. D. Friedmann, A. Hakki, H. Kim, W. Choi and D. Bahnemann, Green Chem., 2016, 18, 5391–5411.
4. A. Fujishima, T. N. Rao and D. A. Tryk, J. Photochem. Photobiol., C, 2000, 1, 1–21.
5. S. Cao, J. Low, J. Yu and M. Jaroniec, Adv. Mater., 2015, 27, 2150–2176.
6. D. Taylor, S. J. Dalgarno, Z. Xu and F. Vilela, Chem. Soc. Rev., 2020, 49, 3981–4042.
7. A. Bavykina, N. Kolobov, I. S. Khan, J. A. Bau, A. Ramirez and J. Gascon, Chem. Rev., 2020, 120, 8468–8535.
8. X. Wang, K. Maeda, A. Thomas, K. Takanabe, G. Xin, J. M. Carlsson, K. Domen and M. Antonietti, Nat. Mater., 2009, 8, 76–80.
9. M. J. Alonso-Navarro, J. Barrio, S. Royuela, N. Karjule, M. M. Ramos, J. I. Martínez, M. Shalom and J. L. Segura, RSC Adv., 2021, 11, 2701–2705.
10. T. Zhang, C. Si, K. Guo, X. Liu, Q. Liu, J. Fu and Q. Han, Inorg. Chem., 2022, 61, 20657–20665.
11. V. Srivastava, P. K. Singh, A. Srivastava and P. P. Singh, RSC Adv., 2021, 11, 14251–14259.
12. R. Cibulka and M. W. Fraaije, Flavin-Based Catalysis, John Wiley & Sons, Ltd, 2021, pp. 97–124.
13. T. Pavlovska, D. Král Lesný, E. Svobodová, I. Hoskovcová, N. Archipowa, R. J. Kutta and R. Cibulka, Chem. – Eur. J., 2022, 28, e202200768.
14. P. Dongare, I. MacKenzie, D. Wang, D. A. Nicewicz and T. J. Meyer, Proc. Natl. Acad. Sci. U. S. A., 2017, 114, 9279–9283.
15. V. Mojr, E. Svobodová, K. Straková, T. Neveselý, J. Chudoba, H. Dvořáková and R. Cibulka, Chem. Commun., 2015, 51, 12036–12039.
16. O. Cabezuelo, R. Martinez-Haya, N. Montes, F. Bosca and M. L. Marin, Appl. Catal., B, 2021, 298, 120497.
17. P. Tang, B. Ji and G. Sun, J. Hazard. Mater., 2022, 435, 128982.
18. S. F. Castillo Pacheco, M. J. Moran, J. I. Santos, L. Salassa and F. López-Gallego, ChemCatChem, 2023, 15, e202300140.
19. F. Parmeggiani, E. Brenna, D. Colombo, F. G. Gatti, F. Tentori and D. Tessaro, ChemBioChem, 2022, 23, e202100445.
20. K. Stott, K. Saito, D. J. Thiele and V. Massey, J. Biol. Chem., 1993, 268, 6097–6106.
21. X. Shu, V. Lev-Ram, T. J. Deerinck, Y. Qi, E. B. Ramko, M. W. Davidson, Y. Jin, M. H. Ellisman and R. Y. Tsien, PLoS Biol., 2011, 9, e1001041.
22. O. Azpitarte, A. Zudaire, J. Uranga, X. Lopez, L. Salassa, E. Formoso and E. Rezabal, ChemPhysChem, 2023, 24, e202300091.
23. J. Gurruchaga-Pereda, V. Martínez-Martínez, E. Formoso, O. Azpitarte, E. Rezabal, X. Lopez, A. L. Cortajarena and L. Salassa, J. Phys. Chem. Lett., 2021, 12, 4504–4508.
24. E. A. Souslova, K. E. Mironova and S. M. Deyev, J. Biophotonics, 2017, 10, 338–352.
25. J. Torra, C. Lafaye, L. Signor, S. Aumonier, C. Flors, X. Shu, S. Nonell, G. Gotthard and A. Royant, Sci. Rep., 2019, 9, 2428.
26. J. Chen, R. Prinsloo and X. Ni, Technologies, 2024, 12, 29.
27. K. Ohkubo, T. Nanjo and S. Fukuzumi, Bull. Chem. Soc. Jpn., 2006, 79, 1489–1500.
28. J. Al-Nu’airat, I. Oluwoye, N. Zeinali, M. Altarawneh and B. Z. Dlugogorski, Chem. Rec., 2021, 21, 315–342.
29. S. Mondal, R. B. Jethwa, B. Pant, R. Hauschild and S. A. Freunberger, Faraday Discuss., 2024, 248, 175–189.
30. J. A. Navio, J. Fuentes Mota, M. A. Pradera Adrian and M. Garcia Gomez, J. Photochem. Photobiol., A, 1990, 52, 91–95.
31. R. Ruiz-González, A. L. Cortajarena, S. H. Mejias, M. Agut, S. Nonell and C. Flors, J. Am. Chem. Soc., 2013, 135, 9564–9567.
32. D. V. Yuzhakova, M. V. Shirmanova, V. V. Klimenko, M. M. Lukina, A. I. Gavrina, A. D. Komarova, D. A. Gorbachev, N. V. Sapogova, K. A. Lukyanov and V. A. Kamensky, Biochim. Biophys. Acta, Gen. Subj., 2021, 1865, 129978.
33. F. Hilgers, N. L. Bitzenhofer, Y. Ackermann, A. Burmeister, A. Grünberger, K. E. Jaeger and T. Drepper, Int. J. Mol. Sci., 2019, 20, 4608.
34. A. Karich, K. Scheibner, R. Ullrich and M. Hofrichter, J. Mol. Catal. B: Enzym., 2016, 134, 238–246.
35. E. Broumidis and F. Paradisi, Angew. Chem., Int. Ed., 2024, 63, e202401912.
36. P. Püllmann, D. Homann, T. A. Karl, B. König and M. J. Weissenborn, Angew. Chem., Int. Ed., 2023, 62, e202307897.
37. S. Willot, E. Fernández-Fueyo, F. Tieves, M. Pesic, M. Alcalde, I. W. C. E. Arends, C. B. Park and F. Hollmann, ACS Catal., 2019, 9, 890–894.
38. P. De Santis, D. Wegstein, B. O. Burek, J. Patzsch, M. Alcalde, W. Kroutil, J. Z. Bloh and S. Kara, ChemSusChem, 2023, 16, e202300613.
39. D. J. Mogensen, M. Etzerodt and P. R. Ogilby, J. Photochem. Photobiol., A, 2022, 429, 113894.
40. M. Westberg, L. Holmegaard, F. M. Pimenta, M. Etzerodt and P. R. Ogilby, J. Am. Chem. Soc., 2015, 137, 1632–1642.
41. S. Talekar, A. Joshi, G. Joshi, P. Kamat, R. Haripurkar and S. Kambale, RSC Adv., 2013, 3, 12485–12511.
42. C. Blanco-Llamero, P. García-García and F. J. Señoráns, Front. Bioeng. Biotechnol., 2021, 9 DOI:10.3389/fbioe.2021.794672.
43. K. C. Duong-Ly and S. B. Gabelli, in Methods in Enzymology, ed. J. Lorsch, Academic Press, 2014, vol. 541, pp. 85–94.
44. O. Savateev and Y. Zou, ChemistryOpen, 2022, 11, e202200095.
45. E. D. Getzoff, D. E. Cabelli, C. L. Fisher, H. E. Parge, M. S. Viezzoli, L. Banci and R. A. Hallewell, Nature, 1992, 358, 347–351.
46. E. Churakova, M. Kluge, R. Ullrich, I. Arends, M. Hofrichter and F. Hollmann, Angew. Chem., Int. Ed., 2011, 50, 10716–10719.
47. M. E. Barnett, T. M. Baran, T. H. Foster and A. P. Wojtovich, Free Radical Biol. Med., 2018, 116, 134–140.
48. F. M. Pimenta, R. L. Jensen, T. Breitenbach, M. Etzerodt and P. R. Ogilby, Photochem. Photobiol., 2013, 89, 1116–1126.
49. C. S. Sampaio, J. A. F. Angelotti, R. Fernandez-Lafuente and D. B. Hirata, Int. J. Biol. Macromol., 2022, 215, 434–449.
50. R. A. Sheldon, Biochem. Soc. Trans., 2007, 35, 1583–1587.
51. A. Borham, M. Bkhit, J. Wang and X. Qian, Environ. Technol. Innovation, 2025, 38, 104143.
52. J. S. Cobb, V. Zai-Rose, J. J. Correia and A. V. Janorkar, ACS Omega, 2020, 5, 8403–8413.
53. M. S. M. Razib, R. N. Z. R. A. Rahman, F. M. Shariff and M. S. M. Ali, Catalysts, 2020, 10, 55.
54. S. A. Chaudhari and R. S. Singhal, Int. J. Biol. Macromol., 2017, 98, 610–621.
55. S. S. Tükel, F. Hürrem, D. Yildirim and Ö. Alptekin, J. Mol. Catal. B: Enzym., 2013, 97, 252–257.
56. R. Schoevaart, M. W. Wolbers, M. Golubovic, M. Ottens, A. P. G. Kieboom, F. van Rantwijk, L. A. M. van der Wielen and R. A. Sheldon, Biotechnol. Bioeng., 2004, 87, 754–762.
57. M. Sadauskas, R. Statkevičiūtė, J. Vaitekūnas, V. Petkevičius, V. Časaitė, R. Gasparavičiūtė and R. Meškys, Dyes Pigm., 2020, 173, 107882.
58. R. Ullrich, M. Poraj-Kobielska, O. M. Herold-Majumdar, J. Vind and M. Hofrichter, Catalysts, 2021, 11, 1495.
59. J. Ribes, P. Cossard, K. Al Yaman, I. Bestel and E. Badarau, RSC Adv., 2023, 13, 2355–2364.
60. K. Bachosz, J. Zdarta, M. Bilal, A. S. Meyer and T. Jesionowski, Sci. Total Environ., 2023, 868, 161630.
61. K. Kinastowska, J. Liu, J. M. Tobin, Y. Rakovich, F. Vilela, Z. Xu, W. Bartkowiak and M. Grzelczak, Appl. Catal., B, 2019, 243, 686–692.
62. Y. Li, D. Si, W. Shang, J. Wang, J. Guo, N. Zhang, C. Hao and Y. Shi, New J. Chem., 2022, 46, 6360–6365.
63. G. Yao, K. Liu, S. Wang, H. Huo and S. Wang, J. Chem. Technol. Biotechnol., 2021, 96, 1049–1056.
64. A. L. Ahmad, E. M. Low and S. R. A. Shukor, J. Mol. Catal. B: Enzym., 2013, 88, 26–31.
65. I. E. Touahar, L. Haroune, S. Ba, J.-P. Bellenger and H. Cabana, Sci. Total Environ., 2014, 481, 90–99.
66. L. T. Nguyen and K.-L. Yang, Enzyme Microb. Technol., 2017, 100, 52–59.
67. D. Yang, Y. Zhang, S. Zhang, Y. Cheng, Y. Wu, Z. Cai, X. Wang, J. Shi and Z. Jiang, ACS Catal., 2019, 9, 11492–11501.
68. R. Fernández-Penas, C. Verdugo-Escamilla, S. Martínez-Rodríguez and J. A. Gavira, Cryst. Growth Des., 2021, 21, 1698–1707.
69. S. Simić, M. Jakštaitė, W. T. S. Huck, C. K. Winkler and W. Kroutil, ACS Catal., 2022, 12, 14040–14049.
70. C. L. Fernández Regueiro, D. Roura Padrosa and F. Paradisi, C. R. Chim, 2025, 28, 349–359.

---