Work · CO₂ Fixation

Product synthesis in a 25-protein cell-free cycle

25-plus enzymes fixing CO₂ outside any cell.


The dream: a molecular printer

Imagine a machine like a coffee maker, but for molecules — you feed it gas, water, and a trace of minerals, and it prints whatever compound you want. That is the vision this project lives under for me. Protein engineering already carries a lot of modern chemistry and medicine, but always in tiny amounts. The reason we don’t yet have the molecular printer is brutally simple: proteins are intrinsically unstable, and a printer would need to run its biochemical engine continuously — powered, ideally, by nothing more than water, air, trace minerals, and electricity.

The paper frames the same system around a bigger problem — large-scale CO₂ fixation for the climate. But those two framings meet in the same place: the challange of protein aggregation.

Three opto-sensing and control modules for NAD(P)H, ATP and FAD/Fc+, each reading a cofactor by its optical signal and feeding substrate back to hold it at a setpoint.
The three cofactor opto-sensing & control modules — NAD(P)H read by UV absorbance (T340), ATP by luciferase luminescence (565 nm), and FAD/Fc⁺ by absorbance (T300). Each holds its cofactor at a setpoint by feedback. Figure 2a from Luo et al., A cell-free self-replenishing CO₂-fixing system, Nature Catalysis 5, 154–162 (2022).

What we built: a metabolism in a test tube

In the Liao lab I worked on a cell-free CO₂-fixing system alongside Paul and Liang-Yu - a full metabolism running outside any living cell. It sounds innocent, but it is all or nothing. It sounds innocent, but it is all or nothing. The big benefit one taks from this is it massively simplifies the system you study, but you loose that proteins are nurtured and cared by and the energy could be supplied conveniently through sugar and lets you manipulate the fixation rate directly, decoupled from growth and regulation. It was published in Nature Catalysis in 2022.

Under constant adjustment we kept the cycle turning and the metabolite pools balanced, and sustained it for six hours at a CO₂-fixation rate comparable to or greater than photosynthetic and lithoautotrophic organisms. The proof of principle held — the only thing we couldn’t beat was the slow degradation of the proteins themselves.

Designing around nature’s weak enzymes

The system is an oxygen-insensitive, self-replenishing cycle that produces acetyl-CoA (C2), pyruvate (C3), and malate (C4) directly from CO₂. Its cofactors (NAD(P)H, ATP, FAD) are held in balance by real-time opto-sensing modules.

Nature already fixes carbon, but its key enzymes come with built-in problems, and the design was really an exercise in routing around them:

The elegant part is that the cycle’s products are also its own intermediates. It replenishes its own building blocks as it runs — self-replenishing, and effectively autocatalytic: run it, and it makes more of what it needs to keep running.

The real enemy: instability

Oxygen-insensitive does not imply oxygen-proof. Even with tolerant enzymes we still got aggregation: These proteins aren’t perfectly folded outside their native cell, and they degrade faster for it. The system’s own chemistry attacked them, too: the electron shuttle that we used to regenerate FAD relies on ferrocenium (Fc⁺) and H₂O₂, and both oxidise the enzymes. After six hours, several of the workhorses — PhaA, Ccr, Epi, Ecm, Mcd, Pps, Ppc, Mtk, Mcl — had lost a notable amount of activity and we were able to attribute the damage to Fc⁺ and H₂O₂. Protein oxidation was the culprit. (Free CoA suffers the same fate — oxidised to its disulfide by O₂ and H₂O₂.)

Seeing it happen — control by light

Real-time MS/MS needs equipment we didn’t have — pulling a sample, quenching it, centrifuge-filtering and loading the machine cost 10 minutes on the centrifugation step alone. So we read the cycle optically instead: NAD(P)H by UV absorbance (T₃₄₀), ATP by a luciferase luminescence flash at 565 nm, FAD/Fc⁺ by absorbance at 300 nm. When a signal drifts below setpoint, the controller injects the matching substrate.

The catch is that the signals overlap. NAD(P)H (≈340 nm) and the Fc⁺ shuttle (≈300 nm) sit close together in the UV, and more than one species absorbs in each band. Assigning the right wavelength to the right cofactor — so the readout meant what we thought it meant — was the difference between a controller that stabilised the cycle and one chasing a phantom. It is what turned a reaction we could merely watch into one we could run, for six hours.

The supplies, measurements and analyses

Running the reaction was preconditioned on having the proteins available — and after Paul, Lin and Liang-Yu incepted the protocols, I developed them further, increasing the amount and quality of proteins that we had for the experiments. Roughly half the enzymes I produced and purified by hand (laborious protein manufacture), while the more common ones were ordered in. Beyond that, a lot of my work was in measuring a system where most of the intermediates are invisible:

HO O OH O SuccinateC4 · polyesters, BDO, solvents
HO O OH O OH MalateC4 · TCA entry, C4 acids
O OH O PyruvateC3 · amino acids, lactate
O H O OH GlyoxylateC2 · glycine, C–C couplings
HO OH OH O GlycerateC3 · serine, PEP, sugars
O SCoA OH (R)-3-hydroxy­butyryl-CoAC4 · PHB / bioplastic monomer
The metabolites, not the enzymes — six of the C2–C4 pools the cycle can be tapped at, and what each is a starting material for. Every carbon in them arrives as CO₂. (Hashed and wedged bonds mark the natural stereocentres: S-malate, R-glycerate, R-3-hydroxybutyryl-CoA.)

The Roads we found Science Locked

Read the paper ↗ Back to the biochemistry wall