Christof Martin Jäger

The selection of appropriate reaction conditions is paramount in chemical synthesis and can greatly impact the success of experimental campaigns. Machine learning models for reaction outcome prediction have focused primarily on reactant representations, whereas the representation of reaction conditions (the chemical components defining the reaction environment, e.g. activators, bases or solvents) has received comparatively less attention. Here, we present a novel machine learning framework designed to generate chemically relevant representations for sets of conditions. We evaluate these representations using literature Amide Coupling High-Throughput Experimentation (HTE) datasets, testing whether distances in embedding space reflect local yield similarity, physical-property trends, and experimentally useful condition diversity. Our results demonstrate that the learnt representations capture reactivity-relevant trends and support downstream tasks including condition visualization, diversity-aware HTE plate design, and selection of alternative conditions.

The selection of appropriate reaction conditions is paramount in chemical synthesis and can greatly impact the success of experimental campaigns. Machine learning models for reaction outcome prediction have focused primarily on reactant representations, whereas the representation of reaction conditions (the chemical components defining the reaction environment, e.g. activators, bases or solvents) has received comparatively less attention. Here, we present a novel machine learning framework designed to generate chemically relevant representations for sets of conditions. We evaluate these representations using literature Amide Coupling High-Throughput Experimentation (HTE) datasets, testing whether distances in embedding space reflect local yield similarity, physical-property trends, and experimentally useful condition diversity. Our results demonstrate that the learnt representations capture reactivity-relevant trends and support downstream tasks including condition visualization, diversity-aware HTE plate design, and selection of alternative conditions.

Biocatalysis in ionic liquids enables novel routes for bioprocessing. Enzymes derived from extremophiles promise greater stability and activity under ionic liquid (IL) influence. Here, we probe the enzyme alcohol dehydrogenase 2 from the halophilic archaeon Haloferax volcanii in thirteen different ion combinations for relative activity and analyse the results against molecular dynamics (MD) simulations of the same IL systems. We probe the ionic liquid property space based on ion polarizability and molecular electrostatic potential. Using the radial distribution functions, survival probabilities and spatial distribution functions of ions, we show that cooperative ion–ion interactions determine ion–protein interactions, and specifically, strong ion–ion interactions equate to higher enzymatic activity if neither of the ions interact strongly with the protein surface. We further demonstrate a tendency for cations interacting with the protein surface to be least detrimental to enzymatic activity if they show a low polarizability when combined with small hydrophilic anions. We also find that the IL ion influence is not mitigated by the surplus of negatively charged residues of the halophilic enzyme. This is shown by free energy landscape analysis in root mean square deviation and distance variation plots of active site gating residues (Trp43 and His273) demonstrating no protection of specific structural elements relevant to preserving enzymatic activity. On the other hand, we observe a general effect across all IL systems that a tight binding of water at acidic residues is preferentially interrupted at these residues through the increased presence of potassium ions. Overall, this study demonstrates a co-ion interaction dependent influence on allosteric surface residues controlling the active/inactive conformation of halophilic alcohol dehydrogenase 2 and the necessity to engineer ionic liquid systems for enzymes that rely on the integrity of functional surface residues regardless of their halophilicity or thermophilicity for use in bioprocessing.

‘Bacterial‐type’ ferredoxins host a cubane [4Fe4S]^2+/+ cluster that enables these proteins to mediate electron transfer and facilitate a broad range of biological processes. Peptide maquettes based on the conserved cluster‐forming motif have previously been reported and used to model the ferredoxins. Herein we explore the integration of a [4Fe4S]‐peptide maquette into a H₂‐powered electron transport chain. While routinely formed under anaerobic conditions, we illustrate by electron paramagnetic resonance (EPR) analysis that these maquettes can be reconstituted under aerobic conditions by using photoactivated NADH to reduce the cluster at 240 K. Attempts to tune the redox properties of the iron‐sulfur cluster by introducing an Fe‐coordinating selenocysteine residue were also explored. To demonstrate the integration of these artificial metalloproteins into a semi‐synthetic electron transport chain, we utilize a ferredoxin‐inspired [4Fe4S]‐peptide maquette as the redox partner in the hydrogenase‐mediated oxidation of H₂.

The periplasmic chaperone SilF has been identified as part of an Ag(I) detoxification system in Gram negative bacteria. Sil proteins also bind Cu(I), but with reported weaker affinity, therefore leading to the designation of a specific detoxification system for Ag(I). Using isothermal titration calorimetry we show that binding of both ions is not only tighter than previously thought, but of very similar affinities. We investigated the structural origins of ion binding using molecular dynamics and QM/MM simulations underpinned by structural and biophysical experiments. The results of this analysis showed that the binding site adapts to accommodate either ion, with key interactions with the solvent in the case of Cu(I). The implications of this are that Gram negative bacteria do not appear to have evolved a specific Ag(I) efflux system but take advantage of the existing Cu(I) detoxification system. Therefore, there are consequences for how we define a particular metal resistance mechanism and understand its evolution in the environment.

The recovery and valorization of metals and rare earth metals from wastewater are of great importance to prevent environmental pollution and recover valuable resources. Certain bacterial and fungal species are capable of removing metal ions from the environment by facilitating their reduction and precipitation. Even though the phenomenon is well documented, little is known about the mechanism. Therefore, we systematically investigated the influence of nitrogen sources, cultivation time, biomass, and protein concentration on silver reduction capacities of cell‐free cultivation media (spent media) of Aspergillus niger, A. terreus, and A. oryzae. The spent medium of A. niger showed the highest silver reduction capacities with up to 15 μmol per milliliter spent medium when ammonium was used as the sole N‐source. Silver ion reduction in the spent medium was not driven by enzymes and did not correlate with biomass concentration. Nearly full reduction capacity was reached after 2 days of incubation, long before the cessation of growth and onset of the stationary phase. The size of silver nanoparticles formed in the spent medium of A. niger was influenced by the nitrogen source, with silver nanoparticles formed in nitrate or ammonium‐containing medium having an average diameter of 32 and 6 nm, respectively.

Pyruvate formate-lyase (PFL) is a glycyl radical enzyme (GRE) playing a pivotal role in the metabolism of strict and facultative anaerobes. Its activation is carried out by a PFL-activating enzyme, a member of the radical S-adenosylmethionine (rSAM) superfamily of metalloenzymes, which introduces a glycyl radical into the Gly radical domain of PFL. The activation mechanism is still not fully understood and is structurally based on a complex with a short model peptide of PFL. Here, we present extensive molecular dynamics simulations in combination with quantum mechanics/molecular mechanics (QM/MM)-based kinetic and thermodynamic reaction evaluations of a more complete activation model comprising the 49 amino acid long C-terminus region of PFL. We reveal the benefits and pitfalls of the current activation model, providing evidence that the bound peptide conformation does not resemble the bound protein-protein complex conformation with PFL, with implications for the activation process. Substitution of the central glycine with (S)- and (R)-alanine showed excellent binding of (R)-alanine over unstable binding of (S)-alanine. Radical stabilization calculations indicate that a higher radical stability of the glycyl radical might not be the sole origin of the evolutionary development of GREs. QM/MM-derived radical formation kinetics further demonstrate feasible activation barriers for both peptide and C-terminus activation, demonstrating why the crystalized model peptide system is an excellent inhibitory system for natural activation. This new evidence supports the theory that GREs converged on glycyl radical formation due to the better conformational accessibility of the glycine radical loop, rather than the highest radical stability of the formed peptide radicals.