*2.8. MW and Distribution Analysis*

The *M*<sup>W</sup> and distribution of levan were detected by high-performance gel filtration chromatography (HPGFC). The system was supplemented with a refractive-index detector and an UltrahydrogelTM Linear column (7.8 mm × 300 mm). The mobile phase is 0.1 N NaNO3 with a 0.5 mL/min flow rate. The *M*<sup>W</sup> reference standards are Dextran T-2000 (*M*<sup>W</sup> = 2 × <sup>10</sup><sup>6</sup> Da, retention time is 14 min), Dextran T-300 (*M*<sup>W</sup> = 3.0 × <sup>10</sup><sup>5</sup> Da, retention time is 15.8 min), Dextran T-150 (*M*<sup>W</sup> = 1.4 × 105 Da, retention time is 16.4 min), Dextran T-10 (*M*<sup>W</sup> = 9.7 × 103 Da, retention time is 19 min) and Dextran T-5 (*M*<sup>W</sup> = 2.7 × 103 Da, retention time is 21 min). The detected temperature was 40 ◦C.

The degree of polymerization (DP) of FOS was determined by high-performance anion-exchange chromatography with pulsed amperometric detection (HPAEC-PAD). The column used in this system was a CarboPac PA200 column (3 mm × 250 mm) with the guard column CarboPac PA200 (3 mm × 50 mm). The eluent was 100 mM NaOH and 40 mM NaAc at the first 40 min, 100 mM NaOH and 400 mM NaAc at 40.1 min, and 100 mM NaOH and 40 mM NaAc between 40.1 and 60 min. The flow rate was 0.5 mL/min at 30 ◦C, and pH-Ag/AgCl as a reference electrode.

#### *2.9. Molecular Dynamics Simulation*

High-temperature molecular dynamics were performed using GROMACS (Version 2020.6) with an AMBER ff14SB force field [20,21]. The enzymes were solvated in TIP3P water and were relaxed through energy minimization to eliminate the error water insert. The system was heated to a pressure and temperature of 1 bar and 500 K using NVT and NPT ensemble balance with position constraint. The LINCS algorithm was used to constrain the hydrogen bonds in the system. The heavy atom of the protein is subjected to a position inhibition force with a constant of 1000 kJ mol−<sup>1</sup> nm−2. After equilibrium, the final output of the NPT simulation was subjected to without-position limitation. The root-mean-square deviation (RMSD) of backbone atom positions per 5 ps were calculated and analyzed using GROMACS analysis tools.

The root-mean-square fluctuation (RMSF) was determined at 280 K, and the other conditions were the same as above. The final structure was simulated by molecular dynamics of 100 ns.

All assays were performed in triplicate. Data management and analysis were performed using GraphPad Prism 5.0 (GraphPad Software, San Diego, CA, USA). All data are presented as the mean ± standard error of the mean.

#### **3. Results and Discussion**

#### *3.1. Computer-Aided Enzyme Screening*

Unlike the traditional BLAST tool, a computer-aided enzyme screening method combined with the Enzyme miner online server (https://loschmidt.chemi.muni.cz/enzymeminer/ custom-sequences) [22] was employed to screen out the potential thermostable LS. Hightemperature molecular dynamics simulations predicted the flexibilities of enzyme orthologs and thermostability. The RMSD value of the different microbial FSs is shown in Figure 1B, which can quantify the backbone atom movements of the protein. The Genbank accession numbers of these FSs are listed in Table S1. As a result, the LS from *C. diazotrophica* (Cedi-LS) exhibited the lowest RMSD value among all the enzymes, which means that Cedi-LS with the most rigid structure might have the best thermostability. Subsequently, a novel LS from mesophilic bacteria *P. orientalis* (Psor-LS) was screened on the template of Cedi-LS.

**Figure 1.** The structure and molecular dynamic simulation of Cedi-LS and Psor-LS. (**A**) The structure comparison of Cedi-LS and Psor-LS. The structures of Cedi-LS and Psor-LS are shown as carton. The superposition structure of Cedi-LS and Psor-LS are made by Pymol. (**B**) RMSD simulations of different microbial FSs at 500 K. (**C**) RMSF of the backbone Cα of modeled Cedi-LS and Psor-LS from the molecular dynamic simulation at 280 K.

The 3D structures of LSs were homologically modeled employing the crystal structures of *E*. *tasmaniensis* LS (PDB: 6FRW) as their template in the SWISSMODEL online server (https://swissmodel.expasy.org/) [23]. As shown in Figure 1A, Psor-LS showed a high coincidence degree in structure with Cedi-LS, except in loops 1 and 8. The RMSD value between two of the LSs was 0.117. The RMSF was used to study the movement of each residue in the enzyme and determine the flexibility of a particular region in the protein. As shown in Figure 1C, the RMSF results of Psor-LS and Cedi-LS showed a similar pattern, and the maximum D-value did not exceed 0.06 nm. Meanwhile, Cedi-LS and Psor-LS both showed low RMSF values in the whole structure (below 0.35 nm), indicating a very close relationship between the crystal structures of Cedi-LS and Psor-LS.
