Samples were thawed at room temperature. Next, 500 µl sample were transferred to 1.5 ml Eppendorf reaction tubes. Each sample underwent a thermal-mechanical treatment of 10 min at 90°C and 1,000 rpm in a BioShakeiQ (Quantifoil Instruments GmbH, Jena, Germany) for cell lysis. Cell debris was separated by centrifugation, the supernatant was directly used for LC-MS analysis.

LC-MS measurements were performed as described by Hamer et al. (2015): samples of 2 µl (1 µl for standards) were injected by an autosampler and subsequently separated by hydrophilic interaction chromatography (HILIC) using an Acquity UPLC BEH Amide column (1.7 µm, 2.1 mm × 150 mm; Waters Corporation, Milford, MA, United States) in combination with a VanGuard pre-column (1.7 µm, 2.1 mm × 35 mm; Waters Corporation, Milford, United States). With a flow rate set to 0.4 ml/min at 35°C, the samples were eluted in a 15-min method with a gradient from A and B. Eluent A consisted of 80:20 acetonitrile:water and eluent B of 20:80 acetonitrile:water. In addition, both eluents contained 10 mM NH₄HCO₂ and 0.1% (v/v) formic acid. The method consisted of the following steps: 0–2.5 min: isocratic, 100% A; 2.5–12.5 min: linear gradient reducing the concentration to 25% A; 12.5–13.5 min: linear gradient back to 100% A; 13.5–15 min: re-equilibration of the column with 100% A. The measurements were performed in positive mode with a target mass of 500 m/z.

For the quantification of COS, the arbitrary signal intensities of the ions from the LC-MS chromatograms were converted to molar fractions and shown as a percentage share from the total amount of product. To this end, different concentrations of fully acetylated chitin oligomers from Megazyme Ltd. (Bray, Ireland) ranging from DP 3–6 (A3–A6) were injected into the UHPLC-ELSD-ESI-MS system to create linear standard curves (see Figure 1) for the conversion from arbitrary intensities into molar fractions based on the masses given in Table 1.

First, the alignment of NodC_Sm (Gene ID at NCBI: 61599275) with the bacterial cellulose synthase subunit A from Rhodobacter sphaeroides [BcsA; PDB code 4HG6_A (Morgan et al., 2013)] was taken from Dorfmueller et al. (2014). Next, NodC_GRH2 (UniProt: A0A0N7ARR3) was manually aligned to BcsA, analogous to NodC_Sm, and both NodC-BcsA alignments were uploaded to SWISS-MODEL for a “target-template-alignment” (Guex et al., 2009; Benkert et al., 2011; Mariani et al., 2013; Bienert et al., 2017; Waterhouse et al., 2018; Studer et al., 2020). The newly generated NodC models were aligned with the pre-translocation state of the bacterial cellulose synthase (PDB code 5EJ1) in PyMOL (Morgan et al., 2013; Schrodinger, 2020). Subsequently, a chitin hexamer built with GLYCAM (Woods Group) was inserted into the NodC models using PyMOL’s pair fitting function (Schrodinger, 2020): the position of the non-reducing end GlcNAc unit is derived from the sugar unit at the non-reducing end of the cellulose chain. For the generation of smaller COS, one or two of the reducing end units were removed.

The molecular dynamics simulations were performed with GROMACS (Van Der Spoel et al., 2005; Abraham et al., 2015; Pall et al., 2015) version 2019.3 as described in tutorials 2 and 5 (Lemkul, 2019) with a few changes: forcefield gromos 54a7 (Schmid et al., 2011) was applied for all components and the parametrization of the ligand was automatically performed with the Automated Force Field Topology Builder [ATB; (Malde et al., 2011; Canzar et al., 2013; Koziara et al., 2014)]. The protein was not oriented along the z-axis using GROMACS commands, but the protein-ligand complex was positioned with its annotated intramembrane sections in the dipalmitoylphosphatidylcholine (DPPC) membrane using the 3-Button Editing Mouse Mode of PyMOL (Schrodinger, 2020), before executing GROMACS command pdb2gmx on the protein PDB-file. While the tutorial features a 1 ns long production simulation, 100 ns simulations were performed instead. Subsequently, VMD 1.9.3 (Humphrey, W., Dalke, A. and Schulten, 1996) was used to analyze the trajectories concerning average distances between certain atoms and hydrogen bonds between selected interactions partners with a distance and angle cutoff of 3.5 Å and 35°, respectively. Based on these data and the in vivo results, individual frames were selected that represent the trends of the distance and hydrogen bond analyses. These frames provide a possible explanation for the observed preferences of NodC_GRH2 and NodC_Sm to produce A5 or A4, respectively.

First, the wildtype NodC_GRH2 was heterologously expressed in both E. coli and C. glutamicum. Notably, the growth measured via the optical density (OD) reaches different maxima (Figures 5A,B): the values for the E. coli cultures (varying around 17) are clearly lower than for the C. glutamicum strains (ca. 50), but both follow the typical pattern of lag, log, and stationary phase (Wang et al., 2015). Furthermore, all cultures were induced at the beginning of their respective exponential phase upon which COS were detected. This observation was also made for the muteins.

In E. coli, the main product for wildtype NodC_GRH2 throughout the whole cultivation is A5 (Figure 5C). At the beginning of the exponential phase after 13 h, it is nearly the only product detectable, accounting for 95% of all COS, with the remaining share being A3 and A4. After 18 h, 64% of COS are A5, nearly 20% are A3, with the rest being A4 (7%) and A6 (9%). In the stationary phase, pentamer is accounting for 75% of the total product, and the remaining portion consists mainly of A4 (19%) and A6 (6%).

In contrast, C. glutamicum expressing the same wildtype enzyme produces predominantly A6 (Figure 4D), with overall only about a quarter of the total production being A5, and A4 accounting for 1% at most, while there was no A3 at all. The share of A6 changes from 65% after 17 h at the beginning of the exponential phase, to 75% after 36 h in the stationary phase.

In a second step, E. coli and C. glutamicum were cultivated expressing the NodC_S19L or the NodC_R346S muteins. Their growth and corresponding COS production are shown in the following Figures 6, 7. Like in the experiment before, the total OD600 differs strongly between E. coli and C. glutamicum, but the induction was always performed at the beginning of the exponential phase, after 9 h for E. coli and 13 h for C. glutamicum.

All strains synthesized oligomers with rather consistent product compositions. The yields, however, varied between the expression hosts and, though to a smaller extent, also between the single cultivations of a given strain. As an example, the total COS yields of two individual cultivations each for both E. coli and C. glutamicum with the wildtype NodC_GRH2 are shown in Figure 8.

For both expression hosts, the COS production clearly is in line with bacterial growth. The protein expression of all cultures was induced at the beginning of their respective exponential phase and only then, they started with COS synthesis. The product concentration rose with increasing OD600; hence, COS production was highest during the exponential growth phase and stagnated (at later timepoints for C. glutamicum, see Supplementary Figure S2) or decreased (E. coli, Figure 8A) during the stationary phase. While for E. coli, product concentration and OD600 develop in parallel, the slope of product formation is delayed for C. glutamicum, and the maximum is not reached at the beginning of the stationary phase but rather during it.

For E. coli, typical yields of 0.01 mmol/L COS were obtained, but the results were not completely uniform, as higher concentrations of up to 0.05 mmol/L were also reached with the same strain in one experiment (not shown). C. glutamicum produced 0.03 mmol/L, but again yields as high as 0.1 mmol/L were also obtained with the wildtype NodC_GRH2 (not shown).

The introduction of the two mutations did not alter the growth of E. coli or C. glutamicum considerably (see Supplementary Figure S3 in comparison to Figure 8). For both expression hosts, the amount of produced COS was roughly estimated half with the NodC_GRH2 S19L mutein compared to the wildtype, but—as for the wildtype—with strong variation between different cultivations. In contrast, the COS yield with the NodC_GRH2 R346S mutein was comparable to that of the wildtype in E. coli and even slightly higher than that of the wildtype in C. glutamicum.

The generation of in silico models for NodC from S. meliloti and Rhizobium GRH2 based on an alignment with a bacterial cellulose synthase subunit (BcsA) was successful. While the cytosolic part of the enzyme contains conserved motifs and is thus similar to other crystallized glycosyltransferases, no closely related crystallized template was available for the transmembrane part (Figure 2). Even though crystal structures for two chitin synthases were recently published (Chen et al., 2022; Ren et al., 2022), this new data could not improve our model: both chitin synthases show an even lower sequence identity to NodC_Sm and NodC_GRH2 than the cellulose synthase BcsA (see Supplementary Table S1), and while the nine motifs discussed by Chen et al. are strongly conserved between the chitin polymer synthases of oomycetes, fungi, and arthropods, this applies to a lesser extent only for the rhizobial NodC chitin oligomer synthases. Especially the transmembrane parts of NodCs and the chitin polymer synthases differ from each other, as shown representatively for the chitin synthase from Candida albicans and NodC_GRH2 and NodC_Sm in Supplementary Figure S4. This is not surprising given that the transmembrane part of the chitin polymer synthases builds a tunnel to allow secretion of the nascent chitin polymer chain, while NodCs produce chitin oligomers intracellularly and their transmembrane domain possibly serves subcellular localization of the enzymes only. Therefore, we decided to solely rely on the experimentally verified data of Dorfmueller et al. (2014) as the best current model for rhizobial NodCs.

When starting with the NodC_GRH2 sequence only, dependent on the server used for protein prediction, either just the cytosolic part was modelled [SWISS-MODEL (Guex et al., 2009; Bertoni et al., 2017; Bienert et al., 2017; Waterhouse et al., 2018; Studer et al., 2020)] or proteins with less reasonable conformations and unlikely features [I-TASSER (Roy et al., 2010; Yang et al., 2014; Yang and Zhang, 2015); RaptorX (Ma et al., 2015; Wang et al., 2016, 2017); AlphaFold (Jumper et al., 2021)] were generated.

Therefore, the “target-template-alignment” (Guex et al., 2009; Benkert et al., 2011; Mariani et al., 2013; Bienert et al., 2017; Waterhouse et al., 2018; Studer et al., 2020) with the cellulose synthase still seems to be a reasonable approach. Yet, despite a generally homologous structure, there are obvious differences between BcsA and NodC. These include the open tunnel of the cellulose synthase compared to the cleft-like structure of NodC or the difference in size between glucose and GlcNAc. Furthermore, with 3.25 Å (Morgan et al., 2013), the resolution of the template, i.e., the cellulose synthase crystal structure, is rather low (Cohen et al., 2009; Zardecki et al., 2022). At this resolution, usually only the basic scaffolding, the backbone and bulky sidechains can be assigned with a high degree of certainty, but not the atomic structure (Martz et al., 2014; Zardecki et al., 2022).

Still, our model supports the previously published hypothesis (Dorfmueller et al., 2014) and fits our experimental data: the potentially relevant amino acids identified in silico actually play an important role in determining COS length in vitro, as show below.

As NodC is membrane-embedded (Barny et al., 1996; Dorfmueller et al., 2014), the model’s environment was adjusted for the simulations. The enzyme’s position is derived from experimental data (Dorfmueller et al., 2014), but the flexibility and composition of the in silico membrane may not properly represent reality. Lacking experimental data, neither the position of the whole COS nor the orientation of the N-acetyl groups is fully validated. The COS can move, but full rotations of the strand, which are e.g., expected to happen during COS elongation (Dorfmueller et al., 2014; Orlean and Funai, 2019) could not be observed. Still, a stable positioning of the units during the simulations tends to indicate correct COS positioning.

Yet, the triplicates of each simulation progressed differently. The outcome was often determined at the beginning of each run, possibly due to the many players involved in a simulation of an oligomer in a membrane-embedded protein. For example, for NodC_GRH2 with the pentamer, the R346 either found S19 at the start of the simulation and built a strong bond, or it engaged with the COS in a less stable interaction. Once the arginine found an interaction partner, a rearrangement became highly unlikely. This is likely an artefact of the simulation.

Still, some frames give a good explanation for the experimental data: for the wildtype NodC_GRH2 with the pentamer, S19 and R346 are close (Table 2) and form hydrogen bonds over considerable parts of the simulation, thereby corroborating the hypothesis of them limiting the depth of the cleft (Dorfmueller et al., 2014). NodC_GRH2 is known as a producer of pentamers in its original organism, the Gram-negative Rhizobia (Lopez-Lara et al., 1995), just as when expressed heterologously in E. coli (Hamer, 2014). Yet, the synthase can also produce larger COS, which might then be stabilized by R346 or induced by high UDP-GlcNAc pools (Orlean and Funai, 2019). Our experimental results showed an increased proportion of A6 once the bulky arginine was removed, further supporting that this amino acid restricts the products’ length.

For NodC_Sm, the fitting of A6 requires a different rotamer of R349 to give more space first. In Figure 4A, there is clearly enough space for a fifth GlcNAc unit (7.2 Å between R349 and the COS), but not a sixth. This is in line with the reported production (Ardourel et al., 1994; Kamst et al., 2000) and fits our observations: the first mutein (NodC_S19L), which is more similar to NodC_Sm than the wildtype, produces drastically less hexamer (C. glutamicum NodC_S19L) or none at all (E. coli NodC_S19L).

Further interesting observations from the simulations were an unexpected hydrophobic interaction between L19 and I348 in NodC_Sm. These two are closer in NodC_Sm than R349 and L19 (Table 2) and seem to link two helices (TM1 and TM3, Figure 5). Thus, one might assume that L19 and I348 in NodC_Sm are the true equivalents of S19 and R346 in NodC_GRH2. In contrast, the corresponding I345 in NodC_GRH2 is probably not involved in any interactions with the substrate as it is located on the outside surface of the protein, pointing towards the membrane.

However, this hydrophobic interaction does not seem to be the factor limiting the COS length for NodC_Sm, which is rather brought about by the bulky arginine which is “unoccupied”. In NodC_GRH2, the corresponding amino acid is interacting with S19 by forming H-bonds, thus its position is rather fixed. In NodC_Sm, the arginine is without a partner and can move down the cleft, thereby limiting the room for COS elongation.

In conclusion and in line with the initial hypothesis, the simulations show for NodC_Sm and NodC_GRH2 that the arginine is particularly relevant for COS length: it either interacts with S19, then its position is quite fixed which gives room for COS, or it is more flexible and reduces the depth of the tunnel. Thus, once this amino acid is removed, larger DPs can be synthesized. In addition, I348 seems to play an important role in NodC_Sm as an interaction partner for L19.

Engineering of NodC_GRH2 to produce certain COS was successful. E. coli produces mainly pentamers with the wildtype protein, but roughly 25% consist of a varying mixture composed of A3, A4, and A6. C. glutamicum is more consistent with 75% A6 and 25% A5 over the course of the cultivation. The downsizing mutation (S19L) changed the production range for both hosts, but more severely for C. glutamicum. Yet, the range of COS also became more uniform for E. coli.

In general, the product range is quite consistent for all strains tested. The strongest variations can be observed between the earliest and later time points. At the beginning, the rarer COS might still be below the detection limit. Therefore, later timepoints are more reliable regarding product composition. Still, one can influence the outcome by prematurely terminating the fermentation, as the product range differs, e.g., between the exponential and stationary phase.

Synthesis of longer COS (R346S mutation) was most successful for C. glutamicum which produced almost exclusively A6, but also for E. coli whose main product was approx. 85% A6. While the production of tetramers only was not achieved, strains synthesizing either mainly pentamers (E. coli NodC_GRH2 or the NodC_S19L muteins with both expression systems) or nearly pure hexamers (C. glutamicum NodC_R346S) were successfully generated. Especially the practically pure hexamer is interesting for several applications: Firstly, purification is simpler if the synthesized COS are uniform, and A6 is still water-soluble, making it industrially manageable and applicable (Bonin et al., 2020; Chapelle et al., 2021). Secondly, A6 is promising for several biological applications, e.g., plant strengthening (Gubaeva et al., 2018; Shi et al., 2019) or biomedical and pharmaceutical applications (Panda et al., 2012; Zhu et al., 2020).

To broaden the product range, the implementation of other nod genes (Mergaert et al., 1995; Samain et al., 1999) or the addition of different chitin deacetylases is also feasible (Hamer et al., 2015). Thereby, the cell factory approach might directly provide defined paCOS with known PA and/or other decorations.

As mentioned before, the COS composition is broader with E. coli (E. coli WT, NodC_S19L) compared to the more uniform product range of C. glutamicum (esp. NodC_R346S, Table 3), with the latter facilitating subsequent purification. Moreover, the obtained oligomers were not only longer, but the yield was also higher for C. glutamicum. One further drawback of COS production with E. coli could be the production of endotoxins, which makes them unsuitable for certain applications (Lieder et al., 2013), or requires costly removal (de Vries et al., 2018). In contrast, C. glutamicum does not produce endotoxins, is non-sporulating, and generally regarded as safe (GRAS) (Blombach and Eikmanns, 2014; Lee et al., 2016). Unlike E. coli, C. glutamicum shows only weak catabolite repression (Baritugo et al., 2018) and cannot metabolize COS naturally (Uhde et al., 2013; Matano et al., 2014).

In summary, C. glutamicum is a more promising candidate for COS production. As a known workhorse from industrial applications, biotechnological tools for modification, knowledge about metabolic pathways, and efficient cultivation systems are readily available (Lee et al., 2016; Baritugo et al., 2018). Moreover, with a possible addition of further genes modifying COS, the product range can be expanded, potentially allowing cost-effective large-scale production of customized oligomers.