A genome-scale metabolic model for Methylococcus capsulatus predicts reduced efficiency uphill electron transfer to pMMO

Background Genome-scale metabolic models allow researchers to calculate yields, to predict consumption and production rates, and to study the effect of genetic modifications in silico, without running resource-intensive experiments. While these models have become an invaluable tool for optimizing industrial production hosts like E. coli and S. cerevisiae, few such models exist for one-carbon (C1) metabolizers. Results Here we present a genome-scale metabolic model for Methylococcus capsulatus, a well-studied obligate methanotroph, which has been used as a production strain of single cell protein (SCP). The model was manually curated, and spans a total of 877 metabolites connected via 898 reactions. The inclusion of 730 genes and comprehensive annotations, make this model not only a useful tool for modeling metabolic physiology, but also a centralized knowledge base for M. capsulatus. With it, we determined that oxidation of methane by the particulate methane monooxygenase is most likely driven through uphill electron transfer operating at reduced efficiency as this scenario matches best with experimental data from literature. Conclusions The metabolic model will serve the ongoing fundamental research of C1 metabolism, and pave the way for rational strain design strategies towards improved SCP production processes in M. capsulatus.

The data that support the biomass equation constructed in this study 1 7 are available from Unibio at 1 8 • http://www.unibio.dk/end-product/chemical-composition-1 1 9 4 • http://www.unibio.dk/end-product/chemical-composition-2 1 Restrictions may apply to the availability of these data. Data are 2 however available from the authors upon reasonable request and with 3 permission of Unibio.

4
Since the three modes relate to how the pMMO receives electrons, we 5 focused on the data generated by growing M. capsulatus in high-copper 6 medium, which is the condition in which pMMO is predominantly active.

3
We altered the efficiency of the three modes to determine whether the 4 fit could be improved. For the redox arm, we gradually decreased the 5 mol protons required for the synthesis of 1 mol ATP, thereby improving 6 the efficiency. This did not change the O 2 /CH 4 ratio (See Supplement 7 Table 2). We decreased the efficiency of the direct coupling mode  Table 2). Because of this, we consider a 1 4 direct coupling to be possible, albeit unlikely. Lastly, we iteratively 1 5 constrained the lower bound of the reaction associated with the 1 6 ubiquinol-cytochrome-c reductase (CYOR_q8ppi), to reduce the 1 7 efficiency of the uphill-electron transfer.

8
As Figure 7B shows, by constraining the reverse flux through this 1 9 reaction, it is possible to modulate the ratio of O 2 /CH 4 consumption. We to avoid an overlap with the effects of NO 2 production. Leak and Dalton 1 pointed out, that the unexpectedly high O 2 /CH 4 ratio of 1.6 was the 2 product of latent NH 4 oxidation rather than assimilation [19] leading to 3 elevated levels of NO 2 . They were uncertain whether this increase could 4 be attributed to the energetic burden of reducing NH 4 or the uncoupling 5 effect of NO 2 .

6
To investigate this effect, we introduced a ratio constraint (see Methods) 7 coupling the uptake of NH 4 to the excretion of NO 2 and explored a 8 number of values for this ratio ( Figure 7C). According to the simulations, 9 the energy spent on reducing about 50% of incoming NH 4 to NO 2 is 1 0 sufficient to account for the observed, high O 2 /CH 4 ratio of 1.6. Although 1 1 this shows that the loss of energy could be significant enough to 1 2 account for the increased ratio, this does not exclude a potential   Table 1).

2
Unsurprisingly, the automated draft generally performs quite poorly in 3 Neither the draft model nor iMb5G(B1) make this distinction.

4
In the automated draft, the oxidation of methane was only possible 5 through a reaction representing the sMMO (MNXR6057). In iMcBath,

4
Throughout the reconstruction process, we iteratively tested and 1 5 validated the reconstruction. For instance, we checked the mass and 1 6 charge balance of each reaction, attempting to manually balance those 1 7 that weren't balanced. In the majority of cases metabolites were missing 1 8 formula or charge definitions. In order to remove energy generating 1 9 cycles, problematic reactions were manually constrained to be 2 0 irreversible. Validation was carried out against growth data [19], which 2 1 was also the point of reference for the parameter fitting.

9
Biomass Composition 1 For the principal components of biomass, measurements were made 2 available through the website of our collaborators Unibio [23]. This 3 included specifically the content of RNA (6.7%), DNA (2.3%), crude fat 4 (9.1%), and glucose (4.5%) as a percentage of the cell dry weight. We 5 did not use the percentage values for crude protein (72.9%) and N-free 6 extracts (7.6%) as these measurements are fairly inaccurate relying on 7 very generalized factors. The percentage value of Ash 550 (8.6%) 8 measurements was inconsistent with the sum of its individual 9 components (4.6%) and was hence excluded.

0
On the homepage of UniBio, we were also able to find g/kg 1 1 measurements of all amino acids except for glutamine and asparagine, 1 2 trace elements and vitamins, which could directly be converted into 1 3 mmol/g DW. However, we omitted some of data: The stoichiometries for 1 4 Selenium, Cadmium, Arsenic, Lead and Mercury were not included in 1 5 the biomass reaction as their values were negligibly small. Beta-

6
Carotene (Vitamin A) and Gama-Tocopherol (Vitamin E) were omitted 1 7 because no genes were found supporting their biosynthesis, in addition 1 8 to both being reportedly below the detection limit [79].

9
For the lack of better measurements, and assuming that M. buryatense

7
We filtered the results and focused only on proteins with a final score 1 8 larger than 7.5, which the authors of PSORTb consider to be 1 9 meaningful. We combined this list with the M. capsulatus specific 2 0 entries from the TransportDB, which allowed us to use the PSORT-2 1 3 2 scores as an additional measure of confidence. At this point, 242 1 putative transport proteins were identified. From this list we then 2 selected all proteins which were predicted to transport metabolites and 3 were already included in the model, which reduced the number to 133.

4
Since for many of the entries, the exact mechanism of transport is 5 unknown, we combined the previously selected transporters with the 6 results from running BLAST against the TCDB. The e-value and 7 bitscore from BLAST provided yet another measure to confidently 8 assess the correctness of the automatic TransportDB predictions, and 9 the Transporter Classification-Numbers (TC-numbers) allowed us to 1 0 gather information on the mechanism of transport. This led to a list of 97 1 1 transport proteins with high confidence, which was filtered once more as 1 2 follows.

3
We checked the general description for a given specific TC-number, 1 4 and then considered the BLAST result to read about a given transporter 1 5 in more detail, especially with regards to finding the specific substrates.

6
When we were able to identify the corresponding transport reaction in 1 7 the BiGG database, we incorporated only the simplest, smallest set of 1 8 reactions. In cases of conflicts between our own BLAST results and the 1 9 automatic TransportDB transporter annotation, we preferentially trusted 2 0 the BLAST results. Thus we ultimately added 75 transporter-encoding 2 1 genes connected via GPR to 56 distinct transport reactions.    Ethics approval and consent to participate: 1 6 "Not applicable"