A novel high-throughput method for kinetic characterisation of anaerobic bioproduction strains, applied to Clostridium kluyveri

Hexanoic acid (HA), also called caproic acid, can be used as an antimicrobial agent and as a precursor to various chemicals, such as fuels, solvents and fragrances. HA can be produced from ethanol and acetate by the mesophilic anaerobic bacterium Clostridium kluyveri, via two successive elongation steps over butyrate. A high-throughput anaerobic growth curve technique was coupled to a data analysis framework to assess growth kinetics for a range of substrate and product concentrations. Using this method, growth rates and several kinetic parameters were determined for C. kluyveri. A maximum growth rate (µmax) of 0.24 ± 0.01 h−1 was found, with a half-saturation index for acetic acid (KS,AA) of 3.8 ± 0.9 mM. Inhibition by butyric acid occurred at of 124.7 ± 5.7 mM (KI,BA), while the final product, HA, linearly inhibited growth with complete inhibition above 91.3 ± 10.8 mM (KHA of 10.9*10−3 ± 1.3*10−3 mM−1) at pH = 7, indicating that the hexanoate anion also exerts toxicity. These parameters were used to create a dynamic mass-balance model for bioproduction of HA. By coupling data collection and analysis to this modelling framework, we have produced a powerful tool to assess the kinetics of anaerobic micro-organisms, demonstrated here with C. kluyveri, in order further explore the potential of micro-organisms for chemicals production.


S.1.2. Production kinetics 44
A randomized sampling scheme (Figure S.1) was constructed using R 1 . One scheme ( Figure  45 S.1) was made for 3 rows and 10 columns, and applied to both conditions in the 96-WP. This 46 experiment was performed in triplicate, i.e. 3 96-WP, of which one was in the platereader for 47 continuous readings (620 nm) every 15 min. Actual time into the experiment for each timepoint 48 is given in Table S   Quantifying biomass based on OD can give valuable information, considering the large amount 85 of OD data available in the high-throughput growth curve experiments. To link these 86 parameters, a culture grown on standard DSM52 medium in 200 mL batches was taken after 2 87 and 4 days. These cultures were centrifuged for 8 min at 8610g, supernatans was removed and 88 filtered over 0.20 µm filters to obtain cell-free spent medium. This spent medium was first used 89 to resuspend the centrifuged biomass in 1/5 th of the original volume and subsequently make a 90 six-fold serial dilution (1:1 cell suspension:spent medium). For each dilution, total cell count, 91 VSS and OD were determined. Total cell counts were determined by flow cytometry 2 . This 92 experiment was performed in triplicate, i.e. a dilution series was made from 3 independently 93  was put at 35°C for one minute, after which it increased from 35°C to 230°C at a rate of 10°C 111 min -1 where it was kept for 2 minutes. FID temperature was set at 250°C, carrier gas was 112 nitrogen gas at a flow rate of 1.09 mL min -1 . The GC was externally calibrated with a minimum 113 detection limit of 25 mg/L. 114

S.1.6. Model development for chemostat simulations 116
Ordinary differential equations (ODE) described in main text (Section 3.2.) were modified for 117 simulation of chemostat conditions. The modified equations for substrate consumption or 118 product formation (eq. (S1)) and biomass production (eq. (S2)) are given below. 119 = ( * 1 * µ 1 6 * + * 2 * µ 2 6 * ) * + * ( − ) (S1) For simulation of chemostat experiments 3 , the reported conditions were applied to the model 121 (see Table S.II). The concentrations of substrates (EtOH, AA) and products (BA, HA) reached 122 after simulation for 500h of chemostat operation were then compared to the reported average 123 concentration for each condition and plotted in Figure 4 of the main text. 124 H2 accumulating in the headspace can limit the metabolism of C. kluyveri by reducing the Gibbs 130 free energy (ΔG 1 ), i.e. the amount of energy liberated during the reaction. To evaluate the 131 importance of this limitation, thermodynamical calculations were performed, using data and 132 methods from literature 4 . Two situations were assessed: (1) ΔG 1 if only the first reaction step 133 is considered, and (2) ΔG 0 if only HA is an end product, i.e. the sum of both reaction steps (see 134   Table S.II.). For both situations, initial conditions were used cf. the DSM 52 medium, and ΔG 1 135 was calculated as the reaction progressed according to the stoichiometry of the reaction. The 136 limiting substrate was used to assess the reaction progression, where the first reaction was 137 limited by AA, and the second by EtOH availability.
To assess the impact of H2, this calculation was performed with and without accumulation of 139 H2 in the headspace. To calculate accumulation of H2, a Balch tube with a total volume of 26 140 mL was considered, with 11.83 mL of liquid medium. Futhermore it was assumed that: (i) no 141 substrate was used for biomass production, (ii) no H2 was lost, through diffusion or in the 142 metabolism, (iii) all H2 produced immediately dissolved from the liquid into the headspace, (iv) 143 pH does not change during the reaction, and, (v) initially a very low H2 partial pressure of 144 0.0001 atm (10.1 Pa) was present in the headspace. For the case without H2 accumulation, H2 145 partial pressure was assumed to stay constant at this same 0.0001 atm. 146 147 EtOH can be transferred through the headspace, according to concentration gradients. It appears 159 an equilibrium was established across all wells, and even in part through a physical barrier 160 (petroleum jelly). Diffusion to the environment might occur too, but 91.9% of all EtOH initially 161 present was recovered at the end of the experiment, not taking into account potential transfer to 162 the wells acting as evaporation buffer. Evaporation appears not to be troublesome for organic 163 acids, since (1) they usually behaved similarly in 96-WP and balch tubes, (2) no apparent 164 stoichiometric inconsistencies were observed as was the case with EtOH, and (3) initial pH was 165 7.85±0.14, implying all organic acids to be present in their anionic, non-volatile form. When a plastic film seal was used to physically separate each well, EtOH no longer migrated 172 between wells in an abiotic experiment. Furthermore, the seal also allowed the outer rims to be 173 used without evaporation issues. The seal used in this experiment was not sterilized, yet no 174 contamination of the medium was observed, likely due to the very restrictive medium 175 compostion. When richer media are used, or axenity is strictly necessary, sterile films could 176 also be used, but were not tested here. In a dilution series experiment, an OD was linked to VSS and total cell countsby flow 219 cytometryas well as VSS and total cell counts to each other. For a given OD, biomass 220 concentrations expressed as VSS will be lower after 4 days than at 2 days. However, the 221 opposite is true when looking at total cell counts; the same OD at 4 days implies more cells 222 present than it does at 2 days. This apparent contradiction implies the amount of cells per g VSS 223 increases between day 2 and 4confirmed by Figure

S.2.6. Model selection 233
The proposed models for each organic acid were calibrated using the data from the experiments 234 for that acid, i.e. the proposed models for AA were calibrated using data from experiments A 235 and B, BA using data from experiments E and F and HA using data from experiment G (Table  236 1, main text),. The calibrations of these models are summarised in Table S

*Bold text indicates best fit 242
Residuals are the sum of the difference between model prediction and experimental data for all data points 243

S.2.7. HA-toxicity 244
To confirm increased salinity was not the mechanism of HA-toxicity, a control experiment was 245 included in experiment G (main text, Table I that condition (n=3). 259

S.2.8. Parameter estimation 260
Fit appropriateness of the kinetic model was evaluated by a combination of 95% confidence 261 intervals (Table 3) estimated by linear approximation of the covariance matrix with the 262 inverse of the Fisher Information Matrix -, correlations between parameters (Table S.VI) and 263 normality of the model output deviation (Figure S.11). 264 The correlation matrix is shown in Table S

S.2.10. Production kinetics of Clostridium kluyveri in 96-WP 293
To validate the dynamic model for growth of and production by Clostridium kluyveri data was 294 collected of a batch growth experiment in 96-WP in the anaerobic closet. 295 A data quality control was performed by looking at replicability within a plate, as well as 296 replicability between plates. The first shows whether there is any influence due to the 297 disturbances during sampling (i.e. measuring OD, removing lid, sampling, refilling emptied 298 wells and putting back the lid), while the latter shows how replicable the experiment as a whole 299 is. 300 The replicability within a plate is excellent; average OD of the sampled wells and those

S.2.11. Assessment of buffering capacity 345
The medium used in these experiments contained less buffer capacity than the conventional 346 DSM52 medium. In Balch tubes, no CO2 was present in the headspace for buffering, while the 347 anaerobic closet contained 10% CO2 instead of the recommended 20% CO2. It was already 348 shown in the manuscript that final product concentrations in Balch tubes and 96-WP were very 349 similar, giving a first indication C. kluyveri was not majorly affected by this change. Secondly, 350 the final pH (Table S.VII), shows a similar trend. pH at the end of the experiment is slightly 351 higher in the anaerobic closet, but always within the pH range for optimum growth of C. 352 kluyveri 8 . Additionally, because the growth curve is log transformed before fitting the growth 353 curve, the early exponential phase will carry greater weight in determination of µ. In this phase 354 of growth, conditions are still close to initial conditions, and the buffering capacity is of lower 355 importance. 356