Application to Developing a Medium for Biomass Accumulation
To assess the utility of this blending-based approach, we next aimed to identify and optimize the concentration of media components beneficial for rapid biomass accumulation of P. pastoris in batch cultivation. We previously described a rich defined medium (RDM) (Matthews et al., 2017a), capable of high growth rates during biomass accumulation. One challenge encountered with this formulation, however, was that precipitates can form at higher pH values that require filtering during bulk preparations. Nonetheless, this medium provided a relevant comparison for assessing the medium realized with our new approach due to its prior demonstrated benefits relative to complex media. Following our modular approach, we improved biomass accumulation by optimizing the accumulated optical density at 600 nm after 24 hours of cultivation.
Algorithms for optimizing systems based on multiple dimensions are often sensitive to initial conditions used (Zakharova & Minashina, 2015). Given this potential confounding effect here, we tested first the effects of the types of carbon source, nitrogen source, and pH set point on biomass accumulation, using 1x YNB without amino acids or ammonium sulfate (YNB) to satisfy minimum requirements for the concentrations of trace elements. We conducted a full-factorial DOE using glycerol, glucose, and fructose as carbon sources; urea and ammonium sulfate as nitrogen sources; and potassium phosphate as a buffer with pH values of 5, 5.75, and 6.5. We selected initial concentrations of 40 g/L, 4 g/L urea or the N-mol equivalent for ammonium sulfate, and 10 g/L potassium phosphate, similar to values used in other media for Pichia pastoris (Matthews et al., 2017a). A least squares regression model, including individual, combination, and quadratic effects was fit to the log of optical density after 24 hours, a proxy variable for the average growth rate (R2 = 0.81). We determined that the two most significant model terms were the type of carbon source and the interaction of the nitrogen source with pH (Figure 2A ). We found that cells grew significantly faster on metabolically related sugars (glucose and fructose) than on the polyol (glycerol) commonly used for Pichia during biomass accumulation (Figure 2B ). This result affirms prior reports where glucose has been used for biomass accumulation of Pichia (Guo et al., 2012; Moser et al., 2017).
The model also suggested that poor biomass accumulation during cultivation resulted from a combination of ammonium sulfate as a source of nitrogen with low buffer pH (Figure 2B ). This outcome may result from the production of acidic species associated with cellular ammonium metabolism in the batch cultivation (Villadsen, 2015). Interestingly, the model indicated slightly greater biomass was achieved with urea instead of ammonium sulfate. The biomass accumulation of cultures grown with urea as a source of nitrogen were less sensitive to reduced pH values (~5). We observed, however, that cultivations at pH 5 showed extensive flocculation compared to those at 6.5. Given the insensitivity of urea-fed cultivations to buffer pH and the high solubility and potential for low-cost sourcing of fructose, we therefore chose to include fructose, urea, and a potassium phosphate buffer with a pH of 6.5 in our initial media formulation.
With this basal formulation determined, we next screened for concentration-dependent interactions of other key additives to the media and then optimized concentration-dependent parameters. Following the same approach for screening effects, we conducted a full factorial DOE over a broad range of media component concentrations: YNB (0.5, 1, 2x), fructose (10, 30, 50 g/L), urea (1, 4, 7 g/L), and potassium phosphate adjusted to a pH of 6.5 (4, 10, 16 g/L). The resulting model identified fructose as a concentration-sensitive parameter (R2=0.73) (Figure 2D ). Terms involving the concentration of YNB were also highly ranked, but not statistically significant. No significant interactions between components were identified in the model. We therefore sought to better understand the concentration dependence of fructose and YNB independently (Figure 2E ), over an 8-fold range of concentrations. As expected, biomass accumulation was highly sensitive to fructose concentration, with an optimum around 22.5 g/L of fructose. The concentration of YNB had minimal effect on biomass accumulation; the presence of trace elements supplied by YNB, however, was essential to growth. Based on these results, we chose concentrations of 22.5 g/L fructose, 1x YNB, 7 g/L urea, and 10 g/L potassium phosphate buffer. We reasoned that although biomass accumulation was relatively insensitive to the concentrations of YNB and urea, higher concentrations could provide improved media depth in future applications. We named this basal formulation DM1_dev0.
We next assessed what additional media components could improve biomass accumulation. To test over 60 different components individually would require over 60 individual solutions. Such an approach would scale linearly with new components; instead, we chose to screen groups of related components, using commercially available pre-mixed supplements. We compiled a library of 16 commercial supplements and industrially-relevant surfactants containing more than 60 unique components and screened their individual effect on biomass accumulation after 24 hours. In this way, we reasoned we could efficiently identify critical classes of components related to the phenotype of interest and potentially deconvolve specific individual additives of interest by inference. We used the recommended concentrations of each supplement as supplied in product information, or critical micelle concentrations, and prior knowledge for broad classes in yeast media to set reasonable screening concentrations (Supporting Information). We identified five beneficial and two detrimental supplements that significantly impacted biomass accumulation (padj < 0.02; 1-way-ANOVA) (Figure 2F ). In general, the results suggest that supplementation with amino acids and trace metals were beneficial for accumulating biomass, while two surfactants, Tween 20 and CHAPS, were detrimental. For this phenotype, the effects of vitamin and lipid supplements were minor; supplements from either supplement category were not significantly beneficial or detrimental to biomass accumulation. Our earlier experiments suggest that vitamins are essential but concentration agnostic (Figure 1E ), while lipid supplementation provides no clear benefit for biomass accumulation.
Based on these results, we chose to test whether combinations of supplements of amino acids and trace salts could yield synergistic improvements in biomass accumulation. We screened pairwise combinations of the five beneficial supplements of mixed composition and ranked the performance of our supplementation strategies (Figure 2G ). A combination of 1x MEM amino acids with 0.1 v/v% PTM1 salts resulted in the highest yield of biomass, though we observed strong performance from other combinations of amino acid and trace metal supplements. Based on these data, we chose to add MEM amino acids and PTM1 salts in our basal medium and optimized their concentrations (Figure 2H ).
Based on these results, we elected 0.1 v/v% PTM1 salts and 1x MEM amino acids, in order to balance the moderate benefits and potentially high costs of amino acids. We found, however, that the inclusion of the PTM1 salts in liter-scale preparations produced fine precipitates, which can impede sterile transfers in use. To overcome this challenge, we screened a broad range of PTM1 salts concentrations to identify the minimum concentration required for improved outgrowth performance (Figure 2I ). We found that PTM1 addition at concentrations as low as 0.0005 v/v% led to increased biomass accumulation. We therefore revised our PTM1 salts concentration to 0.01 v/v%, a concentration high enough to obtain the benefits of PTM1 supplementation without inducing precipitate formation. This formulation we named DM1.
Completing this series of optimizations with our iterative modular approach to define a new formulation of medium, we then compared with other common media used to grow P. pastoris . We evaluated the performance of this new optimized medium (DM1) relative to the unsupplemented basal medium (DM1_dev0), the rich defined medium (RDM) we had previously developed, and a common medium 4 v/v% glycerol BMGY. We found that DM1 yielded the highest biomass accumulation, with significantly higher biomass accumulation relative to RDM and BMGY (Figure 2J ). This result demonstrates the utility of our modular strategy here for media development that yielded an improved formulation for biomass accumulation compared to other common media with minimal time and labor investment, and without requiring complex analytical methods like mass spectrometry or RNA-sequencing.
Identifying media conditions important to heterologous protein production in K. phaffii
In addition to the time and labor savings of modular media development, our proof-of-concept experiments demonstrated that this approach creates a flexible medium that can be rapidly adapted to new growth phenotypes, as well as a data package that the identifies media conditions important to the phenotype of interest. We reasoned that these additional benefits could be particularly relevant for optimizing production of heterologous proteins. Understanding which media components contribute most significantly to productivity could improve culture performance and help identify important metabolic pathways or physiological effects for further study.
To develop a medium for improved production of a recombinant protein, we chose to use a strain engineered to secrete a rotavirus-derived subunit vaccine component, VP4-P[8], as a model protein. We have previously demonstrated that this viral antigen can be expressed at high titer under the control of the methanol-inducible pAOX1 promoter in BMMY media (Dalvie et al., 2020). Similar to our initial approach to optimize a medium for growing biomass, we first determined and optimized the concentrations of the sources for carbon and nitrogen, along with the pH. The expression of P[8] in the strain tested uses the methanol-dependent pAOX1 promoter for inducible expression, so we selected methanol as the initial carbon source. We then examined the impact of the source of nitrogen and buffer pH on titer. We conducted a full-factorial DOE using identical concentrations as those used to create a medium for accumulating biomass. The resulting model was visualized by ranking combinations of sources of nitrogen and buffer (Figure 3A ). The effects showed no interaction between these two factors. Urea was again identified as the preferred source of nitrogen while higher pH values led to improved secreted P[8] productivity. Unlike biomass accumulation, this pH dependence was observed across both nitrogen sources.
We next applied the same DOE to identify important concentration-dependent interactions that impact the production of P[8]. Unsurprisingly, the concentration of methanol was the most important factor, with possible minor effects from other components (Figure 3B ). We decided to screen further a 20-fold range in methanol concentrations using two formulations for remaining media components—the one determined for optimal cell growth (DM1) and the optimal base media formulation predicted by the quadratic model here (2x YNB, 1 g/L urea, 4 g/L potassium phosphate adjusted to a pH of 6.5). We found that production was relatively insensitive for concentrations of methanol ranging from 1-4 v/v%, with an optimum around 2% (Figure 3C ). We postulated that the rapid decline in productivity observed in these milliliter-scale cultures using concentrations >6 v/v% methanol was likely due to excess formation of toxic metabolic byproducts such as formaldehyde and hydrogen peroxide (Wakayama et al., 2016). Interestingly, the predicted optimal medium from this set of studies outperformed the medium we determined for accumulating biomass, suggesting that certain components of the basal medium may benefit protein expression more than cellular growth and underscores the value of optimizing media for specific phenotypes of interest. Based on these data in total, we defined a basal medium for production including 2x YNB, 2 v/v% methanol, 1 g/L urea, and 4 g/L potassium phosphate buffer adjusted to a pH of 6.5 (DM2_dev0).
Next, we examined which supplements could improve the performance of DM2_dev0. We added three chemical chaperones (TUDCA, sodium deoxycholate monohydrate (SDM), and valproic acid) (Kuryatov, Mukherjee, & Lindstrom, 2013; Uppala, Gani, & Ramaiah, 2017), two antioxidants (reduced glutathione (GSH) and N-acetyl cysteine (NAC)), and the chelator, K-ETDA, to the list of 16 supplements included in our original screen defined for biomass accumulation. Concentrations for these components were chosen based on product specifications, literature data, and prior experience (Supporting Information). Many of the 22 supplements screened improved production of P[8] (Figure 3D ). The top four ranking supplements comprised surfactants or lipids, which could modulate membrane fluidity and lipid metabolism (Butler, Huzel, Barnab, Gray, & Bajno, 1999; Degreif, Cucu, Budin, Thiel, & Bertl, 2019; Ritacco, Frank V; Yongqi Wu, 2018).
We then screened combinations of lipid supplements and surfactants to identify potential synergistic effects. We ranked the individual supplements and their combinations (Figure 3E ) according to the measured titers of P[8]. We found that the addition of a cholesterol-rich supplement yielded the highest secreted titers of P[8] (~50% improvement compared with supplement-free condition in initial screens). Interestingly, a synthetic cholesterol supplement alone did not substantially improve performance, suggesting the benefit results from a combination of fatty acids and surfactant components in the supplement (Supporting Information ). This conclusion is consistent with similar improvements observed from other supplements, such as linoleic acid-oleic acid-albumin (Figure 3D ).
Since no other synergistic effects were observed in the combination screen, we assessed the dependence of titer on the concentration of the cholesterol-containing supplement identified (Figure 3F ). Similar to our observations with cellular YNB used in the outgrowth media, we found that concentrations of the supplement as low as 0.2 v/v% were beneficial for protein expression, but that production was relatively insensitive to concentration (Figures 3F, 3G ). We then directly compared the supplemented medium to the original composition; the new supplemented media provided a 25% improvement in titer (p = 0.0006, one-tailed Welch’s T test). This new formulation with 1x cholesterol supplement, which we named DM2_dev1, was the result of one cycle of optimization using our method.
Components of the cholesterol supplement included fatty acids, cholesterol, and cyclodextrin, which are all are known to modulate membrane fluidity, a key parameter in vesicle trafficking (Cooper, 1978; Degreif et al., 2019; Mahammad & Parmryd, 2015). We reasoned that the addition of this supplement could therefore have synergistic effects with other supplements, but did not find any further supplementation that improved P[8] titers within our original screen (Figure 3H ). We, therefore, considered if there could be additional classes of beneficial supplements, absent from the original screen. Previous experiments demonstrated that P[8] productivity is highly sensitive to methanol concentration (Figure 3C ), so we wondered whether further modulation of central carbon metabolism could yield additional productivity gains.
Modification of central carbon metabolism is best accomplished by feeding cells alternative carbon sources, either entirely or as co-feeding substrates. Four co-fed substrates have previously been shown to be non-repressive of pAOX1: sorbitol, mannitol, trehalose, and alanine (Inan & Meagher, 2001). These substrates can be co-utilized with methanol without repressing the pAOX1 promoter, which controls expression of P[8]. We hypothesized that the introduction of supplemental carbon sources could enable further optimization of central carbon metabolism. We screened co-fed substrates individually and in 1:1 combinations at a total concentration of 20 g/L (a concentration similar to the optimal fructose and methanol concentrations observed in previous carbon source optimizations) (Figure 2E,3C ). Sorbitol co-feeding had the most beneficial effect, resulting in a ~80% increase in P[8] titer (Figure 3I ). Mannitol supplementation was also beneficial (~70% increase), while alanine and trehalose co-feeding were detrimental to productivity. While co-feeding carbon sources led to increased biomass yield during production, these differences did not account for the improved titer, as improvements in specific productivity (qp) of ~60% and ~45% were also observed for the sorbitol and mannitol co-fed conditions, respectively (Supporting Information ). Based on these data, we chose to include sorbitol as a supplemental carbon source for further study.
The addition of a supplemental carbon source could significantly impact central carbon metabolism. We, therefore, wondered how the inclusion of sorbitol might impact the optimal carbon feeding strategy. Examining total carbon source concentrations from 20 – 70 g/L, we compared the performance of cultures co-fed with sorbitol:methanol ratios of 3:1, 1:1, and 1:3 to a methanol-only control (Figure 3J ). All co-fed conditions outperformed the methanol-only control, suggesting that the presence of sorbitol is highly beneficial for producing P[8]. The titer was relatively insensitive to sorbitol:methanol ratios and carbon concentrations. Based on the data, we decided to use 2 v/v% methanol and 20 g/L of sorbitol for the final sorbitol-supplemented media named DM2.
Finally, we compared the P[8] titer obtained using DM2_dev0, DM2_dev1, and DM2 to other common production media for P. pastoris : BMMY and RDM. We found that DM2 led to a ~2x improvement in P[8] titers, relative to BMMY and RDM, up to 97\(\pm\) 2 mg/L.