Lignin conversion to β-ketoadipic acid by Pseudomonas putida via metabolic engineering and bioprocess development

Abstract

Bioconversion of a heterogeneous mixture of lignin-related aromatic compounds (LRCs) to a single product via microbial biocatalysts is a promising approach to valorize lignin. Here, Pseudomonas putida KT2440 was engineered to convert mixed p-coumaroyl– and coniferyl-type LRCs to β-ketoadipic acid, a precursor for performance-advantaged polymers. Expression of enzymes mediating aromatic O-demethylation, hydroxylation, and ring-opening steps was tuned, and a global regulator was deleted. β-ketoadipate titers of 44.5 and 25 grams per liter and productivities of 1.15 and 0.66 grams per liter per hour were achieved from model LRCs and corn stover-derived LRCs, respectively, the latter representing an overall yield of 0.10 grams per gram corn stover-derived lignin. Technoeconomic analysis of the bioprocess and downstream processing predicted a β-ketoadipate minimum selling price of $2.01 per kilogram, which is cost competitive with fossil carbon-derived adipic acid ($1.10 to 1.80 per kilogram). Overall, this work achieved bioproduction metrics with economic relevance for conversion of lignin-derived streams into a performance-advantaged bioproduct.

An engineered bacterium efficiently converts plant biomass-derived aromatic compounds to a performance-advantaged bioproduct.

INTRODUCTION

Lignin, an aromatic polymer that comprises 15 to 45% of terrestrial plant biomass (1), is an underutilized by-product in lignocellulosic biorefineries. Conversion of lignin to valuable chemicals is therefore of interest to maximize the carbon efficiency, economic viability, and sustainability of lignocellulosic biorefining (2, 3). A central challenge in lignin valorization is the intrinsic chemical heterogeneity of lignin and its depolymerization, regardless of the approach, results in a mixture of chemicals (4–7). Biological funneling, the use of a microbial biocatalyst to convert a chemical mixture to a convergent bioproduct, has emerged as a promising strategy toward overcoming the challenge of heterogeneity (8–12). In addition, using this approach, the bioproduct is tunable by engineering the microbial host (9–12).

Biological funneling for the conversion of lignin-related aromatic compound (LRC) mixtures to performance-advantaged bioproducts (13, 14) has been demonstrated, and some of the associated bioproducts include 2-pyrone-4,6-dicarboxylic acid (15–17), itaconic acid (18), polyhydroxyalkanoates (8, 19, 20), cis,cis-muconic acid (21–24), vanillin (25, 26), substituted styrene molecules (27), pyridine-2,4-dicarboxylic acids (28, 29), and β-ketoadipic acid (30), among others, reviewed recently by Weiland et al. (12). β-Ketoadipic acid is a six-carbon dicarboxylic acid with a β-ketone (Fig. 1) that can be polymerized with hexamethylenediamine into a nylon-6,6 analog with reduced water permeability relative to nylon made using petroleum-derived adipic acid (31, 32), making it a promising chemical precursor for a performance-advantaged bioproduct.

Fig. 1. Metabolic pathway for the biological conversion of p-coumarate and ferulate to β-ketoadipate.

The genetic modifications applied to Pseudomonas putida are depicted with an “X” or a filled circle with a “+” for gene deletion or gene overexpression, respectively. N/E, nonenzymatic.

The β-ketoadipate pathway is found in diverse soil microbes (33) and enables convergent and atom-efficient (1 mol/mol) conversion of aromatic LRCs, such as the hydroxycinnamic acids p-coumarate and ferulate (34), to β-ketoadipate, which is further catabolized to enter the tricarboxylic acid cycle (Fig. 1). Pseudomonas putida KT2440 (hereafter P. putida) is a Gram-negative soil bacterium with fast growth, high tolerance to cytotoxic compounds (35), extensive genome engineering tools (36, 37), and a native β-ketoadipate pathway. For these reasons, P. putida is widely regarded as a promising chassis for industrial bioconversion processes (38, 39), including for complex substrates such as those derived from lignin (12) and plastics (40, 41). P. putida was previously engineered for the conversion of protocatechuate (42), 4-hydroxybenzoate (31), or p-coumarate (43) to β-ketoadipate by the deletion or repression of pcaIJ, encoding the 3-oxoadipate coenzyme (CoA) transferase PcaIJ. However, metabolic bottlenecks now limit β-ketoadipate productivity, and therefore, further strain and bioprocess development are necessary to achieve economically viable LRC bioconversion to β-ketoadipate.

In this work, we engineered P. putida and developed bioprocesses for the conversion of mixed LRCs to β-ketoadipate to improve cell performance and overall productivities, as this has been identified as a primary cost driver (6). Specifically, we report that overexpressing genes that encode enzymes mediating 4-hydroxybenzoate hydroxylation and vanillate O-demethylation and deletion of the gene encoding the Crc global regulator reduces intermediate accumulation. Bioprocess development with strains that harbor these changes resulted in conversion of p-coumarate and ferulate to β-ketoadipate at productivities exceeding 1 g/liter per hour and yields of 1.0 mol/mol (100% of theoretical). Last, LRCs extracted from lignin-rich liquor derived from the alkaline pretreatment of corn stover [alkaline pretreated liquor (APL)] were fed as solids to bioreactor cultivations, resulting in β-ketoadipate production of 25 g/liter at 0.66 g/liter per hour, with an overall yield of 0.10 g/g from corn stover–derived lignin. Given these process parameters, technoeconomic analysis (TEA) predicted a β-ketoadipate minimum selling price (MSP) of $2.01/kg and life cycle assessment (LCA) predicted a greenhouse gas (GHG) emission of 1.99 kg carbon dioxide equivalent (CO_2e)/kg, a cumulative fossil energy consumption of 24.8 MJ/kg, and a water usage of 32.4 liter/kg with the process.

RESULTS

Fed-batch bioreactor cultivations with constant feeding rates of p-coumarate maximize β-ketoadipate productivity

Previously, we engineered P. putida to convert 4-hydroxybenzoate to β-ketoadipate by the deletion of pcaIJ, generating strain CJ263 (Fig. 1 and Table 1) (31). To establish maximum titers and productivities with CJ263 and identify metabolic bottlenecks in this base strain, we compared two cultivation strategies in pH-controlled bioreactors that differ in the feeding mode: dissolved oxygen (DO)–stat and constant fed-batch. p-Coumarate, the most abundant aromatic compound in corn stover–derived APL (8, 34, 44, 45), was provided as the substrate for β-ketoadipate production, and glucose was supplemented as a carbon and energy source. Both feeding modes initiated when glucose was fully consumed in the batch phase and the DO increased to 75%. The feeding of p-coumarate and glucose in the DO-stat fed batch mode provided p-coumarate at 1 mM per pulse (concentration based on the volume in the batch phase). Alternatively, in the constant fed-batch mode, the feeding rate was fixed throughout the cultivation at 9, 12, 15, or 18 mM p-coumarate/liter per hour.

Table 1. Selected strains used in this study.

Table S1 shows the complete strain list and corresponding construction details.

Strain name	Genotype	Reference
P. putida	Wild-type P. putida KT2440	ATCC 47054
CJ263	P. putida ∆pcaIJ	Johnson et al. (31)
AW124	P. putida ∆pcaIJ ∆crc	This study
AW271	P. putida ∆pcaIJ fpvA:Ptac:praIJJ-1b:vanAB ∆pobAR	This study
AW297	P. putida ∆pcaIJ Ptac:pcaHG	This study
AW289	P. putida ∆pcaIJ ∆gacA	This study
AW287	P. putida ∆pcaIJ ∆gacS	This study
AW293	P. putida ∆pcaIJ ∆gacA ∆gacS	This study
AW156	P. putida ∆pcaIJ ∆lvaE	This study
AW299	P. putida ∆pcaIJ fpvA:Ptac:praIJJ-1b:vanAB ∆pobAR ∆lvaE ∆crc	This study
AW311	P. putida ∆pcaIJ fpvA:Ptac:praIJJ-1b:vanAB ∆pobAR ∆lvaE ∆crc Ptac:pcaHG	This study

DO-stat fed-batch enabled β-ketoadipate production of 43.5 ± 1.6 g/liter at 0.55 ± 0.02 g/liter per hour without accumulation of substrate or intermediates (Fig. 2 and fig. S1A). Increasing feeding rates in the constant fed-batch mode correspondingly increased productivity (fig. S1, B to D), with maximum β-ketoadipate production of 44.5 ± 1.85 g/liter at 0.85 ± 0.04 g/liter per hour observed at a feeding rate of 15 mmol/liter per hour (Fig. 2). At 18 mM/liter per hour, CJ263 cultivations were not reproducible and accumulated p-coumarate at varying levels (fig. S2), suggesting that CJ263 cannot efficiently use p-coumarate at this feeding rate. To improve the reproducibility and enhance productivity at 18 mM/liter per hour, the starting glucose concentration was increased from 15 to 30 mM to increase cell density in the batch phase before feeding. However, this change did not alleviate the inconsistency issues (fig. S2). Last, it is worth noting that at feeding rates of 15 mM/liter per hour, 4-hydroxybenzoate and p-coumarate accumulated in CJ263 cultivations (Fig. 2). These results indicated that constant fed-batch was the optimal strategy for maximizing productivity and that further strain development was warranted to relieve metabolic bottlenecks toward increased β-ketoadipate productivities.

Fig. 2. Production of β-ketoadipate from p-coumarate by CJ263 in bioreactors using DO-stat and constant fed-batch feeding modes.

The profiles show (A) β-ketoadipate titer, (B) β-ketoadipate instantaneous productivity, (C) p-coumarate accumulation, and (D) 4-hydroxybenzoate accumulation in bioreactors. Data show the average of biological duplicates, and error bars represent the absolute difference between biological duplicates. Single plots with all metabolites and cell growth are provided in fig. S1. Numerical data are provided in data S1.

Overexpression of the p-hydroxybenzoate-3-hydroxylase praI and the vanillate O-demethylase vanAB

To alleviate the metabolic bottlenecks observed in CJ263, we first considered key enzymes in the conversion of p-coumarate and ferulate, the second-most abundant aromatic compound in corn stover–derived APL, to β-ketoadipate (Fig. 1). We recently showed that replacing the native p-hydroxybenzoate-3-hydroxylase, pobA, with the heterologous praI_JJ-1b from Paenibacillus sp. JJ-1b improves the conversion of p-coumarate to muconate, another diacid derived from protocatechuate (46). Whereas the p-coumaroyl–type LRC p-coumarate requires 4-hydroxybenzoate hydroxylation, the coniferyl-type LRC ferulate instead requires vanillate O-demethylation to converge at protocatechuate (47). Previously, significant improvements in muconate production were demonstrated by constitutively overexpressing the native vanillate monooxygenase, vanAB, which increased the ferulate conversion by decreasing vanillate accumulation (22). Thus, both praI_JJ-1b and vanAB were expressed by the strong and constitutive tac promoter (P_tac) from the chromosome at the fpvA locus in CJ263, and pobAR was deleted, generating strain AW271 (Table 1 and tables S1 to S3).

To evaluate the effect of these changes, AW271 and CJ263 were cultivated in shake flasks with M9 minimal media supplemented with equimolar p-coumarate, ferulate, and glucose (20 mM each), and glucose was fed every 24 hours to a concentration of 20 mM. CJ263 incompletely used ferulate, accumulated a substantial amount of vanillate, and produced 26.8 ± 1.2 mM (4.29 ± 0.19 g/liter) β-ketoadipate at a molar yield of 0.68 ± 0.03 at 48 hours (Fig. 3 and fig. S3A). Compared to CJ263, AW271 exhibited a reduction in vanillate accumulation but similar substrate utilization (Fig. 3A and fig. S3B). At 48 hours, AW271 produced 29.4 ± 1.2 mM (4.71 ± 0.19 g/liter) β-ketoadipate at a molar yield of 0.79 ± 0.04. Protocatechuate accumulation was higher in AW271 compared to CJ263, suggesting that the slight reduction in vanillate accumulation may have exacerbated the downstream protocatechuate bottleneck.

Fig. 3. Strain engineering for improved conversion of p-coumarate and ferulate to β-ketoadipate.

Metabolite abundances as a function of time in the base strain CJ263 and further engineered strains including (A) overexpression of vanAB and replacement of pobAR with praI_JJ-1b in CJ263 to generate AW271, (B) overexpression of pcaHG in CJ263 to generate AW297, and (C) deletion of crc in CJ263 to generate AW124. Data show the average of three biological replicates. Error bars represent the SD among biological triplicates. Single plots including all metabolites and cell growth are provided in fig. S3. Numerical data are provided in data S1. Genotypes are provided in Table 1.

Overexpression of the 3,4-dioxygenase pcaHG

Protocatechuate ring-cleavage to 3-carboxy-cis,cis-muconate is mediated by the protocatechuate 3,4-dioxygenase PcaHG (Fig. 1). To mitigate the protocatechuate bottleneck, we constituitively overexpressed this enzyme by chromosomally integrating P_tac upstream of pcaHG in CJ263, generating strain AW297 (Table 1). AW297 displayed complete utilization of both p-coumarate and ferulate by 48 hours (Fig. 3B and fig. S3C). In AW297, transient accumulation of 4-hydroxybenzoate and vanillate were both significantly reduced, whereas protocatechuate was not; this result suggests that increased flux from protocatechuate to 3-carboxy-cis,cis-muconate relieved the accumulation of upstream intermediates, but that this alone was not sufficient to completely ameliorate the observed bottlenecks. Regardless, the strain improvements enabled higher β-ketoadipate productivity and at higher yields compared to CJ263, with AW297 producing 35.5 ± 2.6 mM (5.68 ± 0.42 g/liter) β-ketoadipate at a molar yield of 0.94 ± 0.07.

Deletion of the catabolite repression control regulator crc

We next evaluated the effect of global metabolic regulators on the bacterial performance, starting with the catabolite represson control (Crc) regulator. Crc is a global regulator of carbon assimilation in P. putida (48), where it acts as a repressor by inhibiting ribosome binding to mRNA to prevent translation (49). Deletion of crc has been shown to improve conversion of p-coumarate to muconate in P. putida when glucose was provided as a carbon and energy source (22, 50), as was done in this study. To evaluate the effect of Crc repression on β-ketoadipate production, crc was deleted in CJ263, generating strain AW124 (Table 1).

AW124 displayed faster substrate utilization compared to CJ263, with complete utilization of both p-coumarate and ferulate by 24 hours (Fig. 3C and fig. S3D). Faster p-coumarate and ferulate utilization are likely due to derepression of fcs, ech, and vdh, all of which have putative Crc binding sites near their initiation codon and convert these substrates to 4-hydroxybenzoate and vanillate, respectively (49). AW124 also displayed reduced accumulation of 4-hydroxybenzoate and vanillate, the latter of which is likely due to derepression of vanAB. β-Ketoadipate production was improved in AW124, with a titer of 38.1 ± 0.4 mM (6.10 ± 0.06 g/liter) and a molar yield of 0.93 ± 0.01 at 24 hours. These results indicate that, among the genetic modifications tested, the knockout of Crc is a useful modification to enhance β-ketoadipate production.

Deletion of global regulators gacA and/or gacS

The two-component system response regulator GacA and the sensor kinase GacS have been reported to regulate many physiological functions in Gram-negative bacteria (51), including cell adhesion and biofilm formation via LapA and LapF in P. putida (52). Deletion of gacS in P. putida improved conversion of glucose to muconate (53), and deletion of gacAS in P. putida improved conversion of p-coumarate to indigoidine (54). Given these findings, we deleted gacA (AW289), gacS (AW289), or gacA and gacS (AW293) (Table 1). GacA deletion resulted in reduced substrate utilization, with less than half of the initial substrate converted (figs. S3E and S4A); conversely, gacS deletion resulted in faster p-coumarate and ferulate utilization but suffered higher accumulations of vanillate and thus no improvement in β-ketoadipate production (figs. S3F and S4B). The ∆gacS ∆gacA double mutant displayed a phenotype similar to deletion of gacS alone (figs. S3G and S4C). Given the reduced β-ketoadipate production, these targets were not pursued further.

Deletion of levulinic acid catabolic gene lvaE

β-Ketoadipate can undergo abiotic decarboxylation to levulinic acid (32). P. putida can catabolize levulinic acid (55), and therefore, we considered whether carbon loss via levulinate (which could lead to innacurate metabolic yield calculations) could be mitigated by knocking out this capacity. LvaE, annotated as a short/medium chain acyl-CoA synthetase in BioCyc, was shown by Rand et al. (55) to catalyze the conversion of levulinate to levulinyl-CoA; the authors report that LvaE is sufficient to conduct this biochemical step, but that redundant and unidentified enzyme(s) additionally catalyze this reaction. Here, we deleted lvaE in CJ263 to generate AW156; as anticipated, no beneficial or detrimental effects on growth, substrate utilization, intermediate accumulation, or β-ketoadipate production were observed in AW156 compared to CJ263 (figs. S3H and S5). This result indicates that the levulinic acid production and/or utilization by P. putida in these experimental conditions is insignificant, as evinced by the theoretical product yields. Nevertheless, the conversion of β-ketoadipate to levulinic acid and its subsequent catabolism could be important in other cultivation conditions and have included this deletion in strains further engineered for β-ketoadipate production.

Integration of beneficial genetic interventions into a single production strain

We next sought to combine the beneficial genetic modifications—overexpression of praI_JJ-1b, overexpression of vanAB, overexpression of pcaHG, deletion of crc, and deletion of lvaE—into a single strain. Deletion of crc and overexpression of pcaHG showed similar phenotypes (Fig. 3), and therefore, we considered whether ∆crc alone would prove advantageous given the metabolic burden of protein overexpression. To evaluate this, two strains were constructed and compared to CJ263 in shake flasks: AW299 (with ∆crc, among other modifications) and AW311 (with ∆crc and P_tac:pcaHG, among other modifications) (Table 1).

AW299 produced β-ketoadipate at a titer of 37.5 ± 1.7 mM (6.00 ± 0.27 g/liter) and a molar yield of 0.99 ± 0.04 at 35 hours, and AW311 produced β-ketoadipate at a titer of 35.9 ± 2.4 mM (5.75 ± 0.38 g/liter) and a molar yield of 0.91 ± 0.06, both of which are improvements relative to CJ263 (Figs. 4 and fig. S6). Overexpression of pcaHG alone in AW297 did not reduce protocatechuate accumulation (Fig. 3B), whereas it did reduce protocatechuate accumulation when combined with additional modifications in AW311 (Fig. 4C and fig. S6B), suggesting interplay between intermediates in the creation and alleviation of metabolic bottlenecks. Given these similar performance improvements in AW299 and AW311, we proceeded to bioreactor evaluations of both strains.

Fig. 4. β-Ketoadipate production from a mixture of p-coumarate and ferulate by the base strain (CJ263) and the integrated engineered strains (AW299 and AW311) in shake flasks.

Profiles show bacterial growth, substrate utilization, accumulation of aromatic intermediates, and β-ketoadipate production by the (A) base strain CJ263 and integrated engineered strains (B) AW299 and (C) AW311. Data show the average of three biological replicates. Error bars represent the SD among biological triplicates. Numerical data are provided in data S1. Genotypes are provided in Table 1. OD₆₀₀, optical density measured as absorbance at 600 nm.

Fed-batch cultivations with constant feeding rates of single LRCs in bioreactors

Constant fed-batch cultivations with either p-coumarate or ferulate were conducted to compare the performance of strains CJ263, AW299, and AW311. At a feeding rate of 18 mmol p-coumarate/liter per hour, and as described above (fig. S2), CJ263 performed inconsistently. Thus, we selected this feeding rate to evaluate if the engineered strains exhibit improved performance. At 18 mmol p-coumarate/liter per hour and an initial glucose concentration of 30 mM in the batch phase, the performance of AW299 and AW311 was more consistent than that observed in CJ263 and produced a β-ketoadipate maximum titers of 39.1 ± 0.06 and 30.4 ± 8.52 g/liter at productivites of 1.09 ± 0.002 and 0.85 ± 0.24 g/liter per hour, respectively, and at quantitative yields (Figs. 5A and 6, and fig. S7, A and B). To identify the maximum productivity possible with these strains, we increased the feeding rate to 21 mM/liter per hour. At this rate, p-coumarate accumulated rapidly, reducing titers, productivities, and yields (Figs. 5A and 6, and fig. S7, C and D). No protocatechuate accumulated from p-coumarate fed at any rate tested (fig. S7). These results indicate that a constant feeding rate of 18 mmol p-coumarate/liter per hour (or 2.95 g/liter per hour) with AW299 and AW311 is a viable fed-batch strategy and that AW299 has a slightly improved performance as compared to AW311.

Fig. 5. β-Ketoadipate production from p-coumarate and ferulate individually or in a mixture by the base strain (CJ263) and the engineered strains AW299 and AW311 in bioreactors.

β-Ketoadipate was produced from (A) p-coumarate, (B) ferulate, or (C) a 3:1 molar mixture of p-coumarate:ferulate. (A) β-Ketoadipate, p-coumarate, and 4-hydroxybenzoate accumulation as a function of time in fed-batch cultivations with constant feeding of p-coumarate. CJ263 is omitted as it did not grow consistently at substrate feeding rates ≥ 18 mmol p-coumarate/liter per hour. (B) β-Ketoadipate, ferulate, and vanillate accumulation as a function of time in fed-batch cultivations with constant feeding of ferulate. AW311 is omitted as it did not grow consistency under the 21 mmol ferulate/liter per hour condition. (C) β-Ketoadipate, p-coumarate, and ferulate as a function of time in fed-batch cultivations with constant feeding of 3:1 (molar) p-coumarate:ferulate. Error bars represent the absolute difference between duplicates or SD for experiments with more than two biological replicates. Single plots with all metabolites and cell growth are provided in figs. S7 to 10. Numerical data and number of replicates for each experiment are provided in data S1.

Fig. 6. β-Ketoadipate titers, productivities, and yields for AW299 bioreactor cultivations with various LRCs.

Productivities and yields corresponding to the maximum titer are displayed in solid bars; maximum instantaneous productivity and yield data are shown in dotted bars. The data show the average of two biological replicates. Error bars represent the absolute difference between duplicates. Numerical data are provided in data S1. A summary and comparison to previously reported titers and productivities is provided in table S5. n/a, not applicable. APL, alkaline-pretreated liquor.

We then evaluated β-ketoadipate production from ferulate. Because of the reduced ferulate solubility [250 mM at a pH of 7.5 (22)] compared to p-coumarate (520 mM at pH 7) and bioreactors volume limitations, the maximum product titer we could theoretically achieve from ferulate in the same cultivation conditions as those reported above for p-coumarate was ~22 g/liter, which is a 55% reduction compared to bioreactor cultivations fed with p-coumarate only. To maximize titers with a diluted feed, the batch volume was reduced to the minimum volume that still allowed the submersion of the two impellers at the start of the fermentation (from 250 to 200 ml). At a feeding rate of 18 mmol ferulate/liter per hour, AW299 displayed reduced intermediate accumulation compared to CJ263, whereas AW311 accumulated ferulate (Fig. 5B and fig. S8). The improved performance of AW299 was also observed at a feeding rate of 21 mM/liter per hour: CJ263 and AW299 produced maximum β-ketoadipate titers of 7.2 ± 1.95 and 17.0 ± 0.84 g/liter at productivies of 0.28 ± 0.08 and 0.65 ± 0.03 g/liter per hour, respectively (Figs. 5B and 6). AW311 grew inconsistently at 21 mmol ferulate/liter per hour (fig. S9), and thus it is not shown in Fig. 5. Under all conditions, CJ263 accumulated vanillate, whereas none was detected in AW299 or AW311 cultivations. Protocatechuate did not accumulate under any condition (fig. S8). These results demonstrate that the genetic modifications in AW299 enable faster production of β-ketoadipate from ferulate at a substrate feeding rate of 21 mM/liter per hour (4.08 g/liter per hour) compared to both CJ263 and AW311, although with some ferulate accumulation.

Fed-batch cultivations with constant feeding rates of mixed LRCs in bioreactors

We next evaluated the effect of mixed p-coumarate and ferulate feeds on the performance of AW299 and AW311. The ratio of LRCs in lignin-derived streams depends on the biomass source and the pretreatment or lignin depolymerization conditions; in the case of the APL from the corn stover batch we used, p-coumarate and ferulate are present in a ~ 3:1 molar ratio (8). Thus, we evaluated a feeding solution that contained a molar ratio of 3:1 p-coumarate:ferulate. At a feeding rate of 18 mM/liter per hour total LRCs, both AW299 and AW311 grew consistently and achieved titers of 35.1 ± 0.20 and 29.8 ± 0.34 g/liter at productivites of 0.98 ± 0.01 and 1.01 ± 0.01 g/liter per hour, respectively (Figs. 5C and 6, and fig. S10). In both strains, the substrates accumulated at lower p-coumarate:ferulate ratios than those used in the feed (pCA/FA = 1.35 and 1.95 at 36 hours in AW299 and AW311 at 18 mM/liter per hour, respectively; table S4), indicating that p-coumarate is used faster than ferulate. No accumulation of the intermediates 4-hydroxybenzoate, vanillate, or protocatechuate were observed (fig. S10).

We further evaluated the effect of increased feeding rates (24 mM/liter per hour) on strain performance. Substrates accumulated almost immediately, and β-ketoadipate titer and productivity decreased substantially (Figs. 5C and 6). Again, p-coumarate and ferulate accumulated in ratios < 3 (pCA/FA = 2.17 and 2.01 at 23.8 hours in AW299 and AW311, respectively; table S4), indicating that p-coumarate was used faster than ferulate in these conditions. In all tested cases, AW299 outperformed AW311. In addition, no vanillin or 4-hydroxybenzaldehyde was detected (data S1), suggesting that transport and/or the first two CoA-ester intermediates (Fcs and Ech-catalyzed reactions; Fig. 1) are potential bottlenecks that hinder further β-ketoadipate productivity improvements from mixed LRCs feeds.

β-Ketoadipate production from LRCs extracted from APL in bioreactors

Last, we sought to convert LRCs from APL to β-ketoadipate. APL has been previously used for direct conversion to muconate (21, 22), but titers were limited (0.65 g/liter) due to the low concentrations of LRCs (45, 56). In this study, a separation protocol was developed to overcome this challenge by (i) removing high–molecular weight lignin from APL, (ii) removing the salts, and (iii) concentrating the LRCs in a solid state.

The composition of both the corn stover batch used to generate APL and the LRCs in APL are shown in tables S6 and S7, repectively. LRCs were extracted and concentrated from APL by acidification, liquid-liquid extraction, and filtration through a silica plug (hereafter termed APL extractives) (Fig. 7A and fig. S11) (45). Gel permeation chromatography showed that the APL extractives consisted of low–molecular weight lignin (fig. S12). In addition, heteronuclear single quantum coherence nuclear magnetic resonance spectroscopy peaks did not identify lignin linkages. These results indicate that oligomeric lignin was not present in the APL extractives (fig. S13). As expected, the primary LRCs in the APL extractives were p-coumarate [41.2% (w/w)] and ferulate [5.0% (w/w)], along with other less abundant LRCs (table S8). Overall, the extraction process yielded 0.15 g of LRCs/g total lignin in APL (table S9). We note that p-coumaric acid is not integrated into the lignin polymer, but rather it is ester pendent-linked to lignin and hemicellulose, as is a portion of the ferulate (34). In any case, both aromatic monomers are included as part of the total lignin content in the intial compositional analyses of corn stover.

Fig. 7. β-Ketoadipate production from corn stover–derived LRCs.

(A) Overall process flow and yields. A summary of yields is provided in table S9. (B) Profile of bioreactor cultivation conducted with AW299 and fed with solid APL extractives. LRC amount (g) fed in each pulse and accumulation of LRCs, aromatic intermediates, and β-ketoadipate are plotted as a function of time. Because of the substrate limitation, a single cultivation was performed. Numerical data are provided in data S1. LRCs quantified in APL and APL extractives are shown in tables S7 and S8.

Given the titer and productivity benefit observed in AW299, this strain was used for conversion of APL extractives to β-ketoadipate. The extracted solids were pasteurized and fed directly to the bioreactor in 0.5 g or 1 g pulses (Fig. 7B). Trace amounts of substrate and intermediates accumulated, with rapid conversion of p-coumarate and ferulate to β-ketoadipate. After eight solid feedings, a titer of 25 g/liter was achieved with a maximum productivity of 0.66 g/liter per hour and yield of 1.2 g/g LRCs in APL extractives (Fig. 6). This represents a yield of 0.10 g β-ketoadipate/g total corn stover lignin when including p-coumarate and ferulate in the lignin content (table S9).

TEA and life cycle assessment of LRC bioconversion to β-ketoadipate

To benchmark the process improvements and identify key economic and environmental drivers, TEA and LCA were conducted. We focused the process modeling only on the bioconversion and downstream processing section such that the TEA and LCA results could directly inform efforts associated with the bioprocess. We modeled the LRC cost based on previous work for APL generation and p-coumarate and ferulate production through a molecular weight fractionation process using membranes (57). The bioprocess was modeled as an aerobic cultivation in a bubble column fed with LRCs and glucose followed by chromatographic separation of β-ketoadipate via simulated moving bed (SMB) chromatography and concentration via evaporation (Fig. 8A and tables S10 to S13). Ammonium hydroxide was used for β-ketoadipate neutralization in the cultivations, and sulfuric acid was used as a desorbent in SMB adsorption columns. Experimentally determined process metrics were used for the bioconversion base case (40 g/liter, 1.15 g/liter per hour, and 1.0 mol/mol).

Fig. 8. Economic and environmental assessments for the production of 100,000 metric tons of β-ketoadipate per year from LRCs.

(A) Process flow diagram of the β-ketoadipate production process. (B) Breakdown of MSP of β-ketoadipate (βKA). (C) Single-point sensitivity analyses for MSP around main process parameters and configurations. Base case values are shown in parentheses along the y axis. (D) Sensitivity analysis of the isolated effect of productivity and titer on the MSP of β-ketoadipate. (E) Effect of productivity and LRC price on the MSP of β-ketoadipate. (F) Cradle-to-gate LCA of the scaled-up production of LRC-derived β-ketoadipate comparing without (“base”) and with ammonium sulfate recovery as a coproduct for GHG emissions, cumulative fossil energy consumption, and water consumption. Results for the scenarios with ammonium sulfate recovery were derived using the purpose-driven system-level allocation method based on products’ market values. All β-ketoadipate yields are on a molar basis (mol/mol, %). Sensitivity results for coproduct handling approaches, namely, mass-based allocation and coproduct displacement, are presented in fig. S14 and table S14.

In the base case model, the MSP of β-ketoadipate was predicted to be $2.01/kg, mainly driven by LRC ($0.96/kg), glucose ($0.24/kg), and capital ($0.28/kg) costs (Fig. 8B). Ammonium hydroxide and sulfuric acid accounted for another $0.17 and $0.10/kg, respectively. Historical prices (2010–2020) for adipic acid—the conventional diacid monomer for nylon-6,6—range between $1.10 and 1.80/kg in the United States (32), generally tracking the price of oil. Overall, the TEA predicts a β-ketoadipate MSP near to the historical price range of adipic acid.

Single-point sensitivity analyses were conducted to quantify the impact of selected process parameters and variables on the MSP of β-ketoadipate (Fig. 8C). As would be expected considering the high LRC cost, product yield was a key economic driver with a 20% reduction in yield incurring a penalty of $0.40/kg in the MSP of β-ketoadipate. The 100% yield observed here (Fig. 6) counterbalanced the higher cost of LRCs [$0.90/kg; (57)] compared to glucose [$0.32/kg; (58)] such that the MSP of $2.01/kg is similar to that where glucose is the sole substrate [$1.94/kg, 40% yield; (32)]. Still, the ratio between LRC and glucose and the price of glucose exhibit a linear impact on economics. Specifically, a variation of ± $0.10/kg in feedstock price led to a similar ± $0.10/kg impact on the MSP of β-ketoadipate.

To consider the effect of productivity on β-ketoadipate MSP, we performed sensitivity analyses at varying titers (10, 40, and 60 g/liter) and considered the combined effect of LRC price and productivity on the MSP of β-ketoadipate (Fig. 8, D and E). While there are limited economic gains for improving productivity beyond the baseline of 1.15 g/liter per hour, rates below ~0.5 g/liter per hour sharply increased the MSP. An improvement in productivity from the base up to 2 g/liter per hour would only decreased the MSP by $0.07/kg (3.5%). Similar sensitivitity analyses were performed to determine the effect of titer on MSP at fixed productivities (0.90, 1.15, 1.40 g/liter per hour; Fig, 8D). Notably, as residence time in bubble columns was not fixed in this study and rather calculated as a function of productivity and yield, the titer was only modulated through the amount of water entering alongside LRCs and glucose in production bioreactors. In this modeling scheme, the impact of titer was substantial below the baseline but subdued at higher levels. This assessment estimates that MSP will decrease by only $0.02/kg if titer is increased from 40 to 60 g/liter, after which MSP plateaus.

Recovery of ammonium sulfate was found to be another key economic driver. Ammonium hydroxide neutralized with sulfuric acid in the SMB columns forms ammonium sulfate that could be recovered and sold for use as a fertilizer (59) (table S11). The bioprocessing scenario modeled here sent ammonium sulfate to wastewater treatment, but recovering ammonium sulfate and selling this salt as a coproduct would reduce the MSP of β-ketoadipate to $0.31/kg.

Last, cradle-to-gate LCA was conducted for the scaled-up production of LRC-derived β-ketoadipate to estimate GHG emissions, fossil energy consumption, and life cycle water consumption (Fig. 8F). The LCA does not account for the environmental impacts from the feedstock but rather solely focuses on the modeled conversion process studied here, with the aim to provide insights (such as the TEA) on the bioconversion and downstream processing steps. Two common approaches for handling coproducts were considered: systems-level allocation, which divides inputs and outputs between the products of a system based on mass or market value, and coproduct displacement, which replaces an existing product with a new one. Sensitivity analysis on coproduct handling was conducted to avoid potential artifacts in product-specific LCA and to address the coproduct issue associated with the integrated biorefinery design.

The impact of ammonium sulfate recovery was considered for each analysis. Without ammonium sulfate recovery, GHG emissions were 1.99 kg CO_2e/kg, which was improved to 1.69 kg CO_2e/kg with recovery of ammonium sulfate as a coproduct. The key GHG drivers were chemical inputs—predominantly ammonium hydroxide (40%) and glucose (29%)—and process electricity (16%). Fossil energy consumption (FEC) profiles were consistent with GHG emissions results: Chemical input was the major contributor. Similarly, recovering ammonium sulfate as a coproduct also reduced the life cycle water consumption for the current bioprocess from 32.4 to 27.4 liter/kg; this reduction is mainly attributed to embedded chemical inputs and process water. Coproduct handling methods, sensitivity results using mass-based allocation, and coproduct displacement for ammonium sulfate are presented in fig. S14 and table S14.

DISCUSSION

This work demonstrates a bioconversion process from mixed model aromatic compounds and lignin-derived substrates to β-ketoadipate at high titers, rates, and yields. The high productivity coupled with theoretical yields from LRCs to β-ketoadipate led to an estimated MSP of $2.01/kg in a facility with an ouptut of 100,000 metric tons (MT)/year and operating with the best experimentally demonstrated metrics reported in this study. This price should be competitive with most diacid precursors used in the manufacturing of nylon-6,6, either fossil- or bio-based, such as adipic acid [global market of >3 million MT/year (32)]. Cost competitiveness was also observed for biological conversion of glucose to β-ketoadipate (32). Moving forward, reduced glucose demand, by-product recovery, and increased yields of bioavailable monomers from lignin are among the top process considerations for improvement. In addition, previous studies in our team demonstrated that β-ketoadipate can be isolated and purified from bioreactor broths for downstream polymerization efforts (31, 32).

In this work, we achieved β-ketoadipate productivities higher than 1 g/liter per hour from p-coumarate and ferulate (Fig. 7). At the highest feeding rate tested, these LRCs accumulated immediately and, to a greater extent, than quantified intermediates, which suggests that to exceed a productivity of ~1 g/liter per hour, the potential bottlenecks derived from Fcs- or Vdh-mediated reactions and/or import of aromatic compounds to the cell need to be relieved. To evaluate the former bottleneck, free CoA and adenosine 5′-triphosphate should be quantified, along with the CoA-esters, as these may be limiting feruloyl-CoA or p-coumaryl–CoA biosynthesis. To evaluate substrate import-related bottlenecks, transporter engineering to modulate the expression of the MFS sympoter hcnK (60) and/or the putative porin PP_3350 (61) could be considered. Previously, the deletion of PP_3350 was found to be beneficial at high (20 g/liter) concentrations of p-coumarate (61), although ideal expression levels may be dependent on the bioprocess strategy (e.g., batch versus fed-batch).

As expected, many of the genetic interventions previously shown to improve muconate production were also beneficial for β-ketoadipate production. An exception was deletion of gacA/S, which has been shown to be beneficial for production of muconate and indigoidine from p-coumarate or glucose (53, 54), but was unsuccessful here likely due to the mixing of substrates [e.g., fitness defects are observed in p-coumarate and glucose (62)]. While titers are similar for β-ketoadipate from p-coumarate and ferulate, β-ketoadipate from 4-hydroxybenzoate (31), and muconate from p-coumarate (46), all at ~40 g/liter, β-ketoadipate productivity was >2× higher in this study compared to muconate productivity [0.5 g/liter per hour; (46)]. The underlying reason for this difference in productivity is unknown; postulations include that cofactors uniquely required for muconate production [e.g., prenylated FMN for the protocatechuate decarboxylase, AroY; (63, 64)] hinder cellular productivity.

Further process development efforts could follow multiple tracks, many of which focus on reduction of consumable (e.g., LRC, glucose, and base) costs. To reduce glucose costs, using alternative carbon sources already present in the feed, such as acetate or glycerol in APL, for cell biomass generation should be explored. To reduce LRC costs, which can vary greatly by biomass source, biorefining configuration, depolymerization strategy (1, 7, 65), and the use of lower-purity lignin sources should be considered (66). Aqueous LRC feeds require NaOH, or another base, to solubilize the substrates, which is both an added cost and potential source of cytotoxicity. Strain engineering for sodium tolerance and the use of powder (solid) feeders may help reduce the amount of salt required, thereby reducing cost and mitigating sodium stress. In situ product removal is another approach to increase production by mitigating product toxicity (67).

TEA and LCA results predict that ammonium sulfate recovery could reduce the MSP and improve the bioprocess environmental sustainability by >15%. Recycling part of the ammonium sulfate back to bioreactors is a potential strategy to offset fresh nitrogen demand. However, to validate the GHG and FEC results for ammonium sulfate as a coproduct, a consequentional LCA that takes into account the market saturation and the rebound effect would need to be performed. Alternatively, the current baseline market value–based system-level allocation method for this multifuntional system can better capture each respective process train’s technical attributes in quantifying each individual product’s GHG emissions and other impacts on its own accord (68).

The development and integration of scalable upstream processes that provide high yields of bioavailable monomers will enable increased overall lignin conversion, a critical factor in the process economics (6). While this study displays some of the highest product titers and productivities from a lignin stream, the LRC extraction/separation approach used here was not optimized. The emergence of industrially relevant lignin monomer-oligomer separations methods based on membrane technologies could further advance the industrial viability of this approach (69). In addition, dimer cleavage has been reported in soil bacteria and engineered into P. putida for increased LRC conversion to product as a biological strategy to further use oligomeric lignin fractions and enhance conversion yields (11, 70). Thermal and catalytic methods have been shown to enable rapid, extensive, and scaleable lignin deconstruction (6). Chemocatalytic oxidative depolymerization of lignin provides bioavailable monomers at higher yields than alkaline pretreatment methods (71). Notably, recent advances in oxidative chemistry have enabled the cleavage of lignin C─C bonds, not accessible by traditional catalytic lignin depolymerization approaches (72). Complete lignin conversion to bioavailable monomers would be a transformational advance.

MATERIALS AND METHODS

Chemicals

All chemicals were purchased from Sigma-Aldrich unless otherwise noted, except for sodium chloride (Thermo Fisher Scientific), p-coumaric acid (AK Scientific, lot LC56543R), and ferulic acid (AK Scientific, lot LC57665).

Bacterial strains, media, and chemical preparations

All strains are derivatives of P. putida KT2440 ATCC 47054 (P. putida). P. putida derivatives were cultivated in Miller’s LB or M9 minimal media [Na₂HPO₄ (6.78 g/liter), KH₂PO₄ (3 g/liter), NaCl (0.5 g/liter), NH₄Cl (1 g/liter), 100 mM CaCl₂, and 18 mM FeSO₄] supplemented as described for each individual experiment. p-coumarate and ferulate stock solutions were prepared to a concentration of 100 mM in water by adding weighed solid and pH-adjusting slowly and without exceeding pH 9.5 by adding 5 N NaOH to a final pH of 7.5 to 8.0. The solubilized neutral pH solution was 0.2 μM filtered to sterilize before addition to media. Glucose was added from a filtered 2 M solution in water.

Plasmid and strain construction

Plasmids, oligonucleotides, and strains constructed and/or used in this study are described along with construction details in tables S1 to S3. Generally, plasmids were either synthesized directly by Twist Biosciences or assembled using HiFi Assembly (New England Biolabs) and directly transformed into Escherichia coli DH5a F′I^q (New England Biolabs). Plasmid construction was confirmed by colony polymerase chain reaction (cPCR) and Sanger sequencing (GENEWIZ) of the homology regions and insert. Sequence-confirmed plasmids were transformed into P. putida by electroporation, and markerless chromosomal integration was achieved by the sacB/KanR counterselection method, as previously described (73). Correct integration in resulting transformants was confirmed by cPCR, and correct colonies were stored as 20% (v/v) glycerol stocks at −80°C.

Shake flask cultivations

Cultures were revived from glycerol stocks in Miller’s LB medium for 12 to 16 hours, washed in M9 salts, and inoculated into fresh media at an optical density measured as absorbance at 600 nm (OD_600nm) of ~0.1. Glucose and aromatic compounds (p-coumarate and ferulate) were supplemented to the concentrations indicated for each experiment. Samples for metabolite quantification were taken by removing 1 ml of culture, centrifuging 2 min at >15,000g, passing through a 0.2 μM syringe filter into an amber silanized vial, capping, and storing at −20°C until analysis. Glucose was fed from a 2 M solution in water after sampling for OD₆₀₀ and metabolite analysis.

Bioreactor seed preparation

Cells from glycerol stocks were inoculated in 250-ml baffled flasks containing 50 ml of Miller’s LB medium, incubated overnight (~16 hours) at 30°C and 225 rpm, harvested by centrifugation at 6400g for 10 min, and resuspended in 5 ml of M9 medium of the corresponding bioreactor media. Resuspended cells were inoculated in 500-ml bioreactors (Sartorius Stedim Biotech) at OD₆₀₀ ~ 0.2.

Batch phase media preparation and bioreactor conditions

M9 media used consisted of glucose (2.7 or 5.4 g/liter) (15 or 30 mM, respectively), Na₂HPO₄ (13.5 g/liter), KH₂PO₄ (6 g/liter), NaCl (1 g/liter), (NH₄)₂SO₄ (2.25 g/liter), 2 mM MgSO₄, 0.1 mM CaCl₂, and 0.018 mM FeSO₄. A glucose concentration of 2.7 g/liter was only used for the initial comparison of DO-stat and constant fed-batch feeding strategies with CJ263. A glucose concentration of 30 mM was used in the rest of the experiments. For bioreactor cultivations with p-coumarate and mixed aromatic compound feeds, the starting volume was 250 ml; for cultivations with ferulate feeds, the starting volume was reduced to 200 ml while maintaining the same amount (g) of individual nutrients than that used at 250 ml to avoid nutrient dilution at maximum working volumes. Bioreactor conditions were maintained at 30°C and pH 7 with an air flow rate of 0.3 liter/min. The culture pH was automatically controlled by adding 4 M NH₄OH. Initial agitation and DO levels in the batch phase were 350 rpm and 100%, respectively. When DO levels reached approximately 30%, the agitation was automatically increased to maintain the DO setpoint at 30%. When DO levels reached ~75% by the depletion of glucose in the batch phase, either a DO-stat or continuous fed-batch phase began. At 4 hours, single aromatic feed cultivations were induced with 2 mM corresponding aromatic compound from 250 mM stocks, whereas mixed aromatic feed cultivations were induced with 1 mM ferulate and 1 mM p-coumarate from 125 mM stocks. Bioreactor experiments were performed in at least duplicate, and 1.5 ml of samples analyzed for OD₆₀₀ and metabolites. Cultivations were terminated when the feeding solutions were fully consumed, the DO level increased at the minimum agitation value (350 rpm), or if the maximum volume of the reactor was reached.

DO-stat fed-batch cultivations in bioreactors

The feeding solution contained glucose (67.5 g/liter), (NH₄)₂SO₄ (13.5 g/liter), p-coumaric acid (109 g/liter), and Antifoam 204 (4 ml/liter) and was fed in 1 mM p-coumarate pulses (based on the initial media volume in the batch phase) when the DO reached a value of 75%. The DO level was maintained between 15 and 75% with manual agitation adjustments. Glucose–to–p-coumaric acid ratio (~1:2 mol) was selected on the basis of a previous study conducted in our team that demonstrated that a 1:2 ratio was beneficial to maximize product titers from aromatic compounds and avoid the accumulation of catabolic intermediates compared to 1:4 and 1:8 ratios (22).

Constant fed-batch cultivations in bioreactors

A fixed feed rate based on the initial media volume in the batch phase and substrate feed concentration was provided. Feed rates were between 9 and 24 mM/liter per hour and depended on the experiment, as detailed in Results. p-Coumaric acid feed solution contained p-coumarate (109 g/liter), glucose (67.5 g/liter), (NH₄)₂SO₄ (13.5 g/liter), and Antifoam 204 (4 ml/liter) and was adjusted to pH 9.3 with 10 N NaOH. Ferulate feed solutions contained of ferulic acid (49 g/liter), glucose (22.5 g/liter), (NH₄)₂SO₄ (4.5 g/liter), and Antifoam 204 (4 ml/liter) and was adjusted to pH 7.5 with 10 N NaOH. Combined aromatic feeds consisted of a 3:1 molar ratio of p-coumaric acid to ferulic acid, containing p-coumaric acid (82.06 g/liter), ferulic acid (32.36 g/liter), glucose (60 g/liter), and (NH₄)₂SO₄ (12 g/liter) and was adjusted to pH 7.5 with 10 N NaOH. All feeds containing ferulate were prepared the same day as the experiment initiated, and all feeds were wrapped in foil during the cultivations.

Bolus fed-batch cultivation with solid APL extractives in a bioreactor

Cultivations proceeded as described above with exceptions noted here. The solid APL extractives were pasteurized for 1 hour at 70°C before use. The bioreactor batch media not only contained the same total nutrient content as that detailed before but also included 0.5 g of APL extractives. This cultivation was scaled down to 150 ml, and only the bottom impeller was submerged. A 50% (w/v) solution of glucose with ammonium sulfate (20 g/liter) and Antifoam 204 (4 ml/liter) was prepared as a liquid feed to support growth. This feed was added after the batch phase ended and was adjusted to add 1 mmol glucose/liter every 15 min during the cultivation. The agitation rate was maintained between 650 and 700 rpm during the fed-batch phase. Additional antifoam was added as needed directly in the bioreactor. The dosing of the APL extractives consisted of 0.5 g pulses at 13.7, 17.4, 20.6, 22.5, and 24.2 hours and 1 g pulses at 26.5, 30.6, and 35.2 hours. The cultivation was ended when the DO no longer dropped after the glucose feeds.

Quantification of substrates, intermediates, and products

p-coumaric acid, ferulic acid, 4-hydroxybenzoic acid, vanillic acid, and protocatechuic acid were analyzed as previously described (74) unless vanillin was also included as an analyte. Because of the analyte coelution of p-coumaric acid and vanillin with the previously developed method (74), samples that were analyzed for vanillin in addition to the analytes listed above were analyzed by the following ultrahigh-performance liquid chromatography method: Quantitation was performed using an Agilent 1290 Infinity II System (Agilent Technologies) coupled with a diode array detector. Samples and standards were injected onto a Luna Omega PS C18 100 A°, 1.6 μm, 2.1 × 100 mm column (Phenomenex) at a volume of 0.5 μl. The column temperature was held constant throughout the run at 35°C. Chromatographic separation was achieved using (A) acetonitrile and (B) 10 mM phosphoric acid in water as the mobile phases at a constant flow rate of 0.5 μl/min and the following gradient program: (A) = 5% and (B) = 95% at time = 0; (A) = 7.5% and (B) = 92.5% at time = 0.1 min; (A) = 10% and (B) = 90% at time = 1.67 min; (A) = 25% and (B) = 75% at time = 2.92 min; and (A) = 28% and (B) = 72% at time = 4.50 min; return to initial conditions at time = 4.51 min, (A) = 5% and (B) = 95% and hold for a total run time of 8 min. Calibration curves for each analyte of interest were constructed from a minimum of six calibration standards and had an r² coefficient ≥ 0.995. A quantitative wavelength of 265 nm was used for all analytes except for 4-hydroxybenzaldhyde, for which 280 nm was used. A calibration verification standard was run every 10 to 20 samples to monitor instrument drift and ensure calibration stability. β-Ketoadipate was analyzed as previously detailed (32). Representative chromatograms for aromatic substrates, intermediates, and product are provided in fig. S15.

Preparation of APL

Briefly, the APL was produced as follows: 5.0 kg of dry corn stover was loaded into a 90-liter paddle reactor and 27.4 kg of water plus 2600 g of 50% aqueous NaOH solution was added while it was mixing. The reactor was heated to 122°C for 30 min before being cooled and removed from the reactor. The APL was dewatered using a small Vincent screw press with a piston pressure of 10 psi.

Extraction of LRCs from APL

A representative procedure for the extraction of LRCs from APL is as follows (schematic representation in fig. S12): 973 g of APL at pH 14 was centrifuged for 1 hour at 38,400g and 4°C. The supernatant (“basic liquor”; 939 g) was collected, and the basic solid (35 g) was removed. The basic liquor was acidified to pH 1 via addition of 109 g of 37% HCl (aq), resulting in the formation of solid precipitate in the liquor. The acidic liquor was then centrifuged for 1 hour at 16,000 rpm and 4°C. The supernatant (“acidic liquor”; 870 g) was decanted, and the pellet (“acidic solid”; 173 g) was collected and partially dried in a vacuum oven overnight at 40°C. Both the acidic liquor and the acidic solid were extracted with ethyl acetate (EtOAc), producing 6.2 and 3.1 g of extract, respectively. Both extracted fractions were subsequently filtered through 100 g of silica plugs with 1:1 hexane:EtOAc producing 3.3 g of filtered acidic liquor extract and 1.5 g of filtered acidic solid extract. Both filtered extracts were combined and partially dissolved in hot water, and the solution decanted into another vessel and allowed to cool to room temperature before being placed in a cold room overnight. The resultant yellow/brown precipitate was filtered and washed with ice-cold water and dried overnight in a vacuum oven (1.9 g). For bioreactor cultivations, the filtered acidic extracts obtained from 7.3 kg of APL were combined for the precipitation step resulting 10.6 g of LRC solids. LRCs were quantified after solubilization and accounted for 44.5% of the extracted material; the composition was primarily p-coumarate and ferulate but included other aromatics (fig. S12 and table S6).

Gel permeation chromatography

Samples of 10- to 20-mg quantities were acetylated using 0.5 ml of pyridine (anhydrous; 99.8%; Sigma-Aldrich) and 0.5 ml of acetic anhydride (Sigma-Aldrich reagent plus ≥99%) sealed and heated to 40°C for 24 hours while stirring. Subsequently, 1-ml aliquots of methanol were then added to each sample and dried under nitrogen, N₂. This was repeated seven times. Samples were then diluted in tetrahydrofuran (THF) and stirred for 10 min. The THF solution was filtered through a 0.2-μm syringe filter into an high-performance liquid chromatography (HPLC) vial. A total of 20 μl of sample was injected on an HPLC fitted with three PLgel 7.5 × 300 mm columns in series: 10 μm × 50 Å, 10 μm × 103 Å, 10 μm × 104 Å (Agilent Technologies, Stockport, UK) at ambient temperature with an isocratic 100% tetrahydrofuran (1 ml min⁻¹; inhibitor-free; ≥99.9%; Sigma-Aldrich) for 40 min. Analytes are monitored at 210, 260, and 270 nm on the diode array detector (fig. S12).

NMR spectroscopy

Samples for nuclear magnetic resonance (NMR) analysis were dissolved in dimethyl sulfoxide–d₆. Heteronuclear single-quantum coherence NMR spectra were acquired at 25°C on a Bruker Avance 400 MHz spectrometer using a broadband cryo-probe. Spectra were acquired with 2048 points and a spectral width of 12 parts per million (ppm) in the F2 (1H) dimension and 256 points and SW of 220 ppm in the F1(13C) dimension using a standard phase-sensitive, gradient-selected pulse sequence. All spectral processing used Mnova 14.2, and prior work was used for peak identification (57, 75).

Process simulation and TEAs

The baseline scenario for conversion of LRC to β-ketoadipate in bubble columns was conducted at the best strain performance metrics achieved here. Reactions for biomass growth and product formation are shown in table S10.

Process simulations were carried out using the Aspen Plus V10 software (AspenTech, Bedford, MA), centered around the fermentation and purification sections of a biorefinery producing 100,000 MT of β-ketoadipic acid/year. A simplified process flow diagram is shown in Fig. 8A, following a similar approach as in (32). After cultivations, in which LRCs are converted to β-ketoadipate in bubble columns and glucose supports biomass growth, the acid is further recovered from the fermentation broth using SMB chromatography and evaporation to yield a final product with 99 weight % purity. Stoichiometries related to the bioconversion of LRCs and sugars in bioreactors, as well as general performance metrics of the downstream processing, are further detailed in the Supplementary Materials.

Material and energy balances obtained from process simulation efforts were then used as inputs to determine capital expenditures and operational expenses associated with the integrated biorefining process. Last, discounted cash flows were established to assess the economic performance of the bioprocess in large scale, aiming at estimating the MSP of β-ketoadipic acid, which corresponds to the sales price required to balance the economics of the facility (i.e., net present value of zero) over a 30-year lifetime. The main parameters used in the TEA of the bioconversion process are shown in the Supplementary Materials.

Life cycle assessment

The current attributional environmental LCA concentrates on life cycle GHG emissions, cumulative fossil energy consumption, and water consumption for the β-ketoadipate production described above. Figure 8A shows the LCA system boundary, which includes the bioprocess and downstream process to purified β-ketoadipate and excludes the upstream LRC production. The scope of the LCA focused on cradle-to-gate life cycle GHG emissions in kilograms of CO_2e using 100-year GHG emission factors, fossil energy consumption in megajoules, and water in liters. The emission factors for the underlying processes, such as feedstock chemicals, were obtained from various sources, including the ecoinvent database (76) and Greenhouse gases, Regulated Emissions, and Energy use in Transportation (GREET) software (77). The foreground data are the material and energy flows for the β-ketoadipic acid production step that captures the impacts of input raw materials and outputs, such as emissions, wastes, and coproducts as predicted by the process model.

For the scenario with the coproduction of ammonium sulfate, we adopted a purpose-driven system-level allocation method to handle the environmental burden attributed to the bioprocess inputs according to their destined fate contributing to β-ketoadipic acid production, ammonium sulfate production, or both. Moreover, the burdens shared by the main product and coproduct are allocated on the basis of their respective masses or the market values of both products ($1.70/kg for β-ketoadipate and $0.26/kg for ammonium sulfate). The latter is the baseline case, and the former is a sensitivity case. The system-level allocation approach ensures a reasonable estimation of the environmental footprint associated with different input streams that purposefully contribute to different products (68). As a second sensitivity study, the coproduct displacement approach was also carried out.

Supplementary Materials

This PDF file includes:

Figs. S1 to S15

Tables S1 to S14

Legend for data S1

Other Supplementary Material for this manuscript includes the following:

Data S1