Figure 3. Biosynthesis of methyl ketones and FAEEs.
(A-B) We optimized our first set
of models for methyl ketone biosynthesis with simultaneous fits to two
datasets: (A) the product profiles and titers of five strains that
overexpress a thioesterase (TE), a fatty-acid-CoA ligase (FadD), an
acyl-CoA dehydrogenase (FadE), an oxidoreductase (FadB), and an acyl-CoA
thioesterase (FadM; (Yan et al., 2020)), and (B) the ratio of
C12 fatty acids to C11 methyl ketones for a sixth strain
with similar enzymes overexpressed (Goh et al., 2012). (C-D) We
optimized models that generate fatty acid methyl esters (FAMEs) and
fatty acid ethyl esters (FAEEs) by fitting the product profiles and
titers of strains that overexpress (C) an O-methyltransferase (MT;
(Sherkhanov et al., 2016)) or (D) both a wax ester synthase (WS) and a
TE, (Steen et al., 2010)). Table S9 describes the compositions of the
modeled pathways.
well-characterized enzymes (e.g., an alcohol pathway that uses
carboxylic acid reductase, or
CAR, and a thiolase-based methyl ketone pathway discussed later). For
these pathways, we fit experimental data by adjusting either a single
kinetic parameter or multiple enzyme concentrations (but not both). (ii)
Pathways with partially characterized enzymes (e.g., the alkane, FAME,
and FAEE pathways). Here, we optimized a single substrate-specific
kcat and either (a) multiple enzyme concentrations or
(b) an overall kcat vector (i.e., a single factor
multiplied by all substrate-specific kcats). (iii)
Pathways with poorly characterized enzymes (e.g., the β-oxidation-based
methyl ketone pathway and several alcohol pathways discussed later).
Here, we optimized multiple enzyme-specific kinetic parameters. For the
methyl ketone pathway, which is poorly characterized, we used a
sensitivity analysis to identify the most influential kinetic parameters
before fitting. Overall, the reasonable fits afforded by our models
indicate that Michaelis-Menten parameters provide an adequate means of
modeling the activities of enzymes from different oleochemical pathways
(Figs. 2-3).
We parameterized enzymes consistently across models. For each enzyme, we
adjusted kinetic parameters only once, and we maintained consistent
concentrations between pathways. For example, when optimizing pathway
that generates alcohols via a CAR and an aldehyde reductase (AHR), we
changed the concentrations of FabA, FabZ, FabB, and FabF—all core FAS
enzymes—and retained these concentrations for all models; this
constraint is consistent with the native FAS expression levels exploited
in most engineered strains. For heterologously expressed enzymes, by
contrast, we occasionally adjusted concentrations. TesA, in particular,
is common in engineered strains, but its overexpression relies on
different promoters and ribosome binding sites (Barrick et al., 1994;
Lozano Terol et al., 2021); variability in its concentration between
strains is reasonable. In general, the consistent parameterization of
enzymes in different pathways suggests that our models can capture the
activities of important pathway enzymes in different biochemical
contexts.