When a metabolic engineering project stalls at the flux measurement stage — good titers in small-scale, poor titers at scale-up, or good titer of the target compound but unexpectedly high accumulation of intermediate metabolites — the diagnosis most teams reach for first is expression level tuning: promoter strength, codon optimization, gene dosage. These are real levers. But when the expression tuning doesn't fix the problem, and when the issue manifests primarily as redox imbalance (NADPH depletion in the cytoplasm, or NADH accumulation in pathways with competing catabolic reactions), the root cause is almost always upstream in the design phase.
Specifically: the enzymes were selected and optimized individually, without a stoichiometric model of cofactor demand across the full pathway. Individual optimization produces locally excellent enzymes — high kcat/KM for their specific substrate, good expression, adequate thermostability — that collectively impose a cofactor burden the cell can't sustain at production titer levels.
The cofactor imbalance problem in pathway context
Most enzyme engineering happens one enzyme at a time. You have a target reaction, you identify or engineer an enzyme for it, you validate it, and you move to the next step in the pathway. This is rational from a project management perspective — you need individual enzyme functionality before you can test pathway-level functionality. But it creates a systematic blind spot.
Consider a four-step synthesis pathway where steps 1 and 3 each require NADPH as the hydride donor, and step 2 produces NADH as a byproduct of an oxidative decarboxylation. In an isolated enzyme screen, each of steps 1 and 3 looks fine — the enzymes are active with NADPH at their individual Km values. In the assembled pathway expressed in E. coli, NADPH becomes limiting because the pentose phosphate pathway and TCA-linked transhydrogenase can't regenerate NADPH fast enough to sustain the flux through both steps 1 and 3 simultaneously at production levels (typically 0.5–5 g/L titer targets). Step 2 produces NADH but not NADPH, so the redox balance is negative in the NADPH pool.
The engineering response is usually to redesign the NADPH-dependent steps to accept NADH instead — a cofactor specificity switch from NADP+ to NAD+. But if that switch was not anticipated during individual enzyme optimization, you now need to reopen each of those enzyme engineering campaigns, which adds 2–4 months of design-build-test cycle time to a project that thought it was in pathway assembly.
Pathway-aware co-design: what it means in practice
Co-design, in our usage, means building a stoichiometric cofactor model for the full pathway before committing to specific enzyme variants at each step. This is not the same as full constraint-based metabolic modeling (like flux balance analysis on a genome-scale model) — it's a much simpler calculation that maps the cofactor species consumed and produced at each step, identifies net imbalance across the pathway, and flags steps where the enzyme design needs to accommodate a cofactor specificity different from the natural enzyme.
The Fermvyne pathway design module accepts a pathway specification as a sequence of reaction SMARTS or EC numbers with assigned substrates, then computes a cofactor stoichiometry table before running any generation. That table identifies three categories of steps:
- Balanced steps — where the cofactor species produced and consumed within the reaction are net neutral (e.g., oxidoreductases that consume NADPH at steps where an equivalent NADPH regeneration reaction is co-expressed).
- Tension points — where a cofactor is consumed but the pathway has no regeneration mechanism assigned. These are the steps where cofactor specificity is most important to get right during enzyme generation.
- Potential conflict steps — where the natural enzyme for that reaction uses a cofactor that conflicts with the net cofactor balance of the rest of the pathway. These are flagged for specificity switch consideration before synthesis.
The output of the cofactor analysis becomes a constraint on the generation phase. For tension-point steps requiring NADPH regeneration, we generate enzyme variants that either specifically reduce NADPH demand (higher kcat/KM with NADH if the enzyme is tolerant to the switch) or that don't compromise NADH/NADPH discrimination so severely that flux leaks through both cofactor forms inefficiently.
A concrete scenario: terpenoid alcohol synthesis in E. coli
Consider a five-step pathway producing a sesquiterpene alcohol from acetyl-CoA, an increasingly common target in fragrance and specialty chemical biosynthesis. Steps 1–3 build the isoprenoid chain via the methylerythritol phosphate (MEP) pathway; step 4 is a terpene synthase (cyclase) that doesn't require cofactors; step 5 is a cytochrome P450-mediated hydroxylation that requires NADPH and a P450 reductase.
In the MEP pathway, two steps (IspG and IspH in the native pathway) require reduced ferredoxin, which is regenerated from NADPH via ferredoxin-NADP+ reductase. Under overexpression conditions, the combined NADPH demand from the MEP pathway and the final P450 hydroxylation can deplete NADPH to levels that rate-limit the entire pathway. Titer plateaus at 200–400 mg/L even when expression of all enzymes is confirmed adequate.
A co-design approach applied to this pathway identifies the NADPH tension before enzyme selection. When we ran this pathway specification through our cofactor analysis, the tool flagged the P450 hydroxylation step as the highest-risk tension point (it's the step with the largest NADPH stoichiometric demand per pathway flux unit, and it's not reducible by cofactor specificity switching because P450 reductase is NADPH-obligate). The recommendation was to either co-express a soluble NADPH regeneration system (e.g., phosphite dehydrogenase for phosphite-to-phosphate oxidation with NADP+ reduction) or to select a P450 variant with significantly higher coupling efficiency (reducing NADPH wasted through uncoupled oxygen reduction without hydroxylation).
The latter is tractable with generation: Fermvyne generated P450 variants with modified proximal loop geometry predicted to improve coupling efficiency, reducing nonproductive NADPH consumption per productive hydroxylation event. Rather than discovering the NADPH limitation after assembly and then reopening the P450 engineering campaign, the cofactor analysis surfaced it at the design brief stage.
Where individual optimization is still the right approach
We're not arguing that individual enzyme optimization is wrong — in many cases it's exactly right, and co-design adds complexity that isn't warranted. Single-enzyme applications (whole-cell biocatalysis with a single heterologous enzyme, biotransformation reactions where cofactor regeneration is handled by the host metabolism), pathways where all steps use identical cofactor species (fully NAD+-dependent chains where host NADH regeneration via respiration is adequate), and scenarios where the bottleneck is clearly catalytic rate rather than cofactor supply all justify straightforward single-enzyme optimization.
The co-design approach matters most when: (a) the pathway has mixed NADH/NADPH steps, (b) total cofactor demand across the pathway at target titer is likely to exceed what the host regenerates at steady state, or (c) you're targeting high titers (above ~1 g/L) where cofactor pool depletion effects are measurable rather than theoretical. At early feasibility stage with 50–200 mg/L titer targets, cofactor imbalance is rarely the limiting factor. At process development scale, it frequently is.
The compounding effect of getting this right in design
What we've seen consistently is that cofactor imbalance, when it surfaces, surfaces late — usually at the first bioreactor run after pathway assembly, when someone finally has enough material to measure intracellular NADPH/NADH ratios and notices the NADPH:NADH ratio in producing cells is far below the 0.1–0.3 range typical for healthy E. coli under glucose supplementation. By that point, the team has already spent 4–8 months on pathway assembly and strain construction.
The stoichiometric cofactor analysis that surfaces this risk takes roughly 20 minutes on our platform for a typical 4–6 step pathway. That's not a promise that accounting for cofactor balance during design solves all pathway flux problems — metabolic flux has many determinants beyond cofactor supply. But it removes one predictable failure mode from the list, and that's a failure mode that historically has cost months of iteration to diagnose and fix after assembly. Getting the cofactor specification into the enzyme design brief costs almost nothing; discovering it at the bioreactor stage costs a lot.