Every batch cultivation ends with a harvest decision, and that decision determines how much product comes out of the fermenter. The processes most directly served by this article fall into two related groups. The first is processes where the cells are themselves the product — bacterial vaccines, probiotics, agricultural inoculants such as nitrogen-fixing bacteria, and seed cultures for downstream cultivation. The second is processes where the product is a cell-associated component whose accumulation tracks the bacterial growth phase, such as a surface antigen or a secondary metabolite expressed predominantly in stationary phase as a biological response.
What both target groups have in common is that the harvest decision is timed relative to a single moment in the run: the transition from exponential growth into stationary phase — when cell division stops. For cell-as-product workflows this transition is the harvest target itself. For stationary-phase-product workflows, the harvest sits a defined time after the transition, but the transition is still the anchor against which that defined time is measured. In both cases, knowing precisely when the transition occurred is what makes the harvest decision rational rather than habitual.
The interpretive principles below apply to any direct count of structurally intact bacterial cells, regardless of the method used to obtain it. For the foundational explanation of what a BactoBox® cell count is and how it compares to other enumeration methods, see Understanding BactoBox® cell counts. For the OD600 limitations summarised below in their fuller form, see Understanding OD600. For the end-of-fermentation crosscheck against CFU that this article does not cover, see End of fermentation crosscheck.
Anchoring the harvest decision
For processes where the cells are themselves the product, the typical harvest target is the peak in culturable cell concentration, which is reached at the transition from exponential growth into stationary phase. Holding the run past this peak either accumulates cost in extended stationary or begins to lose culturable cells in early decline. Stopping the run before the peak leaves culturable product in the fermenter.
For some processes the optimal harvest sits a defined time after the transition rather than at it. The onset of stationary phase in Escherichia coli triggers a σS-dependent gene-expression program that activates dozens of genes encoding stress-response, transport, and metabolic functions not active during exponential growth[1]; the σS regulator is conserved across γ-proteobacteria more broadly[2], and analogous stationary-phase regulatory programs operate in other bacterial groups under different alternative sigma factors. A product that accumulates as a consequence of this regulatory shift — a surface antigen whose expression turns on as cells stop dividing, a secondary metabolite that builds up during stationary phase — is best harvested some hours after the transition, not at it. The mechanism in these cases is the natural biology of stationary phase, not heterologous induction. Conversely, a seed culture intended for downstream transfer is typically harvested some time before the transition, when cells are still actively dividing. In every case, the harvest decision is timed relative to the moment cell division stops, and a clear view of when that moment occurred is what makes the timing rational rather than habitual.
The same logic extends to fed-batch and cell-factory processes in a limited way: even when the harvest decision is dominated by product titre, knowing the precise moment cells stopped dividing can help anchor decisions about feed-rate transitions and induction-strategy timing. The argument below focuses on the cell-as-product and stationary-phase-product cases, where the timing relationship to the cell-count trajectory is most direct.
Why OD600 cannot answer the question reliably
OD600 is a turbidity measurement, not a cell count. The signal at 600 nm is dominated by light scattering, which depends on cell size, intracellular composition, the refractive properties of the cells in their current physiological state, and any non-cell particulates in the light path[3][4]. Through exponential growth, when cells are dividing actively and morphology is comparatively stable, OD tracks cell concentration reasonably well. As the culture approaches and enters stationary phase, the relationship breaks down — cells change size and intracellular composition even when division has slowed or stopped, and OD continues to respond to those changes[3]. The OD reading drifts on the basis of cell properties rather than cell number.
In practice, when an OD curve is compared with a direct cell count on the same run, three patterns are possible.
OD plateaus before the cell count does. The OD trace flattens while cells are still dividing. A harvest called on this plateau stops the run before the transition into stationary phase is actually reached.
OD continues to climb after the cell count has plateaued. The cells stop dividing, but OD keeps rising because cell size, intracellular composition, or refractive properties continue to change. A harvest called on this plateau sits in the fermenter long after the population has stopped producing, with the run either accumulating cost in extended stationary or beginning to lose culturable cells in early decline. A decision intended to be timed relative to the transition — for example, a harvest at three hours into stationary phase — is also pushed late, because the reference point itself has moved.
OD plateaus at the same point the cell count plateaus. Possible, and some strains and media behave this way. Across the runs we have measured, this alignment is the exception rather than the rule.
The practical consequence is that the moment cell division actually stops is not visible from inside an OD trace alone, and any decision timed against it inherits the OD reading's drift.
A worked example
The figure below shows an internal batch of a Pseudomonas strain, with OD600 and CFU/mL measured in triplicate at every timepoint. The two curves are read on different axes: OD600 on the left, CFU/mL on a log axis on the right.
What the data shows is straightforward. The OD curve approaches a plateau at approximately 13.6 hours; from there it stays within a few percent of its maximum value for the remainder of the run. A reasonable reading of OD alone would conclude that the culture has entered stationary phase at that point. The CFU curve, taken on the same samples, says otherwise — culturable cells continue to be added for nearly two more hours. The CFU value at 15.4 hours is roughly 40% higher than the value at the OD plateau.
A harvest called on the OD signal here ends the run before the population has finished dividing. Over a production calendar, the cumulative effect of a gap of that size can be a material fraction of recoverable product, and the operator running the OD signal will never see the gap from inside the run because the OD reading itself looks like a clean stationary plateau. This is one of the three patterns described above; on a different run, the same operator might face the opposite failure mode, with OD continuing to climb after cell division has already stopped.
What a direct cell count makes possible
A direct count of structurally intact cells does not carry the OD signal's interpretive ambiguity. The number rises only when new cells appear in the sample, and a sequence of measurements through the slowdown into early stationary phase shows directly when the addition of new cells stops. The cell-count plateau is the transition into stationary phase, observed in real time.
What that visibility supports is broader than a single harvest decision. For a process where the cells are the product, the plateau is the target. For a process whose product accumulates after cell division has stopped, the harvest sits a defined time after the plateau, and the timing of that defined time is now grounded in the biology rather than in elapsed time from inoculation. For seed-culture transfers and other workflows where the harvest sits before the transition, BactoBox plays a different role: during process development it characterises when the transition actually occurs and how other available metrics behave in the hours leading up to it, so the production decision can be timed against that calibrated relationship rather than against habit or elapsed time. The shift is from harvesting against OD and hoping the proxy aligns, to harvesting against a known position relative to the moment cell division stopped — known in real time when the harvest sits at or after the transition, and known by prior characterisation when it sits before.
BactoBox counts bacterial cells in approximately two minutes per sample using impedance flow cytometry, with a lower detection limit of 30,000 cells/mL. Two minutes is short enough to make a sequence of measurements across the slowdown practical without dedicated analytical staff. The published equivalence between BactoBox and CFU is worth being explicit about. BactoBox cells/mL and CFU/mL were compared across six bacterial species spanning a range of envelope types and cell sizes, with log-log R² values between 0.9974 and 0.9998 through exponential growth, deceleration, and stationary phase[5]. Within those phases, at least 88% of paired measurements differed by less than 0.1 log10 units across the panel, meeting the equivalence threshold of the USP <1223> framework for alternative quantitative microbiological procedures[5][6]. The two methods diverge in the decline phase, where BactoBox continues to count structurally intact cells that have lost culturability[5]. Through the regime where the harvest decision is actually anchored, BactoBox produces a result consistent with what a CFU plate count would produce a day or two later.
The practical workflow is to sample through the deceleration of the OD curve, run a BactoBox reading on each sample, and watch the cell-count trajectory directly. The plateau is the transition into stationary phase. The decision to harvest at, before, or after that point depends on the process, but the decision is now anchored.
Closing
The harvest decision in a cell-as-product fermentation turns on a single question: has cell division stopped. OD600 does not answer that question reliably across runs, because the OD signal can plateau before division stops, continue to climb after division stops, and only sometimes track the cell-count plateau directly. A sequence of direct cell counts through the slowdown answers the question by measuring the thing the question is about — when new cells stop appearing in the sample.
The cell-count peak is not universally the harvest target. For products that accumulate in stationary phase, the harvest sits later; for seed-culture transfers, it sits earlier. The argument of this article is that knowing precisely when the transition into stationary phase occurred is what enables those decisions to be timed accurately, whether the harvest lands at the transition or at a defined point relative to it. The resulting decision is anchored in the dynamics of the culture rather than in a proxy whose relationship to those dynamics is not fixed.
For help applying this on a specific process, including which sampling cadence and which medium-handling pattern are appropriate for your organism and your fermenter, reach out. The crosscheck at the end of a run — comparing the cell-count plateau with a CFU plate from the same sample — is described in End of fermentation crosscheck.
References
- Lacour S, Landini P. SigmaS-dependent gene expression at the onset of stationary phase in Escherichia coli: function of sigmaS-dependent genes and identification of their promoter sequences. J Bacteriol. 2004;186(21):7186-95. https://journals.asm.org/doi/10.1128/jb.186.21.7186-7195.2004
- Bouillet S, Bauer TS, Gottesman S. RpoS and the bacterial general stress response. Microbiol Mol Biol Rev. 2024;88(1):e0015122. https://journals.asm.org/doi/10.1128/mmbr.00151-22
- Stevenson K, McVey AF, Clark IBN, Swain PS, Pilizota T. General calibration of microbial growth in microplate readers. Sci Rep. 2016;6:38828. https://doi.org/10.1038/srep38828
- Mira P, Yeh P, Hall BG. Estimating microbial population data from optical density. PLoS One. 2022;17(10):e0276040. https://doi.org/10.1371/journal.pone.0276040
- Jordal PL, Díaz MG, Aalund F, Skands G. Performance qualification of impedance flow cytometry as a rapid in-process control proxy for colony-forming units in bacterial fermentation processes. J Microbiol Methods. 2025;238:107284. https://doi.org/10.1016/j.mimet.2025.107284
- United States Pharmacopeia. General Chapter <1223> Validation of Alternative Microbiological Methods. USP-NF. https://www.uspnf.com/