Rapid development of clone‐specific, high‐performing perfusion media from established feed supplements

Abstract Perfusion cultivation of recombinant CHO cells is of substantial interest to the biopharmaceutical industry. This is due to increased space–time‐yields (STYs) and a short residence time of the recombinant protein in the bioreactor. Economic processes rely on cultivation media supporting rapid growth in the exponential phase and high protein production in the stationary phase at minimal media consumption rates. To develop clone‐specific, high‐performing perfusion media we present a straightforward and rapid two‐step approach combining commercially available basal media and feed supplements using design‐of‐experiment. First, the best performing feed supplements are selected in batch cultures. Then, the mixing ratio of selected feed supplements is optimized in small‐scale semicontinuous perfusion cultures. The final media formulation is supported by statistical response surface modeling of a set of cultivation experiments with blended media formulations. Two best performing novel media blends were finally applied to perfusion bioreactor verification runs to reach 200 × 106 c/ml within 2 weeks at minimum cell‐specific perfusion rates as low as 10–30 pL/c/d. Obtained STYs of 0.4–1.2 g/L/d represent a 10‐fold increase compared to batch cultures. This general workflow is universally applicable to any perfusion platform combining a specific cell line, basal medium, and established feed solutions.

are advantageous due to lower volumes during media preparation (thus reduced tank sizes) and higher product concentrations in the harvest.
Minimum CSPR as low as 50 to 300 pL/c/d were reported in literature 2,5 but some groups even achieved 15-20 pL/c/d. 6,7 Despite numerous cell culture media are commercially available for batch and fed-batch cultivation, this is not the case for perfusion media, which must be designed individually in a complex and timeconsuming process. Here, we present a straightforward strategy for perfusion media development by screening and optimizing the spike concentration and blending of eight commercially available feed supplements. In a first step, the best feed supplements are identified by spiking the basal media with different feed combinations according to a fractional factorial design. The supplement concentrations were then optimized in a face-centered central-composite design using semicontinuous small-scale (10 ml) perfusion cultures ("pseudo-perfusion") in shaking tubes. This two-step and empirical design-ofexperiment (DoE) workflow resulted in two novel perfusion media.
The final perfusion media were used in bench-scale bioreactor runs at 500 ml working volume that achieved VCDs of more than 200 × 10 6 c/ml during a 2 weeks bioprocess. Bioreactor perfusion cultures were successfully operated at CSPRs as low as 10-30 pL/c/d.

| Cell lines, media, feed supplements, and process analytics
A recombinant CHO K1 cell line producing an IgG1 antibody under the glutamine-synthetase (GS) selection system was used throughout this study. Cells were passaged every 3-4 days for routine cultivation and adapted to CDM4NS0 or ActiPro basal media. Eight different HyClone™ Cell Boost™ feed supplements, referred to as CB1 to CB7b in this study, were supplemented into basal media and used in "spiked batch" or semicontinuous perfusion cultures in shaking tubes. Feed supplements are chemically defined mixtures of diverse nutrient substance groups as summarized in Table 1 according to the manufacturer. 8 Cultures were incubated in a Kuhner shaker incubator at 37 C, 80% relative humidity, 220 rpm, 90 angle, and 50 mm orbital shaking diameter.
Cell concentration, viability, and cell size was measured with a Vi-Cell XR. Glucose, lactate, glutamine, glutamate, and ammonia concentrations were quantified using a BioProfile 100 Plus. Osmolality was analyzed with Osmomat 030 and product concentrations were determined using an OctetRed.

| Screening design-of-experiment in spiked batch cultures (DoE1)
A screening experimental design (Figure 1a) was used to identify the main effects of eight different feed supplements representing continuous variables ("factors") at three concentrations ("factor levels") on cell culture parameters ("responses"). A total of 19 spiked batch cultures were used to establish a fractional factorial design at two levels to resolve linear (main) effects (run #1 to #16) and a triplicate center point to detect nonlinear curvature in the response surface (run #17 to #19).
Three different factor levels were defined for each feed supplement. For definition of the maximum factor Level +1 all feed supplements were mixed according to their total molar amino acid concentration and spiked into CDM4NS0 or ActiPro basal medium to reach a final target osmolality of 400 mOsm/kg. Accordingly, in the first DoE1 screening design, CB 1 to 6, 7a, and 7b were spiked into CDM4NS0 and ActiPro at 7.2, 4.3, 13.0, 4.1, 11.4, 11.1, 2.5, and 0.3%, respectively. DoE factor Level −1 was defined as no supplement addition and factor Level 0 as the half-maximum spike concentration. Media were prepared by spiking the basal medium with feed supplements according to the DoE matrix depicted in Figure 1a. Spiked batch experiments were seeded at 0.3 × 10 6 c/ml at a working volume of 30 ml in 50 ml shaking tubes.
Once glucose levels reached 3 g/L in spiked batch cultures a 250 g/L glucose stock solution was added to increase glucose levels to 6 g/L. A 60% viability threshold was defined as batch termination criterion.

| Semicontinuous small-scale models ("pseudoperfusion") in shaking tubes
Ten milliliters small-scale semicontinuous perfusion models were seeded at 10 × 10 6 c/ml in fresh medium in 50 ml shaking tubes.
T A B L E 1 Substance classes included in eight commercially available HyClone™ feed supplements ("Cell Boosts," CB1-7b) as disclosed by the manufacturer 8 Hypoxanthine / thymidine Lipids Cholesterol A daily, complete medium exchange was performed every 24 hr by daily sampling followed by centrifugation (300 g, 7 min) and resuspension of the cell pellet in 10 ml of fresh medium (one reactorvolume change per day, vvd). cultures were employed to establish valid regression models by response surface modeling and to define the optimal mixing ratios.

|
Two novel perfusion media were compared to the corresponding basal media in small-scale models and finally applied to perfusion bioreactor runs with the goal to reach high VCD of more than 200 × 10 6 c/ml at low CSPR.

| Screening design of feed-spiked batch cultures (DoE1)
A fractional factorial design (Figure 1a) was used for the initial screening approach to select those supplement combinations that supported the highest IVCD and peak titers (pTiters) in spiked batch cultures.
The pVCD values ranged from 3 to 39 × 10 6 c/ml in CDM4NS0 or  In contrast, enrichment of the basal media with all feed supplements at half-maximum and maximum levels maintained cell viabilities above 80% for 11 days (Figure 3b,d). Similar to batch cultures, any combination of feed supplements decreased cell-specific growth in the exponential growth phase until day four. However, higher cellspecific production rates were maintained in the stationary phase from day 5 to 11 (Supporting Information S6). Importantly, although the steady-state VCDs were also decreased by 1-5% in half-and maximum supplemented cultures, mean steady-state titers were boosted by 30-40% compared to nonsupplemented cultures to F I G U R E 2 Statistical analyses of spiked batch cultures (DoE1) by Pearson correlation and regression analysis. Pearson correlation coefficient (r) and p-values (*<.05, **<.01, ***<.001) were used to detect positive (green) or negative (red) correlations between culture parameters of spiked batch cultures and are exemplarily shown for CDM4NS0 (a) or ActiPro (c). Culture parameters for the exponential phase or the entire process are indicated by the superscript "exp" or "process," respectively. Linear (main) model coefficients of individual feed supplements are exemplarily shown for peak titer (pTiter) and integrated viable cell density (IVCD) in CDM4NS0 (b) or ActiPro (d). High quality models indicated by R 2 and Q 2 by partial least squares (PLS) regression analyses were generated by removing small and nonsignificant coefficients ( †). The complete data set can be found in the Supporting Information S3-S5 0.5-0.9 g/L. Supplemented cultures also showed highest cell-specific productivities maintained over the entire culture process duration. The average stationary qGluc stat consumption and qGlu stat con- hybridoma-, 11,12 myeloma-, 13 and CHO cells. 14,15 Especially at low glucose concentrations, glutamine plays an essential role as primary energy source. 16  demonstrating the validity of the established regression models.

Suppression of cell growth at high viabilities and productivities is
especially important when bleed rates in steady-state perfusion need to be minimized in order to prevent product loss. Such attempts to decouple productivity from growth were already proven by media additives such as small molecule cell growth inhibitors, or altered sodium-potassium levels to arrest cells in the G0/G1 phase. 17 Alternatively, process control strategies were used to keep cells in a physiological state using lactate consumption and the resulting rise in culture pH to regulate perfusion media addition 18 or minimizing growth and maximize productivity by reduced culture temperatures 19,20 or limiting CSPR. 21 Culture pH, pCO 2 levels, and osmolality were also shown to profoundly impact cell growth and productivity 22,23 and may be used for future perfusion control strategies.
Indeed, valid regression models confirm the general observation that supplementation with CB1, 3, 7a, or 7b suppresses VCD exp by 1.0-3.0 × 10 6 c/ml and daily viable cell days (ΔIVCD exp ) by 0.8-2.3 × 10 6 c × d/ml during exponential phase in CDM4NS0 (Table 2A). Growth was also reduced by CB1 in ActiPro resulting in strong suppression of VCD exp by 6.9 × 10 6 c/ml and ΔIVCD exp by 4.7 × 10 6 c × d/ml. Supplement addition did not substantially alter specific productivities or viabilities in the exponential growth phase. Only a small increase of qP exp with CB7b in CDM4NS0 and a minor reduction of viability with CB1 in ActiPro was induced. As qP exp between nonsupplemented Regression analysis was also used to quantify the impact of supplement addition on primary metabolic pathways. Because of decreased growth, cell-specific glucose (qGluc exp ) uptake was also reduced by CB1, CB3, and CB7a in the exponential phase of both basal media (Table 2). Despite lower qGluc exp , specific lactate production (qLac exp ) was increased with CB1 or CB7a.
In the subsequent CDM4NS0 stationary phase, a positive CB1 main effect on VCD stat by 2.5 × 10 6 c/ml (Table 2A)  Exp. Stat.
Stat. resulting in lower daily (ΔIVCD stat ) and cumulative cell days (IVCD). A significant positive stationary titer effect of 33 mg/L for CB1 or 59 mg/L for CB3 was modeled in CDM4NS0 but provoked by different routes (Table 2A). CB1 increased stationary VCD stat but not qP stat , while CB3 caused increasing qP stat but suppressed VCD stat . CB3 similarly enhanced stationary ActiPro titer by 100 mg/L through increased specific productivity qP stat and viability (Table 2B).

VCD (×10
Metabolic activity was enhanced by CB3 marked by a higher qGluc stat consumption and higher qLac stat secretion in both media. Part of this energy is channeled into higher production capability indicated by a significant increase in qP stat . Beside metabolic changes also cell morphological alterations were induced by all feed supplements, except CB7b, indicated by increased cell diameters throughout all culture phases ( Table 2).
F I G U R E 4 Contour plot of regression models for the stationary mean titer mTiter stat of semicontinuous perfusion cultures in CDM4NS0 (a) and ActiPro (b). The final two feed combinations for the CDM4NS0 or ActiPro perfusion media are indicated by blue stars. Note that CB7b was suggested for the ActiPro model, but was omitted in the final media formulation because of solubility issues and to reduce the complexity but at a reasonably highpredicted stationary titer Established regression models for stationary titer values were finally leveraged to define the best supplement factor levels that were then transformed into the respective spike concentrations (Figure 4).
Optimal combination of CB1 and CB3 at Level +1 was calculated for CDM4NS0 keeping CB7a and CB7b at −1 resulting in a predicted titer of 520 mg/L (Figure 4a). CB1 and CB3 at Level +1 translates into a F I G U R E 5 Perfusion media performance verification in semicontinuous small-scale models. The CB1-and CB3 spiked ActiPro (orange triangles) and CDM4NS0 (blue circles) perfusion media were compared to the respective basal media that was spiked only with glucose to the level of the respective perfusion medium (purple squares for ActiPro and green diamonds for CDM4NS0). Total cell concentration (a) and antibody titer (b) was recorded to investigate differences in specific growth and productivity (c) in the exponential-and stationary phase. Highly similar product quality profiles were confirmed by N-glycosylation (d), charge-(e) and size distribution (f) for the basal CDM4NS0 and CB1-and CB3-spiked perfusion medium. Each condition was run in duplicate final spike concentrations of 11.06% CB1 and 19.9% CB3 for CDM4NS0. For ActiPro optimal factor levels were determined for CB1 at +1, CB3 at +0.26, and CB7b at +0.1. However, since the model suggested only minor CB7b amounts of 0.26%, this feed supplement was omitted with keeping reasonable high predicted titer values above 750 mg/L (Figure 4b). Consequently, the final F I G U R E 6 Bioreactor perfusion verification with the final perfusion media. Glucose-spiked CDM4NS0 control (green diamonds) was compared to the CB1-and CB3 spiked CDM4NS0 perfusion medium (blue circles) at constant volumetric perfusion rates of one to two vvd. The performance of CB1-and CB3 spiked ActiPro perfusion medium (orange triangles) was tested at constantly low CSPR of 15 to 30 pL/c/d. Exceptionally high total cell numbers and viabilities (a) were reached, accumulating in high titer and space-time-yields (STYs) (b) at very low CSPR (c). Highly similar product quality profiles as assessed by N-glycosylation (d), charge-(e) and size (f) distribution were obtained for the two CDM4NS0 runs. CSPR, cell-specific perfusion rates ActiPro perfusion medium was spiked with 13.94% CB1 and 15.8% CB3.

| Verification of performance by semicontinuous small-scale models and bioreactor perfusion experiments
The optimized CDM4NS0 and ActiPro perfusion media were finally applied to 10 ml semicontinuous perfusion models ( Figure 5) and two 500 ml bioreactor verification runs using two different strategies ( Figure 6) to identify possible media limitation and to demonstrate applicability to bioreactor perfusion cultures.
Mean stationary VCD stat of semicontinuous perfusion cultures was increased up to 53% by spiking CDM4NS0 with CB1 and CB3 to maintain higher viabilities after day four compared to the nonspiked control ( Figure 5a). As a combined effect of higher VCD stat and higher qP stat (Figure 5c), the mean daily harvested product titer was increased by 56% up to 0.5 g/L ( Figure 5b) and was accurately predicted by the mTiter regression model (Figure 4a). Similarly, mean daily product titer was increased by 22% up to 0.6 g/L when spiking ActiPro with CB1 and CB3. The increased titer is a consequence of increased specific productivity (Figure 5c) rather than an increase in stationary VCD stat (Figure 5a).
Usually the primary and first objective during media optimization is to increase VCD and harvest titer. Despite the two markers are often used for describing best cell performance, the focus of critical parameters is shifted toward increased product quality. [24][25][26] Thereby, critical quality attributes have to be defined for each molecule of interest individually and quality characteristics are predominantly defined by the process operation mode/conditions and harvest time. 22,[27][28][29][30][31][32][33] In semicontinuous small-scale models, N-glycosylation and charge profiles changed distinctly along different culture harvest times, but not by the addition of the CB1 and CB3 feed supplements (Figure 5d, e). The variation of product quality attributes was higher between different harvest times than between different media. The strong decrease of galactosylated N-glycan species over cultivation time can be explained by increasing levels of ammonia in semicontinuous perfusion models (Supporting Information S11E). It is known that elevated ammonia levels inhibit activity of glyco-attaching enzymes including galactosyl transferases. [34][35][36] Final ammonia levels and specific ammonia production rates in the stationary phase were increased by feed supplementation (Supporting Information S11E and J), whereas specific glucose consumption and lactate production were consistent throughout all cultures at high viabilities (Supporting Information S11F and G).

| CONCLUSION
The main goal of improved perfusion media compositions is the reduction of cell-specific media volume in combination with high viable cell concentrations, high cell-specific productivities, and consistent product qualities. In this study, we investigated the benefit of a rapid and empirical DoE-driven workflow to develop novel perfusion media by blending chemically defined basal media and established feed supplements. Using this empirical approach, a larger design space can be investigated when the exact media ingredients are unknown, compared to a more rational and targeted optimization approach. 37