Experiments were performed with two panels of proprietary clonally-derived Chinese hamster ovary (CHO) cell lines provided by AstraZeneca. The first CHO cell line panel expressed a proprietary monoclonal antibody (mAb1) and comprised eight clones (mAb1_C1 – C8), whilst the second CHO cell line panel expressed a proprietary bispecific antibody (bspAb1) and comprised six clones (bspAb1_C1 – C6). Cells were cultivated in two commercially available media: (1) fed-batch specific CD CHO medium (Gibco^®, Thermo Fisher Scientific, Massachusetts, United States) and (2) perfusion-specific High Intensity Perfusion (HIP) medium (Gibco^®, Thermo Fisher Scientific, Massachusetts, United States), both supplemented with 4% 50X HT supplement (Gibco^®, Thermo Fisher Scientific, Massachusetts, United States). Cell banks were established for each cell clone and medium prior to the clone screening experiments. Cell suspensions were cultivated in 125 mL non-baffled shake flasks (Corning^®, United States) placed in an incubator (MCO-19AIC, Sanyo, JP) at 37°C, 5% CO₂, and were agitated at a shaking speed of 180 rpm, using an orbital shaker with an orbital diameter (OD) of 25 mm (CO₂ resistant shaker, Thermo Fisher Scientific). Cells were passaged every 3–4 days and expanded into 500 mL non-baffled shake flasks (Corning^®, United States) for use for inoculation.

MWP cultures were performed using standard round well ultra-low attachment 24-well MWPs (CLS3473, Corning^®). The MWPs were sealed with a Duetz sandwich lid (CR1524, EnzyScreen, Heemstede, NL) to reduce evaporation while maintaining headspace gas exchange. All cultures were cultivated in an incubator at 37°C, 5% CO₂, an agitation speed of 250 rpm, and an OD of 19 mm, unless otherwise indicated. The plates were held in place by a Duetz MWP clamp system (CR1801h, EnzyScreen, Heemstede, NL).

For fed-batch (FB) cultures, MWPs were inoculated at 1 × 10⁶ cells mL⁻¹ at a working volume of 1.2 mL in CD CHO medium (day 0). Following a 3-day batch phase, feeding commenced on day 3 for a period of 5 consecutive days. Feeding was conducted as previously described by Silk et al., 2010 and involved a 6% v/v bolus addition of nutrient supplement (EfficientFeed™ B), followed with a 2.5% v/v bolus of 10x diluted bicarbonate solution (0.75 M Na₂CO₃, 0.5 M NaHCO₃) to control pH. For semi-perfusion (SP) cultures, MWPs were inoculated at 1 × 10⁶ cells mL⁻¹ at a working volume of 1.2 mL (day 0) in CD CHO or HIP medium. Following a batch phase, semi-perfusion was started on day 3, where total or partial medium exchanges equal to 1 RV d⁻¹ and 0.75 RV d⁻¹, respectively, were performed. A sacrificial well methodology was used for both process operations, where triplicate samples were taken every 24 h from day 3 to day 10.

Viable cell concentrations (VCCs) and percentage of viability were determined using a ViCell™ XR cell viability analyser (Beckman Coulter, United States). Due to the small sample size for MWP cultures, the number of extracellular metabolites was limited to three. Glucose, lactate, and ammonium were measured using an Optocell CuBiAn VC biochemistry analyser (4BioCell, Bielefeld, Germany) as described in Dorn et al. (2024a) and Dorn et al. (2024b). Osmolality was determined using a freeze point osmometer (Gonotec^® Osmomat 3000).

For titre quantification, an HPLC (HPLC Agilent 1100 series; Agilent, United States) with a 1 mL Protein G column (HiTrap™ Protein G HP, Cytiva) was used with loading buffer A (20 mM phosphate, pH 7.0) and elution buffer B (20 mM glycine, pH 2.8) to evaluate the protein concentration. An IgG standard with a concentration of 1.9 g L⁻¹ was used for both cell lines (determined by Nanodrop 1000, Labtech). The IgG standard was diluted in PBS and measured in triplicate to obtain a standard curve in range of 0–1.9 g L⁻¹.

For comparative analysis between different operation modes and perfusion rate strategies, the following Equations 1–4 were used to determine cell specific rates, and yields. q = ∆ c ∆ t ⋅ X ¯ + H + B ⋅ c i X ¯ (1) S T Y = Y i V W ⋅ t i − t 0 (2) Y i = ∫ 0 i c P , i ⋅ H i ⋅ V W   d t (3) Y P / G l u c = S T Y i ⋅ ∆ t c G l u c , 0 − c G l u c , i 1 V w (4) where q is the cell specific consumption/production rate, H is the daily harvest rate, B is the daily bleed rate, Δt is the time interval between two sampling time points, X ¯ is the daily average of the VCC, and c is the metabolite or product concentration, STY is the space-time-yield, Y _i the yield equal to the accumulated mass produced since start of the cultivation, Y _P/Gluc the product-over-glucose yield, c _P the antibody concentration at time i, c _Gluc the glucose concentration at time i, V _W the working volume, and t the cultivation time.

The cell line manufacturability index (MI_CL) leverages available data on growth characteristics, metabolites, and productivities by combining these into a single metric for the evaluation of m cell lines according to n criteria (Goldrick et al., 2023). The selection of the lead clone is formulated as a multi-criteria decision-making problem to allow assessing the performance of each individual clone (Goldrick et al., 2023). In this work, the MI_CL was adapted to MWP operation and for the comparison of fed-batch and semi-perfusion operations. To achieve a fair comparison with an equal number of parameters and to account for the differences in feeding and medium exchange protocols, the following parameters and criteria were selected: maximum values of viable cell concentration (VCC), growth rate, space-time-yield (STY) and product-over-glucose yield, minimum values of viability, and lactate concentration, as well as average values of cell specific production and consumption rates of the product and metabolites (q_P, q_Gluc, q_Lac, q_Amm). The parameters and criteria were fixed in this work, but the selection could be adjusted according to individual user requirements.

The calculation and data visualisation were performed in MATLAB 2021b based on the following equations previously described by Goldrick et al., 2023, where the MI_CL was calculated using Equation 5. M I C L , i = ∑ j = 1 j = n w j × r i j   f o r   i = 1 , 2 , 3 , … m (5)

Where w _i is the normalised weight of each criteria j, r _ij is a dimensionless rating per clone i and selection criteria j: r i j = x i j − x j , w o r s t x j , b e s t − x j , w o r s t (6)

In Equation 6 x _ij is the individual ranking of cell clone i for criteria j, x _j,best is the best overall ranking, x _{j, worst} is the worst overall ranking for criteria j. x _j,best and x _{j, worst} were defined as maximum or minimum values based on expertise from industry, where some parameters are considered “best” for the maximum value and “worst” for the minimum value of the parameter j (e.g., VCC or q_P) while others have a reverse rating where the minimum value is considered “best” and the highest value considered “worst” (e.g., lactate concentrations). The normalised weights were set to 1 for this study but could be adjusted in the future or depending on individual process requirements.

While the previous investigations showed the impact of the operation mode on growth, metabolism and clone ranking, a crucial step is to ensure the reproducibility of these results. Therefore, an investigation was conducted to assess the reproducibility of growth performance and ranking results. A second run of SP-HIP-75% was performed, with cells thawed from a second vial for each clone and compared to the SP-HIP-75% described in Section 3.1.2, now referred to as Run 1.

The growth performance and external metabolites of both runs are shown in Figure 4 and in general showed similar dynamics. In particular, metabolite concentrations were nearly identical between both runs (Figure 4: Row 2–4), where only for two clones (mAb1_C2 and mAb1_C6) some differences of lactate concentrations were observed, with a delayed onset of the reduction of lactate concentration. However for the growth performance, larger differences between Run 1 and Run 2 were observed (Figure 4: Row 1). The maximum VCCs of Run 2 were lower than those of Run 1. For example, mAb1_C2 reached 29.44 × 10⁶ cells mL⁻¹ and 17.51 × 10⁶ cells mL⁻¹ in Run 1 and 2, respectively. An exception was mAb1_C7, which reached around 45 × 10⁶ cells mL⁻¹ in Run 2 in contrast to around 35 × 10⁶ cells mL⁻¹ in Run 1.

The clone ranking was evaluated for a single parameter strategy as well as with the MI_CL strategy. The raw values of q_P ranging from 16.46–46.57 pg cell⁻¹ d⁻¹ for Run 1 and from 12.21–40.98 pg cell⁻¹ d⁻¹ for Run 2 and MI_CL ranging from 0.43–1.00 for Run 1 and 0.43–1.00 for Run 2 are reported in Supplementary Table S1. The ranges of q_P and MI_CL for both runs were similar.

The clone ranking based on the q_P is shown in Table 2. Both runs shared the same two top clones (#1: mAb1_C5 with q_P: 46.57 and 40.98 pg cell⁻¹ d⁻¹ for Run 1 and 2 respectively; #2: mAb1_C4 with q_P: 32.14 and 35.13 pg cell⁻¹ d⁻¹ for Run 1 and 2, respectively) as well as the same lowest-ranking position (#8, mAb1_C3), while the middle ranking position showed some variation with several clones changing by multiple positions. For example, mAb1_C6 ranked at positions #7 (q_P: 22.13 pg cell⁻¹ d⁻¹) and #4 (28.77 pg cell⁻¹ d⁻¹) for Run 1 and Run 2, respectively, or mAb1_C2 ranked at #3 in Run 1 (q_P: 27.58 pg cell⁻¹ d⁻¹) and #5 in Run 2 (q_P:23.66 pg cell⁻¹ d⁻¹). However, when taking multiple parameters into account (Figure 5; based on MI_CL), the ranking of Run 2 showed that ranking positions differed only by one position relative to Run 1. In both runs the same clone mAb1_C5 was identified as top clone (#1). Further, the positions #7 and #8 (mAb1_C2 (MI_CL: 0.63 and 0.43) and mAb1_C7 (MI_CL: 0.43 and 0.54)) as well as #5 and #6 (mAb1_C1 (MI_CL: 0.69 and 0.57) and mAb1_C3 (MI_CL: 0.68 and 0.65)) were shared by the same clones. Only for mAb1_C8, a greater change by 2 positions was observed, where for Run 1 it ranked at position #4 (MI_CL: 0.69 (0.43–1.00)) and for Run 2 it ranked at #2 (MI_CL: 0.86 (0.43–1.00)).

Moreover, comparing across ranking methods from Table 2 (based on q_P) and Figure 5 (based on MI_CL), it can be seen that for both methods the same clone mAb1_C5 ranked at #1 (q_P: 46.57 and 40.98 pg cell⁻¹ d⁻¹) and mAb1_C4 was among the top 3 candidates (q_P: 32.14 and 35.13 pg cell⁻¹ d⁻¹; MI_CL: 0.89 and 0.83). However, some clones ranked considerably worse when using the MI_CL than when using the q_P strategy. For example, mAb1_C2 ranked at #7 and #8 for the MI_CL strategy but at middle positions (#3, #5) for q_P. Similarly for mAb1_C3, with positions #5 and #6 for MI_CL (0.68 and 0.65 (0.43–1.00 and 0.44–1.00)), but at position #8 for the q_P based ranking strategy (q_P: 20.18 and 18.81 pg cell⁻¹ d⁻¹ (16.46–46.57 and 12.21–40.98 pg cell⁻¹ d⁻¹)), while for mAb1_C6 the reverse was observed, ranking better using the MI_CL at #3 and #4 (0.75 and 0.746 (0.43–1.00 and 0.43–1.00)) than the q_P based strategy at #7 and #4 (q_P: 22.13 and 28.77 pg cell⁻¹ d⁻¹ (16.46–46.57 and 12.21–40.98 pg cell⁻¹ d⁻¹)).

To evaluate reproducibility, two independent SP-HIP-75% experiments were conducted, demonstrating comparable growth and metabolite dynamics. The q_P-based ranking yielded identical top two clones but exhibited some variability in lower-ranked positions. In contrast, the MI_CL-based ranking identified the same top clone with minimal positional shifts in subsequent rankings, resulting in greater overall consistency across the two runs.

Experiments comparing the different operation modes and media were repeated with 6 clones expressing a bispecific antibody (see Figure 6 and Supplementary Material).

The comparison of FB, SP-CD CHO and SP-HIP showed similar results regarding growth, and metabolism to those obtained for mAb1 (Figure 6). The most prominent difference was the much shorter duration of the FB cultures for the bspAb1 clones of just 7 days instead of 10 days, due to an earlier reduction of viabilities around day 5. Semi-perfusion cultures ran for 10 days, and although clones in SP-HIP obtained higher maximum VCCs than clones in SP-CD CHO, the latter had higher viabilities until the end of culture as seen with the mAb1 cell clones. Furthermore, the dynamics of the metabolite concentrations were similar for the mAb1 and bspAb1 cell clones. The glucose concentrations depleted earlier for the SP-HIP cultures around day 5, while for SP-CD CHO cultures glucose was depleted from around day 7. For lactate, only the SP-HIP cultured clones showed a reduction in concentration around days 5 and 6, which coincided with the depletion of glucose, while for clones in FB and SP-CD CHO lactate concentration remained stable throughout. Overall, lactate concentrations remained below 30 mmol L⁻¹ and ammonium concentrations remained below 10 mmol L⁻¹ with the sudden increase of concentration on day 5–6 for clones in SP-HIP. Moreover, the comparison of growth and metabolite dynamics for semi-perfusion with different perfusion rates (SP-HIP vs. SP-HIP-75%) were nearly identical, as seen for the mAb1 cell clones. The only exception was bspAb1_C3 where VCCs were much lower for SP-HIP-75% than for SP-HIP.

The ranking was performed as described before, first based on the q_P and second on the MI_CL strategy. In Supplementary Table S2 of the Supplementary Material, the q_P based clone rankings are presented, where a colour code was employed, assigning a colour to each clone ranging from dark blue (bspAb1_C1) to dark red (bspAb1_C8), aiming to simplify the visualisation of the ranking. In addition, the raw q_P values are reported in Supplementary Table S3, in range of 1.37–6.82, 11.07–23.90, 6.31–14.93 and 6.90–26.62 pg cell^-1 d^-1 for FB, SP-CD CHO, SP-HIP and SP-HIP-75%, respectively. The MI_CL was in range of 0.06–1.00, 0.39–1.00, 0.30–1.00 and 0.27–1.00 for FB, SP-CD CHO, SP-HIP and SP-HIP-75%, respectively.

While for mAb1, the same clone was ranked top for all operation modes (FB vs. SP-CD CHO vs. SP-HIP) when using the q_P strategy, the opposite was true for the bspAb1 cell line, where the worst clone was identical (bspAb1_C6; Supplementary Table S3). Similarly, some ranking differences between SP-CD CHO and SP-HIP were observed. For the bspAb1, the top clone differed while the following ranking positions were shared by the same clones, while for the mAb1 clones, the lower ranking positions differed, and the higher positions were the same. Nonetheless, general differences between operation modes (FB vs. SP) were found, for example, bspAb1_C2 was consistently ranked lower in SP (#5; q_P: 14.2 pg cell⁻¹ d⁻¹ (11.07–23.90 pg cell⁻¹ d⁻¹) and 12.3 pg cell⁻¹ d⁻¹ (6.31–14.93 pg cell⁻¹ d⁻¹) for SP-CD CHO and SP-HIP, respectively) than in FB (#2; q_P 5.3 pg cell⁻¹ d⁻¹ (1.37–6.82 pg cell⁻¹ d⁻¹)) while bspAb1_C5 consistently ranked higher in SP (#3) than in FB (#5). For the comparison between partial and total medium exchanges different rankings were obtained where particularly higher ranks showed variation while the same two clones (bspAb1_C2 and bspAb1_C6) ranked at #5 and #6 for SP-HIP and SP-HIP-75% (Supplementary Table S3). The q_P for bspAb1_C2 was 12.34 and 11.22 pg cell⁻¹ d⁻¹, and for bspAb1_C6, q_P was 6.31 and 6.90 pg cell⁻¹ d⁻¹ for SP-HIP and SP-HIP-75%, respectively. It is worth noting that the q_P ranges achieved were larger for SP-CD CHO (11.07–23.9 pg cell⁻¹ d⁻¹) and SP-HIP-75% (6.90–26.62 pg cell⁻¹ d⁻¹) than for SP-HIP (6.31–14.93 pg cell⁻¹ d⁻¹), while for the mAb1 cell line all experiments in SP resulted in similar ranges of q_P (see Section 3.2.1; Supplementary Table S1).

When considering multiple parameters and using the MI_CL as basis for the ranking, FB and SP as well as SP with total and partial medium exchanges resulted in similarly different rankings as seen for q_P (Figure 7). However, the comparison across ranking strategies showed some interesting results, i.e., for both methods the FB ranking was identical. Further the worst performing clone was the same for all operation modes and perfusion rates and across ranking strategies (bspAb1_C6). However, for SP cultures some clones ranked consistently worse with the q_P strategy compared to the MI_CL strategy or vice versa. For example, bspAb1_C2 ranked at #5 and #3 for the q_P and MI_CL strategies, respectively, while bspAb1_C3 ranked at #5 for MI_CL and at the top two positions for the q_P strategy.

To assess applicability across cell lines and protein products, the study was replicated with a second cell line producing bspAb1. Similar to the mAb1 cell line, growth dynamics varied between FB and SP. While glucose, ammonium, and osmolarity concentrations exhibited comparable patterns across methods, lactate levels differed in SP-HIP and SP-HIP-75% cultures compared to FB and SP-CD CHO. Clone rankings revealed shared low-ranking clones, contrasting with the mAb1 cell line. Nevertheless, distinct rankings emerged between FB and SP with varying media and perfusion rates, corroborating previous findings.

In a first step, a screening of cell clones was carried out with a MWP methodology in fed-batch (Silk et al., 2010) and semi-perfusion at 1 RV d⁻¹ (Tregidgo et al., 2023a) using fed-batch as well as perfusion specific media. The experiments aimed to understand what impact the media might have on the clone performance in addition to the shift in the operation mode.

For the mAb1 cell lines, the growth dynamics of SP cultures with total medium exchanges in both media were comparable and achieved a 2- to 5-fold increase in maximum VCCs compared to FB cultures. Similar patterns were observed for the bspAb1 cell clones (Figure 6). Further, viabilities of FB cultures decreased earlier than for SP cultures, which was most notable for the bspAb1 cell clones where the cultivation time of FB cultures was significantly shorter (7 days instead of 10 days).

The initial hypothesis was that glucose depletion, which was observed in all clones for all operation modes, in combination with the accumulation of toxic by-products such as lactate and ammonium, led to the reduced viability. Previous studies on CHO cells have shown a 25% reduction in growth at lactate concentrations of 60 mmol L⁻¹ or ammonium concentrations greater than 10–15 mmol L⁻¹ (Lao and Toth, 1997). However, lactate and ammonium concentrations of all clones in FB cultures (for both products) were shown to be similar or lower compared to SP cultures (in both media). Further, lactate and ammonium concentrations remained below 30 mmol L⁻¹ and 10 mmol L⁻¹, respectively (for all cultivations and cell lines) and thus below levels typically considered toxic. Moreover, in SP cultures a reduction of lactate concentration was observed, which coincided with the depletion of glucose and indicated a shift of metabolism from lactate production to consumption, as previously described in the literature for CHO cells (Altamirano et al., 2000; 2004; Mulukutla et al., 2012). Interestingly, only SP cultures in HIP medium (of both cell lines) showed a drastic reduction and depletion of lactate at the end of cultivation. The depletion of both glucose and lactate is the most likely cause for the reduced growth and greater reduction of viability in SP-HIP cultures compared to SP-CD CHO cultures, where lactate was not depleted.

In addition to the evaluation of operation modes, SP was investigated in greater detail by applying different perfusion rates. This aimed to understand potential differences of clone performance and resulting cell clone rankings between a typical manual SP operation with 1 RV d⁻¹ whilst a prospective automated workflow would use a partial medium exchange (often at a fixed rate, e.g., not more than 75% of the working volume – 0.75 RV d⁻¹) due to the operation with a liquid handling arm and the use of sedimentation as the cell retention method (Bielser et al., 2019; Jin et al., 2021). For both perfusion rates, the growth and metabolic performance was nearly identical for all the mAb1 and bspAb1 cell lines. A faster reduction and depletion of glucose in combination with an earlier change of metabolism to lactate consumption was expected for SP-HIP-75% compared to SP-HIP cultures, however the profiles obtained were very similar. The reason for this observation could be that a change in metabolism for the SP-HIP-75% cultured clones occurred, where higher lactate concentrations led to a reduction in lactate production rate and higher consumption rate even in the presence of glucose. Similar observations were previously made for batch and fed-batch CHO cultures (Altamirano et al., 2004; Dorai et al., 2009; Lao and Toth, 1997; Li et al., 2012).

For the cell clone ranking, two strategies were used and compared. Previous studies conducted at mL-scale showed that clone ranking is often based on a single productivity parameter (Bielser et al., 2019; Markert et al., 2019). In batch and fed-batch operations the use of accumulated end-point titre is very common. However, in this work two distinctly different operation modes, FB and SP, were being compared. Thus, the selected parameter needed to account for the accumulation of the product in FB operation as well as the medium exchange leading to product removal in SP. The cell-specific production rate, q_P, was suggested in the literature for early-stage screening at low cell concentrations to avoid early exclusion of clones (Markert et al., 2019). The calculation of q_P carried out in this work took the variable flow rates of SP and FB into account, allowing for a fair comparison between operation modes. In the first instance a clone ranking based on the q_P was carried out.

For mAb1, the ranking of clones for SP-CD CHO and SP-HIP was nearly identical with minor differences of low performing clones at positions #6 to #8. Thus, it was initially hypothesised that the medium had little impact on the outcome of the ranking. The comparison between operation modes (FB vs. SP-HIP) showed greater differences. Some clones ranked either high or low, while the top clone was identical for both operation modes. Some clones changed ranking positions quite significantly, and by more than 2 positions (e.g., mAb1_C6). Similar observations can be made regarding the comparison between the different perfusion rates (SP-HIP vs. SP-HIP-75%), where, for example, the top clone for SP-HIP-75% ranked at position #4 for SP-HIP. These results on the clone ranking based on the q_P led to the hypotheses that both (i) the operation mode and (ii) the perfusion rate impact the clone ranking. Consequently, the selection of clone candidates for transfer to perfusion bioreactors for further selective screening differs depending on the chosen method (operation mode) and the use of a manual or automated workflow (perfusion rate).

In a complementary study, the bspAb1 cell clone ranking between FB and SP showed more ranking position changes than observed for the mAb1 cell line ranking. For example, bspAb1_C2 performed considerably worse in SP culture (#5) than in FB culture (#2). It should be noted that the rankings comparing SP-CD CHO and SP-HIP (1 RV d⁻¹) showed some differences in the better performing clones as opposed to the worst performing for mAb1. It is noteworthy that this observation could have been caused by the smaller number of clones investigated for bspAb1. When comparing the clone rankings for SP cultures with total and partial medium exchanges (1 RV d⁻¹ vs. 0.75 RV d⁻¹), ranking differences for top performing clones were similar to the ranking of the mAb1 clones. Even though the number of clones investigated for bspAb1 was smaller than for mAb1, a trend seems to emerge showing that the cell clone ranking was being impacted by the operation mode and perfusion rate. However, these results also led to a re-evaluation of the impact of medium on the cell clone ranking. The exact compositions of the commercially-available media (ThermoFisher) are unknown, but it is stated that the HIP medium has a higher native concentration of many components compared to a FB medium (e.g., CD CHO), which is intended to be paired with a feed for best performance (ThermoFisher Scientific, 2024). While further studies testing additional media with larger panels of clones would give more insight on the impact of media on ranking, the influence is likely to be minor in comparison to the impact the different operation modes and the perfusion rates have. Additionally, investigations under glucose access should be considered. A preliminary clone screening with additional glucose supplementation in the perfusion-specific medium (HIP) and total medium exchange was performed with the mAb1 clones (referred to as SP-HIP20; see Supplementary Figure S1; Supplementary Table S2). The investigation showed only minor differences in single-parameter ranking between conditions with or without glucose supplementation similar to differences between different media while significant differences remained evident between operation modes. Given these findings, we chose to proceed with glucose-limited conditions, as real-time glucose control is generally not feasible at this scale during cell clone screening. Medium optimisation is typically carried out in subsequent studies focused on the lead clone, conducted at bioreactor scale, where glucose concentrations will be measured, monitored, and controlled in real-time to maintain a predetermined range.

The small working volume of well plates in the initial cell clone screening often limits the number of studied parameters to VCCs and titre. However, the need for greater predictability and process robustness has driven the efforts to develop high-throughput analytical technologies using less volume per single analysis, thus allowing the analysis of additional parameters (e.g., metabolites). Although the process operation of well plate systems often differs from later stage bioreactor operations (i.e., orbital shaking vs. stirring), early insights in growth and metabolic behaviour as well as process performance can be valuable for process development and clone selection. Previous studies have shown that the MWP and a perfusion bioreactor compare well regarding the growth and production performance (Tregidgo et al., 2023a). We also performed comparability studies for 3 of the 8 clones, comparing MWP data (1 RV d⁻¹) with historical data for 7 L perfusion bioreactors (1.2 RV d⁻¹). These showed consistently good comparability for both growth and productivity metrics (data not shown).

The MWP methods used in this study allowed the collection of enough material to measure several parameters of growth (i.e., VCC, viability), productivity (i.e., titre), and metabolism. Based on these measurements, additional parameters such as cell specific rates were determined. Some studies have included growth parameters (e.g., VCC or growth rate) into their ranking (Gagliardi et al., 2019) by assigning a score to each parameter and determining ranks based on an overall score calculated with equally weighted parameters. However, this approach is often performed manually and is limited to a small number of clones and parameters to be feasible. Industrial cell clone screening processes assess several hundreds of candidates in MWPs; therefore, the manual assessment of each parameter, the assignment of scores and determination of ranks for all clones is an impractical and daunting task. To address this need, Goldrick et al. have developed a new index, the manufacturability index (MI_CL), to summarize the cell clone performance by combining several parameters into a single numerical value that considers both the ranking within each parameter as well as the importance of each parameter through weighting (Goldrick et al., 2023).

In a second instance, the MI_CL approach was adapted to the investigation in MWPs comparing the different operation modes and another clone ranking based on the MI_CL was performed. The parameters, such as yields and cell-specific rates, were carefully selected considering the different feeding and medium exchange protocols that lead to accumulation or dilution of metabolites and product, to allow a fair comparison between FB and SP. Ten parameters were selected in total (see Section 2.4), where the lactate concentration was included as the only volumetric parameter impacted by the differences in feed and medium exchange protocols. This decision was made based on the dual nature of lactate as a toxic by-product potentially inhibiting cell growth but also serving as carbon-source through metabolism shift from lactate production to consumption (Luo et al., 2021; Mulukutla et al., 2012; Pereira et al., 2018).

The ranking obtained using the MI_CL strategy showed a similar overall outcome as the ranking based on the q_P with regards to different rankings between operation modes as well as between perfusion rates (for both panels of cell lines). The results emphasise that the chosen methods (FB vs. SP) at MWP scale impact the candidate selection for further screening in larger scales, leading to exclusion of clones more suitable for perfusion processes while including clones more suitable for fed-batch.

The comparison across ranking methods (q_P vs. MI_CL) gave some interesting insight. For example, the ranking for FB using either the single parameter q_P or the MI_CL metric was identical for both cell line panels, which suggests that for FB cultures the q_P is a good enough indicator for early stage ranking at MWP scale where availability of data is limited. However, for SP the same cannot be said, as some greater differences between ranking methods (q_P vs. MI_CL) could be observed. The most prominent example was mAb1_C7 in SP-CD CHO, which ranked low (#6) for q_P but on top position for MI_CL ranking. The most likely explanation for this finding was the better growth observed for mAb1_C7 in SP-CD CHO, where viabilities were high throughout and the highest maximum VCCs was achieved, compared to FB, SP-HIP, or SP-HIP-75%. Another possible reason is that the medium does have some impact on the ranking. This was supported by the ranking results for the bspAb1 cell clones, where some differences between SP-CD CHO and SP-HIP were already observed for the ranking based on q_P and the ranking based on q_P and MI_CL for SP-HIP20 with higher glucose concentrations (see Supplementary Figures S1, S2; Supplementary Table S2). Another interesting example was clone mAb1_C6, which ranked overall better in SP cultures compared to FB, including SP-HIP-75% where it had ranked much lower for the q_P based ranking. This also led to an interesting observation comparing ranking methods (q_P vs. MI_CL) for SP cultures. It seems that ranking results of SP cultures, regardless of the medium or the perfusion rate, are more consistent using the MI_CL strategy. With some exceptions, the same mAb1 clones cluster to rank at either higher or lower positions with mAb1_C1 being a “barrier” in the middle. For the bspAb1, the greater differences can likely be attributed to the smaller number of clones. Nonetheless, a clustering of clones at low-ranking positions can also be observed. This could be seen as supporting the hypothesis of a more consistent ranking across perfusion rates and media when using the MI_CL strategy for SP cultivations.

Lastly, the reproducibility of the results was evaluated in this work. Particularly the semi-perfusion screening was of importance as the different ranking strategies showed the ranking positions of individual clones differed. To further investigate this, the SP-HIP-75% condition and the mAb1 clones were selected and experiments repeated to better understand the statistical variability.

The comparison of the two runs showed a close similarity of growth and metabolic performance, even though the second run showed slightly slower growth. A possible reason for this could be that Run 2 (P16) was initiated at a later passage number than Run 1 (P10). However, previous studies found no significant impairment of performance for passage numbers up to P20 (Kaur et al., 2023; Qian et al., 2020). Another likely cause for this observation might be the manual operation of this clone screening, which is more prone to operator error and day-to-day performance variation, thus potentially leading to less accurate medium exchanges or accidental disturbance of the cell pellet during the medium exchanges. It is expected that automating the workflow would lead to higher precision and less difference between identical runs, ultimately increasing reproducibility and robustness.

The clone ranking of both runs using the single parameter strategy based on the q_P is reported in Table 2 and showed that 4 out of 8 clones were ranked at identical positions including the two top clones. Of the remaining 4 clones, 2 showed changes by two ranking positions (from #3 to #5 and vice versa) while the remaining 2 showed even greater position changes. It is possible that the growth differences of Run 2 caused the observed variation in the q_P based ranking, though it is unlikely this is the only cause. For example, the clones mAb1_C2 and mAb1_C6 both achieved higher VCC in Run 1 than in Run 2, which would suggest that the productivity was higher in Run 1. However, while this assumption seems to be accurate for mAb1_C2, which ranked at position #3 in Run 1 and dropped by 2 ranking positions to #5 (Run 2), mAb1_C6 improved by 3 positions from #7 (Run 1) to #4 (Run 2).

For Run 1, the comparison of both ranking strategies (q_P vs. MI_CL) had shown significant position changes for some clones (Section 3.2.2). Although the top two clones had been identical, some clones such as mAb1_C2 were ranked much lower while others (e.g., mAb1_C6) ranked higher with the MI_CL strategy. For Run 2 similar observations were made, where, for example, mAb1_C2 dropped from position #5 for q_P to #8 for MI_CL. Comparing the rankings of the MI_CL strategy of both runs showed that only the top clone was identical in both runs. While the remaining positions did not share identical clones, a clustering of clones was observed, such that mAb1_C4, mAb1_C6 and mAb1_C8 ranked “high” in both runs, whereas the remaining clones ranked consistently “low”. In addition, the clones appeared in pairs, where two clones shared the same two positions in both runs (e.g., mAb1_1 and mAb1_C3 on positions #5 and #6). The small difference by one ranking position, can be attributed to the more error-prone manual operation of this investigation.

To summarize, although multiple clones ranked in the same position for the q_P strategy, as opposed to only one for the MI_CL strategy, the overall ranking based on q_P showed more variation than that based on MI_CL. This was due to the fact that several clones changed their ranks by two or more positions for the q_P strategy, while for the MI_CL strategy, position changes were limited to only one rank (with the exception of mAb1_C8 which changed by two ranks). Consequently, a more consistent ranking pattern was obtained using the MI_CL strategy. This suggests that ranking based on MI_CL is generally more consistent, which is a hypothesis that was previously developed based on results from other SP conditions (Sections 3.2.2, 4.2).