Conclusion
The methods presented in this study were consolidated as a single platform approach to rapidly optimize amino acid concentrations as CD media supplements (Fig. 8). Since amino acids are one of the most important building blocks for cellular biomass and protein production, they also comprise an essential component group in CD media. In addition, since the relative amino acid requirement is similar across CHO cell lines and similar between various mAbs, the amino acid modeling platform presented here can be applied to various CHO cell cultures expressing various mAbs (Széliová et al., 2020). Nevertheless, Xie and Wang’s original stoichiometric model was inclusive of all media nutrients and thus, our empirical modeling approach and methods to calculate stoichiometric balances could be generalized to other nutrients as well, for example, vitamins, co-factors, trace elements, organic, and inorganic salts among others. Although raw measurements of nutrient concentrations can be directly input into MVDA approaches (Fig. 8 (1a)), calculation of secondary variables such as cell growth rate, specific productivity, and consumption rates can help formulate stoichiometric balances which in turn can serve as enhanced inputs to derive causal linkages (Fig. 8 (1b-1c)). Within the modeling platform, MVDA dimensional reduction algorithms such as PLS or OPLS can be utilized to understand the correlation between the dynamics of time-dependent stoichiometric balances and key response variables such as cell growth or mAb productivity. VIP statistics and correlation coefficients from multivariate algorithms can help generate importance and directionality criteria for the input variables. Key nutrients towards specific response variables or cellular phenotypical states can be further translated into experimental designs for validation studies (Fig. 8 (2a)). However, in the case of stoichiometric balances, a feature selection decision tree approach can be implemented to choose the appropriate stoichiometric balances for process developing. Bolded here is the positive correlated balances that are negative in value which potentially represent limiting nutrients as shown in Criterion 2 (Fig. 8 (2b-2c)). However, negatively correlated stoichiometric balances could also be considered for media formulation development. To support multivariate statistical models, experimental validation is necessary to not only verify the performance of the model, but also measure the success criteria of the desired phenotypical state. Model outputs can dictate supplementation with modified nutrient feeds to be validated in a targeted DoE, which in turn can generate new cell culture data to refine against the MVDA model (Fig. 8 (3 – 4)).
Since the prediction of the statistical model is largely based on the variance of the training dataset, high variance datasets can provide a larger prediction space for the model and a reduction in model bias. However, models training on datasets with high variance rely heavily on the upper and lower bounds of the variance space and thus, any validation datasets that fall outside of the training dataset space can lead to poor success. Therefore, as new processes are adapted based on model-driven conditions, future datasets can be combined with training datasets to iteratively improve the model (Fig. 8 (5)) (Luo et al., 2020). For instance, important amino acid stoichiometric balances based on high nutrient feeds were also fed to the moderate and low nutrient conditions resulting in significant increases in cell growth and mAb productivity. However, in future iterations the newly generated validation datasets could help improve the original training dataset and help increment towards an improved biologics production process and optimized CD media.
References
Ahn, W. S., & Antoniewicz, M. R. (2011). Metabolic flux analysis of CHO cells at growth and non-growth phases using isotopic tracers and mass spectrometry. Metabolic Engineering, 13 (5), 598-609. doi:https://doi.org/10.1016/j.ymben.2011.07.002
Akarachantachote, N., Chadcham, S., & Saithanu, K. (2013). Cutoff threshold of variable importance in projection for variable selection.International Journal of Pure and Applied Mathematics, 94 (3)
Alden, N., Raju, R., McElearney, K., Lambropoulos, J., Kshirsagar, R., Gilbert, A., & Lee, K. (2020). Using metabolomics to identify cell line-independent indicators of growth inhibition for chinese hamster ovary cell-based bioprocesses. Metabolites, 10 (5), 199. Retrieved from https://pubmed.ncbi.nlm.nih.gov/32429145
Balandras, F., Olmos, E., Hecklau, C., Blanchard, F., Guedon, E., & Marc, A. (2011). Growth and death kinetics of CHO cells cultivated in continuous bioreactor at various agitation rates. BMC Proceedings, 5 Suppl 8 (Suppl 8), P101-6561-5-S8-P101. eCollection 2011. doi:10.1186/1753-6561-5-S8-P101 [doi]
Behrouz, H., Molavi, B., Tavakoli, A., Askari, M., Maleknia, S., Mahboudi, F., & Khodadadian, M. (2020). Multivariate optimization of the refolding process of an incorrectly folded fc-fusion protein in a cell culture broth. Current Pharmaceutical Biotechnology, 21 (3), 226-235. doi:10.2174/1389201020666191002144424 [doi]
Brosnan, J. T. (2003). Interorgan amino acid transport and its regulation. The Journal of Nutrition, 133 (6 Suppl 1), 2068S-2072S. doi:10.1093/jn/133.6.2068S [doi]
Bylesjo M, Rantalainen, M., Cloarec, O., Nicholson, J. K., Holmes, E., & Trygg, J. (2006). OPLS discriminant analysis: Combining the strenghts of PLS-DA and SIMCA classification. Journal of Chemometrics,(20), 341-351.
Carrillo-Cocom, L. M., Genel-Rey, T., Araiz-Hernandez, D., Lopez-Pacheco, F., Lopez-Meza, J., Rocha-Pizana, M. R., . . . Alvarez, M. M. (2015). Amino acid consumption in naive and recombinant CHO cell cultures: Producers of a monoclonal antibody. Cytotechnology, 67 (5), 809-820. doi:10.1007/s10616-014-9720-5 [doi]
Cuperlovic-Culf, M., & Culf, A. S. (2016). Applied metabolomics in drug discovery. Expert Opinion on Drug Discovery, 11 (8), 759-770. doi:10.1080/17460441.2016.1195365
Del Val, I. J., Polizzi, K. M., & Kontoravdi, C. (2016). A theoretical estimate for nucleotide sugar demand towards chinese hamster ovary cellular glycosylation. Scientific Reports, 6 , 28547. doi:10.1038/srep28547 [doi]
Dietmair, S., Hodson, M. P., Quek, L. E., Timmins, N. E., Chrysanthopoulos, P., Jacob, S. S., . . . Nielsen, L. K. (2012). Metabolite profiling of CHO cells with different growth characteristics.Biotechnology and Bioengineering, 109 (6), 1404-1414. doi:10.1002/bit.24496 [doi]
Duarte, T. M., Carinhas, N., Barreiro, L. C., Carrondo, M. J., Alves, P. M., & Teixeira, A. P. (2014). Metabolic responses of CHO cells to limitation of key amino acids. Biotechnology and Bioengineering, 111 (10), 2095-2106. doi:10.1002/bit.25266 [doi]
Dutton, R. L., Scharer, J., & Moo-Young, M. (2006). Cell cycle phase dependent productivity of a recombinant chinese hamster ovary cell line.Cytotechnology, 52 (1), 55-69. doi:10.1007/s10616-006-9041-4 [doi]
Fan, Y., Jimenez Del Val, I., Muller, C., Wagtberg Sen, J., Rasmussen, S. K., Kontoravdi, C., . . . Andersen, M. R. (2015). Amino acid and glucose metabolism in fed-batch CHO cell culture affects antibody production and glycosylation. Biotechnology and Bioengineering, 112 (3), 521-535. doi:10.1002/bit.25450 [doi]
Fomina-Yadlin, D., Gosink, J. J., McCoy, R., Follstad, B., Morris, A., Russell, C. B., & McGrew, J. T. (2014). Cellular responses to individual amino-acid depletion in antibody-expressing and parental CHO cell lines. Biotechnology and Bioengineering, 111 (5), 965-979. doi:10.1002/bit.25155 [doi]
Gangadharan, N., Turner, R., Field, R., Oliver, S. G., Slater, N., & Dikicioglu, D. (2019). Metaheuristic approaches in biopharmaceutical process development data analysis. Bioprocess and Biosystems Engineering, 42 (9), 1399-1408. doi:10.1007/s00449-019-02147-0 [doi]
Harcum, S. W., & Lee, K. H. (2016). CHO cells can make more protein.Cell Systems, 3 (5), 412. doi:https://doi.org/10.1016/j.cels.2016.11.007”
Hassan, S., Farhan, M., Mangayil, R., Huttunen, H., & Aho, T. (2013). Bioprocess data mining using regularized regression and random forests.BMC Systems Biology, 7 Suppl 1 (Suppl 1), S5-0509-7-S1-S5. Epub 2013 Aug 12. doi:10.1186/1752-0509-7-S1-S5 [doi]
Hoshan, L., Jiang, R., Moroney, J., Bui, A., Zhang, X., Hang, T. C., & Xu, S. (2019). Effective bioreactor pH control using only sparging gases. Biotechnology Progress, 35 (1), e2743. doi:10.1002/btpr.2743 [doi]
Hsu, W. T., Aulakh, R. P., Traul, D. L., & Yuk, I. H. (2012). Advanced microscale bioreactor system: A representative scale-down model for bench-top bioreactors. Cytotechnology, 64 (6), 667-678. doi:10.1007/s10616-012-9446-1 [doi]
Insaidoo, F., Banerjee, S., Roush, D., & Cramer, S. (2017).Computational modeling in bioprocess developmentdoi:https://doi.org/10.1002/9781119031116.ch6
Kanehisa, M., & Goto, S. (2000). KEGG: Kyoto encyclopedia of genes and genomes. Nucleic Acids Research, 28 (1), 27-30. doi:gkd027 [pii]
Kelley, B. (2009). Industrialization of mAb production technology: The bioprocessing industry at a crossroads. mAbs, 1 (5), 443-452. doi:9448 [pii]
Kim, O. D., Rocha, M., & Maia, P. (2018). A review of dynamic modeling approaches and their application in computational strain optimization for metabolic engineering. Frontiers in Microbiology, 9 , 1690. doi:10.3389/fmicb.2018.01690 [doi]
Kirdar, A. O., Conner, J. S., Baclaski, J., & Rathore, A. S. (2007). Application of multivariate analysis toward biotech processes: Case study of a cell-culture unit operation. Biotechnology Progress, 23 (1), 61-67. doi:10.1021/bp060377u [doi]
Kroll, P., Hofer, A., Ulonska, S., Kager, J., & Herwig, C. (2017). Model-based methods in the biopharmaceutical process lifecycle.Pharmaceutical Research, 34 (12), 2596-2613. doi:10.1007/s11095-017-2308-y [doi]
Kumar, V., Bhalla, A., & Rathore, A. S. (2014). Design of experiments applications in bioprocessing: Concepts and approach.Biotechnology Progress, 30 (1), 86-99. doi:10.1002/btpr.1821 [doi]
Kyriakopoulos, S., Polizzi, K. M., & Kontoravdi, C. (2013). Dynamic profiling of amino acid transport and metabolism in chinese hamster ovary cell culture. BMC Proceedings, 7 (Suppl 6), P97-6561-7-S6-P97. eCollection 2013. doi:1753-6561-7-S6-P97 [pii]
Lee, K. M., & Gilmore, D. F. (2006). Statistical experimental design for bioprocess modeling and optimization analysis: Repeated-measures method for dynamic biotechnology process. Applied Biochemistry and Biotechnology, 135 (2), 101-116. doi:ABAB:135:2:101 [pii]
Li, F., Vijayasankaran, N., Shen, A. Y., Kiss, R., & Amanullah, A. (2010). Cell culture processes for monoclonal antibody production.mAbs, 2 (5), 466-479. doi:12720 [pii]
Lim, J. M., Kim, G., & Levine, R. L. (2019). Methionine in proteins: It’s not just for protein initiation anymore. Neurochemical Research, 44 (1), 247-257. doi:10.1007/s11064-017-2460-0 [doi]
Luo, Y., Lovelett, R. J., Price, J. V., Radhakrishnan, D., Barnthouse, K., Hu, P., . . . Ogunnaike, B. A. (2020). Modeling the effect of amino acids and copper on monoclonal antibody productivity and glycosylation: A modular approach. Biotechnology Journal, , e2000261. doi:10.1002/biot.202000261 [doi]
Martínez, J. A., Bulté, D. B., Contreras, M. A., Palomares, L. A., & Ramírez, O. T. (2020). Dynamic modeling of CHO cell metabolism using the hybrid cybernetic approach with a novel elementary mode analysis strategy. Frontiers in Bioengineering and Biotechnology, 8 , 279. doi:10.3389/fbioe.2020.00279
Martins Conde Pdo, R., Sauter, T., & Pfau, T. (2016). Constraint based modeling going multicellular. Frontiers in Molecular Biosciences, 3 , 3. doi:10.3389/fmolb.2016.00003 [doi]
Mevik, B. H., & Wehrens, R. (2007). The pls package: Principal component and partial least squares regression in R. Journal of Statistical Software, 18 (2)
MKS Umetrics AB (Ed.). (2013). Multi- and megavariate data analysis (3rd ed.). Sweden:
Morris, C., Polanco, A., Yongky, A., Xu, J., Huang, Z., Zhao, J., . . . Yoon, S. (2020). Bigdata analytics identifies metabolic inhibitors and promoters for productivity improvement and optimization of monoclonal antibody (mAb) production process. Bioresources and Bioprocessing, 7 (1), 31. Retrieved from https://doi.org/10.1186/s40643-020-00318-6
Nitish, B., & Rathore, A. (2011). Use of multivariate data analysis (MVDA) for generating process understanding from manufacturing data of biotech processes. Proceedings of the Indian National Science Academy, 77 , 133-142.
Noguchi, Y., Nishikata, N., Shikata, N., Kimura, Y., Aleman, J. O., Young, J. D., . . . Stephanopoulos, G. (2010). Ketogenic essential amino acids modulate lipid synthetic pathways and prevent hepatic steatosis in mice. PloS One, 5 (8), e12057. doi:e12057 [pii]
Noll, P., & Henkel, M. (2020). History and evolution of modeling in biotechnology: Modeling & simulation, application and hardware performance. Computational and Structural Biotechnology Journal, 18 , 3309-3323. doi:10.1016/j.csbj.2020.10.018 [doi]
Pan, X., Dalm, C., Wijffels, R. H., & Martens, D. E. (2017a). Metabolic characterization of a CHO cell size increase phase in fed-batch cultures. Applied Microbiology and Biotechnology, 101 (22), 8101-8113. doi:10.1007/s00253-017-8531-y [doi]
Pan, X., Dalm, C., Wijffels, R. H., & Martens, D. E. (2017b). Metabolic characterization of a CHO cell size increase phase in fed-batch cultures. Applied Microbiology and Biotechnology, 101 (22), 8101-8113. doi:10.1007/s00253-017-8531-y [doi]
Pan, X., Streefland, M., Dalm, C., Wijffels, R. H., & Martens, D. E. (2017). Selection of chemically defined media for CHO cell fed-batch culture processes. Cytotechnology, 69 (1), 39-56. doi:10.1007/s10616-016-0036-5 [doi]
Park, S. Y., Reimonn, T. M., Agarabi, C. D., Brorson, K. A., & Yoon, S. (2018). Metabolic responses and pathway changes of mammalian cells under different culture conditions with media supplementations.Biotechnology Progress, 34 (3), 793-805. doi:10.1002/btpr.2623 [doi]
Pascoe, D. E., Arnott, D., Papoutsakis, E. T., Miller, W. M., & Andersen, D. C. (2007). Proteome analysis of antibody-producing CHO cell lines with different metabolic profiles. Biotechnology and Bioengineering, 98 (2), 391-410. doi:10.1002/bit.21460 [doi]
Pereira, S., Kildegaard, H. F., & Andersen, M. R. (2018). Impact of CHO metabolism on cell growth and protein production: An overview of toxic and inhibiting metabolites and nutrients. Biotechnology Journal, 13 (3), e1700499. doi:10.1002/biot.201700499 [doi]
Powers, D. N., Trunfio, N., Velugula-Yellela, S. R., Angart, P., Faustino, A., & Agarabi, C. (2020). Multivariate data analysis of growth medium trends affecting antibody glycosylation.Biotechnology Progress, 36 (1), e2903. doi:10.1002/btpr.2903 [doi]
Prieto, B. G., Eriksson, L., & Tyrgg, J. (2014). Variable influence on projection (VIP) for orthogonal projections to latent structures (OPLS).Journal of Chemometrics, 28 (8), 2627.
Pybus, L. P., Dean, G., West, N. R., Smith, A., Daramola, O., Field, R., . . . James, D. C. (2014). Model-directed engineering of ”difficult-to-express” monoclonal antibody production by chinese hamster ovary cells. Biotechnology and Bioengineering, 111 (2), 372-385. doi:10.1002/bit.25116 [doi]
Rencher, A. C. (1993). The contribution of individual variables to hotelling’s T2, wilks’ lambda, and R2. Biometrics, 49 (2), 479-489.
Robitaille, J., Chen, J., & Jolicoeur, M. (2015a). A single dynamic metabolic model can describe mAb producing CHO cell batch and fed-batch cultures on different culture media. PloS One, 10 (9), e0136815. doi:10.1371/journal.pone.0136815 [doi]
Robitaille, J., Chen, J., & Jolicoeur, M. (2015b). A single dynamic metabolic model can describe mAb producing CHO cell batch and fed-batch cultures on different culture media. PloS One, 10 (9), e0136815. doi:10.1371/journal.pone.0136815 [doi]
Rouiller, Y., Perilleux, A., Collet, N., Jordan, M., Stettler, M., & Broly, H. (2013). A high-throughput media design approach for high performance mammalian fed-batch cultures. mAbs, 5 (3), 501-511. doi:10.4161/mabs.23942 [doi]
Roush, D., Asthagiri, D., Babi, D. K., Benner, S., Bilodeau, C., Carta, G., . . . Zavalov, O. (2020). Toward in silico CMC: An industrial collaborative approach to model-based process development.Biotechnology and Bioengineering, doi:10.1002/bit.27520 [doi]
Salazar, A., Keusgen, M., & von Hagen, J. (2016). Amino acids in the cultivation of mammalian cells. Amino Acids, 48 (5), 1161-1171. doi:10.1007/s00726-016-2181-8 [doi]
Schaub, J., Clemens, C., Kaufmann, H., & Shulz, T. W. (2011). Advancing biopharmaceutical process development by system-level data analysis and integration of omics data. Advanced in biochemical engineering biotechnology (pp. 133-133-163)
Sha, S., Huang, Z., Wang, Z., & Yoon, S. (2018). Mechanistic modeling and applications for CHO cell culture development and production.Current Opinion in Chemical Engineering, 22 , 54-61. doi:https://doi.org/10.1016/j.coche.2018.08.010
Sharma, V., Tripathi, M., & Mukherjee, K. (2016). Application of system biology tools for the design of improved chinese hamster ovary cell expression platforms. Journal of Bioprocess Biotech, 6 (284)
Smiatek, J., Jung, A., & Bluhmki, E. (2020). Towards a digital bioprocess replica: Computational approaches in biopharmaceutical development and manufacturing. Trends in Biotechnology, 38 (10), 1141-1153. doi:S0167-7799(20)30142-6 [pii]
Suarez-Zuluaga, D. A., Borchert, D., Driessen, N. N., Bakker, W. A. M., & Thomassen, Y. E. (2019). Accelerating bioprocess development by analysis of all available data: A USP case study. Vaccine, 37 (47), 7081. doi:https://doi.org/10.1016/j.vaccine.2019.07.026”
Széliová, D., Ruckerbauer, D. E., Galleguillos, S. N., Petersen, L. B., Natter, K., Hanscho, M., . . . Zanghellini, J. (2020). What CHO is made of: Variations in the biomass composition of chinese hamster ovary cell lines. Metabolic Engineering, 61 , 288-300. doi:S1096-7176(20)30101-4 [pii]
Templeton, N., Smith, K. D., McAtee-Pereira, A. G., Dorai, H., Betenbaugh, M. J., Lang, S. E., & Young, J. D. (2017). Application of (13)C flux analysis to identify high-productivity CHO metabolic phenotypes. Metabolic Engineering, 43 (Pt B), 218-225. doi:S1096-7176(17)30033-2 [pii]
Templeton, N., Xu, S., Roush, D. J., & Chen, H. (2017a). (13)C metabolic flux analysis identifies limitations to increasing specific productivity in fed-batch and perfusion. Metabolic Engineering, 44 , 126-133. doi:S1096-7176(17)30187-8 [pii]
Templeton, N., Xu, S., Roush, D. J., & Chen, H. (2017b). (13)C metabolic flux analysis identifies limitations to increasing specific productivity in fed-batch and perfusion. Metabolic Engineering, 44 , 126-133. doi:S1096-7176(17)30187-8 [pii]
Torkashvand, F., Vaziri, B., Maleknia, S., Heydari, A., Vossoughi, M., Davami, F., & Mahboudi, F. (2015). Designed amino acid feed in improvement of production and quality targets of a therapeutic monoclonal antibody. PloS One, 10 (10), e0140597. doi:10.1371/journal.pone.0140597 [doi]
Traustason, B., Cheeks, M., & Dikicioglu, D. (2019). Computer-aided strategies for determining the amino acid composition of medium for chinese hamster ovary cell-based biomanufacturing platforms.International Journal of Molecular Sciences, 20 (21), 5464. doi: 10.3390/ijms20215464. doi:10.3390/ijms20215464 [doi]
Traustason, B. (2019). Amino acid requirements of the chinese hamster ovary cell metabolism during recombinant protein production.bioRxiv, doi:10.1101/796490
Vajargah, K. F., Sadeghi-Bazargani, H., Mehdizadeh-Esfanjani, R., Savadi-Oskouei, D., & Farhoudi, M. (2012). OPLS statistical model versus linear regression to assess sonographic predictors of stroke prognosis. Neuropsychiatric Disease and Treatment, 8 , 387-392. doi:10.2147/NDT.S33991 [doi]
Worley, B., & Powers, R. (2013). Multivariate analysis in metabolomics.Current Metabolomics, 1 (1), 92-107. doi:10.2174/2213235X11301010092 [doi]
Worley, B., & Powers, R. (2015). A sequential algorithm for multiblock orthogonal projections to latent structures. Chemometrics and Intelligent Laboratory Systems : An International Journal Sponsored by the Chemometrics Society, 149 (Pt B), 33-39. doi:10.1016/j.chemolab.2015.10.018 [doi]
Xie, L., Nyberg, G., Gu, X., Li, H., Möllborn, F., & Wang, D. I. (1997). Gamma-interferon production and quality in stoichiometric fed-batch cultures of chinese hamster ovary (CHO) cells under serum-free conditions. Biotechnology and Bioengineering, 56 (5), 577-582. doi:10.1002/(SICI)1097-0290(19971205)56:53.0.CO;2-9 [doi]
Xie, L., & Wang, D. I. (1994a). Applications of improved stoichiometric model in medium design and fed-batch cultivation of animal cells in bioreactor. Cytotechnology, 15 (1-3), 17-29. doi:10.1007/BF00762376 [doi]
Xie, L., & Wang, D. I. (1994b). Fed-batch cultivation of animal cells using different medium design concepts and feeding strategies.Biotechnology and Bioengineering, 43 (11), 1175-1189. doi:10.1002/bit.260431123 [doi]
Xie, L., & Wang, D. I. (1994c). Stoichiometric analysis of animal cell growth and its application in medium design. Biotechnology and Bioengineering, 43 (11), 1164-1174. doi:10.1002/bit.260431122 [doi]
Xing, Z., Kenty, B., Koyrakh, I., Borys, M., Pan, S., & Li, Z. (2011). Optimizing amino acid composition of CHO cell culture media for a fusion protein production. Process Biochemistry, (46), 1423-1423-1429.
Yamamoto, H., Yamaji H, Abe, Y., Harada, K., Waluyo, D., Fukusaki, E., . . . Fukuda H. (2009). Dimensionality reduction for metabolome data using PCA, PLS, OPLS, and RFDA with differential penalties to latent variables. Chemometrics and Intelligent Laboratory Systems, (98), 136-142.
Figure Legends
Fig. 1. Variance and distribution of training dataset: A 25-batch CHO cell culture experiment was designed to obtain a training dataset to test the amino acid stoichiometric balance MVDA model in which the culture was supplemented with various amounts of high nutrient feeds within a defined fed-batch process. Data on cell growth and mAb titer from all batches was analyzed to obtain peak growth and productivity time points and a PCA model was generated to analyze the distribution and variance of training batches with respect to the amino acid stoichiometric balances. (a) VCD peaked between day 6 to day 9 across the 25 training batches with a majority peaking near day 9 (Black Lines). Similarly, CVC at each time point was also plotted along side VCD to show a similar distribution (Green dashed lines). Cell growth data up to day 9 represented the log-growth phase of the cell culture, thus, day 9 CVC was chosen as the response variable for the growth model. (b) Titer data was normalized to the peak titer value of the fed-batch process control. Since mAb titer represented an accumulation of extracellular secreted mAb in the spent media, peak titer occurred at the end of the process on day 14. Accordingly, day 14 titer was chosen as the response variable for the production model. (c). Distribution of the batches based on principle components 1 and 2, which cumulatively described 53.5% of the variance did not show any significant grouping or clustering of the batches thereby eliminating the possibility of any inherent bias in the data. (d) Similarly, the loadings plot of the time-dependent amino acid stoichiometric balances showed a wide distribution without any collinearity of variables at specific time points.
Fig. 2. Criteria for Model-based Amino Acid Selection: VIP values and coefficients from the OPLS models along with stoichiometric balance directionalities from the process control were utilized as selection factors in 3 unique criteria for amino acids. (a-b) Criteria 1 amino acid SBs represented positive correlation to either Day 9 CVC or Day 14 Titer, a VIP value greater than 1, and a positive stoichiometric balance sign. (c-d) Criteria 2 amino acid SBs represented positive correlation to either Day 9 CVC or Day 14 Titer, a VIP value greater than 1, and a negative stoichiometric balance sign. (e-f) Criteria 3 amino acid SBs represented positive correlation to Day 9 CVC, a VIP value less than 0.5, and a negative stoichiometric balance sign. Coefficient values were normalized and scaled to unit variance by the model, and only positively correlated amino acid SBs were selected. Since time-dependent amino acid SBs were used in the model, various selected time points that met each respective criterion were plotted. For each model and criteria, amino acid cocktail feeds were developed grouped by day of importance. For instance, in the growth model for criteria 1 (a), Asp, Ser, Pro, and Val were highlighted as important on Day 3, and thus, a cocktail of all 4 amino acids was developed and fed on day 3 of the culture.
Fig. 3. Criteria 1 Validation of Culture Performance with respect too Cell Growth and mAb Productivity: Amino acids for the growth model (Day 9 CVC) and the Production model (Day 14 Titer) were supplemented to CHO cell culture batches that ran with the high nutrient fed-batch process according to the selection factors from criteria 1 (Corr > 1, VIP > 1, SB > 0). Day 9 and Day 14 values for CVC and Titer were normalized to the maximum value of the process control (black). Error bars represent standard error to mean. (a) Amino acids supplementation for models based on criteria 1 did not result in a significant increase in total cells by day 9. Only the growth model showed a nominal increase of ~10% at day 14 when compared to control (p < 0.15). (b) Similarly, no significant increase of mAb productivity was observed when compared to the process control for either model and in the case of the growth model, mAb productivity decreased by ~20% by day 14 (p < 0.05).
Fig. 4. Criteria 2 Validation of Culture Performance with respect too Cell Growth and mAb Productivity: In addition to the high nutrient fed-batch process, model informed amino acid supplements based on criteria 2 (Corr > 1, VIP > 1, SB < 0) were also tested on moderate and low nutrient fed-batch processes. Accordingly, Day 9 and Day 14 values for CVC and Titer were normalized to the maximum value of each respective process control (Gray Line). Error bars represent standard error to mean. (Relative CVC) Amino acids supplementation based on criteria 2 showed ~20% increase in total cells for high nutrient, but close to a ~30% increase in moderate nutrients conditions (p < 0.05). The relative total cells was even more profound in low nutrient reaching near ~55% increase in total cells (p < 0.01). (Relative Titer) Antibody titer in contrast increased only for the production model for high nutrient feeds but drastically for both models in moderate and low nutrient feeds reaching close to ~80% increase in titer by day 9 and a 60% increase in titer by day 14 with respect to each process control (p < 0.001).
Fig. 5. A comparison of total glucose consumption and specific glucose consumption: To determine if criteria 2 amino acid supplementation for both, the growth and the production models modified the metabolic state of the cells, the total glucose consumption of each culture was compared to the specific glucose consumption per cell. The total glucose consumption was calculated by the change in extracellular glucose concentrations between feed events in the process. Specific glucose consumption was calculated like total glucose consumption however, the difference was divided by the change in CVC between each time point. Although total glucose consumption showed a slight decrease in moderate and low nutrient conditions, a more profound decrease was noticed in the specific glucose consumption rates for moderate and low nutrients feeds suggesting that cells under moderate and low nutrient conditions shifted consumption for carbon source and energy from glucose to other preferred metabolites such as the stoichiometrically balanced amino acids.
Fig. 6. Criteria 3 Validation of Culture Performance with respect too Cell Growth and mAb Productivity: To test the effectiveness of the VIP value to select biologically important amino acids only, amino acids were supplemented according to the selection factors from criteria 3 (Corr > 1, VIP > 0.5, SB < 0). Supplementation based on criteria 3 was assumed to result in a minimal change in cell growth and mAb productivity for both models (a) Accordingly, no significant change in D14 CVC was observed across all nutrient conditions, with some cases resulting in a decrease of cell growth. (b) In contrast, there was a significant in increase in titer for both moderate and low nutrient conditions in the production model (p < 0.05), presumably due to the lack of nutrient levels in those conditions. However, this increase was not as significant as supplementation with criteria 2 amino acids.
Fig. 7. Driving consumption of model informed amino acids: To validate the statistical model, it was important to assess if the model driven amino acid supplementations led to an increased consumption of supplemented amino acids and/or activated the consumption of other amino acids. Specific consumption rates were calculated from the change in extracellular amino acid levels between feed events per cell. The resulting specific consumption rates within model conditions were compared to the consumption rates of the respective process control for each nutrient level. Positive changes are reflected by green boxes and negative by red. Gray boxes indicate missing values that were not captured by amino acid assay for various reasons. In addition, amino acids were grouped between glucogenic and ketogenic groups.
Fig. 8. Media optimization platform with Stoichiometric Balances: The modeling approach presented in this study was a proof of concept to utilize stoichiometric balances for amino acids to rapidly optimize media development efforts. However, stoichiometric balances and the methods presented could also be applied to other macro-nutrients necessary in CD media. Accordingly, figure 8 presents our approach as a platform to iteratively improve various stages of media development. Utilization of MVDA can easily highlight key stoichiometric balances that can be filtered through a feature selection decision tree to identify the limiting nutrients. Experimental validation studies can then be conducted to assess the change in phenotypical behavior with the modified media. Lastly, the newly generated data can be combined with the original dataset to further inform and improve media development efforts.
Supplementary Figure Text
Fig. S1. Training Model Statistics: MVDA models using the Orthogonal Partial Least Squares (OPLS) algorithm in SIMCA-P+ were generated using 25 training batches. Models were built towards day 9 CVC or day 14 titer in the batch level model format. In addition to the goodness of fit (R2) parameter, a goodness of prediction (Q2) parameter was also generated that measured the cross-validated prediction accuracy. (a) Observed vs predicted plot for day 9 CVC showed a strong R2 of 0.912, and a strong Q2 of 0.726. (b). Observed vs predicted plot for day 14 titer showed a strong R2 of 0.832 but a predictable accuracy of 0.422. Although the Q2 was relatively lower, the day 14 titer model was able to highlight information on key variables for media optimization.
Fig. S2. Criteria 2 Relative Growth Rate and Relative Specific Productivity. Relative growth rate for high nutrient condition showed a slight increased by day 3 and maintained the increase by day 7 which could explain the slight increase (~20%) in total cells for high nutrient feeds. Relative growth rates for moderate and low nutrient feeds also showed only slight increases, however remained slightly higher for a longer duration of time than the high nutrient condition. As a result, there was a more significant increase in total cells for moderate and low nutrient conditions. Similarly, relative productivity (qP) showed a more profound increase in moderate and low nutrient conditions. Moderate nutrient conditions showed a 20 increase in qP for the growth model (p < 0.05) and showed near a 30% increase for the production model (p < 0.01). In contrast, the low nutrient condition showed about a 40% increase in qP starting from day 7 (p < 0.05) suggesting the increased mAb production was not only an artifact of increased cell growth, but a higher drive for these cells to produce more monoclonal antibody.
Fig. S3. Cell Growth and Productivity Changes resulting from Process Controls: Since amino acid cocktails were made at a 100x concentration for each of the specific amino acids, increased pH by NaOH was needed to solubilize the mixtures. In addition, highly concentrated cocktail mixtures also resulted in a high osmolality. Therefore, to measure the effects of high pH and high osmolality (osmo), a pH water solution was generated at the highest pH among the amino acid cocktail mixtures and fed to the cultures at the same volume as the cocktail mixture. In addition, a high osmo solution in water was generated using NaCl and fed at the same volume as the amino acid cocktail with the highest osmo. pH controls were only measured against the high nutrient conditions and resulted in a slight decrease of about 20% in total cells. Similarly, pH controls showed about a 30% decrease in relative titer. Osmo controls for high nutrient conditions showed similar trends as the pH controls but showed minimal differences for moderate and low nutrient conditions. The minimal overall effects of the pH control and the osmo control on total cells and relative titer suggest that the significant increases in cell growth and mAb production from criteria 2 cultures was attributed to the amino acids within the cocktail mixtures.