INSIGHT SERIES
Elsa Gorre 1, Javier D. Gomez 1, Kathryn Dorst 1, Bo Zhai 1, Andrew Mahan 1, Jack Howland 2, Geoffrey K. Feld 3, Baljit K. Ubhi 2 1. Johnson & Johnson Innovative Medicine, Spring House, PA · 2. Matterworks, Somerville, MA · 3. Geocyte, Dublin, OH
Metabolite profiling reads out cellular nutrition and metabolism directly, covering both the analytes a process already monitors and the mechanistic biochemicals that determine whether a culture stays productive. What has kept it out of process control is turnaround. Conventional quantitative LC-MS requires method development, procurement or synthesis of standards, calibration curves, and manual peak integration, and typically takes six to nine weeks to return a result a process engineer can act on.1,2,3
The conventional quantitation workflow
Adjusting media composition and interpreting cellular metabolism both depend on knowing concentrations rather than peak areas. Reaching concentrations by the conventional route means committing to a target list before the run, since the method is built around isolated calibration standards and labeled internal standards for each analyte of interest. Annotation is the slowest step, drawing on several software tools, spectral libraries, and reference databases to establish which features are worth reporting. The resulting panel is bounded by the compounds for which standards can be procured or synthesized (Figure 1A).
Identification and concentration without calibration standards
Pyxis is built on the Large Spectral Model (LSM), a foundation model trained natively on more than 10 billion raw spectra across millions of biological contexts. Rather than matching spectra against a reference library, LSMs learn structural representations directly from fragmentation, which allows compounds outside existing libraries to be identified. Expert models fine-tuned on curated, PhD-labeled training data supply the identification and concentration determinations on top of that representation.
Scientists run their analytical samples and submit the raw files. Named compounds and their concentrations return as a table within 15 minutes, with no calibration curves, no labeled internal standards, and no reference library in the loop (Figure 1B). Concentrations are reported in micromolar, with no per-analyte calibration standard, and only for analytes above a pre-validated limit of quantitation.4 A 96-well plate of samples and blanks yields named metabolites with concentrations within 48 hours.
Figure 1. Workflow comparison. (A) The conventional LC-MS route, spanning analyte selection, standard procurement or synthesis, isolated calibration runs, chromatographic method optimization, sample acquisition, and manual peak integration and compound assignment. (B) The Pyxis route, in which sample preparation and overnight acquisition are followed by cloud processing that returns compound identities and concentrations in under 15 minutes.
Bioreactor runs and metabolite coverage
Johnson & Johnson Innovative Medicine ran two 13-day, 10-L CHO bioreactors producing the NIST monoclonal antibody reference material, sampling spent media and cell pellets daily in three technical replicates. Proteins were precipitated in 50:30:20 methanol:acetonitrile:water and the samples centrifuged. Spent media supernatants were diluted 150-fold. Acquisition used HILIC chromatography, optimized for polar metabolites, on a Thermo Scientific Vanquish Horizon UHPLC coupled to an Orbitrap Exploris 120. The resulting 224 raw files were processed in the cloud in under 15 minutes.
Culture viability and viable cell density served as the performance indicators for the runs (Figure 2). Viability held near 100 percent through Day 6 and declined from Day 7 onward. Viable cell density rose to a peak at Day 8 and fell steadily thereafter.
Across the two runs, 72 biochemicals in spent media and 115 in cell pellet extracts cleared their pre-validated limit of quantitation and were reported. Limits of quantitation for individual analytes ranged from 5 nM, for carnitine and pyridoxine, to 100 nM for valine.
224 raw LC-MS files converted to named compounds with concentrations.
Named and quantified in spent media and in cell pellet extracts.
Range across reported analytes, pre-validated per analyte.
Figure 2. Performance indicators for the CHO bioprocess. (A) Culture viability and (B) viable cell density over the 13-day time course.
Choline depletion preceded the decline in viability
Differential analysis compared cell pellet extracts at Day 12, when viability was falling, against Day 6, when it was still near 100 percent. Choline was among the analytes most reduced at Day 12 (Figure 3A). Tracking it across the full time course separates the two compartments: choline in the spent media fell from approximately 1,000 µM to the limit of quantitation by Day 9, and the intracellular pool, stable through Day 8, declined from Day 9 onward and continued to fall through the end of the run (Figure 3B). Both bioreactor runs followed the same trajectory.
Choline is an integral component of membrane phospholipids, with phosphatidylcholine the primary lipid of CHO cell membranes.5 The consequences of running short of it have been characterized directly: in a chemically defined fed-batch medium development study, choline limitation produced lower cell viability, lower mAb titer, higher aggregate content, and higher mannose-5 content, and optimizing the choline to glucose ratio in the feeds resolved all four.6
The ordering observed here, medium exhaustion first and intracellular decline after, is consistent with the medium supply limiting the intracellular pool. It remains an observational association across two runs of one cell line and one process. No choline supplementation arm was run, so the association was not tested by intervention, and no titer or product quality data accompany these measurements.
Figure 3. Choline in the medium reached the limit of quantitation before the intracellular pool declined. (A) Volcano plot comparing CHO cell pellet metabolite concentrations at Day 12 against Day 6. Blue and red mark analytes significantly lower or higher at Day 12 respectively (p < 0.05 and |log2 fold change| > 1). (B) Choline concentration in µM in spent media (above) and cell pellets (below) across the bioreactor course. Black and red trace the two runs; error bars are ±1 standard deviation of three technical replicates.
Central carbon metabolism and lactate
Lactate and the tricarboxylic acid (TCA) cycle intermediates are the standing surrogates for cellular energy production, and the balance among them reports on mitochondrial energetics as well as on lipid and amino acid synthesis (Figure 4A). Concentrations for lactate, citrate, succinate, fumarate, and pyruvate were returned daily in both compartments (Figure 4B). Medium lactate rose to roughly 10 mM by Day 3, fell through the middle of the run, then rose again through Day 13. Medium citrate and succinate both climbed to a peak around Day 9.
Controlling lactate is a common route to improved titer, and the timing of any intervention matters as much as its composition. Adding α-ketoglutarate, malic acid, or succinic acid directly to the basal medium had no significant effect on culture performance in a controlled study, whereas feeding the same intermediates during the stationary phase improved lactate consumption, reduced ammonium accumulation, and raised cell-specific productivity and antibody titer.7 A daily concentration readout across both compartments is what makes that timing decision available during a run rather than after it.
The two compartments were not equally reproducible. Medium concentrations tracked closely between the runs across all three analytes shown. Intracellular citrate diverged: run 1 rose to approximately 16 µM at Days 7 and 8 while run 2 stayed near 6 µM, a difference well outside the technical replicate error. Intracellular pools of central carbon intermediates were the less stable readout of the two here.
Figure 4. Central carbon metabolism across the bioreactor course. (A) Overview of mitochondrial energetic metabolism. (B) Concentrations in µM of representative energetic metabolites in spent media (left) and cell pellets (right). Black and red trace the two runs; error bars are ±1 standard deviation of three technical replicates.
Glutathione ratio and cellular redox state
Protein production raises reactive oxygen species (ROS) in CHO cells, and the balance between ROS and free thiols such as cysteine and glutathione (GSH) sets the cellular redox state.8 Left uncontrolled, the consequences are product aggregation, endoplasmic reticulum stress, and reduced productivity, and most process development analytics observe only the extracellular side.9 Enzymatic kits for intracellular GSH and glutathione disulfide (GSSG) are standard in cell line development. An LC-MS account extends the same picture to the amino acid constituents of glutathione, namely cysteine, glycine, and glutamate, to the GSH/GSSG ratio, and to other markers and mitigators of ROS, inside and outside the cell.
Daily intracellular GSH and GSSG concentrations yielded the ratio shown in Figure 5A, and cystine, the oxidized form of cysteine, was quantified in both compartments (Figure 5B). Medium cystine fell from approximately 550 µM to below 200 µM by Day 7 and remained low. The glutathione ratio was the noisiest measurement in the dataset: it ranged from below 10 to above 120 across the time course, moved substantially day to day, and diverged between the runs at Day 9, where run 2 reached roughly four times the run 1 value with a replicate spread spanning half that range again. Reported here, the ratio serves as a daily monitoring readout. These data do not establish that the intracellular redox environment was stable.
Tracking redox mediators on both sides of the membrane and adjusting the corresponding media components has been reported to raise viability and productivity, improve titer, and reduce antibody aggregation.8,10 The margin is largest for disulfide bond-containing therapeutics such as bispecific antibodies.9
Figure 5. Redox monitoring by concentration determination of glutathione and related compounds. (A) Intracellular reduced-to-oxidized glutathione ratio (GSH/GSSG) across the two bioreactor time courses. (B) Cystine concentration in µM in spent media (left) and cell pellets (right). Black and red trace the two runs; error bars are ±1 standard deviation of three technical replicates.
What these measurements establish
This dataset establishes coverage and turnaround. Raw HILIC LC-MS files from a 13-day bioprocess, in two sample matrices, returned 72 and 115 named compounds with concentrations in under 15 minutes, with no analyte selected in advance and no calibration standard run alongside. It establishes that a depletion event in the medium was visible days before the viability curve moved, and that the two bioreactor runs agreed on that trajectory.
It does not establish a link to product yield. No titer or product quality measurements accompany these samples, so no metabolite reported here is tied to output in this dataset. The choline account is an association supported by external literature rather than by an intervention run in these bioreactors.
Limitations
One cell line, one process, and two bioreactor runs. Reproducibility across runs is shown; generalization across cell lines and processes is not.
Replicates are technical rather than biological, so the error bars describe measurement precision and not culture-to-culture variation.
Concentrations are reported in micromolar with no per-analyte calibration standard run alongside these samples. Per-analyte accuracy rests on prior benchmarking and on the pre-validated limit of quantitation.4
Cell pellet values are per-extract concentrations, referencing 5 × 106 cells extracted into 1 mL. They are not corrected for cell volume and are not intracellular concentrations.
Coverage is bounded by the HILIC method, which is optimized for polar metabolites. Analytes outside that chemical space are absent from these counts.
Summary
Quantitative metabolic analysis guides bioprocess decisions only when it arrives inside the run. Here the biochemical account of a 13-day process, across both the medium and the cells, was available daily from raw files, and it surfaced a nutrient depletion several days ahead of the viability curve that reflected it. The measurements support three uses across the development pipeline:
Assessment of cellular metabolism and media composition together, within the run, at the point where a feed strategy can still be changed.
Daily readouts of energetic compounds bearing on mitochondrial function, glucose consumption, and lactate production.
A combined view of cellular health, redox state, and metabolism spanning cell line development, media development, metabolic and genetic engineering, and bioreactor scale-up.
Bioprocess development is a setting where a measurement that arrives after the run has no effect on the run. Returning named compounds with concentrations from raw files on the same day places the biochemical account inside the decision window, next to the viability and density readings that process engineers already act on. Pyxis is the Matterworks co-scientist for interpreting omic data and predicting phenotypic biology.
Data: CHO cell spent media and pellet extracts from two 13-day, 10-L bioreactor runs producing the NIST monoclonal antibody reference material, sampled daily in three technical replicates. HILIC LC-MS on a Thermo Scientific Vanquish Horizon UHPLC coupled to an Orbitrap Exploris 120, 224 raw files. Concentrations are reported in micromolar, with no per-analyte calibration standard. Spent media values are corrected for the 150-fold dilution; cell pellet values reference the extraction of 5 × 106 cells into 1 mL of extraction solution. Questions about this analysis: info@matterworks.ai
References
Partopour B, Pollard D. Advancing biopharmaceutical manufacturing: economic and sustainability assessment of end-to-end continuous production of monoclonal antibodies. Trends Biotechnol. 2025;43(2):462-475. doi:10.1016/j.tibtech.2024.10.007
Yao G, Aron K, Borys M, Li Z, Pendse G, Lee K. A metabolomics approach to increasing Chinese hamster ovary (CHO) cell productivity. Metabolites. 2021;11(12):823. doi:10.3390/metabo11120823
Singh R, Fatima E, Thakur L, Singh S, Ratan C, Kumar N. Advancements in CHO metabolomics: techniques, current state and evolving methodologies. Front Bioeng Biotechnol. 2024;12:1347138. doi:10.3389/fbioe.2024.1347138
Ferro LS, Wong AYL, Howland J, Costa ASH, Pruyne JG, Shah D, Lauterbach JD, et al. A scalable approach to absolute quantitation in metabolomics. bioRxiv. 2024. doi:10.1101/2024.09.09.609906
Vance JE, Vance DE. Phospholipid biosynthesis in mammalian cells. Biochem Cell Biol. 2004;82(1):113-128. doi:10.1139/o03-073
Kuwae S, Miyakawa I, Doi T. Development of a chemically defined platform fed-batch culture media for monoclonal antibody-producing CHO cell lines with optimized choline content. Cytotechnology. 2018;70(3):939-948. doi:10.1007/s10616-017-0185-1
Zhang X, Jiang R, Lin H, Xu S. Feeding tricarboxylic acid cycle intermediates improves lactate consumption and antibody production in Chinese hamster ovary cell cultures. Biotechnol Prog. 2020;36(4):e2975. doi:10.1002/btpr.2975
Ali AS, Raju R, Kshirsagar R, et al. Multi-omics study on the impact of cysteine feed level on cell viability and mAb production in a CHO bioprocess. Biotechnol J. 2019;14(4):1800352. doi:10.1002/biot.201800352
Sinharoy P, Aziz AH, Majewska NI, Ahuja S, Handlogten MW. Perfusion reduces bispecific antibody aggregation via mitigating mitochondrial dysfunction-induced glutathione oxidation and ER stress in CHO cells. Sci Rep. 2020;10:16620. doi:10.1038/s41598-020-73573-4
Handlogten MW, Lee-O’Brien A, Roy G, et al. Intracellular response to process optimization and impact on productivity and product aggregates for a high-titer CHO cell process. Biotechnol Bioeng. 2018;115(1):126-138. doi:10.1002/bit.26460