Multi omics reveals mesodermal fate bias and enables predictive cell state control in human pluripotent stem cell biomanufacturing

Technical Summary: Multi-omics Reveals Mesodermal Fate Bias and Enables Predictive Cell State Control in Human Pluripotent Stem Cell Biomanufacturing

Problem Statement
Despite the potential of human induced pluripotent stem cells (hiPSCs) for regenerative medicine, conventional biomanufacturing approaches fail to fully recapitulate natural embryonic or lineage-committed states. Current strategies often rely on restricted phenotypic characterization that overlooks the complex interplay of epigenetic, transcriptional, and metabolic domains required to maintain pluripotency and drive differentiation. Consequently, therapeutic derivatives generated at scale frequently exhibit only partial fidelity to their natural counterparts. The specific mechanisms by which artificial culture conditions—particularly oxygen levels and hydrodynamic agitation—cause divergence from natural developmental trajectories remain poorly understood.

Methodology
The authors employed an integrative experimental-computational framework to characterize hiPSC cell states under varying oxygen and agitation conditions in stirred-suspension bioprocesses.

• Experimental Design: Two distinct hiPSC lines (TC1133 and PGPC14) were cultured under three conditions: static monolayer, static aggregates, and dynamic aggregates in bioreactors. Agitation rates were varied (below, within, and above optimal ranges), and oxygen conditions included normoxia and hypoxia. Samples were harvested at multiple timepoints (Days 0, 2, and 4) for multi-omics profiling.
• Multi-omics Profiling:
  • Proteomics: Utilized LC-MS/MS (Orbitrap Fusion Lumos) to quantify protein expression and post-translational modifications.
  • Metabolomics: Employed UHPLC-MS (Q Exactive) for semi-targeted analysis of intracellular and extracellular metabolites.
  • Transcriptomics: Performed bulk RNA sequencing (NovaSeq 6000) with ribosomal depletion.
• Computational Framework:
  • Statistical Analysis: Differential expression and accumulation were assessed using DESeq2, Wilcoxon rank-sum tests, and DGSEA (Directional Gene Set Enrichment Analysis) against MSigDB and SMPDB databases.
  • Developmental Trajectory Mapping: Experimental data were compared against reference single-cell RNA-seq datasets representing natural human preimplantation and early gastrulation. A "reinforcement metric" was applied to quantify the direction and strength of gene expression changes in the experimental data relative to specific embryonic cell types (e.g., epiblast, trophectoderm, mesoderm) without requiring explicit data harmonization.
  • Machine Learning: A Random Forest Regressor (RFR) framework was implemented to model population dynamics across metabolic and transcriptional programs. The model was trained to predict cell state outcomes based on environmental setpoints (oxygen and agitation) and time, with feature sets pruned to ensure high prediction confidence and propagation to biological signatures.

Key Results

• Translational and Metabolic Shifts: Dynamic culture conditions selectively engaged molecular programs associated with early embryonic tissue organization, mechanosensation, and growth factor-mediated patterning. Agitation significantly enriched pathways related to basement membrane assembly and cytoskeletal remodeling. Metabolic profiling revealed that hypoxia induced an early shift toward phenotypic adaptation rather than heightened proliferation, while normoxia supported DNA replication and synthesis. Agitation rates modulated translational control via mTORC1-dependent pathways and stress responses (e.g., AMPK-linked repression at suboptimal agitation).
• Fate Bias and Reinforcement: The study identified that bioprocess conditions reinforce specific developmental trajectories. Notably, high agitation rates reinforced gene expression signatures associated with the transition of gastrulating epiblast toward early mesodermal lineages. Specifically, the gene PCGF6 showed strong reinforcement toward a nascent mesodermal state under high agitation. Conversely, oxygen variations showed weaker reinforcement of definitive lineage markers but did influence genes related to lysosomal function, tRNA modification, and RNA regulation.
• Predictive Modeling: The Random Forest models successfully captured non-linear interactions between environmental conditions and cell state dynamics within the observed timeframes. The models could differentiate cell state outcomes between specific conditions (e.g., high vs. low agitation) but could not interpolate outcomes for unobserved timepoints, highlighting the non-linear nature of the system. The models identified a subset of features (genes and metabolites) critical for accurate propagation of predictions to cell state representations.

Significance and Claims
The paper claims to provide a framework for guiding the optimization of pluripotent cell state maintenance and differentiation in biomanufacturing. By integrating multi-omics data with machine learning, the authors demonstrate that environmental perturbations in artificial culture niches do not merely cause stress but actively reinforce specific, albeit partial, developmental programs.

The study modestly concludes that while the bioprocess conditions observed did not fully recapitulate a definitive lineage commitment (such as a complete transition to naïve pluripotency or a specific differentiated fate), they revealed a "mesodermal fate bias" driven by agitation. The work suggests that understanding these reinforced programs allows for the refinement of biomanufacturing strategies to better control cell state dynamics. The authors emphasize that this approach offers a method to estimate shared or divergent activity relative to cell types of interest without the need for complex data harmonization, potentially advancing the clinical translation of regenerative medicine therapies by improving the fidelity of cell products.