Single-cell proteomics reveals proteome remodeling and cellular heterogeneity during NGF-induced PC12 neuronal differentiation

Abstract

Single-cell proteomics enables direct measurement of cellular heterogeneity during dynamic biological processes, but its application to fragile and highly adherent neuronal models remains challenging. Here, we developed and applied an optimized single-cell proteomics workflow to characterize proteome remodeling during nerve growth factor (NGF)–induced differentiation of PC12 cells. To enable reliable single-cell analysis, we implemented gentle dissociation, anti-aggregation strategies, and thermal inkjet–based cell dispensing, achieving high accuracy in single-cell isolation. Inclusion of n-dodecyl-β-D-maltoside (DDM) improved recovery of membrane-associated and low-solubility proteins. Coupled with LC–ion mobility–mass spectrometry, this workflow enabled quantification of ~2,000–3,000 proteins per cell across the differentiation time course.

Single-cell proteomic analysis revealed progressive and heterogeneous proteome remodeling during differentiation. While undifferentiated cells formed a relatively homogeneous population, later stages (Days 4–6) exhibited increased variability, including multimodal protein abundance distributions and separation into distinct subpopulations. Dimensionality reduction, clustering, and non-negative matrix factorization identified multiple coexisting proteomic states within the same time points, reflecting asynchronous differentiation trajectories. These subpopulations were characterized by coordinated differences in pathways related to intracellular trafficking, protein translation, cytoskeletal organization, and neuronal maturation.

Comparison with bulk proteomics demonstrated that proteins associated with differentiated neuronal states, including those involved in neurite formation and structural remodeling, are underrepresented in population-averaged measurements but enriched within specific single-cell subpopulations. Temporal and cluster-resolved analyses further revealed distinct protein expression trajectories, including early decreases in cell cycle and metabolic pathways and later increases in neuronal structural and regulatory proteins.

Together, this study establishes an optimized workflow for single-cell proteomics of neuronal systems and demonstrates that NGF-induced PC12 differentiation proceeds through heterogeneous and divergent proteomic states that are not resolved by bulk analysis.

Graphical Abstract

Introduction

Proteomics provides a direct readout of cellular state by measuring protein abundance and regulation across biological systems. Advances in mass spectrometry–based proteomics have greatly expanded coverage and quantitative performance, enabling detailed characterization of complex cellular processes. However, conventional bulk proteomic measurements report population-averaged signals, which can obscure biologically meaningful variability among individual cells. Single-cell proteomics addresses this limitation by enabling protein-level analysis at cellular resolution, allowing heterogeneity to be directly measured rather than inferred^1–3.

Recent progress in mass spectrometry instrumentation, chromatographic separations, and computational analysis has made it possible to quantify thousands of proteins from individual mammalian cells^4,5. These advances have revealed functional heterogeneity that is not always predicted by transcriptomic measurements, highlighting the importance of direct protein-level analysis. Despite this progress, single-cell proteomics remains technically challenging. Low protein abundance, labor-intensive sample preparation, and the need for highly sensitive and reproducible workflows continue to limit throughput and robustness^1,6. Ongoing efforts to miniaturize and automate sample handling, through nanodroplet processing, microfluidic platforms, and improved acquisition strategies, have begun to address these challenges^7–9.

Accurate isolation and handling of individual cells are critical determinants of quantitative performance in single-cell proteomics. Several approaches have been developed, including fluorescence-activated cell sorting (FACS), laser microdissection, and microfluidic or droplet-based platforms^10–12. Each method involves trade-offs between throughput, cell viability, and compatibility with low-volume proteomic workflows. Optimizing cell isolation strategies is particularly important for fragile or highly adherent cell types, such as neurons, where aggregation and cell loss can compromise data quality.

Neuronal differentiation is a tightly regulated process that underlies nervous system development and plasticity. To study proteome remodeling during this process, we used the PC12 cell line, a well-established model of NGF-induced neuronal differentiation. Upon NGF treatment, PC12 cells exit the cell cycle and extend neurite-like processes, adopting features characteristic of sympathetic neurons¹³. This system provides a controlled experimental framework for examining how protein expression programs change as cells transition from a proliferative state to a neuron-like phenotype.

Overall, our optimized single-cell proteomics workflow revealed the phenotype heterogeneity of PC12 cells, resolving transitional states and identifying proteomic signatures that define early, intermediate, and late phases of neuronal differentiation, features that would otherwise be obscured by population averaging.

Results

Nanoliter-scale single-cell proteomics workflow for PC12 cells

To establish robust and reproducible conditions for single-cell proteomics, key experimental parameters were selected based on prior systematic optimization using the same analytical platform (Tecan Uno dispenser, NanoElute 2 nanoLC, and timsTOF SCP), as described in Chi et al. (2026)¹⁴. In that study, digestion conditions, sample preparation strategies, plate formats, chromatographic flow rates, and cell isolation approaches were systematically evaluated using metrics including protein coverage, peptide length distribution, missed-cleavage rates, and coefficients of variation (CV).

These experiments demonstrated that moderate amounts of trypsin (~1 ng) provided efficient digestion with fewer missed cleavages while maintaining reproducible protein identification. Low-binding 384-well plates improved protein completeness and reduced sample loss, and a simplified one-step digestion workflow performed comparably to multi-step protocols while reducing variability. Chromatographic flow rates near 300 nL min⁻¹ provided an optimal balance between sensitivity and reproducibility. In addition, Tecan Uno–based single-cell dispensing produced comparable or improved protein completeness relative to FACS while minimizing sample handling.

Building on this, our earlier work established a nanoliter-scale proteomics workflow using HP’s thermal inkjet dispenser¹⁵. Optimized in HEK-293 cells, this non-contact approach proved robust and broadly applicable to other mammalian cell types, reducing cell damage and contamination and supporting efforts toward accessible, high-throughput single-cell proteomics^16–17.

Based on these findings, the workflow used in this study employs moderate digestion conditions, a one-step preparation in low-binding 384-well plates, optimized nanoLC flow rates, and Tecan Uno–based single-cell isolation.

For high-sensitivity proteomic analysis, we employed data-independent acquisition (DIA) combined with parallel accumulation–serial fragmentation (PASEF), an approach previously optimized for single-cell proteomics to maximize proteome coverage and quantitative reproducibility¹⁴. This acquisition strategy integrates comprehensive ion sampling with ion mobility separation, enabling consistent peptide detection across single cells while maintaining high sensitivity and throughput. Specifically, DIA enables fragmentation of all ions within defined m/z windows, while PASEF leverages trapped ion mobility spectrometry (TIMS) to accumulate ions in parallel and fragment them serially based on ion mobility.

The single-cell proteomics workflow was applied to PC12 cells treated with NGF and sampled across the differentiation time course (Figure 1A). Over time, cells exhibited morphological changes consistent with neuronal differentiation, including the appearance of neurite extensions and increased cell body size, which were more apparent at Day 4 and Day 6. Within each time point, individual cells showed variable morphological features, with some cells displaying extensive neurite outgrowth while others retained a more undifferentiated appearance (Figure 1B–E).

Figure 1.

Single-cell proteomics workflow and morphological characterization of PC12 neuronal differentiation. (A) Schematic overview of the single-cell proteomics workflow, including NGF-induced differentiation of PC12 cells, cell dissociation and anti-clumping handling, single-cell isolation using the Tecan Uno dispenser, nanoliter-scale lysis and digestion (TEAB, DDM, trypsin), and LC–MS/MS analysis using dia-PASEF on a timsTOF SCP platform. The schematic was created with BioRender.com. (B-E) Representative bright-field microscopy images (20× magnification) of PC12 cells undergoing NGF-induced differentiation at Days 0, 2, 4, and 6 are shown in panels B–E. Images represent selected microscopic fields of view and are intended to illustrate progressive neuronal differentiation and neurite outgrowth rather than overall cell density. Cells were seeded at low density (~2x10³ cells/cm² cells per culture vessel) to facilitate neurite extension and were harvested and counted prior to normalization for downstream single-cell proteomic analysis. (B) Undifferentiated PC12 cells at Day 0, showing typical round morphology with expected variability in size and adherence. (C) Day 2 following NGF treatment, showing early neurite initiation. (D) Day 4, with increased neurite extension and partial neuronal morphology. (E) Day 6, showing extensive neurite outgrowth and mature neuron-like features. Neuronal differentiation was distinguished from cell aggregation based on morphology: differentiated cells exhibit elongated neurite structures and branching extensions, whereas aggregates appear as dense, rounded clusters lacking defined projections. Scale bar = 50 μm.

Single-cell handling and dispensing optimization for PC12 cells

Processing PC12 cells for single-cell proteomics is particularly challenging due to their fragile nature, variable size, and increased tendency to aggregate following differentiation. To minimize aggregation and preserve cell integrity, differentiated PC12 cells were harvested using a gentle EDTA-based dissociation method (Versene), followed by washing in EDTA-containing buffer to disrupt cell–cell interactions. Prior to dispensing, cell suspensions were filtered through a 35 μm strainer to remove aggregates and debris.

Single-cell dispensing parameters were optimized to ensure accurate isolation of individual cells while minimizing mechanical stress. Microscopic validation confirmed that approximately 85% of wells contained single, viable cells. Occasional doublets were identified based on both imaging and elevated precursor counts during data analysis and were accounted for during downstream quality control.

Importantly, neuronal differentiation was distinguished from aggregation based on morphological features. Aggregated cells appear as dense, rounded clusters lacking defined structure, whereas differentiated PC12 cells exhibit characteristic neurite outgrowth and branched morphology. These features were consistently observed in NGF-treated cells at later time points, confirming successful differentiation rather than aggregation artifacts.

Another challenge with differentiated neurons, such as PC12, is their irregular sizes and fragile neurite extensions. To solve this, we systematically tested nozzle settings to ensure precise single-cell deposition without damaging cells (Figures S1 and S2). Using the “sticky cell” mode in the Tecan UNO software further improved droplet adhesion, stabilized dispensing of larger or semi-adherent cells, and prevented doublet formation (Table S2). Dispensing accuracy was evaluated with a Keyence microscope (Figure S1). Microscopic inspection confirmed that approximately 85% of wells contained single, viable PC12 cells, demonstrating the high precision and reproducibility of the optimized Tecan UNO settings. These improvements, combining mechanical, software, and biological adjustments, enabled reliable single neuronal cell dispensing, reduced doublet formation, and provided high-quality input material for single-cell proteomic profiling of neuronal differentiation. A total of 163 single PC12 cells were analyzed across four differentiation stages, including Day 0 (N = 35), Day 2 (N = 36), Day 4 (N = 42), and Day 6 (N = 50) (Table S3). To ensure data quality and instrument stability throughout the acquisition, blank injections and 250 pg HeLa digest quality control samples were acquired after every 10 single-cell injections.

Single-cell proteomics dataset overview and quality control

Quality control of the DIA data across all time points demonstrated consistent, reliable instrument performance. Figure S2A shows that most proteins were identified by a single peptide, with fewer proteins supported by multiple peptides, which is common in single-cell datasets. The distribution of precursor-level coefficient of variation (CV) across time points (Figure S2B) showed consistent quantitative performance throughout differentiation. All stages, from undifferentiated (Day 0) to fully differentiated cells (Day 6), displayed CV distributions peaking around 30-40%, indicating reliable reproducibility and stable precursor quantification across replicates. The similar shapes and widths of these distributions over time confirm minimal technical variability, while slight broadening at later stages likely reflects increased biological heterogeneity as cells develop neuronal phenotypes. A small percentage of precursors with CV values above 75% correspond to low-abundance features near the detection limit, which is common in single-cell datasets^1,2,5.

To evaluate the quantitative dynamic range of the dataset, representative proteins across high, medium, and low abundance levels were analyzed (Figure S2C). High-abundance proteins, such as Tuba1a, Ehhadh, and Thap1, mainly involved in cytoskeletal structure and energy metabolism, were consistently detected at all time points. Medium-abundance proteins (Eif2b3, Gtf2f2, Hook3), linked to transcriptional and translational functions, and low-abundance signaling or regulatory proteins (Samd10, Ankrd46, Senp7) were also reliably quantified. Overall, the dataset covers a dynamic range of approximately four orders of magnitude, capturing both highly abundant and low-abundance proteins in a single analysis.

The number of precursors identified per cell (Figure S3A) confirms the reproducibility and consistency of DIA acquisitions across the neuronal differentiation timeline. Median precursor counts remained stable from Day 0 to Day 6, indicating uniform detection depth and consistent instrument performance. A small subset of Day 2 cells had unusually high precursor counts, likely due to two-cell dispenses or doublets that increased signal intensity. Differences in precursor counts per single cell across differentiation time points (Day 0, Day 2, Day 4, and Day 6) were assessed using a nonparametric Kruskal–Wallis test because of non-normal data distributions and unequal variances among groups. When a significant overall effect was observed, Dunn’s multiple-comparisons test was applied for post hoc pairwise comparisons with adjusted p-values. Statistical significance was evaluated at α = 0.05 (Table S4). Log₂ precursor intensity distributions differed significantly across differentiation stages (Kruskal–Wallis test, H = 57.68, p < 0.0001). Post hoc Dunn’s multiple-comparisons test identified significant differences between Day 6 and all other time points (adjusted p < 0.0001), whereas no significant differences were observed among Day 0, Day 2, and Day 4 (adjusted p > 0.05) (Table S5). The log₂ intensity distributions of precursor abundances across all DIA runs (Figure S3B) were highly overlapping, suggesting consistent signal ranges and minimal variation between runs.

Single-cell proteomics reveals neuronal proteins masked in bulk measurements at late stages of differentiation

Protein ranks from the bulk, and single-cell datasets were converted to percentile scores using the formula percentile = 1 – ((rank – 1) / N), where N represents the total number of detected proteins in each dataset. This transformation places both datasets on the same 0–1 scale (1 indicating highest relative abundance), allowing direct comparison despite differences in proteome depth. The percentile values were then matched by protein identifier and visualized in a scatter plot with bulk percentiles on the x-axis and single-cell percentiles on the y-axis. A dashed identity line (y = x) indicates proteins with equivalent relative abundance between modalities. Proteins were ranked according to the difference between bulk and single-cell percentiles to identify those that appear highly abundant in bulk but relatively low in single-cell measurements (Figure S4).

Among the proteins with the largest bulk–single-cell divergence, several are directly relevant to neuronal differentiation and synaptic biology. Cend1 has been shown to promote neuronal lineage progression and differentiation¹⁸. Baiap2 (IRSp53) is a regulator of actin dynamics and filopodia formation and has established roles in neurite outgrowth and synaptic structure¹⁹. Pick1 is involved in AMPA receptor trafficking and synaptic plasticity²⁰, and Mapk8ip3 (JIP3) functions in axonal transport and neuronal development²¹. At the same time, several ribosomal (Rpl29, Rpl37-ps4, Rps7-ps23) and mitochondrial-associated proteins (e.g., Ndufab1, Mrpl3, Ptcd3, Slirp) were also enriched among the bulk-high group, suggesting that structural and bioenergetic components contribute strongly to the averaged bulk signal. Taken together, these results indicate that bulk proteomics amplifies signals driven by either structural programs or subpopulations of differentiating cells, whereas single-cell proteomics preserves heterogeneity and reveals that many proteins associated with neuronal growth and trafficking are not uniformly abundant across individual cells (Table S6).

By contrast, single-cell measurements captured these proteins at higher relative abundance, indicating their enrichment within specific subsets of differentiated, neurite-bearing cells. Similar discrepancies between bulk and single-cell proteomic measurements have been reported previously, where proteins linked to defined cellular states are underrepresented in population-averaged data^1,2. Together, these observations indicate that bulk measurements underestimate a subset of neuronal proteins that become evident only when cellular heterogeneity is preserved.

Protein identification and protein abundance distribution across differentiation time points

Protein counts were compared across differentiation time points. The number of identified proteins differed between time points (Figure 2A), with fewer proteins detected at Day 4 and Day 6 compared with Day 0 (H = 38.81, p < 0.0001; Table S7). A gradual decrease in the number of proteins identified per cell was observed throughout the differentiation time course (Figure 2A, D). This trend likely reflects progressive proteome specialization accompanying neuronal differentiation, as cells transition from a proliferative state toward a more defined neuronal phenotype. Importantly, quality control measurements using HeLa digest standards demonstrated stable instrument performance throughout the study, indicating that the observed reduction in protein identifications was biological rather than analytical in origin.

Figure 2.

Protein group quantification, Principal Component Analysis (PCA), and density distribution across differentiation time points. (A) Protein group identification at different differentiation stages. The number of protein groups identified varies between individual cells at the same time, reflecting cell-to-cell differences in proteome depth during differentiation. Kruskal–Wallis testing was applied due to non-normal distributions. Dunn’s multiple comparisons test was used for post hoc analysis with adjusted p-values. Significance: ns, not significant; **p < 0.01; ***p < 0.0001. (B) Principal component analysis (PCA) of single-cell proteomes collected at Days 0, 2, 4, and 6, performed after imputation. (C) Violin plots showing the distribution of protein group identifications per single cell for samples prepared with DDM and without DDM. (t = 2.61, df = 60, p = 0.011) (D) Density plot of single-cell proteomic profiles across four differentiation time points. Kernel density estimation summarizes the distribution of proteomic profiles for each time point.

Protein abundance values were visualized by principal component analysis (Figure 2B). Day 0 replicates form a compact cluster. Samples from later time points occupy a larger area of the PCA space. Dispersion is most pronounced at Day 4 and Day 6.

Protein identifications were also compared between samples prepared with and without n-dodecyl-β-D-maltoside (DDM). Samples prepared with DDM show, on average, higher protein counts (Figure 2C). A difference between the two groups was observed by an independent samples t-test (p = 0.011; Table S8). DDM has been reported to improve recovery of membrane-associated and low-solubility proteins in LC–MS workflows^22–24. Protein abundance distributions are shown as density plots in Figure 2D. Day 0 and Day 2 show relatively simple distributions. Day 4 and Day 6 show broader distributions with multiple peaks. The distributions at Day 4 and Day 6 are shifted toward lower relative abundance values.

Single-cell protein abundance heatmaps are shown in Figure S5 without clustering (A) and with clustering (B). When ordered by time point, samples show similar overall trends from Day 0 to Day 6. After clustering, subsets of proteins with opposing abundance patterns are visible within Day 2, Day 4, and Day 6 samples. These features are not apparent in the unclustered heatmap.

Effect of DDM on protein recovery in neuronal samples

Comparison of neuronal samples prepared with and without n-dodecyl-β-D-maltoside (DDM) showed clear differences in the proteins recovered (Figure S6A). Proteins enriched in the DDM condition included vesicular and lysosomal proteins such as Cathepsin D and Transferrin, as well as proteins with limited solubility or strong structural associations, including Desmoplakin and Hemoglobin β. Several secreted proteins, including Alpha-fetoprotein and Inter-alpha-trypsin inhibitor heavy chain 4, were also detected at higher abundance when DDM was used (Figure S6B). Samples prepared without DDM had lower abundances of membrane-associated and poorly soluble proteins, indicating reduced recovery of these species under detergent-free conditions^22,24,25.

The beneficial effect of DDM on improving recovery of hydrophobic and membrane-associated proteins has been previously demonstrated. In this study, our objective was not to establish the general utility of DDM, but to evaluate its performance within the context of our single-cell proteomics workflow and neuronal model system. Single-cell proteomics of differentiated PC12 cells presents distinct challenges due to increased membrane complexity, cellular morphology, and protein solubility characteristics associated with neuronal differentiation.

Under the specific conditions used here, including nanoliter-scale sample preparation, optimized digestion workflow, and defined DDM concentration, DDM improved the recovery of membrane-associated, vesicular, and low-solubility proteins. Importantly, we provide a system-specific characterization of proteins enriched with DDM, including those relevant to neuronal structure and function. To improve transparency, the complete list of proteins identified with and without DDM has been included in the Supplementary Information (Table S14). These results represent application-specific validation rather than a general methodological innovation and provide practical guidance for detergent use in single-cell proteomics of neuronal systems.

Comparison of Day 0 and Day 2 proteomes

To examine early-stage proteome remodeling associated with NGF-induced differentiation, protein expression levels between Day 0 and Day 2 were analyzed using a volcano plot (Figure S7A). Proteins were ranked based on adjusted p-value (FDR) and log₂ fold change, and the complete ranked list is provided in Table S11. The top 20 proteins, defined by the lowest FDR values and largest magnitude fold changes, were used for visualization (Figure 3A–C). From this set, representative proteins were selected based on their overlap with significantly enriched pathways.

Figure 3.

Differential protein expression and pathway enrichment between Day 0 and Day 2. (A-C) A representative subset of proteins with the lowest p-values distinguishing Day 0 and Day 2, grouped by functional category (two-tailed Mann–Whitney U test, ****p < 0.001). Horizontal lines indicate group medians. (D) Bar plot of enriched biological terms for the input protein lists, colored by adjusted p-value. (E) Enrichment network visualization with nodes colored by cluster assignment.

The selected proteins map directly to the pathways identified in the enrichment analysis (Figure 3D). Lysosomal proteins (Lamp1, Ctsl) decrease early and are associated with pathways including lysosome, vacuolar transport, and endomembrane system organization, consistent with reduced degradative activity as cells transition away from a proliferative state²⁶. Proteins involved in lipid and sphingolipid metabolism, including Asah1 and Msmo1, also decline and align with pathways related to intracellular chemical homeostasis, ferroptosis, and modified metabolic processes, reflecting changes in lipid turnover and metabolic reprogramming rather than direct lipid biosynthesis²⁵.

Adhesion-related proteins (Thy1, Itga3) correspond to pathways such as protein localization to the cell periphery, membrane organization, and regulation of plasma membrane-bounded cell projection organization, consistent with early detachment and membrane restructuring preceding neurite extension^30,31. Additional proteins contributing to categories such as cellular response to chemical stress and intracellular chemical homeostasis further support coordinated metabolic and stress-related changes during this transition²⁷. All proteins shown in Figure 3A–C are contained within the ranked list in Table S11, and selection required agreement between differential expression and pathway enrichment. The STRING network³² (Figure 3E) further groups these proteins into interconnected functional clusters, indicating coordinated early changes during neuronal differentiation.

Comparison of Day 0 and Day 4 proteomes

To examine how differentiation progresses beyond the early NGF response, we compared Day 0 and Day 4 using a volcano plot framework to identify proteins with significant fold changes and low p-values. (Figure S7B)(Table S12) (Figure 4A–B). By this point, most cells have exited the cell cycle, which is reflected in the significant downregulation of proliferation markers like Mki67 and Cdk1. These proteins typically peak during the G1–S–G2–M phases, so their decrease indicates that cells are no longer dividing and have fully left the cell-cycle program²⁸.

Figure 4.

Differential protein expression and pathway enrichment between Day 0 and Day 4. (A) Single-cell abundances of MKI67 and CDK1 at Day 0 and Day 4. (B) Single-cell abundances of ASAH1 and NIPSNAP3A at Day 0 and Day 4. Points represent individual cells (two-tailed Mann–Whitney U test, ****p < 0.001). Horizontal lines indicate group medians. (C) Bar plot of enriched biological terms for the input protein lists, colored by adjusted p-value. (D) Enrichment network visualization with nodes colored by cluster assignment.

This pattern aligns with the enrichment analysis in Figure 4C, where pathways related to DNA replication, mitotic progression, mismatch repair, and nucleosome assembly are among the most depleted. In contrast, proteins associated with differentiation and cellular remodeling remain elevated at Day 4. Asah1 and Thy1, which were already increased at Day 2, continue to show higher abundance, aligning with pathways involved in membrane organization, lipid metabolism, and protein localization. The rise in Nipsnap3a corresponds with enriched categories related to RNA metabolism, mRNA processing, and intracellular homeostasis, reflecting that transcriptional and metabolic programs are being reorganized as the cells mature²⁹. The STRING network in Figure 4D groups these enriched terms into distinct clusters, illustrating how shifts in RNA processing, lysosomal function, membrane remodeling, and stress responses fit together to define the Day 4 state. Overall, these patterns show that by Day 4, PC12 cells have largely stopped cycling and are engaging in metabolic, membrane, and RNA-regulatory activities that support progression toward a more defined neuronal phenotype³⁰.

Comparison of Day 0 and Day 6 proteomes

To characterize late-stage proteomic changes associated with neuronal maturation, we compared Day 0 and Day 6 using a volcano plot (Figure S7C). Comparison of undifferentiated (Day 0) and NGF-differentiated (Day 6) PC12 cells revealed increased levels of protein markers associated with neuronal cytoskeletal organization and maturation. Neuronal intermediate filament proteins NEFM and NEFL, along with INA, Prph, and TUBB2A, showed higher abundances at Day 6 relative to Day 0. These proteins are key components of neurite structure, microtubule dynamics, and neuronal stabilization, indicating substantial cytoskeletal remodeling during differentiation³⁶ (Figure 5A). Pathway enrichment analysis and protein interaction network revealed strong overrepresentation of RNA metabolic processes, mRNA processing, protein folding, and intracellular transport pathways, indicating increased biosynthetic and regulatory activity associated with neurite extension and stabilization^35,36 (Figure 5B–C, Table S13).

Figure 5.

Differential protein expression and pathway enrichment between Day 0 and Day 6. (A) Boxplots show the distribution of normalized protein abundances for selected neuronal markers (NEFM, NEFL, INA, Prph, and TUBB2A) at Day 0 (undifferentiated) and Day 6 (differentiated) PC12 cells. (two-tailed Mann–Whitney U test, ****p < 0.001). Horizontal lines indicate group medians. (B) Bar graph displaying enriched biological terms across the input gene lists, with bars colored according to adjusted p-value. (C) Enrichment network visualization, where nodes are colored based on cluster ID; terms within the same cluster are positioned near each other, indicating shared biological themes.

UMAP projection of single-cell proteomic profiles

To further evaluate neuronal features emerging at later stages of differentiation, we next examined how known maturation markers distribute across the single-cell proteome using UMAP projections (Figure 6). UMAP showed distinct separation between the later time points (Day 4 and Day 6) and earlier ones (Day 0 and Day 2). When we mapped neuronal maturation markers onto this space, proteins such as NEFL, NEFM, TUBA4A, PRPH, and VGF showed higher intensities primarily in the Day 4 and Day 6 clusters (Table S9).

Figure 6.

Expression maps of neuronal maturation markers. (A) PCA plot of all time-course samples, showing separation of early and late differentiation stages. (B-F) Expression patterns of key neuronal maturation proteins: Nefl, Tuba4a, Vgf, Prph, and Nefm. All five proteins show increased abundance in Day 4 and Day 6 clusters.

Temporal protein expression patterns across UMAP-defined clusters

Single-cell protein abundance data from Day 0, Day 2, Day 4, and Day 6 were visualized using UMAP (Figure 7). The resulting embedding separates cells into seven clusters based on similarity in protein abundance profiles. Cells from different time points are present across multiple clusters. For each cluster, relative protein abundance values were summarized by differentiation day. Proteins within clusters show different patterns across time, including decreases, increases, and transient changes between intermediate time points.

Figure 7.

UMAP-based clustering reveals distinct protein expression trajectories during PC12 neuronal differentiation. (A) Single-cell proteomic profiles collected across the NGF-induced differentiation time course (Day 0, Day 2, Day 4, and Day 6) were embedded using UMAP, revealing seven distinct clusters with separable protein expression patterns (left). Each cluster represents a group of cells sharing similar proteomic trajectories over time. (B-F) Line plots (right) show the relative protein abundance trends across differentiation days for proteins within each cluster, highlighting distinct temporal expression patterns across early, intermediate, and late stages.

Protein-level changes within selected clusters are shown in Figure 8. Subgrouped temporal expression patterns for Clusters 0, 2, 3, and 5 are presented in Figure 8A–D, where proteins were partitioned based on shared trajectories across the NGF-induced differentiation time course. Representative proteins with the largest absolute differences in normalized abundance between Day 0 and Day 6 are shown in Figure 8E–H.

Figure 8.

Subgrouped protein expression trajectories reveal heterogeneity within UMAP-defined clusters during PC12 differentiation. (A–D) Clusters 0, 2, 3, and 5 showing subgrouped protein expression patterns based on shared temporal trajectories across the NGF-induced differentiation time course. Proteins within each cluster were partitioned into subgroups (blue and green) reflecting distinct expression trends. (E) Representative Cluster 0 proteins (Dnajc9, Hmox1, Nasp, Srf1, and Tcea1) showing higher abundance at Day 0, followed by progressive decreases through Day 6. (F) Representative Cluster 2 proteins (Tead1, Ncald, Vrk1, Mcm6, and Stk26) exhibiting reduced abundance by Day 4 with minimal additional change at Day 6. (G) Representative Cluster 3 proteins (Mki67, Fen1, Cdk1, Ubl5, and Aqr) displaying decreased abundance between Day 2 and Day 4, followed by increased levels at Day 6. (H) Representative Cluster 5 proteins (Txndc5, Tax1bp3, Pofut2, Tbl1x, and Gdf15) showing elevated abundance at Day 2 relative to Day 0 and decreased levels at later time points.

In Cluster 0 (Figure 8A, E), proteins including Dnajc9, Hmox1, Nasp, Srf1, and Tcea1 exhibit higher abundance at Day 0, followed by a progressive decrease through Day 6. In Cluster 2 (Figure 8B, F), proteins such as Tead1, Ncald, Vrk1, Mcm6, and Stk26 show a marked reduction in abundance by Day 4, with minimal additional change at Day 6. In Cluster 3 (Figure 8C, G), cell cycle–associated proteins, including Mki67, Fen1, Cdk1, Ubl5, and Aqr, decrease between Day 2 and Day 4, followed by increased abundance at Day 6 relative to Day 4. In Cluster 5 (Figure 8D, H), proteins such as Txndc5, Tax1bp3, Pofut2, Tbl1x, and Gdf15 display elevated abundance at Day 2 compared to Day 0, followed by reduced levels at later time points.

These cluster-level patterns demonstrate that proteins within UMAP-defined groups follow distinct temporal trajectories during differentiation, with subgroup-specific trends captured in Figure 8A–D and representative protein behavior illustrated in Figure 8E–H.

Several proteins defining these temporal clusters are associated with stress response, transcriptional regulation, and cell cycle processes. Stress- and protein homeostasis–related proteins, including Dnajc9, Hmox1, and Txndc5, show higher abundance at early time points followed by decreases over the differentiation time course (Figure 8E, H). Proteins involved in transcriptional regulation, such as Tcea1, Tead1, and Tbl1x, are more abundant at early or intermediate stages (Figure 8E, F, H). In contrast, cell cycle–related proteins, including Mki67, Cdk1, Mcm6, Fen1, and Vrk1, decrease during differentiation (Figure 8F, G). These temporal patterns are not apparent from pairwise comparisons between individual time points^33–37.

Applied non-negative matrix factorization for cluster analysis and identification of proteins contributing to group separation

Application of NMF resolved three protein profiles that captured major sources of variation across the single-cell proteomics data (Figure S8A). When cells were projected into the space defined by these profiles, undifferentiated cells clustered within a narrow region characterized by higher contribution from profile 1. Cells collected at later differentiation stages occupied a broader region of profile space defined mainly by profiles 2 and 3.

At Days 4 and 6, cells separated into two partially overlapping groups distinguished by different combinations of profiles 2 and 3. By Day 6, most cells showed reduced contribution from profile 1. A subset of Day 6 cells also showed reduced contribution from profile 3, indicating additional variation within the late differentiation stage.

Inspection of the protein-by-profile loading matrix identified a limited number of proteins that contributed strongly to each profile (Figure S8B). These proteins differed across profiles and accounted for much of the observed separation between cell groups. Cells within the same differentiation stage often showed different profile combinations, indicating variability in proteomic state among cells sampled at the same time point.

Proteins contributing to early subgroup separation at Day 2

Proteins contributing to the separation of the two Day 2 groups were examined by inspecting loadings along PC1 (Figure S9; Table S10). A subset of proteins showed relatively large positive or negative contributions to PC1, with z-scores close to ±2 and p-values below 0.05. These proteins accounted for much of the variance distinguishing the two groups at this time point.

Several of the highest-loading proteins are associated with translational and RNA-related processes, including Nat10, Eprs1, and Eif3f, along with proteins involved in intracellular trafficking such as Arl1. Proteins were distributed on both sides of PC1, indicating that the two Day 2 groups differ in opposing patterns of protein abundance rather than a shared directional change.

Proteins commonly used as markers of neuronal differentiation were not prominent among the PC1-associated features. Instead, the proteins contributing to group separation are consistent with differences in regulatory and biosynthetic activity, suggesting that variability at Day 2 reflects differences in cellular state rather than commitment to a neuronal differentiation program^38,43.

Proteins contributing to heterogeneity at Day 4 and Day 6

While time-point–based comparisons capture average proteomic changes across differentiation stages (Figures 2–5), clustering-based analyses (Figure 9) resolve subpopulation-specific differences within the same time points. This distinction is particularly important at later stages (Days 4-6), where increased cellular heterogeneity leads to the emergence of distinct proteomic subpopulations that are not apparent when cells are grouped solely by differentiation stage.

Figure 9.

Subpopulation-specific proteomic remodeling during NGF-induced differentiation. (A) PCA plot showing separation of Group A and Group B subpopulations at Days 4 and 6. (B) Volcano plot of proteins differentially expressed between the two groups. |log₂FC| ≥ 1, p < 0.05). Proteins meeting both thresholds are highlighted. (C-D) Relative abundance of VPS26B and SNX1 in Group A and Group B cells. (E) Schematic summary of NGF-responsive pathways associated with the two subpopulations. Created with Biorender.com.

The four differentiation time points (Day 0, Day 2, Day 4, and Day 6) represent defined experimental conditions, whereas the two groups identified in this analysis (Group A and Group B) correspond to emergent biological subpopulations resolved through clustering. At later stages (Days 4-6), cells do not form a single homogeneous population but instead separate into distinct proteomic states, reflecting differences in differentiation progression and NGF responsiveness. Importantly, these subgroup classifications do not replace time-point–based analysis but instead refine it by revealing heterogeneity within differentiation stages, where multiple proteomic states coexist.

To further examine subgroup-specific proteomic features associated with divergent differentiation outcomes, we compared signaling and proteome remodeling patterns between Group A and Group B cells during NGF treatment (Figure 9). Group A cells showed higher abundance of endosomal trafficking proteins SNX1 and VPS26B, along with elevated levels of translational and proteostasis-related proteins (EEF1A2, HSP90AB1, SARNP, RPSA, and RPS21). These features were accompanied by increased abundance of cytoskeletal and neuronal maturation–associated proteins (TUBA1A, MAP1B, NEFL, NEFM, PRPH, and VGF) and coincided with neurite outgrowth and neuron-like phenotypes³⁶.

In contrast, Group B cells exhibited lower abundance of proteins involved in receptor trafficking, translation, and cytoskeletal organization, aligning with weaker NGF-associated proteomic responses and the absence of neurite extension. VPS26B and SNX1 were selected as representative markers because their abundance reflects differences in endosomal trafficking capacity⁴⁵ between subgroups: VPS26B was enriched in Group A cells, whereas SNX1 showed reduced abundance in Group B cells. Collectively, these subgroup-resolved proteomic profiles show how single-cell proteomics enables the identification of divergent molecular states within a heterogeneous population, differences that would be averaged and therefore masked in bulk proteomic measurements (Figure S10).

Discussion

Although technical optimizations in sample preparation and single-cell dispensing are presented, this work is primarily focused on the biological application of single-cell proteomics to resolve proteome remodeling and cellular heterogeneity during neuronal differentiation.

The proteomic changes observed across the differentiation time course reflect coordinated remodeling of cellular programs associated with neuronal maturation. Although the total number of proteins identified per cell decreased throughout differentiation (Figure 2A, D), the abundance of numerous neuronal differentiation markers increased substantially at later stages. This pattern suggests a transition from a broadly expressed proteome characteristic of proliferating PC12 cells toward a more specialized neuronal proteome. Early time points (Day 2) were characterized by reductions in proteins involved in lysosomal function, lipid metabolism, and proliferation, consistent with the transition from a proliferative state toward neuronal differentiation. At later stages (Days 4 and 6), proteins associated with neuronal structure, axonal growth, and maturation, including NEFM, NEFL, INA, DPYSL4, TUBB2A, and VGF, exhibited increased abundance relative to Day 0. The enrichment of neurofilament proteins and microtubule-associated components is consistent with the structural reorganization required for neurite outgrowth and stabilization during NGF-induced differentiation.

In addition to these global trends, the single-cell measurements revealed substantial cell-to-cell variability within each time point, indicating asynchronous progression through differentiation. This heterogeneity was evident in both protein expression patterns and observed cellular morphology, where subsets of cells exhibited advanced neurite outgrowth while others remained less differentiated.

Both morphological assessment and proteomic profiling further demonstrate that PC12 differentiation does not occur uniformly across the cell population. While many cells exhibit neurite outgrowth and neuron-like features by Days 4 and 6, others remain partially or fully undifferentiated. This variability is reflected in the proteomic data, where later time points show increased dispersion in PCA, multimodal protein abundance distributions, and separation into multiple subclusters. In contrast, Day 0 cells cluster more tightly and align with a relatively homogeneous proliferative state, with individual cells occupying multiple proteomic states within the same differentiation stage.

Capturing this heterogeneity required careful optimization of cell handling and sample preparation. Differentiated PC12 cells are larger, more fragile, and prone to aggregation, making gentle dissociation, anti-clumping strategies, and careful tuning of dispensing parameters essential. These optimizations enabled high single-cell dispensing accuracy and reproducibility, providing consistent input for DIA-based proteomic analysis. Quality control metrics confirmed stable quantitative performance across all time points, with similar precursor-level variability and overlapping intensity distributions. Although fewer proteins were identified at later stages, this likely reflects biological changes associated with differentiation, including cell-cycle exit, reduced global protein synthesis, and increased membrane specialization, rather than technical limitations alone. The improved protein recovery observed with DDM further emphasizes the importance of optimized extraction strategies for low-input and differentiated cell types.

Beyond revealing cell-to-cell heterogeneity, single-cell proteomics provides insight into regulatory mechanisms that are inherently obscured in bulk measurements, particularly during dynamic biological processes such as neuronal differentiation. Differentiation involves asynchronous transitions, transient signaling events, and stage-specific remodeling of cytoskeletal, trafficking, and metabolic pathways. Bulk proteomics captures the mean signal across thousands of cells, which can mask transitional states and compress dynamic range, especially for proteins that are tightly regulated or expressed in a subset of cells. In contrast, single-cell approaches preserve the distribution of protein abundance across individual cells, enabling detection of rare states, asynchronous differentiation trajectories, and coordinated remodeling within specific subpopulations^2,5. This is particularly relevant in neuronal systems, where neurite extension, synaptic maturation, and axonal transport are not uniformly activated across all cells at the same time. Moreover, protein-level regulation often diverges from transcript-level patterns due to post-transcriptional control, differences in translation efficiency, and protein⁴⁶. Single-cell proteomics, therefore, captures regulatory complexity at the functional protein level and provides a more direct view of molecular programs governing neuronal growth and maturation.

An important consideration in this study is that the analytical platform used here was specifically designed for high-sensitivity, low-input proteomics. The timsTOF SCP system integrates trapped ion mobility spectrometry with diaPASEF acquisition, enabling efficient separation of singly charged background ions from multiply charged peptide species and improving precursor selectivity and overall sensitivity. The combination of ion mobility filtering, synchronized quadrupole isolation, and rapid duty cycle increases peptide identification rates at the single-cell level. In addition, optimized ion accumulation parameters enhance signal intensity for low-abundance peptides, which is critical for single-cell measurements. These features contribute to the depth and reproducibility observed in this dataset and distinguish this platform from conventional mass spectrometry approaches not optimized for single-cell analysis.

From a broader perspective, continued improvements in instrumentation, including SCP-dedicated platforms and emerging high-sensitivity systems such as Orbitrap Astral, are expected to further expand proteome coverage and throughput at the single-cell level. These advances will be particularly important for complex biological systems where cellular heterogeneity is a defining feature.

An important future direction for this workflow is its application to central nervous system tissue. Ongoing work in our laboratory is focused on single-cell dissociation of brain regions, including cortex and hippocampus, from 5xFAD mouse models. Following dissociation, individual cells can be analyzed using either single-cell proteomics workflows or spatially resolved approaches such as single-cell MALDI. Integration of these datasets with multivariate statistical analysis will enable classification of distinct cell populations and identification of cell-type-specific pathways associated with disease progression. Extending single-cell proteomics to tissue-derived samples will provide a more physiologically relevant view of neuronal heterogeneity and molecular mechanisms underlying neurodegenerative processes.

In summary, this work demonstrates the utility of single-cell proteomics for studying dynamic and heterogeneous biological processes such as neuronal differentiation. This workflow enabled reproducible protein measurements across multiple days of differentiation and captured intermediate proteomic features that are not resolved in bulk analyses. Although current single-cell proteomics platforms remain limited in throughput, continued improvements in sample preparation, chromatographic separation, and mass spectrometry are expected to enhance performance. Application of these approaches to primary neuronal systems and in vivo models, combined with complementary modalities such as imaging or transcriptomics, will further advance understanding of molecular processes underlying neuronal differentiation and maturation.

Materials and methods

Cell culture

PC12 cells were obtained from ATCC (Part No. CRL-1721; ATCC) and differentiated into neuronal-like cells using 100 ng/mL of Nerve Growth Factor (NGF) (Part No. 86923-98-0; Sigma-Aldrich). Cells were seeded at a density of 2x10³ cells/cm^{2 50} in poly-L-lysine (Part No. 0403; ScienCell Research Laboratories) -coated 10 mm culture dishes. After seeding, cells were incubated at 37 °C for 3 hours in RPMI medium (Part No. R8758-100ML; Sigma-Aldrich) supplemented with 5% fetal bovine serum (FBS) (Part No. F4135; Sigma-Aldrich), 10% horse serum (HS) (2 Part No. 6050088; Thermo Fisher Scientific), and 1% penicillin-streptomycin (Part No. 15140122; Thermo Fisher Scientific) to promote adherence to the PLL-coated surface. Following incubation, the medium was replaced with OPTI-MEM (Part No. 15140122; Thermo Fisher Scientific) containing 0.5% FBS to initiate neuronal differentiation. NGF-treated cells were cultured for up to 6 days, with media and NGF replaced every 2 days to ensure consistent and effective differentiation⁴⁹. Cells were collected on days 2, 4, and 6 after treatment and dispensed to be used in downstream proteomics analysis.

Single-cell dispensing proteomics workflow

Single cells and reagents were dispensed using the Hewlett-Packard Digital Dispenser (HP-D100), also known as the Uno single-cell dispenser (Tecan, Switzerland) (Figure 1A, Table S1 and S2). Single cells were deposited into 384-well plates (Part No. EP0030129547-25EA; Sigma-Aldrich) pre-filled with 0.5 μL LC/MS-grade water, using C1a cassettes (Part No. 30230841, Tecan) with the medium cell size setting. Then, 1 μL of trypsin digestion solution (Part No. 90057, Thermo Fisher Scientific), containing 1 ng of trypsin/lysC mixture and 0.03% n-dodecyl β-D-maltoside (DDM) (Part No. D4641-500MG; Sigma-Aldrich), was dispensed using D1 cassettes (Part No. 30230843, Tecan) with the surfactant-free setting in the single cell dispenser software (Table S2). Plates were centrifuged at 2,250 × g for 5 minutes to ensure cells and reagents settled at the bottom of each well, reducing sample loss during injection. Digestion was performed by incubating the plate at 37 °C for 2 hours, followed by centrifugation and LC-MS analysis using 384 well plate pre-slit plate sealer (Part No. NC2469150, Fisher Scientific) to prevent contamination and sample evaporation (Figure S11).

Fluorescence imaging to validate single-cell dispensing and neuronal differentiation

Fluorescence imaging was used to assess cell viability and confirm proper single-cell dispensing during neuronal differentiation. Live cells were labeled with Calcein AM (Thermo Fisher Scientific, Part No. 65-0853-39) at an optimized working concentration and incubated for 15 minutes prior to dispensing. Fluorescent signal from viable cells was used to verify cell integrity and handling throughout the procedure. GFP fluorescence images were acquired on the Keyence BZ-X800 microscope using the GFP filter cube (CH2). Images were collected in monochrome mode at high resolution. Because a single field of view did not capture the full well area at the selected magnification, multiple adjacent fields were acquired per well using the multi-point capture function. Autofocus was applied prior to imaging. Individual fields were stitched after acquisition using the Keyence Analyzer software to generate composite images for each well. When enabled, low-photobleach mode was used to restrict excitation light exposure to the time of image capture. Using the Keyence BZ-X800 microscope and fluorescent and bright-field settings, we confirmed the precise isolation of single cells by the single-cell dispenser and observed neurite outgrowth indicative of neuronal differentiation. Representative images of stained cells and the labeling mechanism of Calcein AM are shown in Figure S1.

Data acquisition using timsTOF SCP mass spectrometer

Single-cell proteomic analysis was performed using a NanoElute 2 UHPLC system coupled to a Bruker timsTOF SCP operated in dia-PASEF mode (Bruker Daltonics, Germany). Peptides were separated on an Aurora Series Gen3 C18 analytical column (25 cm × 75 μm, 1.7 μm; IonOpticks, Australia) maintained at 50 °C. Mobile phase A consisted of 0.1% formic acid and 0.5% acetonitrile in water, and mobile phase B consisted of 0.1% formic acid and 0.5% water in acetonitrile. Chromatographic separation was performed at a flow rate of 0.3 μL minutes⁻¹ using a 15-min gradient.

Mass spectrometric analysis was performed in positive ion mode over an m/z range of 100–1700. The Captive Spray ionization source was operated at 1700 V capillary voltage, 3L/min drying gas, and 200 °C drying temperature. During analysis, the timsTOF SCP was operated with Parallel Accumulation-Serial Fragmentation (PASEF)⁵¹ scan mode for DIA acquisition. Trapped ion mobility spectrometry (TIMS) separates ions in the gas phase by balancing an opposing electric field against a constant gas flow, resulting in separation based on ion mobility. In timsTOF instruments, ion mobility is reported as the inverse reduced mobility (1/K₀, V·s·cm⁻²), which reflects ion size and gas-phase conformation and serves as a proxy for collision cross section ^53,54. The TIMS analyzer operated in custom mode with an ion mobility range of 1/K₀ = 0.66-1.42, with a 100 ms accumulation time and a 100 ms ramp time. The accumulation time, or trap fill duration, is the time during which ions are stored in the TIMS cell before release; longer accumulation enhances ion collection, which is especially important for sensitivity at single-cell peptide levels. The ramp time indicates how long the instrument takes to scan the full mobility range by lowering the electric field, thereby affecting ion mobility resolution; longer ramps offer higher resolution but slow the duty cycle. These settings resulted in a 9.34 Hz ramp rate. In the dia-PASEF acquisition, a precursor mass range of 400-1000 m/z and a mobility window of 1/K₀ = 0.64-1.37 were used, resulting in a cycle time of 0.96 s. Fragmentation was performed with mobility-dependent collision energies ranging from 20 eV at 1/K₀ = 0.60 to 50 eV at 1/K₀ = 1.60. For each TIMS cycle, 11 dia-PASEF scans were used, each with 3-4 steps. A total of 36 dia-PASEF windows were used, spanning from m/z 299.5 Th to m/z 1200.5 Th, and from ion mobility range (1/K₀) 0.7 V·s/cm² to 1.3 V·s/cm², with an overlap of m/z 1 Th between two neighboring windows. The collision energy was ramped linearly as a function of mobility value from 20 eV at 1/K₀ = 0.6 V·s/cm² to 65 eV at 1/K₀ = 1.6 V·s/cm².

Single-cell proteomics data processing

Single-cell proteomic data acquired through dia-PASEF were analyzed using DIA-NN (version 1.8.1) (https://github.com/vdemichev/DiaNN)⁵⁴ against a Rattus norvegicus FASTA database (UniProt release 2024_01)⁵⁵ and common contaminants. Bulk libraries were generated using the same method with 10 ng injections. Single-cell samples were analyzed using a spectral library and the Match-Between-Runs (MBR) feature, leveraging bulk samples for alignment. In-silico digestion of a FASTA database was performed using trypsin/P with allowance for one miscleavage. Deep learning-based predictions of MS/MS spectra, retention times (RTs), and ion mobilities (IMs) were enabled to enhance peptide identification. Peptide length ranged between and 30 amino acid residues, precursor charge ranged 2-4, precursor m/z ranged 300-1200, and fragment ion m/z ranged 200-1800. Precursor FDR was set to 1%, with 0 for settings ‘mass accuracy’, ‘MS1 accuracy’, and ‘scan window’. Settings ‘heuristic protein inference’, ‘use isotopologues’, and ‘no shared spectra’ were all enabled. ‘Gene’ was chosen for protein inference parameter along with ‘double-pass mode’ for neural network classifier. Robust LC (high precision) was used for quantification, the RT-dependent mode for cross-run normalization, and smart profiling mode for library generation.

Bioinformatic and statistical data analysis

Single-cell proteomics data were processed using standard filtering, normalization, and statistical procedures. Proteins detected in fewer than 25% of biological replicates were removed prior to downstream analysis. This filtering step was used to reduce the contribution of low-frequency protein identifications. For analyses requiring complete data matrices, missing values were imputed using a k-nearest neighbor (kNN) method based on similarity between samples.

Protein intensity values were normalized on a per-cell basis by dividing each protein intensity by the total protein signal measured within that cell (sum normalization; Figure S12). Normalized values were log₁₀-transformed. Z-score scaling was then applied across proteins. Z-scores were calculated as ${\text{z}=(\text{x}}_{\text{i}}-\overline{\text{x}})/\text{S}$, where x_i is the protein intensity in a given cell, $\overline{\text{x}}$ is the mean protein intensity across all cells, and S is the standard deviation.

Dimensionality reduction was performed using principal component analysis (PCA) and Uniform Manifold Approximation and Projection (UMAP)⁵⁶. PCA was conducted on imputed data to ensure a complete matrix for multivariate analysis. Clustering was performed using k-means clustering on normalized protein abundance values. For each cluster, mean normalized protein abundance values were calculated at each differentiation time point.

Non-negative matrix factorization (NMF) was also applied to the filtered protein abundance matrix using non-negative intensity values. Proteins with low detection frequency were removed prior to factorization. Multiple factorization settings were evaluated, and three profiles were retained for downstream analysis. No differentiation time-point labels were included during factorization. Cell-level profile weights were used for visualization, and protein loadings were examined to identify proteins contributing to differences between groups ^57,58.

Functional enrichment analysis was performed using Metascape⁵⁹(v3.5.20250701). Enriched terms with −log₁₀(p) ≥ 1.3 were retained for interpretation.

Statistical comparisons of protein abundance between undifferentiated cells and NGF-treated cells at Day 2, Day 4, and Day 6 were performed using unpaired, two-tailed nonparametric tests with α = 0.05. Pairwise comparisons were assessed using the Mann–Whitney U test. Comparisons involving more than two groups were evaluated using the Kruskal–Wallis test followed by Dunn’s post hoc test. Differences between samples prepared with and without n-dodecyl-β-D-maltoside (DDM) were assessed using unpaired, two-tailed Student’s t-tests after confirming assumptions of normality and variance. Precursor-level intensity distributions were compared using log₂-transformed values and Kruskal–Wallis tests with multiple comparisons correction. All statistical analyses were performed in R using the rstatix package.

Supplementary Material

Additional experimental details and methods; supplementary figures (Figures S1–S11) showing extended data analyses and validation; supplementary tables (Tables S1–S10) summarizing protein quantification and statistical analyses; additional tables (Tables S11–S17) provided as an Excel file containing extended datasets and annotations.

Data Availability Statement

The raw single-cell proteomics mass spectrometry data generated in this study have been deposited in the MassIVE repository under accession number MSV000100666.