Design and characterization of the iProChip and streamlined microproteomics workflow
To provide a streamlined microproteomic pipeline for mass-limited samples, we designed a microfluidic device as an integrated proteomics chip (iProChip) to offer all-in-one functionality from cell input to complete proteomic sample processing. The iProChip has a two-layer, push-up geometry and allows accurate fluid manipulation via 34 valves controlled by a custom program, thereby offering an automated protocol for precise and systematic control (Fig. 1a–f and Supplementary Fig. 1a)29. The chip is composed of 9 units to enable multiplexed proteomic experiments running in parallel. Each unit contains a cell capture, imaging and lysis chamber, a protein reduction, alkylation and digestion vessel, and a peptide desalting column (Fig. 1c and Supplementary Fig. 1b). All units share 9 inlets and 2 outlets, allowing programmed delivery of reagents and simultaneous sample processing to increase assay throughput. The cell trap is made up of arrays of 10, 50 and 100 wedge-shaped twin pillars spaced by 5 μm for rapid size-based cell capture in 5, 8.5, and 11 nL chambers, respectively (Fig. 1e)30. A circular chamber with a radius of 1 mm and a height of 100 μm (312 nL) was fabricated to accommodate the entire proteomic workflow, including cell lysis, protein reduction, alkylation and digestion in a single step (Fig. 1f and Supplementary Fig. 2). Note that the calculated surface-to-volume ratio for iProChip is larger than that of existing single-cell devices, such as nanoPOTS, yet it still exhibits a substantially reduced surface area by >90% in comparison to the microscale vial-based workflow (Supplementary Table 1)11. For peptide desalting, a 2.5 cm-long column with a cross-section of 200 μm × 25 μm was fabricated by packing reversed-phase C18 beads into the microchannel prepatterned with 5 μm filters to perform on-chip clean-up of digested peptides (Fig. 1e, f and Supplementary Movie 1). To increase proteome coverage, we applied a deep single-shot profiling strategy that integrates direct- and library-based DIA analysis using an Orbitrap mass spectrometer. We developed a spectral library resource complementarily established by hybrid DDA-DIA datasets using either cancer cell lines or immune cells consisting of different proteome compositions, which can serve as a digital map to theoretically recover all peptides in the m/z and retention time domains of DIA data (Fig. 1g). Specifically, spectral libraries constructed from cell lines with different cell numbers were tested and optimized to maximize the number of identified and quantified proteins.
a A bright-field image of the integrated proteomics chip (iProChip), where cell capture chambers (cyan), reaction vessels (orange), on-chip SPE columns (green), sample collection ports (dark green), and control layers (brown) are shown. b The entire system set-up for iProChip operation. c A close-up view of a single operation unit. Scale bar: 300 μm. d A ready-to-use iProChip mounted on the microscope. e SEM images of cell capturing pillars (left) and C18 filters in the SPE column (right). These images are representative of two chips that were observed using SEM. f Operational procedures of iProChip for streamlined sample preparation, including (1) cell trapping, imaging, and counting, (2) cell lysis, (3) protein digestion, (4) desalting, and (5) peptide collection. g Proteomic analysis using data-independent acquisition-based liquid chromatography–tandem mass spectrometry (LC-MS/MS) and spectral library search.
In the first step of the streamlined workflow, the cell trapping efficiency was determined using non-small-cell lung cancer (NSCLC) PC-9 cells (“Methods”). Using optimal cell density (500 cells/μL), desired numbers of cells (1–100) for each unit can be trapped in 10–60 s. The average percentage of cells captured from traps containing a single cell were 100, 92 ± 3, and 89 ± 8% for chambers with 10, 50, and 100 traps, respectively. The targeted capture efficiency for all units reached ~100% after counting traps containing 1 (~90%) and 2 or 3 cells (~10%), establishing it as an absolute quantifiable module to perform simple and fast size-based cell isolation (Fig. 2a, b, “Methods,” and Supplementary Movie 2). Compared to external stand-alone cell sorters, such a built-in module offers simple, rapid, and efficient cell isolation. Additionally, we also showed that by using a lower cell density (25 cells/μL) operated at 3 psi, such cell chambers allow precise capture of lower numbers of cells at the level of 1 and 5 cells (Supplementary Fig. 3). The cell trapping capability of iProChip was also evaluated to characterize the cell usage efficiency (defined as numbers of trapped cells/numbers of total injected cells) and minimum numbers of cells needed for iProChip operation (“Methods”). Using a total of either 5 or 10 μL cell solution (25 cells/μL), the results showed that cell usage efficiency ranged from ~4 to 44% for capturing 1–100 cells (Supplementary Table 2 and Supplementary Fig. 4). Next, we sought to characterize whether reagents can mix efficiently in the closed vessel during cell lysis and protein digestion. Three mixing approaches, including vortexing, shaking (by a plate shaker), and passive diffusion, were tested (Supplementary Fig. 5). Using imaging analysis, the relative mixing index (RMI) was calculated to assess the mixing performance (“Methods”)31. The results showed that it took 11, 16, and 30 min for vortexing, shaking, and diffusion-mixing to reach 75% RMI, indicating that all three mixing strategies were sufficient to accommodate reactions within minutes to hours of reaction kinetics, which fit the timescale of conducting the proteomics workflow (Fig. 2c, d and Supplementary Fig. 5). Although vortex mixing was found to provide faster mixing, mixing by shaking was used in subsequent experiments due to its flexibility in handling and sufficient reaction timescale.
a Bright-field images of non-small lung cancer PC-9 cells captured in 10, 50, and 100 cell chambers. Top right: a zoom-in image of a trapped cell. b Characterization of the cell capture efficiency for separate capture chambers. Data are presented as mean values ± SD (n = 3 independent measurements). c Representative time-lapse images of a reaction vessel filled with green dye during mixing-by-shaking characterization. Scale bars: 300 μm. d Comparison of mixing efficiency in the reaction vessel. Data are presented as mean values ± SD (n = 3 independent experiments). e A bright-field image of the SPE column packed with C18 beads. Top right: a close-up view near the C18 filter. f Desalting recovery efficiency of the on-chip SPE column. Data are presented as mean values ± SD (n = 3 independent experiments). Source data are provided as a Source data file.
Another integration to the miniaturized device is the on-chip peptide desalting module (Fig. 2e and “Methods”). The effectiveness of desalting was evaluated by processing ~10 cells, where the profile of the desalted sample showed reproducible typical peptide ion profiles, while the nondesalted sample showed the predominant presence of detergent peaks (Supplementary Fig. 6). The loading capacity and peptide recovery (%) of the desalting module showed a linear correlation from 0.125 to 1 μg with ~89% recovery (Fig. 2f and “Methods”). Assuming that a typical mammalian cell contains 200 pg proteins17, the capacity of the desalting column is thus anticipated to capture peptides from approximately 4000 cells. Furthermore, the concern of compromised sample retrieval due to preferential flow was evaluated by flowing a colored dye through the C18 bead-packed column. The results showed that 9 psi was the minimal flow pressure to overcome the preferential flow, and 11 psi was used in the following workflow (Supplementary Fig. 7 and “Methods”).
Integration of iProChip with DIA MS
To operate the iProChip for proteomic workflow, precise numbers of cells, including 1, 5, 10, 50, and 100 cells, were captured using built-in cell traps. Parallel processing of the cells trapped in capture chambers was performed by dispensing and incubating with cocktail buffer containing RapiGest, tris(2-carboxyethyl)phosphine hydrochloride (TCEP), and chloroacetamide (CAA), which was specifically adapted to achieve one-pot cell lysis, protein reduction, and alkylation to minimize sample loss (Supplementary Figs. 2 and 8 and Movie 3). Subsequent protein digestion and acidification were conducted in the reaction vessel, and digested peptides were then subjected to multiplexed desalting by passing through the C18 column for 15 min. Note that before each experiment, the cell chamber and reaction vessel were coated with bovine serum albumin (BSA) to minimize adsorptive losses of peptides. For subsequent liquid chromatography (LC)–MS/MS analysis, single-shot DIA-MS acquisition parameters, including isolation window, resolution, peptide amount, and LC–MS/MS gradient were optimized to enhance proteomic profiling coverage in low numbers of cells (“Methods”)32.
To allow deep profiling and enhanced identification of low abundance and cancer-relevant proteins by DIA, high-quality project-specific spectral libraries were constructed using lung cancer and human chronic lymphocytic leukemia (CLL) cell lines. The protein compositions and dynamic range may vary in bulk samples containing thousands of cells and a single cell, which likely affect the chromatographic time domain and DIA acquisition. Thus, we constructed both large-scale (1 μg) and small-scale (~10 cells) libraries from respective cell types, i.e., PC-9 and MEC-1 cells, and implemented them to analyze various numbers of cells. Specifically, the large-scale project-specific libraries of PC-9 and MEC-1 processed in bulk/dilution with DDA and DIA modes consisted of 6,345 protein groups (83,305 peptides) and 6261 protein groups (60,335 peptides) with 1% precursor and protein FDR, respectively. These large-scale libraries were used to analyze higher cells (i.e., >10 cells). For lower input samples (i.e., ~5 and single cells), we reasoned that a small-scale specific library should be beneficial for identification and quantification. To maximize the proteome profiling sensitivity at the single-cell level, we constructed small-scale spectral libraries from ~10 cells processed through iProChip as well as aliquots of ~1.5 ng (~10 cells) processed through bulk/dilution, yielding a depth of 2231 protein groups (14,054 peptides) and 2440 protein groups (11,720 peptides), respectively.
To evaluate the performance of DIA-based quantitation for iProChip, analytical merits in sensitivity, proteome coverage, and reproducibility were systematically investigated by using iProChip to process 13–14 PC-9 cells and compared to the conventional DDA method (Fig. 3a). The identification and quantification analysis of all the datasets in this study were performed at a statistically stringent criterion of 1% FDR at the peptide-to-spectrum match (PSM) and protein levels. By the DDA method, an average of 869 protein groups (3280 peptides) were identified in the triplicate analysis. In comparison, the direct DIA (dirDIA) approach using the Spectronaut tool identified 1409 protein groups (5174 peptides), whereas the library-assisted DIA (libDIA) approach using the large-scale PC-9 cell library showed significantly higher proteome coverage of 1874 protein groups (6929 peptides). Comparing the dirDIA and spectral library-based results, the superior quantitation of libDIA is likely due to more efficient detection of low-intensity peptide ions in DIA mode to match the corresponding peptide spectra in our library. By combining the complementary dirDIA and libDIA results, the overall identification coverage further increased to 2022 proteins (7757 peptides). The identification of 2.3-fold and 2.4-fold protein groups and peptides, respectively, by the DIA approach revealed its superior profiling coverage over the DDA approach (Fig. 3a). These results demonstrated that a single-shot DIA-based LC–MS/MS, complementarily processed by dirDIA and libDIA, improved the proteome identification coverage for the small-scale sample from the fully automated sample preparation in iProChip.
a Comparison of protein groups and peptides identified by triplicate analysis using the data-dependent acquisition (DDA, light blue) and data-independent acquisition (DIA, light green) modes. Single-shot DIA was acquired and processed by both library-based DIA (libDIA) and direct DIA (dirDIA) approaches by Spectronaut. b Overlap of protein groups identified by DDA and DIA. c Distribution of coefficient of variation (CV%) for quantified protein groups by DDA and DIA. d Evaluation of missing values (%) of proteins identified and quantified in triplicate LC–MS/MS runs by DDA and DIA. e Assessment of dynamic range based on protein abundance rank and annotation of selected proteins related to lung cancer. Source data are provided as a Source data file.
To evaluate the reproducibility of the iProChip-DIA workflow, we calculated the percentage of overlapping proteins between the triplicate analysis of 14 PC-9 cells. The results showed that 84.2% of the 2022 identified proteins and 71.7% of the 869 identified proteins were reproducibly detected by DIA and DDA, respectively, indicating higher reproducibility and proteome coverage with the iProChip-DIA approach (Fig. 3b). For evaluation of reproducible quantitation with CV ≤ 20%, DIA achieved significantly higher coverage of 1160 quantifiable proteins compared to 522 proteins quantified by DDA (Fig. 3c). Previous label-free quantification methods have commonly observed 10–50% between-run missing values, presenting a bottleneck for reproducible quantification across samples33. Additional comparison for run-to-run variabilities revealed fewer missing values, i.e., the number of proteins only quantified in one of triplicate runs, in the DIA result (16%) compared to that of DDA (28%) (Fig. 3d). Meanwhile, the wide dynamic range of proteome compositions presents another major bottleneck for deep profiling, especially for low-abundance proteins. Thus, we assessed the dynamic range based on the protein abundance rank. By DDA, the abundances of the 1014 identified proteins were found to span ~4 orders of magnitude, whereas 2170 identified proteins in DIA span ~5 orders of magnitude with coverage of important and low-abundance oncoproteins related to cancer. Notably, the FDA approved druggable targets for lung cancer, such as EGFR, MAP2K1, and MAP2K2, and proteins involved in the NSCLC pathway, including EGFR, NRAS, MAP2K1, MAP2K2, MAPK1, MAPK3, CDK4, and TP53, were readily identified in DIA, whereas only TP53 and CDK1 were detected in DDA using our approach (Fig. 3e)34 (https://www.cancer.gov/about-cancer/treatment/drugs/lung). In summary, the DIA approach provides higher proteome profiling coverage, lower missing values, reproducible quantification, and a wider dynamic range than DDA at the level of 14 cells.
Quantitative proteome profiling of mass-limited samples by iProChip and DIA-MS
Next, we systematically evaluated the iProChip-DIA performance for proteomic profiling of PC-9 cells in chambers with 10, 50 and 100 traps, which may allow the experimental design of different sample inputs (Supplementary Data 1). As expected, the iProChip provided precise cell counting for each chamber to ensure unambiguous quantification (Fig. 4a). Combining libDIA and dirDIA analysis, on average, 4722 ± 10 protein groups (25,785 peptides) were identified from triplicate analysis from 106 ± 2 cells at 1% FDR. At chambers with lower numbers of cell traps, an average number of 3435 ± 262, 2022 ± 114, and 1638 ± 191 protein groups were identified from 55 ± 1, 14 ± 1 and as low as 5 ± 1 cells, respectively (Fig. 4a and Supplementary Fig. 9). The overlap of identified protein groups in triplicate analyses was 77–93% from all cell numbers, illustrating the high reproducibility of the iProChip-DIA approach (Supplementary Fig. 10). Importantly, to characterize assay reproducibility across experiments, a separate batch of 10-cell samples was processed in an independent chip. The results showed a similar level of protein identification results in the two batches (Supplementary Fig. 11a), with 78–84% within-batch overlap and 71% between-batch overlap (Supplementary Fig. 11b–c), suggesting good reproducibility among separate experiments.
a Identification summary of protein groups across different PC-9 cell numbers by iProChip-DIA. Data are presented as mean values ± SD (n = 3 independent cell samples for each condition). b Single PC-9 cell trapping using 10 cell capture chambers and corresponding cell image and triplicate analysis results of identified protein groups and coverage. SC: single cell; scale bar: 100 μm. c A heatmap showing the reproducibility of protein abundances obtained among different cell numbers and their replicates (R1–R3). d Distributions of total protein abundances in commonly identified proteins across different cell numbers. Each data was analyzed in n = 3 independent measurements. Center lines show the mean; box limits indicate the 25th and 75th percentiles; whiskers, 1.5× interquartile range. e Representative examples of lung cancer-related proteins showing quantitation of protein abundance calculated from peak area. Data are shown as mean ± SD from n = 3 independent measurements. f Identification coverage of proteins within the NSCLC pathway under different cell numbers using Kyoto Encyclopedia of Genes and Genomes (KEGG) database. Source data are provided as a Source data file.
Encouraged by the promising sensitivity, we further pushed the profiling sensitivity at a single PC-9 cell using 10-cell capture chambers. An average of 976 ± 37 protein groups (3069 peptides) were identified from triplicates of a single PC-9 cell (Fig. 4b). The sum of triplicate measurements yielded the identification of 1160 protein groups (3995 peptides) from a single PC-9 cell. Comparing the identification coverage showed that 69% protein groups and 55% peptides were common among triplicate results. To evaluate the background of the single-cell measurement, blank control samples (cell-free supernatant without trapped cells) were prepared for all sample preparation steps using iProChip and analyzed. Compared to the average of 88 proteins from similar studies35, our results of an average of 58 proteins presented comparatively fewer proteins from the blank sample (Supplementary Fig. 12). Furthermore, the comparison showed a significant overlap between the 1-cell and 5-cell samples, both minimally overlapping with the blank sample. These results suggest low cross-contamination and false identifications.
To compare the iProChip with bulk sample processing, we directly processed ~100 cells in vial yet barely identified proteins with only an average of 86 protein groups (Supplementary Fig. 13). Comparison to dilution-based processing in vials (~50 µg lysate) was also performed by injecting small aliquots corresponding to 1.5 ng (~10 cells) and 15 ng (~100 cells) for DIA analysis (“Methods”). While the results showed comparable protein and peptide identification at 15 ng in both iProChip and bulk-dilution preparations, 2-fold more identifications were obtained from the iProChip workflow at 1.5 ng (Supplementary Fig. 14). This higher performance gain of iProChip for smaller numbers of cells further validates the efficiency of our approach for limited cells. Additionally, when the data (1–106 cells) were compared to a conventional proteomics sample (~1 µg PC-9, ~6000 cells, >6000 proteins), 90–96% of the proteins identified from lower cells (1–106) using iProChip were common (Supplementary Fig. 15a). The fractions of the bulk proteins captured with iProChip showed an increasing trend correlated with cell numbers (Supplementary Fig. 15b). Interestingly, 75% of these 6000 proteins from 1 µg PC-9 cells were identified from 106 cells processed by iProChip, suggesting bulk-like comparable identification at a small sample scale. Further analysis of protein abundance showed that mainly abundant proteins were observed in lower cells, suggesting challenges of detecting low abundant proteins in low cell samples (Supplementary Fig. 15c).
To evaluate the analytical reproducibility of our approach for quantitative proteomics, the protein abundances in triplicate datasets were quantitatively compared by pairwise correlation analysis. The results showed good reproducibility (Pearson’s correlation of 0.88–0.98) in the measured protein abundance for protein quantification by our iProChip-DIA workflow (Fig. 4c and Supplementary Fig. 16). To assess the quantitative performance, next, the distribution of overall protein abundances quantified in each cell number was calculated, which showed a log-linear correlation across different cell numbers (Fig. 4d). The capability of quantitative proteomic analysis was further evaluated at the individual protein level. Representative examples were selected from lung cancer-related oncoproteins, and their abundances were computed by Spectronaut36. The average protein abundances among representative lung cancer proteins, EGFR, CDK1, and MAP2K1, revealed good linearity between the measured protein abundance and increasing cell numbers (Fig. 4e). Most importantly, many quantified proteins, such as selected examples of TP53, ITGB1, PGK1, and MAPK1, showed good quantitative dependence (50-100-fold) between the protein abundance and cell number (1–106 cells) (Fig. 4e and Supplementary Fig. 17). Quantification at 50–100-fold magnitudes also demonstrated a wider dynamic range compared to conventional quantitative proteomic results on a bulk scale. In line with the aforementioned quantitative performances, iProChip-DIA enables a high degree of robustness, good reproducibility, and quantitative proteomic measurements down to the level of a single cell, a level of performance only achieved previously for ensemble measurements.
The identification of these proteins enabled us to map lung cancer-related signaling pathways searched against the Kyoto Encyclopedia of Genes and Genomes (KEGG) database34. A total of 329 pathways were enriched, such as the NSCLC pathway, metabolic pathways, pathways in cancer, spliceosome, viral carcinogenesis, proteoglycans in cancer, MAPK signaling, and apoptosis (Supplementary Fig. 18). The major lung cancer pathway, the NSCLC pathway, was enriched with coverage of a total of 29 proteins across different numbers of cells (Fig. 4f). Even at low cell numbers (14 ± 1 cells), 13 proteins, including the drug targets EGFR, MAP2K1, MAP2K2, MAPK1, MAPK3, and KIF5B, the tumor suppressor TP53, and other key signaling components (KRAS, CDK4, CDKN2A, EML4, KIF5B, NRAS, BAX, and RB1), were identified (Fig. 4f). In terms of sensitivity, EGFR, MAPK1, MAP2K1, MAP2K2, CDKN2A, TP53, KIF5B, and GRB2 proteins were still detected in as few as 5 cells, whereas MAP2K1, KRAS, and TP53 were even identified at the single-cell level (Fig. 4f). The confidence of protein identification for the above representative proteins is shown in their MS/MS spectra (Supplementary Fig. 19 and Supplementary Table 3). Based on a lung cancer model study, these results reveal the capability of the developed approach to provide protein coverage to study the cancer proteome and a wide range of cellular pathways at limited cell numbers.
Application of iProChip-DIA for single leukemia cell proteomic profiling
The general applicability of our iProChip-DIA platform was next demonstrated on the human B-CLL cell line MEC-1. From the perspective of methodology development, leukemia cells representing a heterogeneous cancer type are ideal models for developing sensitive proteomics tools, as they could readily complement various existing methods by delineating the system-wide profiles of phenotypic functionality37. When processing MEC-1 cells on-chip, the imaging-based cell trapping feature of iProChip revealed that MEC-1 cells were noticeably smaller than PC-9 cells (Supplementary Fig. 20). Combining dirDIA and libDIA using the MEC-1 spectral library, triplicate analyses of MEC-1 cells by iProChip-DIA analysis yielded averages of 3811 ± 362, 931 ± 72, and 455 ± 98 protein groups from 117 ± 1, 14 ± 1 cells, and a single cell at 1% FDR, respectively (Fig. 5a, Supplementary Fig. 21, and Supplementary Data 2). The protein abundance was found to span ~5 orders of magnitude across different cell numbers, allowing the detection of important B cell surface markers CD20 and HLA molecules from as little as a single cell (Fig. 5b), while other key proteins, including CD19, CD22, CD47, and CD74, were identified from 14 and 117 cells (Fig. 5b). Functional annotation using UniProt showed that many proteins related to adaptive immunity, innate immunity, kinases, phosphatases, and Ig domains were identified from the single MEC-1 cell dataset, where the depth of protein coverage positively correlated with the cell number (Fig. 5c). By mapping 518 human kinases deposited in KinMap38, 114 protein kinases were identified across all major branches of the kinase phylogenetic tree, such as tyrosine kinase (TK), TK-like kinases, serine/threonine protein kinases, casein kinase 1 (CDK1), and Ca2+/calmodulin-dependent protein kinase (Fig. 5d). It was also noted that although MEC-1 cells were smaller than PC-9 cells, protein identification achieved good coverage and overlap (61–81%) using the iProChip-DIA approach, suggesting the versatility and robustness of our platform for different cell types (Supplementary Figs. 9c and 10b).
a Identification summary of protein groups across different MEC-1 cell numbers by iProChip coupled to DIA-MS. Data are presented as mean values ± SD (n = 3 independent cell samples for each condition). b Assessment of dynamic range based on protein abundance rank and annotation of selected proteins related to immune cancer markers. c Enrichment of immune-related and other functional classes against the UniProtKB database. d Kinase tree for mapping 114 kinases from the total cell numbers. The kinase families listed includes TK (tyrosine kinases), TKL (tyrosine kinase-like), CK1 (casein kinase 1), CAMK (calcium/calmodulin-dependent protein kinase), AGC (containing PKA, PKG, PKC families), CMGC (containing CDKs, MAPK, GSK, CLK families), and STE (serine/threonine kinases many involved in MAPK kinases cascade). Source data are provided as a Source data file.
B cell receptor (BCR) signaling is crucial for mounting efficient adaptive immunity and is involved in the survival and growth of malignant B cells in B-CLL39. B cell activation is regulated via the interaction between the surface receptor complexes in BCRs and specific antigens40. The iProChip-DIA approach allowed the mapping of 83% of proteins within the BCR pathway and key BCR coreceptors, including CD19, CD21, CD22, and CD81 (Supplementary Fig. 22). Compared to a human immune cell proteomics study at a depth of >10,000 proteins using 28 primary hematopoietic cell populations by Rieckmann et al., 93% of the 4211 proteins identified from MEC-1 cells were in common, with an additional 309 proteins uniquely identified in this study, including a comparable number of key B cell surface receptors, such as CD19, CD21, CD22, CD81, FcgRIIB, and Igβ41. We further compared our data with B cell leukemia cells from Johnston et al.42. Even though the MEC-1 cell line also belongs to human B cell leukemia, only 51% of the protein groups commonly overlapped between the two datasets (Supplementary Table 4 and Supplementary Fig. 23). The common or unique proteins are likely due to cell type-dependent protein expression according to the Human Protein Atlas database (https://www.proteinatlas.org/). Nevertheless, B cell markers (e.g., CD19, CD20, and CD22) were commonly detected in both datasets and not detected in our PC-9 data. Similarly, lung cancer markers (e.g., EGFR and TP53) were not detected in MEC-1 cells (Supplementary Table 4). This supports the notion of cell-dependent specificity of protein expression. Taken together, these results demonstrate the versatility of this approach for in-depth proteomic characterization of distinct cell types, thereby paving the way toward quantitative exploration of dissecting cellular heterogeneity in complex systems such as the tumor microenvironment.
Enhanced sensitivity by single-cell integrated proteomics chip (SciProChip) and DIA MS
Inspired by promising results in the proteome coverage and quantification of iProChip, we next sought to implement iProChip dedicated to 20-plex processing of single cells, which we termed single-cell iProChip (SciProChip) (Fig. 6a, b). This SciProChip was designed to include 20 chambers, each containing a single-cell trap, to facilitate precise and unattended capture of one cell for proteomic processing (Fig. 6c). The chip showed an improved cell usage efficiency of ~40% for single-cell capture by optimal positioning of the cell traps and the narrower dimension of the chamber (Fig. 6d and “Methods”). Compared to iProChip, the total processing volume for SciProChip was reduced from 312 to 78.5 nL, and the length of the C18-packed column was reduced from 2.5 to 1 cm, both of which helped to reduce sample loss. For compatibility of small cell input, the LC–MS/MS gradient time was reduced to 90 min.
a A photograph of SciProChip for parallel processing of 20 single cells. b A schematic of the overall SciProChip layout. The control layer is shown in pink, while the flow layer is shown in black and blue. Note that SciProChip contains 20 operational units. c A bright-field image of a capturing chamber with a trapped cell. Scale bar: 120 μm. d The cell usage efficiency of SciProChip. Data are presented as mean values ± SD (n = 28 independent measurements). e Protein group identification of 10 single cells from 2 independent experimental batches. SC: single cell. f Overlaps of protein groups identified in replicate analyses and g distributions of total protein abundances in identified proteins of single cells shown in e. Each data was analyzed as an individual measurement. Centerlines show the mean; box limits indicate the 25th and 75th percentiles; whiskers, 1.5× interquartile range. h Pairwise Pearson’s correlation showing the reproducibility of protein abundances obtained across different single cells. i Correlation between the estimated cell sizes and protein groups identified of 20 single cells. Source data are provided as a Source data file.
Using SciProChip, we next performed proteome profiling of a series of single cells acquired from two batches of cell cultures using two chips (Supplementary Data 3). For the two batches, each containing 10 single-cell profiles, similar depths of an average of ~1500 ± 131 protein groups were reproducibly identified across 20 single cells. Compared to single-cell profiling by iProChip, significantly improved (1.53-fold) proteomic coverage was obtained by SciProChip (Fig. 6e). The accumulated number of all 10 single-cell datasets yielded a total identification of 1,792 and 1,995 protein groups from both batches of single cells, respectively. Moreover, the identified protein groups achieved good overlap (81%-86%) between replicates of single-cell runs, indicating high reproducibility in the SciProChip-DIA workflow (Fig. 6f). Next, we evaluated the quantitation performance of single-cell replicates by calculating the overall protein abundances quantified in every single cell, which showed a consistent log abundance distribution across all datasets (Fig. 6g). The analytical reproducibility was assessed by computing the pairwise correlation of protein abundances between replicates, which showed a good Pearson correlation (Fig. 6h). We further explored the potential correlation between cell size and identified protein groups at the single-cell level. Intriguingly, the results showed that the number of identified proteins did not seem to correlate with the approximate size of individual captured cells (Fig. 6i). In summary, compared to iProChip, SciProChip demonstrates higher cell usage efficiency and significantly improved proteomic coverage while maintaining good reproducibility for single-cell proteomic profiling.
