Multicolumn Nanoflow Liquid Chromatography with Accelerated Offline Gradient Generation for Robust and Sensitive Single-Cell Proteome Profiling

Abstract

Peptide separations that combine high sensitivity, robustness, peak capacity, and throughput are essential for extending bottom-up proteomics to smaller samples including single cells. To this end, we have developed a multicolumn nanoLC system with offline gradient generation. One binary pump generates gradients in an accelerated fashion to support multiple analytical columns, and a single trap column interfaces with all analytical columns to reduce required maintenance and simplify troubleshooting. A high degree of parallelization is possible, as one sample undergoes separation while the next sample plus its corresponding mobile phase gradient are transferred into the storage loop and a third sample is loaded into a sample loop. Selective offline elution from the trap column into the sample loop prevents salts and hydrophobic species from entering the analytical column, thus greatly enhancing column lifetime and system robustness. With this design, samples can be analyzed as fast as every 20 min at a flow rate of just 40 nL/min with close to 100% MS utilization time and continuously for as long as several months without column replacement. We utilized the system to analyze the proteomes of single cells from a multiple myeloma cell line upon treatment with the immunomodulatory imide drug lenalidomide.

Graphical Abstract

INTRODUCTION

Mass spectrometry (MS)-based proteomics has recently become compatible with much smaller samples,^1–3 enabling in-depth proteome profiling of single mammalian cells^4–8 and unbiased spatial proteomics at or near single-cell resolution.^9–11,24 In addition to efficient sample preparation, MS acquisition and data analysis, peptide separations play a critical role in enabling low-input bottom-up proteome profiling.^8,12–16 These separations contribute substantially to the overall sensitivity, throughput, and robustness of a proteomic analysis, yet it is difficult to optimize all aspects of the separation simultaneously. For example, liquid chromatography (LC) separations performed at very low flow rates (e.g., ≤50 nL/min) provide dramatically increased ionization efficiencies at the electrospray source by producing smaller charged droplets that are more readily desolvated, thus enhancing overall sensitivity.^17–22 However, such separations operate below the flow specifications of commercial binary pumps, requiring a flow splitter and other accommodations that can compromise robustness. Moreover, many low-input proteomics workflows cannot accommodate offline sample cleanup due to inevitable analyte losses, which poses an additional challenge to the robustness of ultralow flow LC.^9,25 As such, most experimental setups compromise between sensitivity and robustness, using commercial columns and operating at standard ~100–300 nL/min flow rates, even at the expense of reduced sensitivity and proteome coverage.²⁶

Achieving high measurement throughput is also critical for single-cell proteomics, as resolving heterogeneity at the single-cell level generally requires more samples to be analyzed than for standard bulk-scale experiments. Proteomics instrumentation is expensive, so simply scaling up the number of LC–MS systems is not feasible for most laboratories. For example, nanoflow LC systems cost ~$100k, and high-resolution/mass-accuracy mass spectrometers can cost nearly $2 M and depreciate at a rate of hundreds of dollars per day in addition to their substantial operating costs. As such, the overall cost per sample is largely driven by the number of samples that can be analyzed per unit time. LC separations impact measurement throughput through both their active gradient time during which peptides elute, as well as the time during which sample loading, column regeneration, column washing, and mobile phase transport through the system take place. These inactive “overhead” times may comprise a greater portion of the overall LC cycle time when employing faster separations and lower flow rates, as the fixed volumes of the LC system take longer to clear at low flow rates. As with trade-offs between sensitivity and robustness, these factors can lead to compromises between sensitivity and measurement throughput.

It should be possible to have compromise-free separations that simultaneously achieve high sensitivity, robustness, throughput, and peak capacity. One promising solution is to develop multicolumn nanoLC systems to reduce or eliminate mass spectrometer idle time even with rapid, low-flow separations.^27–30 These multiplexed systems enable an active column to elute peptides to the mass spectrometer, while one or more additional columns cycle through their overhead steps. The number of columns can be matched to the duty cycle of the nanoLC system. For example, Livesay et al. produced a 4-column system that could achieve 100% duty cycles even when that of an individual column was <50%.³¹ However, this and other designs can lead to significant compromise in terms of robustness. In the case of the above-mentioned 4-column system, 11 individually operated valves were required. More recently, we developed a 2-column nanoLC system for rapid single-cell proteome profiling.³² The system was demonstrated with overall LC cycle times in the range of 7–30 min with corresponding duty cycles ranging from 58 to 83%. As with other systems, this also included compromise. Short analytical columns were used, reducing peak capacity, and the flow rate was ~80 nL/min, reducing sensitivity relative to the sub-50 nL/min flow rates typically used in our single-cell proteomics analyses.²³

Here, to achieve a more favorable combination of throughput, sensitivity, robustness, and peak capacity, we have developed an improved multicolumn nanoLC system with accelerated offline gradient generation. Offline mobile phase gradient generation and storage were originally developed by Jorgenson et al.^33,34 and more recently incorporated into the Evosep platform.³⁵ Our design requires no flow splitting and only one valve for autosampling plus one additional valve per column. A single binary pump generates gradients in an accelerated manner to support multiple analytical columns, and one trap column interfaces with all analytical columns to reduce required maintenance and simplify troubleshooting. A high degree of parallelization is possible, as one sample undergoes separation while the next sample plus its corresponding mobile phase gradient is transferred into the storage loop and a third sample is loaded into a sample loop. Selective offline elution^36,37 from the μSPE column into the sample loop prevents salts and hydrophobic species from entering the analytical column, thus greatly enhancing column lifetime and system robustness. The design enables rapid, low-flow sample analysis, and months long operation without requiring column replacement. After characterizing accelerated gradient generation with 30–60 min gradients and a 2-column system, a 3-column system was used to analyze the proteomes of single cells from a multiple myeloma cell line upon treatment with the immunomodulatory imide drug lenalidomide.

EXPERIMENTAL SECTION

Sample Preparation.

HeLa protein digest standard and formic acid were purchased from Thermo Fisher Scientific (Waltham, MA). LC–MS grade water and acetonitrile from Honeywell (Charlotte, NC) were used to prepare mobile phase gradient solutions A (0.1% formic acid in water) and B (0.1% formic acid in acetonitrile), respectively.

Multiple Myeloma MM.1S cells (ATCC, Manassas, VA) were cultured and harvested as recommended by ATCC. Approximately 4 × 10⁶ MM.1S cells were seeded into 10 mL of Roswell Park Memorial Institute (RPMI) media 24 h before treatment. The cells were incubated in the presence of the immunomodulatory imide drug lenalidomide. The final concentration of lenalidomide in culture was 2 μM. After reaching the specific incubation time (0, 6, 48 h), the treatment was quenched by spinning down and removing the lenalidomide-containing media. MM.1S cells were then filtered and washed 3× with phosphate-buffered saline (PBS) and had viability >90% before sorting into the 384-well PCR plate using the cellenONE X1 (Cellenion, Lyon, France). A one-step sample preparation protocol was followed in which 500 nL of 10 pg/nL Rapid Trypsin/Lys-C digestion mix (Promega) was dispensed to the isolated cell in the 384-well PCR plate using the Tecan Uno Single Cell Dispenser (Tecan, Männedorf, Switzerland) followed by a 1 h incubation at 70 °C as described previously.²⁶ Alkylation and reduction steps were omitted. The well plate was covered with sealing foil and stored at −20 °C prior to MS analysis.

Separations.

Analytical and trap columns were prepared inhouse as described previously²³ using Dr. Maisch (Ammerbuch, Germany) ReproSil-Pur C18 media having 1.9 μm diameter and 120 Å pore size. Columns were packed in 30 μm-i.d. × 30 cm long fused silica capillaries (Polymicro, Phoenix, AZ). Trap columns were 50 μm-i.d. × 5 cm long. Both columns were fritted using the Kasil Frit Kit (Next Advance, Troy, NY). Trap columns were fritted on both ends to enable bidirectional flow. To complete the connection of the short trap column between two ports on the same valve, one end of the trap column was connected to a 10 cm length of 20 μm-i.d. fused silica tubing. Capillary ends were polished using the Capillary Polishing Station (ESI Source Solutions, Woburn, MA). A 10 μm-i.d. chemically etched nanoelectrospray emitter from MicrOmics Technologies (Spanish Fork, UT) was connected to each analytical column via a PicoClear union (New Objective, Woburn, MA). A 10-port Nanovolume valve (Part no. C72MFSX-4670D, VICI, Houston, TX) replaced the standard autosampler valve in the UltiMate 3000 nanoLC system (Thermo Fisher) to introduce sample as described previously.³⁸ The 6-port Nanovolume storage loop valves described below were controlled by the valve actuators in the UltiMate 3000 column oven. Gradient generation was accomplished with the standard binary pump of the UltiMate 3000 system. Single-cell samples were analyzed directly from the 384-well PCR plate using the UltiMate 3000 autosampler.²⁶

For comparison with accelerated gradient generation, samples were analyzed by nanoLC using a standard trap-and-elute setup.³⁸ Briefly, samples were loaded onto the trap column at a flow rate of ~0.3 μL/min for 10 min using 1% B before the valve was switched to elute the sample from the trap onto the analytical column. The flow rate through the analytical column was ~40 nL/min. The % B was increased from 1 to 2% in 1 min, 2 to 8% in 5 min, 8 to 15% in 15 min, 15 to 20% in 9 min, 20 to 25% in 6 min, and 25 to 45% in 10 min. For column washing and regeneration, solution B was increased from 45 to 80% in 5 min, stepped to 90% for 5 min, and then decreased to 1% and held for 25 min.

For accelerated gradient generation (Figure 1), samples were first loaded onto the trap column at 1 μL/min for 5 min and then eluted from the trap column using a binary pump at a flow rate that was 2–8× greater than the 40 nL/min flow rate of the analytical column. The ~3 μL of eluent, including the sample, mobile phase gradient, and column wash, was stored inside a 570 cm long ×30 μm i.d. fused silica capillary (storage loop). After switching the valve position of the 6-port valve that was connected to the storage loop, an additional UltiMate 3000 pump containing solvent A (0.1% formic acid in water) served as an isocratic pump to push the contents of the storage loop through the 30 cm long, 30 μm i.d. analytical column at a pressure of ~410 bar. Constant-pressure elution was approximated by operating the pump at a total flow rate of 1 μL/min and splitting the flow between the analytical columns and a 400 cm long fused silica capillary that had an inner diameter of 15 μm. This split column helped to minimize pressure variations that would otherwise occur when the mobile phase gradient passed through the analytical columns. The effluent from the split column is pure solvent A and can thus be recycled into the corresponding solvent reservoir if desired.

Figure 1.

Diagram of a 2-column nanoLC system with accelerated offline gradient generation. (A) Sample 3 (red) is drawn into the sample loop, while the binary pump elutes Sample 2 (green) embedded in its mobile phase gradient from the trap column into the top storage loop. An isocratic pump filled with aqueous solution concurrently pushes Sample 1 (blue) and its corresponding gradient through the lower analytical column for peptide separation, while the upper analytical column is regenerated. (B) All samples advance one position, while Sample 4 (purple) is introduced into the sample loop and peptide elution for Sample 2 takes place on the upper analytical column. (C–E) Autosampler subroutines. (C) Binary pump elutes the “green” sample from the SPE column plus the mobile phase gradient to a storage loop, while the autosampler syringe transfers the “red” sample plus any required rinses⁹ of “red” well or vial to the sample loop. (D) Valve is switched and the green sample plus the mobile phase gradient finish transferring to the storage loop, while the red sample is transferred from the sample loop and concentrated on the trap column. At the same time, the autosampler needle is rinsed and refilled in preparation for the next sample. (E) Valve is switched back to its original position to elute the red sample to a storage loop, while the subsequent purple sample is loaded to the sample loop.

Electrospray Ionization.

As described previously,³² chemically etched nanoelectrospray emitters were attached to each column via PicoClear Unions (New Objective) and positioned in front of the mass spectrometer inlet using a custom 3D-printed adapter. An image of the adapter is shown in Figure 2A. The 2.2 kV electrospray potential was alternately applied upstream of each analytical column at a Nanovolume union (VICI, Houston, TX). A photograph of both emitters positioned in front of the mass spectrometer is shown in Figure 2B. The inactive emitter accumulated a solvent droplet, while peptides were electrosprayed from the active emitter.

Figure 2.

Interfacing columns with the mass spectrometer inlet. (A) Custom 3D-printed adapter that enables electrospray emitters corresponding to multiple analytical columns to be accurately positioned. (B) Photograph of two emitters positioned in front of the mass spectrometer inlet. A droplet accumulates on the inactive emitter (bottom), while electrospray potential is applied to the active emitter (top).

MS Acquisition.

The LC column/emitter assembly was interfaced with an Orbitrap Exploris 480 mass spectrometer (Thermo Fisher) via the Nanospray Flex Ion Source. The temperature of the ion transfer tube was 200 °C and the RF lens setting was 50%. For MS1, the Orbitrap resolution was 120,000 (m/z 200) with the normalized AGC target set to 300%. The scan range was from 375 to 1575 Th, and the maximum injection time was set to 118 ms. To trigger MS2 for all experiments, the precursor intensity threshold was set to 5.0 × 10³ and the charge state was 2–4. Dynamic exclusion was set to 20 s for 20 min LC separations and to 25 s for 40 min separations. The cycle time was 1.5 s. For standard data-dependent acquisition (DDA), the isolation window was 1.6 Th, the HCD collision energy was 30%, and the MS2 resolution was, respectively, set to 15,000, 30,000, and 60,000 for 10, 2, and 0.4 ng HeLa protein digest standards. The maximum MS2 injection times for 10, 2, and 0.4 ng samples were 22, 54, and 118 ms, respectively, and the AGC target was 200%. Single cells were analyzed using wide-window data-dependent acquisition (WWA).¹⁶ The same resolution and maximum injection time were used as mentioned above for 0.4 ng HeLa digest standards, but the MS2 AGC target and isolation width were increased to 1000% and 8 Th as described in a previous study.¹⁶

Data Analysis.

For DDA and WWA experiments, raw files were searched using Proteome Discoverer (PD) (Thermo Fisher Scientific, Version 3.0.1.13) with the CHIMERYS identification node. Database searches included human (Uniprot, downloaded on 2022-8-11) as well as common contaminants (PD, 2018-10-26). The enzyme was set as trypsin with a maximum of two missed cleavages. Other parameters included peptide lengths of 7–30 amino acids, a maximum of 3 modifications, and charge states from +2 to +4. Fragment mass tolerance was 20 ppm. Carbamidomethyl (C) was fixed as a static modification in the CHIMERYS software. Results were filtered using the Percolator node with 1% FDR. For match between runs (MBR), the retention time tolerance was set at 0.25 min, and the mass tolerance was set at 5 ppm.

For coefficient of variation (CV) calculations, Pearson r calculations, and quantifiable proteins, all protein data were imported into the Python programming language from the PD “…_Proteins.txt” files. Medium and low confidence proteins were removed, as were contaminants. Next, these proteins were sorted according to gradient lengths and sample size, normalized to the median in the separate groups, and filtered for <33% missing values. For principal component analysis (PCA), K-Nearest Neighbors was used to impute missing values with a k = 5. Analysis of differential protein expression upon treatment of MM.1S cells with lenalidomide was considered significant for proteins with a fold change > 2 and a Welch’s t-test p-value < 0.05 after adjusting for multiple comparisons with Benjamini Hochberg FDR.

RESULTS AND DISCUSSION

Offline gradient generation and storage followed by elution through the analytical column with an isocratic pump was first developed by Jorgensen et al.^33,34 and has since been successfully incorporated into a commercial system by Evosep.³⁵ Here, we adapt this general scheme for use in a multicolumn nanoLC system to achieve a high duty cycle for low-flow separations. Figure 1 shows a schematic for a 2-column system comprising a single autosampling valve with a trap column and an additional storage loop-containing valve for each analytical column. Note that even when only two columns are employed, three samples simultaneously pass through the system at different stages. For example, as shown in Figure 1A, Sample 1 (blue) elutes through the analytical column to the mass spectrometer, while Sample 2 (green) and its corresponding gradient fill a gradient storage loop and Sample 3 (red) is loaded from a microwell or nanowell plate into the sample loop. After eluting Sample 1 (Figure 1B), each sample advances one position, and Sample 4 (purple) is loaded into the sample loop. A single binary pump operates at low pressure to generate the gradient and rapidly push the gradient through the sample-containing trap column into the storage loop such that the sample is entrained in the gradient and both sample and gradient are stored in the loop. A high-pressure isocratic pump delivers aqueous solution (mobile phase component A) to all columns simultaneously, bypassing the storage loop for the inactive column(s) to accomplish column regeneration and driving the sample + gradient from the storage loop through the active column and to the electrospray source at ~40 nL/min. Autosampling for all columns is accomplished with a single 10-port/2-position injection valve, and a subroutine takes place in this autosampler valve to rinse the sampling needle, etc., as shown in Figure 1C–E.

The system is modular such that adding a third or fourth column (e.g., for use when the duty cycle of a corresponding single-column system would be <50%) requires only one extra valve/storage loop per additional column as shown for a 3-column system in Figure S1. To use a single binary pump to serve multiple columns, the gradients need to be generated faster than elution takes place. For example, provided that a storage loop can be filled with sample and mobile phase ~4× faster than the analytical column elutes peptides, a 4-column system is possible. This has the added benefit of operating the gradient-generating binary pump at a higher flow rate and lower pressure than would be required for direct online coupling to the analytical column. For example, for an analytical separation taking place at 40 nL/min, the binary pump can operate at 160 nL/min and at reduced pressure since only open tubing and the short trap column are part of the flow path. This setup avoids the need for flow splitting to achieve ultralow flow rates through the analytical column, as no binary pumps are currently specified to operate at <50 nL/min.

To evaluate the impact of higher-flow-rate gradient generation on chromatographic performance, we analyzed 2 ng aliquots of bulk-prepared HeLa digest with online coupling of the binary pump to the analytical column and with offline gradient storage. In both cases, the analytical separation took place over 60 min at 40 nL/min, but the gradient was generated in just 15 min for the offline coupling. As shown in Figure 3, performance was very similar between the two gradient elution strategies in terms of elution times, peak widths, and proteome coverage. We additionally compared three different gradient generation speeds: 2× (generated at 60 nL/min in 30 min), 4× (generated at 120 nL/min in 15 min), and 8× (generated at 240 nL/min in 7.5 min). Elution through the analytical column was set to be the same in all cases. As shown in Figure S2, all provided similar chromatographic performance.

Figure 3.

Comparison of accelerated offline gradient generation and conventional online gradient generation for the analysis of 2 ng HeLa digest using 60 min gradients at 40 nL/min. (A) Representative base-peak chromatograms. The intensity scale is the same for both separations. (B) Chromatographic peak widths and (C) MS2-based proteome coverage (n = 3).

It is somewhat surprising that peak widths are unaffected by Taylor dispersion while passing through the large distance from the trap to the analytical column. Indeed, when eluting through the 5 cm long trap column in forward flush mode, significant broadening occurs that is not observed when using standard backflushing through the trap (Figure S3). In backflush mode, it appears that the sample was eluted from the trap at a lower solvent strength than is required for elution through the analytical column, such that peaks refocus at head of the analytical column prior to eluting.³⁹ When utilizing a trap-and-elute strategy, it is therefore beneficial to operate it in backflush mode or alternatively to dilute the gradient after elution from the trap as with the Evosep platform.³⁵

In contrast to our previous multicolumn nanoLC system that had a separate trap column for each analytical column,³² here a single trap column interfaces with all analytical columns. This enhances reproducibility by removing a potential variable, and also facilitates troubleshooting. For example, if performance degrades across all analytical columns, this points to an issue with the trap column. It was more challenging to differentiate problems with the trap or analytical columns in previous work.³² However, even with accelerated elution from a single trap column, such a setup may limit throughput in some cases, as elution of one sample from the trap column and loading of another sample to the trap column cannot take place simultaneously. For such instances, e.g., when employing a very large sample loop, two trap columns may be necessary.

Another benefit of eluting samples from a trap column to a storage loop rather than directly to an analytical column is that analytes and the required mobile phase gradient can be selectively sent to the storage loop, while undesirable sample components such as salts, insoluble debris, lipids, and the nonionic surfactant n-dodecyl-beta-maltoside (DDM) used in SCP sample preparation are directed to waste. Since such nonpeptide species never enter the analytical column or mass spectrometer, system robustness is greatly enhanced. Selective elution from the trap column is accomplished as shown in Figure S4. Briefly, once a sample is loaded onto the trap column, the trap column is backflushed with a low%B such that remaining salts and any insoluble debris are dislodged and directed to waste as shown in Figure S4A. Gradient elution of peptides through the trap then takes place, and the storage loop is filled (Figure S4B). The autosampler valve is then switched to bypass the trap column. This enables a high %B to enter the storage loop to wash the analytical column without dislodging hydrophobic species from the trap column (Figure S4C). It should be noted that during this step, the sample loop is disconnected from the autosampler needle, so care should be taken to avoid a conflict with autosampler timing. Finally, the autosampler valve is switched to wash and re-equilibrate the trap column, with the resulting eluent being directed to waste (Figure S4D). To illustrate selective sample elution, we analyzed 2 ng HeLa digest samples and adjusted the conditions of elution from the trap column to the storage loop. In the bottom chromatogram of Figure 4, a full eluent collection from 2 to 75% B was introduced to the storage loop. All peaks, including the late-eluting surfactant DDM, passed through the analytical column and into the mass spectrometer. In the middle chromatogram of Figure 4, the trap column was bypassed after reaching 35% B, so the DDM peak disappears, indicating that it was not introduced to the analytical column. Finally, in the top chromatogram of Figure 4, the trap column was also bypassed early in the gradient (to ~10% B), which resulted in only the late-eluting peptides being introduced to the analytical column. This demonstrates the selective elution from the trap to the storage loop to control which species pass to the analytical column. Using the selective elution strategy in which 2–35% B passes through the trap column while the rest of the gradient bypasses the trap (middle trace in Figure 4), we have successfully analyzed >3000 samples without column replacement.

Figure 4.

Selective eluent collection from the trap column into the storage loop. Representative base-peak chromatograms are shown for the analysis of 2 ng HeLa digest using the same gradient conditions as in Figure 3. The red box highlights a region with a similar elution profile across different conditions. The DDM peak is only present in the bottom chromatogram but overlaps with the chromatogram above it.

An LC system containing a single trap column and two analytical columns was constructed, as shown in Figure 1, and first evaluated using 60 min gradients. The overall cycle time was 65 min (92% duty cycle). As shown in Figure 5A, chromatographic profiles were similar for both analytical columns when analyzing 2 ng aliquots of HeLa digest standard. We then evaluated proteome coverage across 142 replicate analyses of 0.4 ng HeLa digest to evaluate system stability, proteome coverage, and potential column carryover. During the experiment, six analyses were inadvertently performed without alternating columns, which left a total of 136 successful analyses. Figure 5B shows stable proteome coverage across the analyses, averaging 1429 protein groups without match between runs (MBR) and 3529 protein groups with MBR when analyzing samples with a 10 ng library of the same digest, although one column appears to provide slightly greater proteome coverage than the other. While the relative standard deviation in proteome coverage across both columns of 3.3 and 2.1% is quite small, it may be desirable to further reduce technical variability when using a multicolumn nanoLC system by running all samples from one study on one column and all samples from another study on the other.

Figure 5.

Sequential separations of 0.4 ng HeLa digest using 2-column nanoLC with the same separation conditions as those used in Figure 2. (A) Alternate elution from each of the two columns as shown by red and blue chromatograms. (B) Proteome coverage shown with and without MBR for 136 consecutive analyses.

The LC columns were unheated in this study, so there was some retention time variation due to room-temperature fluctuations, as shown in Figure S5. The standard deviation of retention time averaged 0.80 min for these representative peptides. This variability necessitates that a time buffer be included in the retention time window; timing should be more precise when the effects of room-temperature fluctuations are minimized through column heating to minimize viscosity changes in the mobile phase.

The multicolumn nanoLC system can provide additional time for column washing and regeneration while peptides elute from the other column, which is expected to minimize column carryover. We used the same 5 peptides that were tracked in Figure S5 to quantify column carryover between runs. After running repeated blanks to thoroughly clean the column, we measured 0.4 ng and then 10 ng HeLa digest samples. A blank run from the same column as the 10 ng sample then followed. As shown in Figure S6, the peak intensities from extracted ion chromatograms for the 5 peptides were just 0.15% of those from the 10 ng sample, indicating a lower degree of carryover than the typical ~1% often observed for nanoLC. As such, the system can minimize carryover while maintaining a nearly 100% duty cycle.

In addition to the above 2-column system operated at 40 nL/min with 60 min gradients, we also implemented a 3-column system with 20 min separations at the same flow rate of 40 nL/min. While 20 min separations could also be implemented with a 2-column system, potentially using two separate trap columns, we used a 3-column system with a single trap column in the present study, which afforded additional time for washing and regenerating each column. Adding a third column simply involves daisy chaining another storage loop-containing 6-port valve, as shown in Figure S1.

We first analyzed 10 ng HeLa digest samples using WWA¹⁶ with the 20 min, 3-column system and identified an average of 3229, 3503, and 3517 unique proteins on each of the three columns using MS2-based identification (i.e., without MBR). We then utilized the 3-column system to study proteome changes in a multiple myeloma (MM) cell line (MM1S) upon treatment with the immunomodulatory (IMiD) drug lenalidomide. MM is the second most prevalent hematologic malignancy, and it affects more than 20,000 people in the United States each year. Despite new therapies becoming available for MM in recent years, including IMiDs,^40,41 the emergence of multidrug-resistant clones following treatment remains a significant hurdle.⁴² A pivotal aspect of drug development is the identification of the specific protein profiles characterizing these resistant clones. Proteomics technologies provide a promising avenue for pinpointing effector proteins that directly influence the biological processes leading to the emergence of resistant subclones. When considering the unique challenges associated with this hematological malignancy, single-cell proteomics (SCP) offers a distinct advantage by enabling the resolution of cell heterogeneity and providing an in-depth, unbiased profile of protein expression within individual cells.² As previously reported, lenalidomide (marketed as Revlimid) is an immunomodulatory drug that has undergone rapid clinical development in multiple myeloma and was approved by the US FDA for use in select patients.^43,44

We evaluated the ability of our platform to differentiate closely related cells by measuring changes in the proteome of individual MM.1S cells upon treatment with 2 μM lenalidomide for varying durations: 0 h (i.e., cells were washed immediately following treatment), 6 and 48 h. Subsequently, individual cells were isolated using the cellenONE, prepared as described previously,²⁶ and analyzed using the 20 min, 3-channel LC platform. As shown in Figure 6A, all three treatment groups showed similar proteome coverage of approximately 1300 protein groups per cell. Both the volcano plot and PCA plot (Figure 6B,C, respectively) clearly illustrate that MM cells exhibit distinct protein expression profiles when subjected to varying treatment durations. Indeed, a total of 169 proteins were differentially expressed between MM.1S cells treated with lenalidomide for 48 and 0 h (control), with 113 proteins upregulated following 48 h treatment and 56 proteins downregulated (see also Table S1). Using the DAVID⁴⁵ bioinformatics resource system for gene ontology analysis, we found that the molecular functions specific to the upregulated proteins following extended lenalidomide treatment were DNA-binding, lyase, oxidoreductase, and ribonucleoprotein. On the other hand, downregulated proteins were specific to helicase, ligase, and rRNA-binding groups. Cancer cells are known to manipulate metabolism for their survival,⁴⁶ and in this case, lyase and oxidoreductase, being metabolic enzymes, were upregulated in the 48 h treatment group. While preliminary, this pilot study provides insights into adaptations following treatment with lenalidomide at single-cell resolution.

Figure 6.

(A) Number of high-confidence master proteins identified following the treatment of MM-1S cells with lenalidomide. (B) Volcano plot indicating significantly up and downregulated proteins following 48 h treatment (|Fold change| > 2 and adjusted p-value ≤ 0.01). (C) PCA plot of the three treatment groups.

CONCLUSIONS

Here, we reported an LC workflow that uses accelerated offline gradient generation, multiple storage loops, and constant-pressure elution to achieve the goal of robust, sensitive, and high-throughput LC–MS analysis. This workflow is particularly valuable for low-input LC–MS processes in which achieving ultralow flow rates (below 100 nL) is imperative to meet the sensitivity demands for detecting the analytes. It effectively addresses the challenges associated with low throughput and robustness that typically accompany low-flow LC systems.

We were able to run two or three columns with one set of gradient pumps and selectively delivered only the desired analytes to the separation columns. Through repeated analysis of single-cell-sized aliquots of HeLa digest, we verified the repeatability, low carryover, and robustness of our system. Finally, through the utilization of a three-column system and a 20 min gradient, we successfully differentiated lenalidomide treatment groups in MM.1S cells. Remarkably, this sensitivity was accomplished while preserving both the high throughput and robustness that are typically challenging to attain.