Metabolomic, Proteomic, and Single-Cell Proteomic Analysis of Cancer Cells Treated with the KRAS^G12D Inhibitor MRTX1133

Abstract

Mutations in KRAS are common drivers of human cancers and are often those with the poorest overall prognosis for patients. A recently developed compound, MRTX1133, has shown promise in inhibiting the activity of KRAS^G12D mutant proteins, which is one of the main drivers of pancreatic cancer. To better understand the mechanism of action of this compound, I performed both proteomics and metabolomics on four KRAS^G12D mutant pancreatic cancer cell lines. To obtain increased granularity in the proteomic observations, single-cell proteomics was successfully performed on two of these lines. Following quality filtering, a total of 1498 single cells were analyzed. From these cells, 3140 total proteins were identified with approximately 953 proteins quantified per cell. At 48 h of treatment, two distinct populations of cells can be observed based on the level of effectiveness of the drug in decreasing the total abundance of the KRAS protein in each respective cell, with results that are effectively masked in the bulk cell analysis. All mass spectrometry data and processed results are publicly available at www.massive.ucsd.edu at accessions PXD039597, PXD039601, and PXD039600.

Introduction

KRAS mutations are found in up to 25% of solid tumors and typically those with the worst prognosis.^1,2 Pancreatic cancer is the best example of this, as a cancer with a dismal survival rate of only 44% after 5 years. Mutation in KRAS, and specifically the KRAS^G12D mutation, is particularly prevalent in pancreatic cancer. A recently described compound, MRTX1133, builds on recent successes with KRAS small-molecule inhibitors as the first inhibitor described for the KRAS^G12D proteoform.³ MRTX1133 differs from successful KRAS^G12C inhibitors in that it is purported to bind through noncovalent mechanisms.^4,5 To better evaluate the mechanism of action of this exciting new inhibitor, four human pancreatic cancer cell lines were treated with the drug for 48 h and standard label-free proteomics and global untargeted metabolomics were performed.

Work in our laboratory recently described the use of single-cell proteomics (SCP) to better understand the heterogeneity of response in the proteomes of cells treated with sotorasib, a KRAS^G12C inhibitor. While more of a proof of concept, the work led to the development of tools for multiplexed SCP using a TIMSTOF instrument. Multiple limitations existed in the study, including the relatively small number of cells that were analyzed. We were, however, able to recapitulate many of the findings of both bulk proteomic and single-cell transcriptomic studies of drug activity, due largely to the profound effects sotorasib had on the cell cycle. Despite limitations, we were able to highlight new findings that were not previously revealed by other approaches.⁶ Following optimization to alleviate bottlenecks on the sample preparation and data analysis side the resulting methods were applied to two cancer cell lines treated with MRTX1133.

Methods

Cell Culture and Drug Treatment

All cell lines were obtained from ATCC between March and December 2022 and were cultured according to vendor instructions. PANC 0813, PANC 0304, and PANC 0203 were grown in RPMI 1640 (ATCC 30-2001) supplemented with 15% fetal bovine serum (ATCC 30-2020) and 10 units of human insulin (Fisher). ASCP-1 was grown in RPMI 1640 with 10% FBS. All cell culture media contained 10 mg/mL Penn Strep antibiotic solution (ATCC 30-2300). All cell lines were passaged a minimum of 3 times prior to treatment with 10 nm MRTX1133 for 48 h. Cells were harvested by vacuum aspiration of cell culture media. The adherent cells were briefly rinsed in 3 mL of a 0.05% trypsin plus EDTA solution (ATCC 30-2001). This solution was rapidly aspirated off and replaced with 3 mL of the same solution. The cells were examined by light field microscopy and incubated at 37 °C with multiple examinations until the adherent cells had lifted off the plate surface. The active trypsin was then quenched by the addition of 7 mL of the original culture media. The 10 mL solution was transferred to sterile 15 mL Falcon tubes (Fisher) and centrifuged at 300g for 3 min to pellet the cells. The supernatant was gently aspirated off, and the cells were resuspended in PBS solution without calcium or magnesium with 0.1% BSA (both, Fisher Scientific) at 1 million cells per mL as estimated by bright field microscopy. Approximately 2 million cells were taken for bulk proteomic and metabolomic analysis. Cells for single-cell aliquoting were gently dissociated from clumps by slowly pipetting a solution of approximately 1 million cells through a Falcon cell strainer (Fisher, 353420), and the cells were placed on wet ice and immediately transported across the street to the JHU Public Health sorting core. Nonviable cells were labeled with a propidium iodide solution provided by the core facility and briefly vortexed prior to cell isolation and aliquoting.

Global Metabolomics

A solution containing approximately one million cells from each condition was centrifuged at 13,000g for 15 s at speed to obtain a solid pellet. The supernatant was aspirated off and replaced with 300 μL of 70% liquid chromatography mass spectrometry grade LCMS methanol in LCMS water solution. Pellets were resuspended with vigorous vortexing and placed at −80 °C for approximately 48 h. Following storage, the cells and methanol solution were thawed on wet ice and resuspended with vigorous vortexing, followed by centrifugation at 13,000g for 5 min to precipitate proteins and cellular debris. The top 250 μL of supernatant was removed to limit the disturbance of the lower material and was dried to completeness via SpeedVac at 25 °C. The dried metabolite extract was resuspended in 50 μL of 5% LCMS-grade acetonitrile 0.1% formic acid in LCMS-grade water. A pooled sample containing equivalent amounts of all control and treated metabolite extracts was prepared for use as the reference sample for chromatographic alignment. Two microliters of solution were used for each of the four replicate LCMS injections. For LCMS analysis, a Q Exactive “Classic” system coupled to Dionex U3000 UHPLC was used. Separation was performed on a 2.1 mm × 15 cm HyperSil Gold column with 2 μm particle size and a flow rate of 300 μL/min for the active gradient. The LCMS method parameters have been deposited at www.LCMSmethods.org as “Q Exactive Positive Metabolomics” in the 2019 method release. Briefly, MS1 scans were acquired from 90 to 850 m/z at 70,000 resolution, with an AGC target of 3 × 10⁶ charges or a maximum injection time of 50 ms. Data-dependent acquisition with a loop count of 3 was used to fragment ions that were accumulated with a target of 2 × 10⁵ charges or a maximum injection time of 50 ms. A three-step collision energy was used with 10, 20, and 60 CE for each fragment ion. Source conditions were determined through an automated method with Exactive Tune 2.11 SP1 for this flow rate. All resulting output files were processed in Compound Discoverer 3.1 using a combination of 3 human metabolite libraries from mzCloud, ChemSpider, and a local desktop library of 4000 human metabolites provided by Thermo Scientific. Compound identities were prioritized in the case of conflict between databases in the order described above.

Bulk Proteomics

A solution containing approximately one million cells from each condition was rapidly centrifuged at 13,000g for 60 s at speed to obtain a solid pellet. The sorting solution was carefully aspirated off and the cell pellet was resuspended with 200 μL of S-Trap lysis buffer (5% SDS in 100 mM TEAB, ProtiFi). The pellet was suspended with rigorous vortexing prior to being placed in an ultrasonic water bath for 15 min at 37 °C for further cellular lysis. Fifty microliters of each sample were taken for reduction at 95 °C for 5 min in 20 mM DTT followed by benchtop cooling to room temperature and alkylation with 30 mM iodoacetamide for 20 min at room temperature in a light-tight drawer. Pierce “Easy Aliquot” single-use reagents were used for both reduction and alkylation. Another 50 μL of each solution was taken for digestion without reduction and alkylation for the later construction of bulk cell digest carrier channels. Digestion in both cases was performed using an S-Trap mini (ProtiFi) spin column according to vendor instructions with an approximate 200 μg of total protein load and a 2 h digestion at 47 °C. Peptides were eluted following digestion, vacuum-centrifuged to dryness, and resuspended in 0.1% formic acid. Peptides were quantified using a peptide colorimetric assay (Pierce 23290).

All proteomic analyses were performed on a TIMSTOF Flex mass analyzer (Bruker) coupled to an EasyNLC 1200 system (Proxeon) by a 15 cm × 75 μm PepSep column with 1.9 μm Reprosil particles. The data acquisition method employed was a default diaPASEF method called “short gradient diaPASEF” included in TIMS Control 4.0 2023. The chromatography gradient ramped from 8% buffer B (80% acetonitrile in 0.1% formic acid) to 35% B in 25 min with a flow rate of 350 nL/min prior to a rapid increase to 100% B at 500 nL/min by 30 min. Baseline conditions were restored at the beginning of each chromatography gradient by HPLC prior to loading the next sample.

Single-Cell Aliquoting

SCP sample preparation, analysis, and data processing were performed as described previously.⁶ Briefly, single cells were aliquoted using an analog MoFlo sorter into cold 96-well plates containing 2 μL of LCMS-grade acetonitrile. At the completion of each plate aliquot, they were immediately sealed and placed in an insulated box of dry ice with the wells pressed into the material to ensure rapid cooling. After approximately 5 min, each plate was transferred to a cooler of wet ice for transport back to our lab and −80 °C storage. Estimations of cellular viability and cell aliquoting efficiency were provided by the core director who has over 30 years of cell sorting and aliquoting experience and significant experience in the analysis of pancreatic cancer cell lines by flow cytometry.⁷⁻⁹ Digital reports of FACs data were provided by email following the isolation procedure and are available upon request.

Single-Cell Lysis, Digestion, and Combination of Multiplexed Cells

Single cells were placed into storage followed by extensive use of tape to help ensure plate seal integrity during storage. A full protocol describing all steps of cell lysis, digestion, and tandem mass tag (TMT) labeling has been permanently published at Protocols.io and can be accessed at dx.doi.org/10.17504/protocols.io.yxmvm2w16g3p/v1. Cell lysis was further driven to completeness, and acetonitrile was removed by placing each plate directly from cold storage onto a 95 °C hot plate for 90 s. The plates were cooled to room temperature on a benchtop with static-free surface covers. The author performed all aliquots of reagents while standing on a static-free floor mat at all stages of the preparation. Periodically, a Zero Stat “static gun,” clearly labeled in fluorescent pink tape, “NOT A GUN” to help ensure author survival during late night sample preparation, was used to further minimize the negative side effects of static electricity on single cells (Fisher Scientific, NC9663078). The dried lysed cellular lysate was digested using a solution of 5 ng/μL LCMS grade trypsin (Pierce) in 0.1% n-dodecyl-β-maltoside detergent (DDM, Thermo Fisher, 89902) and 50 mM TEAB. One microliter of trypsin solution was used for all single cells and blank wells and four microliters were used for digestion of the carrier channel lanes. The excess trypsin was used as an extension of the “sacrificial carrier” concept recently described in Choi et al.(10) The plates were tightly sealed and incubated for 2 h at 45 °C in an orbital shaker with centrifugation every 30 min to condense evaporated liquid. Following digestion, the plates were cooled to room temperature and centrifuged again to precipitate all liquid solution. TMTPro 18 reagents previously prepared and aliquoted at a concentration of 20 ng/μL were used to label all wells and sacrificial trypsin peptides by adding 1 μL and incubating for 30 min at room temperature with orbital shaking. A solution of 0.5% hydroxylamine in TEAB was used to quench the remaining TMT by adding 0.5 μL to each well with 20 min incubation at room temperature with orbital shaking. Each plate was then dried by SpeedVac at room temperature, which took approximately 5 min for each pair of plates. Ten unit-resolution TMT channels from the TMTPro 18 kits (Thermo Fisher) were used. The method blank was labeled with the 126 reagent in all comparisons and 135n was used for the carrier. To minimize the effects of isotopic impurity carryover, staggered reagents were utilized from the kit, with 127n, 128c, 129n, 130c, 131n, 132c, 133n, and 134c used to label single-cell wells.¹¹ TMT-labeled single cells and carrier channels were combined in a staggered manner to ensure each TMT plex contained a mixture of the control and treated carrier channel digest and both control and treated cells. For example, injection well A1 on the autosampler would contain control cells as 127n, 128c, 129n, and 130c, with labeled treated cells occupying the remainder of the multiplex set. Autosampler well A2 would contain the reversed set, with the first four channels containing treated cells and control cells as the last four channels. The net result is that each well in the autosampler plate contains a carrier channel that is an equal mixture of control and treated cells. In addition, four control cells and four treated cells were measured in each LCMS experiment. By alternating loading of single-cell plates, we can obtain a pseudorandom injection pattern where a different set of single cells from each batch with alternating labels are injected consecutively. The cell identities were deconvoluted during the final data analysis in Proteome Discoverer.

Comparison of FACs Sorted and Bulk Cell Homogenate Carrier Channels

To compare the two competing methods for carrier channel preparation, aliquots from bulk cell lysates that were not reduced and alkylated as mentioned above were labeled with the 135n channel according to the manufacturer’s protocol, with the exception that a 4:1 ratio of label to peptide was utilized. Prelabeling peptide concentrations were used to estimate the postlabeled concentrations. Control and treated lysates from PANC 0813 and ASPC-1, respectively, were combined to create a pooled carrier lysate for each cell line. An estimated 40 ng mixture was used to resuspend each single-cell set using the same combination method described above for the FACs sorted single-cell lysates. The end goal was an equivalent pooled carrier channel, and both control and treated single cells were analyzed in each LCMS injection with the channels flipped in each respective injection.

Single-Cell Data Analysis

As previously described,¹² a single-point recalibration method using the 135n carrier channel signal was utilized to adjust the reporter ion region. This secondary calibration allows tighter mass accuracy tolerances to be used during the final data analysis, resulting in reduced background noise. For all ASPC-1 and PANC 0813 cells to meet the QC requirements for inclusion in this manuscript, the linear calibration shift applied was +0.0141 Da. This was performed using an updated version of the pasefRiQ calibrator with a GUI interface and the ability to batch recalibrate the MGF files. This software is openly available and permanently published at DOI: 10.5281/zenodo.7259511. The recalibrated output files were processed in Proteome Discoverer 2.4SP1 using SequestHT and Percolator. In brief, the MS/MS spectra were binned into 100 Da segments and filtered so that only the top 12 most abundant ions from each bin were retained. The resulting cleaned spectra were searched with a 15 ppm MS1 tolerance and 0.03 Da MS/MS tolerance. TMTPro labels were considered static on the N-terminus and dynamic on lysines to allow for the search for lysine acetylation, methylations, and dimethylation. Methionine oxidation was the only additional dynamic modification. The default cutoffs for Percolator PSM validation and peptide and protein FDR determinations were employed in all analyses. Reporter ions were integrated using a 35 ppm mass tolerance window (17.5 ppm up and down), and quantification values were only used for unique peptides. Multiple consensus workflows were performed on the resulting MSF files to assess the different normalization methods. The final data report for downstream analysis used a sum-based normalization of all PSM signals for TMTPro channels 127–133 as well as the raw non-normalized abundance values. The 134 single-cell channel was discarded in all runs due to apparent inflation from impurities in the 135 reagent channel.

Uniform Manifold Approximation and Projection

The normalized output sheet for all single cells from Proteome Discoverer was converted to .CSV and loaded into Perseus 1.6.17.¹³ Perseus was previously set up to communicate with R 4.1.3 as described previously.¹⁴ All single-cell channels were added into Perseus as “Main” data points, with the Protein Accession and Description added as “Text” variables. All zero data points were replaced with normal distributions with each single cell used as the distribution matrix individually. The cells were then grouped under row categories, describing their individual cell and dose conditions. UMAPs were generated using the default settings for Euclidean distribution with 15 neighbors, 2 components, a random state of 1, and a minimum distance of 0.1. The output was plotted using the multiscatter node in Perseus.

Clustering Drug-Treated Cells by Relative KRAS^G12D Expression

The raw non-normalized abundance values for the KRAS protein for each cell with a measured abundance 3 times the calculated noise level were exported from the Proteome Discoverer output for each control and MRTX1133-treated cell. The output text file was analyzed in GraphPad Prism to determine the median and mean protein abundance. MRTX1133-treated cells with an abundance greater than the mean for all treated cells were manually flagged for clustering. Twenty-five PANC 0813 cells with a level of KRAS expression greater than the treated mean were analyzed in the SimpliFi platform as a single group compared to 107 cells with lower levels of relative expression as the control group. The analysis was performed using default parameters with the exception that missing values were allowed to be up to 50% per protein to account for the missing values inherent in SCP.

Results and Discussion

MRTX1133 Treatment Alters Central Metabolism

Global reversed phase positive metabolomics identified 940 distinct small molecules after the standard adduct reduction processes. Of these, 781 could be assigned a putative molecular formula and 264 could be assigned a putative name within the cutoffs used for this study (Supporting Information File 1). The most striking changes were that drug treatment revealed a universal depletion of ATP precursors in all cell lines following the MRTX1133 treatment. Notably, adenosine (Figure 1A), AMP (Figure 1B), and ADP were observed at reduced relative concentrations. NAD precursors were likewise reduced by drug treatment in all four cell lines. In contrast, many amino acids such as tyrosine and glutamate (Figure 1C) were observed at increased relative concentrations in drug-treated cells. A smaller number of central metabolites appeared to demonstrate mutant copy number-specific effects. PANC 0203 and 0403, which have single copies of the mutant gene, appear to have increased levels of glutamine following treatment, while PANC 0813 and ASPC-1 which have two copies of the protein appear to have decreased levels of this metabolite following treatment (Figure 1D). While this is a small group for comparison, the number of KRAS mutant gene copies has been previously linked to metabolic reprogramming effects.¹⁵

Figure 1.

Representation of global metabolomic alterations following MRTX1133 treatment. (A) Adenosine levels are decreased in all cell lines following treatment. (B) The same trend is observed for AMP. (C) Drug treatment leads to increased intracellular glutamate levels in all cell lines. (D) Glutamine levels following treatment suggest a different effect in the cell lines harboring single or double copies of the mutant gene

Proteomic Bulk Cell Homogenate Data

The bulk cell proteomics data largely produced the expected changes in all cell lines. MRTX1133 acute treatment resulted in a consistent decrease in the abundance of the central MAPK pathway. ERK (MAPK3) was reduced in all cell lines following treatment, as previously reported⁴ as was MAPK1 (Figure S1). SimpliFi pathway analysis identified the Rho GTPase pathway as the seventh most significantly altered pathway between control and treated lines (R-HAS-563220; 2.04 × 10^–60). Similarly, MAPK was the ninth most significant altered pathway (R-HSA-5687128; 7.41 × 10^–53). The MHC Class II antigen presentation system fell (R-HSA-2132295; 9.03 × 10^–60 between these two pathways), which is also expected based on work demonstrating high levels of MHC level expression in KRAS mutant cells.¹⁶

SCP Summary

We have recently described the use of a TIMSTOF mass spectrometer for multiplexed SCP. While we determined that higher carrier channel concentrations could be utilized with little decrease in quantitative accuracy, they came at the expense of increased missing values.¹² In this study, I employed a protocol more similar to the original SCoPE-MS study¹⁷ in which 200 single cells from each condition were used as the carrier channel in most experiments. For a smaller subset of cells, a pooled diluted homogenate was used as the carrier channel. By mixing the carrier channels from each experimental condition and using a pseudorandom mixing and loading protocol, I can combine SCP samples from multiple conditions in each LCMS experiment. While this may seem to be an obvious way to set up a study of this sort, this is not yet possible in commercial products such as the ProteoChip kit designed for use on the CellenOne systems.^18,19 In the CellenOne, one population of cells is picked up, isolated, and aliquoted at a time. Lysis, digestion, and labeling occur in nanoliter-sized droplets for optimal kinetics and minimal sample loss to surface adhesion. While the instrument can maintain precise levels of internal humidity to minimize digestion, it may take hours to load a second population of cells for isolation and sample preparation, and the opening of the system to load new cells may be enough for evaporation to occur. While these are challenges that are likely to be solved soon, they were not when this paper was written, leading to the use of the approach I described in this study.

At 200 cell carriers with an estimate of 200 picograms of protein per cell, the starting protein load for each injection should be approximately 41.6 ng. In the two cell lines described, 953 proteins were quantified, on average, when using a single search engine. In total, 3143 protein groups were identified. We have found that the use of multiple search engines leads to more than a 20% increase in identifications over a single search engine alone. Similar observations have been recently reported by others in SCP,²⁰ and these results track well with previous observations employing multiple search engines for higher concentration proteomic samples.^21,22 In addition, multiple advanced tools for single cell analysis such as isobaric tag enabled match between runs,²³ and posterior error adjustment can lead to a further increase in total identification rates.²⁴ The complete report of all proteins and quantification values for all single cells is provided in Supporting Information File 2.

Protein Post-Translational Modifications Can Be Identified in Nearly All Single Cells

As previously reported for the H358 cell line, I can detect and quantify multiple classes of protein post-translational modifications (PTMs) in single cells in this study. The most abundant and readily observed PTMs are histone modifications, particularly the well-characterized and tryptically convenient K28 and K80 lysine modification sites. Bar plots demonstrating the abundance of the K28 methylated and dimethylated peptides are shown in Figure 2. The former is commonly observed, with a signal in 75.8% (440/580) of the PANC 0813 single cells shown. When considering the possibility of methylation or dimethylation occupying the K28 site, 98.7% of single 0813 cells (573/580) demonstrate a quantifiable signal. These results are consistent for both cell lines in all of the experiments (data not shown). The ability to detect protein PTMs is directly related to both the protein total copy number and the total sequence coverage obtained for the protein. The tryptic peptide containing the K28 modification is identical in at least three different human histone proteins annotated in the SwissProt UniProt database today, all of which exist in excess of one million copies in most mammalian cells.²⁵ As such, it is difficult to interpret the meaning of these PTMs in these data as each signal observed may be a composite of modifications from multiple original histone proteoforms. These results suggest that the analysis of histone PTMs in single human cells would be a practical avenue for research. However, as in traditional proteomics, the use of alternative proteases²⁶ or lysine derivatization prior to tryptic digestion²⁷ would be required for accurate assignment of the PTM to the originating proteoform.

Figure 2.

Measured abundance of two PTMs on histone 3 with 580 single 0813 cells shown. (A) K28 dimethylated peptide was observed in over 75.8% of these cells. (B) Single dimethylation on this lysine was observed less frequently

Unsupervised Analysis Can Discriminate between Similar Pancreatic Cancer Cell Lines

One quality control metric for SCP methods that has been employed by multiple groups is the demonstration of cell-type clustering by unsupervised statistics such as PCA.^18,28,29 If HeLa cells and HEK293 cells form distinct clusters when analyzed together, then it stands to reason that the method is sound. One criticism of this method is that there are considerable differences between the size and protein contents of these two common laboratory cancer cell line strains. ASPC-1 and PANC 0813 are two highly similar pancreatic cancer cell lines that were derived from patient samples using the same immortalization strategy.³⁰ In addition, both cell lines possess two copies of the KRAS^G12D mutant gene. One major difference is that the ATCC recommends supplementing media for growing PANC 0813 with insulin, whereas none is recommended for ASPC-1. In limited experiments in our laboratory, we found this addition to be critical for cell survival (data not shown). As shown in Figure 3, a principal component analysis can resolve these two cancer cell lines when the carrier channel is obtained from 100 FACs sorted cells from each condition following sum-based normalization of the total ion current for each single cell. A recent preprint demonstrated that the largest proteomic effects in single-cell measurements are simply cell size and recommended the normalization of the single-cell signal using histones as the normalization factor.³¹ While normalization with histone H4 does appear to reduce the gap in clustering between these two cell lines, it did not lead to a coalescence of these two groups (data not shown). It is worth noting that previous studies have demonstrated large proteomic effects from insulin stimulation in the media of cancer cell lines.³² It is therefore possible that the true factor driving clustering visualized in this figure is insulin stimulation alone. However, this is still likely a more biologically meaningful metric for SCP method development than unsupervised clustering of HeLa and HEK 293 cells.

Figure 3.

Principal component analysis of random injections of two similarly derived KRAS^G12D mutant pancreatic cancer cell lines (n = 595, ASCP-1; n = 574, PANC 0813 cells)

How Carrier Channels Are Prepared Alters the Proteins That Can Be Detected in SCoPE-MS

There are two main strategies today for the preparation of carrier channels for multiplexed SCP. The original study by Budnik et al. utilized a collection of single cells which were lysed and prepared as the carrier channel.¹⁷ Studies utilizing liquid handling devices such as the CellenOne system, a miniaturized robotics system based on the SciFlex Arrayer,³³ likewise use this strategy.^18,34 In Orsburn et al., we prepared a bulk cell homogenate of cells to be analyzed, labeled the homogenate with the 134c channel, and prepared a dilution series to obtain an experimentally determined ideal carrier channel concentration for analysis.⁶ While our hypothesis was that this consistency in carrier composition would lead to a higher consistency between samples, an analysis of the two preparation methods appeared warranted. To evaluate the differences between these two approaches, carrier channels for PANC 0813 and ASPC-1 were prepared using both approaches. As shown in Figure 4 unsupervised analysis clusters the samples into four groups based on cell line and the carrier channel creation method.

Figure 4.

Principal component analysis of PANC 0813 and ASPC-1 cells analyzed using either a bulk proteomics homogenate for a carrier channel or a FAC sorted 200 cell carrier channel

Bulk Cell Lysate Carriers Increase the Representation of Proteins Involved in Apoptosis

To assess the differences imparted by the two carrier channel preparation strategies, I examined the functions of proteins exclusively identified in each carrier channel experiment using a simple plus/minus filter in Venny.³⁵ Proteins unique to each carrier channel preparation method were analyzed for function in StringDB.³⁶ The most striking difference between groups was the presence of apoptotic and programed cell death proteins, which appeared almost exclusive to files where carrier channels were prepared from bulk cell homogenates. Manual evaluation of the original proteomic data confirmed that proteins such as caspase 6 and caspase 8 were exclusive to these samples (Figure S1).

SCP Data Largely Recapitulates Bulk Proteomics Data

As mentioned above, MRTX1133 treatment at this time point leads to a decrease in MAPK pathway protein levels, such as a 2.38-fold decrease in MAPK1 levels in PANC 0813 and a 3.46-fold decrease in the same levels in ASPC-1. As a 2-fold differential is a frequently used cutoff in preliminary proteomic data analysis and this protein is known to be linked to drug treatment, an analysis of MAPK1 in PANC 0813 was a practical metric for SCP data quality.

During the first analysis of these data, it did not appear that SCP recapitulated these findings, particularly through data reduction methods such at principal component analysis (Figure 5) . This was despite a relatively high detection rate for MAPK1 which was quantified in 61.4% of PANC 0813 cells in this study. This led to the release of a preprint with a considerably pessimistic tone regarding the current state of SCP and this particular method of approach. Helpful discussions originating from the preprint manuscript as well as a complete reanalysis of these files presented at ASMS 2023 by Jim Palmieri (data to be published elsewhere) suggested flaws existed in the quantitative data analysis.

Figure 5.

Principal component analysis of 1498 KRAS^G12D control and MRTX1133-treated cells. Experiments where the carrier channel was prepared from FAC isolated single cells are denoted as “200 cell carrier”. Experiments where the carrier was prepared from a bulk cell homogenate that was labeled and diluted are denoted as “homogenate carrier”

For a more direct analysis, the abundance values for MAPK1 were plotted during each stage of data normalization. Normalization in the Proteome Discoverer 2.4SP1 is similar to other reporter ion-based strategies but is broken into user-determined stages. The first step occurs at the peptide level in which the total observed peptide reporter ion signal for each single cell is assumed to be constant, and the total peptide amount summed normalization strategy sets these as even. At the next stage, scaling can be used, in which the total amount of an individually observed protein in every cell is used to scale the protein abundance for downstream ratio analysis. In this approach, protein A for 500 cells will be set so that the mean value is 100 and protein abundance in each cell is proportional to this mean value.

In PANC 0813, a mean fold change of −1.48 was observed when evaluating the raw unadjusted protein abundance measurements for MAPK1 in control versus treated cells. A student’s unpaired t-test found a significant difference between the control and treated cells (p = 0.031, n = 553). However, both sequential normalization and abundance scaling reduced both the observed ratios and the apparent significance. Sum-based normalization reduced the fold change to −1.11 at a p-value of 0.238 and scaling further reduced the median ratio to −1.09 at a p-value of 0.327, suggesting no change and no significance (Figure 6A). It should be noted that when we previously utilized a two-proteome standard to determine the extent of the carrier channel effects on a TIMSTOF instrument, we observed that known ratios of 5:1 were compressed to approximately 3:1 when using a 200× carrier channel. Using a simple linear extrapolation for MAPK1, the −2.38-fold differential observed in bulk proteomics would represent a strikingly similar expectation of −1.50-fold for MAPK1 at this carrier level on this platform.

Figure 6.

Demonstration of ratio suppression after protein normalization in SCP data, with non-normalized data (RAW) plotted against normalized and bulk cell proteomics findings. (A) Abundance of MAPK1. (B) The same analysis for apolipoprotein B100. (C) Abundance of the KRAS protein in PANC 0813 cells following treatment with MRTX1133

Further spot checks of proteins observed as altered in the bulk proteomic data following treatment likewise found quantitative alterations suppressed by these traditional proteomics normalization and scaling tools. Apolipoprotein B100 was a protein with substantially higher abundance than MAPK1 which followed the opposite trend, with a 2.03-fold increase following drug treatment in PANC 0813 bulk proteomics. Measurements of the mean of the original peptide mirror the differential regulation observed in the bulk proteomics experiment, but this observation is lost following normalization and scaling (Figure 6B). The observation that data handling can reduce quantitative significance in SCP is not new to this study. A label-free SCP work noted that imputation markedly reduced peptide differentials when applied.³⁷ A recent reanalysis of those files further determined that recapitulation of bulk proteomics from SCP was convoluted by both standard normalization and imputation approaches.³⁸ While a full analysis of the effects of commonly used tools for bulk proteomics on SCP data is beyond the scope of this paper, these results do suggest that new informatics should be embraced for SCP, and the results of traditional tools should be interpreted with caution.

Grouping MRTX1133-Treated Cells by Total KRAS Abundance Suggests That Subpopulations Exist

As expected for a KRAS inhibitor, treatment with MRTX1133 leads to an overall decrease in the abundance of KRAS protein in the bulk proteomics data. Likewise, the mean of the non-normalized abundance of KRAS protein in single cells demonstrated a net decrease in protein abundance. The recapitulation of bulk proteomic findings is a useful confidence metric for SCP data interpretation, but it is not the goal. The real power of SCP is the ability to identify and evaluate subpopulations of cells with different phenotypes.³⁹ The mean protein abundance of KRAS in untreated PANC 0813 cells was calculated at 167.8, while the same for treated cells was 124.1 (Figure 6C). Approximately 15.1% of MRTX1133-treated PANC 0813 cells were observed to have a KRAS protein abundance greater than the total population mean. These results suggested that multiple populations of cells may be present following drug treatment at this time point, with at least one population of cells where the drug reduced total KRAS expression and one where it did not. To further evaluate this subpopulation, treated cells with a nonzero measurement for KRAS protein expression were clustered for SimpliFi statistical analysis with the cells divided into two groups, those with above median KRAS protein abundance and those below that line. From this analysis, 62 analytes were determined as significantly differential (log 2 > 1, p-value < 0.05). The MAP kinase 4/6 pathway again appears to be differential in this experiment (R-HSA-5687128, p = 6.88 × 10^–8). In addition, several proteasome subunits, namely, 26S subunits 13 (PSMD13) and both alpha type 3 and 5 (PSMA3 and PSMA5) appear significantly higher in abundance in cells with higher mutant KRAS protein abundance. A complete list of differential proteins and statistical analysis is provided as Supporting Information File 3. When evaluating proteins as differential by fold-change alone, proteins involved in central metabolism and protein translation as well as proteosome degradation appear to be the highest in cells with above median KRAS abundance. Lower relative KRAS expression corresponds to increased levels of ubiquitin 6, 10, and E2 all having 4-fold higher abundance in cells with low KRAS abundance. While full conclusions may be difficult to draw from a relatively small sample set, it is tempting to conclude that in cells where mutant KRAS protein has not been successfully inhibited at this dose and time point, the cells proceed with business as usual. In cells where inhibition has occurred effectively, the lack of the prosurvival cascades caused by the mutant protein has inhibited cell growth and metabolism and is beginning to lead to quiescence or apoptosis.

In Cases Where Drug Treatment May Result in Large Cell Death, SCP May Not Be Possible

PANC 0203 and PANC 0403 have been recently described as two cell lines with extremely high sensitivity to MRTX1133.⁵ My results strongly support these claims as 10 nM treatment for 48 h resulted in both dramatic reductions in viability as well as alteration in cellular morphology as detected by fluorescence scatter. During single-cell isolation in this study, viable cells were collected based on scattering using a simple live/dead cell stain. For both PANC 0203 and PANC 0403 cells, it was not possible to collect 200 cells for the carrier channel for more than a few intended plates due to the reduced number of cells present. Sorting gates required adjustment for both these treated cell lines due to the level of difference in relative scatter compared to control (Figure S2). In this case, the cells that were isolated appeared to be the minority of all cells, leading to concerns regarding the value of data from these cells for this study. Repeated attempts to obtain SCP data on the PANC 0203 cell line have been unsuccessful even at levels of drug treatment significantly lower than those in previous reports for this cell line (data not shown). For treatment conditions where large amounts of cell death may be an outcome, careful titration of the conditions may be necessary to obtain biologically meaningful cells for study. In some cases, this may not be possible at all.

Limitations of the Current Study

Considerable limitations exist in this current work, as described. Limitations in the metabolomics analysis include the lack of internal or external controls for the identification and quantification of the identified metabolites. All identifications were made through a combination of high-resolution MS1 and MS/MS identifications against spectral library databases. As no tools have been fully implemented for the estimation and false discovery rates of metabolomics data today, library-based matches should always be interpreted with caution. SCP data comes with a myriad of limitations, including the fact that a data set of this size should be interpreted by a larger team than a single one. While this work was intended to pressure test our current SCP workflows while evaluating a drug of particular interest to our program, the true goal of our work is to analyze the time courses of drug treatments. While this will require significantly more time, cells, and assistance to execute, the basic workflow appears sound to carry out these studies.

Conclusions

In this work, the application of metabolomics, proteomics, and SCP to the analysis of acute dosage of a promising new small-molecule inhibitor is described. While MRTX1133 is currently in clinical trials and may not succeed, it is inevitably the basis of further compounds to target one of the most deleterious of all human cancer mutations.⁴⁰ To the best of our knowledge, this is the first metabolomic or proteomic analysis of cells treated with this compound. Metabolomics demonstrates that the drug leads to a decrease in both ATP and NAD precursors in all cell lines. In addition, many amino acids appear to accumulate following drug treatment, while some amino acids such as glutamine appear to be differential between cell lines in a manner that corresponds to the number of mutant copies of the gene in each cell line. However, this is the result of only four cancer cell lines, and these results should be interpreted with caution. Proteomic analysis indicates the expected decrease in the MAPK pathway in all cell lines, as well as an alteration in GTPase signaling and MHC expression. In general, results should be expected based on previous results.^1,41 To add further granularity to these measurements, approximately 1500 single cells from the PANC 0813 and ASPC-1 cell lines were analyzed. While dimensional reduction of SCP data did not indicate the existence of subpopulations of cells, in these analyses the direct analysis of the drug target does indicate that they do. At least two populations appear to exist: those in which the inhibitor has led to a decreased concentration of the KRAS protein and those where they do not. Further work in our laboratory is centered on new informatics solutions for the direct analysis of single proteins in single-cell data to help identify new populations based on expected phenotypes such as this.

In addition, I divided the single-cell study into two sections, one in which the carrier channel was a collection of presumably viable FACs sorted cells and a second where the carrier was created from a bulk cell homogenate. These two methods for preparing the carrier channels imparted differences in the proteins that were observed. While many of the protein level differences between cells analyzed with the two carrier channel strategies are difficult to reason through, some are not. The presence of Caspases 6 and 8 was exclusively detected in single cells where a bulk cell lysate carrier channel was employed. This suggests that when FACs isolate viable cells as carrier channels proteins from cells that are dead or dying are being sorted away with the unhealthy cells to which they belong. In the case of drugs such as MRTX1133, where the goal is the death of cells harboring a specific mutation, care must be taken to assess the correct population for each experimental goal.

Considerable effort today is being directed toward the improvement of the field of SCP. The majority of work to date has centered on developing new tools for cellular isolation, peptide recovery, and increased measurement sensitivity.^18,29,42,43 However, a recent multi-institute collaborative effort has provided suggestions for the reporting of SCP data⁴⁴ and informatics has continued to develop to address the unique challenges inherent in this emerging new field.⁴⁵ My takeaway from this study is that equal effort may be necessary to develop ideal experimental designs for a successful SCP study of drug treatment models.

Data Availability Statement

All proteomics results have been deposited and can be viewed in the SimpliFi cloud platform. The cell lines with a single and double copy of the KRAS^G12D mutant form were processed separately for ease of visualization. The PANC 0813 and ASPC-1 proteomics data can be found at the following link: https://simplifi.protifi.com/#/p/4fd12e70-9809-11ed-b0f0-03f62ebcf0cd. PANC 0203 and PANC 0403 can be found here: https://simplifi.protifi.com/#/p/28a07700-b93a-11ed-92f4-b7d04df347c4. The original RAW and processed files are divided among 3 openly available repositories at www.Massive.UCSD.edu: PXD039597, PXD039601, and PXD039600.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jproteome.3c00212.

• Metabolomic alteration in four MRTX1133-treated cell lines (XLSX)

• Proteomic report for all single cells in this study (XLSX)

• Summary of SCP observations of MRTX1133-treated cells with above median KRAS^G12D protein expression versus those with below median expression (XLSX)

• Proteins exclusively observed when a bulk cell homogenate sample is used as a carrier channel and FAC analysis demonstrating increased cell death in MRTX1133-treated PANC 0203 cells (PDF)

Funding was provided by the National Institutes of Health through the National Institute on Aging award R01AG064908 and the National Institute of General Medical Sciences R01GM103853.

The author declares no competing financial interest.