Descriptive Proteome Analysis to Investigate Context-Dependent Treatment Responses to OXPHOS Inhibition in Colon Carcinoma Cells Grown as Monolayer and Multicellular Tumor Spheroids

Abstract

We have previously identified selective upregulation of the mevalonate pathway genes upon inhibition of oxidative phosphorylation (OXPHOS) in quiescent cancer cells. Using mass spectrometry-based proteomics, we here investigated whether these responses are corroborated on the protein level and whether proteomics could yield unique insights into context-dependent biology. HCT116 colon carcinoma cells were cultured as monolayer cultures, proliferative multicellular tumor spheroids (P-MCTS), or quiescent (Q-MCTS) multicellular tumor spheroids and exposed to OXPHOS inhibitors: nitazoxanide, FCCP, oligomycin, and salinomycin or the HMG-CoA-reductase inhibitor simvastatin at two different doses for 6 and 24 h. Samples were processed using an in-depth bottom-up proteomics workflow resulting in a total of 9286 identified protein groups. Gene set enrichment analysis showed profound differences between the three cell systems and confirmed differential enrichment of hypoxia, OXPHOS, and cell cycle progression-related protein responses in P-MCTS and Q-MCTS. Treatment experiments showed that the observed drug-induced alterations in gene expression of metabolically challenged cells are not translated directly to the protein level, but the results reaffirmed OXPHOS as a selective vulnerability of quiescent cancer cells. This work provides rationale for the use of deep proteome profiling to identify context-dependent treatment responses and encourages further studies investigating metabolic processes that could be co-targeted together with OXPHOS to eradicate quiescent cancer cells.

Introduction

Cancer cells are usually characterized by their increased proliferation, resistance to apoptosis, invasiveness, and poor differentiation. However, it has been recently described that many solid tumors harbor nonproliferative, quiescent cells, residing in nutrient-deprived and hypoxic microenvironments, characterized by increased DNA damage and altered metabolism.^1,2 This heterogeneous tumor microenvironment poses a therapeutic challenge, as chemo- and radiation therapy has been demonstrated to be less effective against cells in the poorly vascularized hypoxic niches, the presence of which has been associated with tumor relapse and poor prognosis.³ Thus, except targeting fast-growing tumor cells, there is also a need to find therapeutic strategies aimed at the quiescent tumor regions.

Three-dimensional (3D) cell cultures, in contrast to monolayer cultures, offer the possibility to investigate cell signaling, growth characteristics, and drug response in more in vivo like settings. Of various 3D culture types, multicellular tumor spheroids (MCTS) and tumor organoids have gained the most attention. In general, MCTS are usually generated from cell lines while tumor organoids are formed from primary cancer cells. Organoid cultures, which are grown in an external protein matrices (such as Matrigel), more accurately recapitulate the genetic and morphological characteristics of a primary tumor. However, because of the cost of organoid development and expansion and limited cellular material availability, MCTS can more easily and reproducibly be used in the large experimental setup, for example, high-throughput drug screening,⁴⁻⁶ and have been a valuable model for the studies of clinically relevant aspects of cancer biology, as they resemble morphological, functional, and microenvironmental features of in vivo tumor tissues.⁷ However, most commonly used spheroid models are highly proliferative MCTS (P-MCTS), as they are usually maintained in standard nutrient-rich culture media. P-MCTS experience high nutrient concentrations as the model contains a mixture of proliferating cells in the outer layers and quiescent cells toward the center of the spheroid. Recently, we have demonstrated that quiescent spheroids (Q-MCTS), cultured under in vivo-like conditions mimicking solid tumor microenvironments (i.e., nutrient-deprived and with low extracellular pH), contain a larger fraction of quiescent cells and are therefore profoundly different from P-MCTS and monolayer cultures. In consequence, Q-MCTS enable modeling of quiescent regions of solid tumors and identification of unforeseen vulnerabilities of quiescent cancer cells.^8,9 Furthermore, Q-MCTS allow studying drug responses of quiescent cancer cells in clearer detail, without the results being too obscured by the contribution from proliferating cells. Importantly, we and others have shown that quiescent cancer cells have a context-dependent oxidative phosphorylation (OXPHOS) dependency that can be pharmacologically targeted (recently reviewed in¹⁰). Previously, we showed that pharmacological inhibition of OXPHOS resulted in selective upregulation of key genes of the initial steps in the cholesterol biosynthesis, namely the mevalonate pathway, in Q-MCTS but not in P-MCTS or monolayer cells. Further, we demonstrated that such vulnerabilities can be exploited to design drug combinations effective only in the specific tissue context, such as synergistic cytotoxicity of nitazoxanide (OXPHOS inhibitor) and simvastatin (mevalonate pathway inhibitor) in quiescent cancer cells. Therefore, differentiating various cellular models, such as P-MCTS, Q-MCTS, and monolayer cell cultures, might provide insights into targetable context-dependent vulnerabilities in the phenotypically heterogeneous tumor tissue context. The metabolic differences between proliferative and quiescent cancer cells have also been described in other cell systems such as fibroblasts and mammary epithelial cells.^11,12

Several comparative analyses of monolayer and spheroid cultures addressing their global proteomes have shown differences in adhesion proteins, immune response pathways, matrix metalloproteinase expression pattern, phosphoproteomic pathways, and growth rate.¹³⁻¹⁵ However, the integration of proteome analysis with gene expression data and its functional interpretation in coherence with clinical results remains a challenge.

Here, we present a framework for in-depth proteome analysis of drug-treated 3D cultures. We demonstrate the influence of different culture conditions by profiling the colon cancer cell line HCT116 grown as Q-MCTS, P-MCTS, and monolayer cells. Furthermore, we characterize the effects after treatment with four different OXPHOS inhibitors or simvastatin in the different models. The combination of 3D cell culture and mass spectrometry-based proteome analysis allowed us to complement previous findings on gene expression profiles and to compare the relative protein changes between the three cellular models.

Results and Discussion

The aim of this study was to investigate and compare the proteome of three distinct multicellular colon cancer models (i.e., P-MCTS, Q-MCTS, and monolayer cell cultures) in response to four different OXPHOS inhibitors (nitazoxanide, FCCP, oligomycin, and salinomycin) or the 3-hydroxyl-3-methylglutaryl-coenzym A reductase (HMGCR) inhibitor simvastatin at two different doses. To obtain signatures related to the direct mechanism of action without major feedback signaling, the proteome was studied at a relatively early time point of 6 h and a later time point of 24 h for comparison and related to our previous findings on gene expression data with main respect to metabolic pathways. Q-MCTS did not receive fresh media during the culture period, resulting in a smaller fraction of proliferating cells than P-MCTS (Figure S1), which, in turn, translates into differential drug response.⁸

Quality Analysis of Bottom-Up Quantitative Proteomic Data

To obtain quantitative data on protein abundances, proteins from spheroid samples were extracted, digested, and tandem-mass-tag (TMT) labeled using single-pot solid phase-enhanced sample preparation (SP3), as depicted in Figure 7. For in-depth protein profiling, samples were prefractionated by high pH reverse-phase liquid chromatography prior application to nanoliquid chromatography tandem mass spectrometry (LC–MS/MS). From a total of 268 LC–MS/MS runs, 5174157 MS/MS spectra were obtained yielding 1594878 peptide spectrum matches (PSMS) and 9286 protein groups (Figure 1A). The obtained list of identified proteins was further reduced to proteins with a high-protein false discovery rate (FDR) confidence and at least two unique peptides (Figure 7D). Venn diagram visualization of proteins and protein isoforms present across 60% of the whole data set revealed that 73% of identified proteins is shared between monolayer, P-MCTS, and Q-MCTS and 4.3, 6.2, and 1.2% unique proteins could be found for each cell type, respectively (Figure 1B). When comparing dimethyl sulfoxide (DMSO)-treated cells, a similar distribution could be observed (Figure S3B). Comparing proteins expressed by two of the three cell types, monolayer and P-MCTS share 1247 proteins while PMCTS and QMCTS share 828 proteins and QMCTS and monolayer only share 135 proteins. This indicates that monolayer cells and Q-MCTS are most functionally different, corroborating our earlier findings from gene expression analysis. For further data processing, only proteins with unique sequences (no isoforms) were used. The treatment and control data sets with 6025 and 5965 quantified proteins, respectively, were individually processed for downstream analysis. The number of identified and quantified proteins, mass distributions covered, and the reproducibility of the measured absolute protein abundances shown for the control set DMSO data (Pearson correlation coefficient > 0.9) indicate a high identification quality and reproducibility throughout the data set (Figure 1A,C). However, six out of 60 processed samples did not fulfill our expectations with respect to quantification quality (Figure S2) and were removed from further downstream analysis. In Figure 1D, a principal component analysis (PCA) of the whole data set after removal of outliers shows three clusters corresponding to the three different cell types, suggesting that the protein expression of monolayer cells is different to both P-MCTS and Q-MCTS, independently of drug exposure.

Figure 7.

Workflow and experimental design of nano-LC–MS/MS-based proteomics for 2D and 3D colon cancer cells. HCT116 cells were cultured as 2D monoculture or 3D spheroids in 384 well plates. Cells were treated with indicated concentrations of the respective drugs (A) or vehicle control for 6 or 24 h. For proteome analysis, cells of 28 spheroid replicates were pooled, and proteins were extracted, digested, TMT-labeled, and purified by SP3 paramagnetic bead technology, followed by high-pH reversed-phase prefractionation prior nano-LC–MS/MS analysis (B,C). Acquired peptide spectra data were processed using proteome discoverer software, and quantified proteins were visualized using R and open-source platforms like GSEA and EnrichR (D). This graphic was generated in Microsoft Visio 2007.

Figure 1.

Quality control analysis and data assessment of bottom-up quantitative proteomics. (A) Mass range coverage and distribution of the obtained MS/MS spectrum information (5174157) and resulting PSMS (1594878) throughout the gradient. (B) Venn diagram of the whole data set comparing the proteome of the three cell types (including isoforms). Proteins present across 60% of the samples for each cell type were included in the analysis. (C) Reproducibility of the measured absolute protein abundances for the control set data is presented, where each row/column depicts a control (DMSO treated) sample (Pearson correlation coefficient > 0.9). (D) PCA (PC1, PC3) of quantified proteins throughout the whole data set after removal of outliers.

Protein Expression Pattern of Three Distinct Multicellular Colon Cancer Models

We recently presented a novel MCTS model that allows to specifically study quiescent cancer cells with highly hypoxic core, low proliferation rate, and low glucose levels resembling dormant tumor regions that present strong differences in gene expression profiles and context-dependent drug responses compared to P-MCTS and monolayer cells.^8,9 In order to investigate whether these differences are translated to the protein level, we used mass spectrometry (MS)-based proteomics to compare protein expression profiles in the three distinct cellular models. Normalized protein abundances of vehicle (DMSO treated) controls were hierarchically clustered and represented in a heat map (Figure 2). The cluster analysis shows differences in protein expression between the cellular models, with the largest dissimilarities between monolayer and the spheroid cultures, demonstrating profound differences between 2D and 3D cellular models (Figures 2 and S3A). The profiles of respective cellular models clustered together irrespective of the incubation time. However, profiles from P-MCTS differed between the time points whereby profiles of 24 h incubated P-MCTS clustered with Q-MCTS (6 and 24 h incubated), indicating that prolonged culture of P-MCTS renders them more similar to Q-MCTS (Figure 2), likely because of depletion of culturing medium and higher levels of cellular quiescence.

Figure 2.

Differences in protein expression between the three cellular models. Comparison of the expression profiles of proteins quantified (5965 proteins) in all three cell types (vehicle controls) by hierarchical clustering (1-minus the Pearson correlation, average linkage).

To further characterize the differences between the cell models and time effects, we performed gene set enrichment analysis. As expected from the gene expression data and findings by others,^8,22 P-MCTS and Q-MCTS profiles when compared with monolayer profiles were positively enriched in genes involved in response to hypoxia and OXPHOS and negatively enriched for genes involved in cell cycle progression (E2F targets and G2M checkpoint) (Figure 3). Comparison of Q-MCTS versus P-MCTS revealed that OXPHOS, adipogenesis, fatty acid metabolism, late estrogen response, and peroxisomes were positively enriched, and the pathway G2M-checkpoint, E2F targets, interferon alpha-and-gamma responses, and mitotic spindle were negatively enriched. This was in line with the results obtained in the gene expression analysis in these cellular models and can be related to OXPHOS as one major source of ATP for quiescent as well as proliferative MCTS.²³ Overall the enrichment profiles were similar between the different spheroid models and independently of the time point the cells were harvested. In our previous work, the gene expression analysis showed strongest upregulation of the mevalonate pathway genes for Q-MCTS. When comparing the proteome of P-MCTS as well as Q-MCTS to monolayer cell cultures, cholesterol homeostasis (including the mevalonate pathway proteins) was positively enriched for both time points with slightly higher enrichment scores for P-MCTS (Supporting_Information_XLSX_1). Elevated cholesterol levels have been reported for several solid tumors to play a role in cancer progression and to highly correlate with tumor cell resistance to chemotherapy.²⁴⁻²⁶ However, compared to the strong genetic upregulation in Q-MCTS, protein-based enrichment analysis demonstrated only weak activation of the mevalonate/cholesterol pathway with more positive enrichment in P-MCTS. Indicating that although the mevalonate pathway genes are overexpressed in both Q-MCTS and P-MCTS, only P-MCTS are metabolically fit to synthesize the proteins. In conclusion, our results highlight the need of metabolically challenged cells to initiate the upregulation of OXPHOS to maintain energy demands and the active downregulation of energy-consuming pathways. Moreover, our results suggest that gene expression does not translate to protein expression in metabolically challenged cells.

Figure 3.

GSEA of the three cellular models. Enrichment analysis of vehicle-treated cells in P-MCTS (P) and Q-MCTS (Q) compared to monolayer (M) cells and Q-MCTS vs P-MCTS comparison. The top five positive and negative enriched pathways are listed for 6 and 24 h time points. NES scores (normalized enrichment scores) were generated after 1000 permutations. Estimation of the statistical significance of the enrichment score for each respective gene set is indicated by the nominal p-value.

Protein Expression Pattern of Three Distinct Multicellular Colon Cancer Models Upon Treatment with Four Candidate Compounds Targeting OXPHOS and Simvastatin

In our previous studies,⁸ we showed that Q-MCTS are more sensitive to OXPHOS inhibition, as they experience low glucose levels and are forced to utilize OXPHOS to maintain their energy needs (contrary to P-MCTS and monolayer cells, which persist in high glucose concentrations and can produce enough ATP through glycolysis alone). Based on previous gene expression analysis, we identified the mevalonate pathway as important for the survival of Q-MCTS when inhibiting OXPHOS. Also, we observed potent synergistic toxicity upon combinatory treatment with simvastatin and OXPHOS inhibitors, which highlight the mechanistically coupling of both pathways.⁸ Here, we set out to identify model-specific responses on the protein level upon drug treatment. Therefore, we compared the effects of four different OXPHOS inhibitors: nitazoxanide, FCCP, oligomycin, salinomycin, and simvastatin to DMSO controls at two different drug doses and two different drug exposure times. Gene set enrichment analysis of protein expression after 24 h drug treatment revealed a negative enrichment of proliferation in monolayer, P-MCTS, and partially in Q-MCTS upon all treatments, whereas an upregulation of hypoxia, metabolic, and immune processes could be observed (Table 1). Furthermore, we observed an evident positive enrichment of cholesterol homeostasis upon treatment with simvastatin and salinomycin in monolayer cells and P-MCTS.

Table 1. GSEA of the Three Cellular Models Upon High Dose Drug Treatment for the 24 h Time Point^a.

POS-enriched PW
NTZ 10 μM	FCCP 2 μM	OLIGO 1 μM	SAL 5 μM	SIMV 10 μM
Monolayer
hypoxia	Hypoxia	hypoxia	cholesterol_homeostasis	fatty_acid_metabolism
Adipogenesis	adipogenesis	xenobiotic_metabolism	hypoxia	glycolysis
fatty_acid_metabolism	epithelial_mesenchymal_transition	IL2_STAT5_signaling	coagulation	cholesterol_homeostasis
oxidative_phosphorylation	IL2_STAT5_signaling	coagulation	IL2_STAT5_signaling	hypoxia
xenobiotic_metabolism	glycolysis	bile_acid_metabolism	MTORC1_signaling	coagulation

POS-enriched PW
NTZ 10 μM	FCCP 10 μM	OLIGO 1 μM	SAL 5 μM	SIMV 10 μM
PMCTS
hypoxia	apical_junction	oxidative_phosphorylation	MTORC1_signaling	hypoxia
MYC_targets_V1	hypoxia	protein_secretion	MYC_targets_V1	cholesterol_homeostasis
xenobiotic_metabolism	adipogenesis	xenobiotic_metabolism	cholesterol_homeostasis	apical_junction
MTORC1_signaling	xenobiotic_metabolism	androgen_response	TNFA_signaling_via_NFKB	estrogen_response_early
Adipogenesis	glycolysis	MTORC1_signaling	unfolded_protein_response	PI3K_AKT_MTOR_signaling
QMCTS
MYC_targets_V1	MYC_targets_V1	MYC_targets_V1	MYC_targets_V1	fatty_acid_metabolism
MTORC1_signaling	MTORC1_signaling	MTORC1_signaling	unfolded_protein_response	xenobiotic_metabolism
unfolded_protein_response	unfolded_protein_response	unfolded_protein_response	MTORC1_signaling	peroxisome
protein_secretion	xenobiotic_metabolism	apical_junction	xenobiotic_metabolism	apoptosis
mitotic_spindle	glycolysis	protein_secretion	fatty_acid_metabolism	MTORC1_signaling

NEG-enriched PW
NTZ 10 μM	FCCP 2 μM	OLIGO 1 μM	SAL 5 μM	SIMV 10 μM
Monolayer
MYC_targets_V1	G2M_checkpoint	MYC_targets_V1	G2M_checkpoint	E2F_targets
E2F_targets	E2F_targets	DNA_repair	E2F_targets	MYC_targets_V1
DNA_repair	MYC_targets_V1	E2F_targets	oxidative_phosphorylation	MYC_targets_V2
G2M_checkpoint	DNA_repair	G2M_checkpoint	spermatogenesis	G2M_checkpoint
unfolded_protein_response	spermatogenesis	oxidative_phosphorylation	mitotic_spindle	DNA_repair

NEG-enriched PW
NTZ 10 μM	FCCP 10 μM	OLIGO 1 μM	SAL 5 μM	SIMV 10 μM
PMCTS
G2M_checkpoint	E2F_targets	E2F_targets	E2F_targets	E2F_targets
E2F_targets	G2M_checkpoint	G2M_checkpoint	interferon_alpha_response	MYC_targets_V1
interferon_alpha_response	interferon_alpha_response	interferon_alpha_response	oxidative_phosphorylation	G2M_checkpoint
interferon_gamma_response	spermatogenesis	interferon_gamma_response	G2M_checkpoint	MYC_targets_V2
Spermatogenesis	mitotic_spindle	Spermatogenesis	interferon_gamma_response	spermatogenesis
QMCTS
oxidative_phosphorylation	oxidative_phosphorylation	oxidative_phosphorylation	interferon_gamma_response	E2F_targets
epithelial_mesenchymal_transition	KRAS_signaling_up	interferon_alpha_response	apoptosis	G2M_checkpoint
KRAS_signaling_up	interferon_alpha_response	interferon_gamma_response	P53_pathway	spermatogenesis
P53_pathway	interferon_gamma_response	KRAS_signaling_up	epithelial_mesenchymal_transition	DNA_repair
interferon_gamma_response	TNFA_signaling_via_NFKB	TNFA_signaling_via_NFKB	IL2_STAT5_signaling	reactive_oxigen_species_pathway

^aEnrichment analysis of drug-treated cells in monolayer, P-MCTS, and Q-MCTS compared to vehicle controls. The top five positively and negatively enriched pathways are listed.

In Figure 4, a detailed analysis of cholesterol homeostasis proteins shows an upregulation of key proteins of the mevalonate pathway such as ACAT2, CYP51A1, HMGCS1, HMGCR, and SQLE upon Simvastatin and Salinomycin treatment in dose response and 24 h post treatment in monolayer and P-MCTS but not in Q-MCTS. On the protein level, Q-MCTS do not show an enrichment of the mevalonate pathway, as observed in our previous work on the gene expression level (Senkowski et al., 2016, Figure4B).

Figure 4.

Cholesterol pathway proteins regulated upon drug treatment at 6 and 24 h time points compared to vehicle control. Protein abundance ratios for respective gene IDs extracted from GSEA hallmark cholesterol homeostasis are depicted.

This indicates that Q-MCTS, maintained in culture medium containing low nutrient concentrations, may not be metabolically viable to synthesize proteins from overexpressed transcripts, and that protein expression 24 h post treatment indicates only a modest correlation of transcripts and proteins. As protein levels are highly influenced by post-translational events, such as folding, modifications, or degradation processes, which are depending on the energy status of a cell or the cell cycle, they are increasingly affected in metabolically challenged cells.²⁷

However, for monolayer cells, the upregulation of key proteins such as HMGCS1, HMGCR, and SQLE upon salinomycin and simvastatin treatment after 24 h incubation was in line with the gene expression data. This indicates higher susceptibility of monolayer cells to drug treatment and a reduced ability of MCTS to respond to the treatment. MCTS are characterized by complex multicellular interactions that lead to alternated and adapted gene and protein expression in response to nutrient availability. Protein synthesis is one of the most energy-demanding metabolic processes and subject to a strict regulation in starved cells.^28,29 An overall downregulation of protein synthesis despite selective transcriptional activity upon drug treatment indicates a not fulfilled genetic-intention especially in P-MCTS and Q-MCTS. An upregulation of cholesterol homeostasis proteins upon simvastatin, targeting the rate-limiting enzyme in the mevalonate/cholesterol pathway, (HMGCR) could be explained by a statin-induced feedback upregulation as an intention to maintain stable end-product levels.^24,30

Interestingly, positive enrichment of cholesterol homeostasis upon treatment with salinomycin, but not other OXPHOS inhibitors, in monolayer cells and P-MCTS indicates that the underlying molecular mechanism of the upregulation of the cholesterol pathway by salinomycin seems to be different from the other OXPHOS inhibitors.

To identify proteins that are commonly regulated by the individual cell types, we focused on the top five differentially expressed proteins in the three cellular model types after 6 and 24 h treatment. Frequency analysis comparing different treatments and cell types resulted in the identification of five proteins that were most frequently regulated (Figure 5). We identified dermicidin (P81605), a potential oncogene, which has been associated with tumor growth and increased survival and migration in different tumors,³¹ to be downregulated in all cell types upon drug treatment. The finding of dermicidin has to be interpreted with caution as it had been shown to be present in human skin, sweat, and tears, exhibiting high stability, likely to contaminate throughout the sample work-up.³² Calcium-binding protein (45 kDa, Q9BRK5) and nucleobindin-1 (Q02818) were downregulated in monolayers, whereas eukaryotic elongation factor 2 kinase (O00418), which has been shown to protect cancer cells against nutrient starvation,³³ and tripartite motif-containing protein 26 (Q12899) were downregulated specifically in Q-MCTS (Figure 5).

Figure 5.

Top five differentially expressed proteins upon 6 and 24 h high dose drug treatment. Proteins regulated upon any treatment and appearing at least two times in each cell type are listed in the right part of the figure. The dark gray bars represent the frequency of how often a protein was differentially regulated upon treatment for the respective cell type (a count of five indicates a protein being regulated upon all five treatments). Frequency analysis focusing on proteins with counts three and higher revealed five proteins of interest that show cell type-specific regulations: P81605 (regulated upon all treatments and in all cell types), Q9BRK5, Q02818 (regulated upon nitazoxanid, FCCP, and salinomycin in monolayers), O00418, and Q12899 (regulated upon OXPHOS inhibitors in Q-MCTS).

Identification of Early Responding Proteins in Q-MCTS upon OXPHOS Inhibition

To identify early responding proteins, we performed enrichment analysis using the online tool EnrichR of the most differentially (fold change > 0.6) upregulated proteins after treatment with exclusively OXPHOS inhibitors for 6 h. Table 2 shows the 10 most enriched pathways for the three cell types. Proliferative spheroids react with increased oxidative stress upregulating selenocysteine synthesis and termination of eukaryotic translation. We can observe strong context-dependent enrichment in OXPHOS proteins (p < 0.000001) exclusively in Q-MCTS. This context-dependent drug response points toward OXPHOS being a vulnerability in quiescent, nutrient-deprived cells. These observations are in line with ATP level measurements in Q-MCTS after 6 h treatment with nitazoxanide and salinomycin. ATP levels drop to 61 and 71%, for 10 μM nitazoxanide or 4 μM salinomycin, indicating an early-onset energy crisis (Figure 6). As we found only weak association with OXPHOS protein upregulation for Q-MCTS after 24 h treatment, this suggests that Q-MCTS after 24 h are severely depleted of energy and unable to synthesize OXPHOS proteins anymore (Tables 1 and S1).

Table 2. Pathways Upregulated Upon OXPHOS Inhibition after 6 h Treatment^a.

monolayer		PMCTS		QMCTS
pathway	p-value	pathway	p-value	pathway	p-value
cytosolic sulfonation of small molecules	0.0012	intra-Golgi and retrograde Golgi-to-ER traffic	0.0022	respiratory electron transport, ATP synthesis by chemiosmotic coupling, and heat production by uncoupling proteins	6.91 × 10–6
phase II conjugation	0.0029	retrograde transport at the trans-Golgi network	0.0023	the citric acid cycle and respiratory electron transport	3.57 × 10–5
platelet degranulation	0.0033	membrane trafficking	0.0030	respiratory electron transport	6.37 × 10–5
response to elevated platelet cytosolic Ca2	0.0038	vesicle-mediated transport	0.0053	complex I biogenesis	2.35 × 10–4
complex I biogenesis	0.0086	peptide chain elongation	0.0066	RHO GTPases activate CIT	6.90 × 10–4
apoptotic execution phase	0.0093	viral mRNA translation	0.0066	prefoldin-mediated transfer of substrate to CCT/TriC	1.98 × 10–3
apoptosis	0.0113	selenocysteine synthesis	0.0070	signaling by Rho GTPases	1.99 × 10–3
programmed cell death	0.0118	eukaryotic translation termination	0.0070	cooperation of prefoldin and TriC/CCT in actin and tubulin folding	2.78 × 10–3
signaling by MST1	0.0142	eukaryotic translation elongation	0.0074	RHO GTPase effectors	3.49 × 10–3
amyloid fiber formation	0.0161	nonsense-mediated decay independent of the exon junction complex	0.0074	HIV life cycle	4.99 × 10–3

^aProteins that were upregulated with fold change >0.6 upon OXPHOS inhibition were analyzed using EnrichR.

Figure 6.

ATP measurements in Q-MCTS after 6 h treatment with nitazoxanide and salinomycin. ATP levels (mean ± SD of technical triplicates) are presented as percentage (%) of vehicle-treated control (100%). This graph was prepared using GraphPad Prism v5.

Conclusions

In this study, we combine 3D spheroid cell culture with a global proteomic approach to investigate cellular responses to inhibitors targeting OXPHOS or the mevalonate pathway. Furthermore, we use SP3 paramagnetic bead technology and isobaric peptide labeling to process multiple samples simultaneously and offline high-pH reversed-phase prefractionation to gain high-quality in-depth proteomic data. With a coverage of 9286 protein groups identified, this workflow is suitable for large-scale proteome analysis.^34,35

Our data on vehicle-treated spheroids confirm the results on the gene level with respect to profound differences in the regulation of hypoxia, OXPHOS, and cell cycle progression between the three cell systems. This underlines the importance of choosing an accurate model system when addressing research questions using cell culture. Moreover, the outlined workflow paves the way for proteomic analysis of other 3D cultures, for example, tumor organoids, derived from primary cancer cells. As organoids enable propagation and expansion of patient-derived cellular material from different tumor types, thus providing enough cellular material for proteomic analysis, our workflow could be applied in these models for studying individualized drug responses on the global proteomic level.

Cancer cell metabolic pathways are a highly complex system consisting of multiple feedback loops that contribute to cancer surveillance. Understanding the vulnerabilities of cancer cells will help to develop successful drugs with increased specificity and efficacy. The present study emphasized the importance of taking the metabolic status and culture conditions into account when performing proteome analysis and drug evaluations. We demonstrated that drug-induced changes in gene expression are not always translated on the protein level, likely because metabolically challenged cells are not able to respond beyond gene expression. The metabolic capacity of the cellular models used is of high importance when searching for drug targets using proteomics under different culture conditions.

Finally, our study shows that proteome analysis is a powerful approach to identify context-dependent drug responses, as exemplified by the demonstrated dependency on OXPHOS specifically seen in Q-MCTS 6 h post treatment. Insights from this study contribute to a better understanding of the metabolic complexity of proliferative and quiescent cancer cells, and the presented workflow might be applied to future investigations regarding their targetability.

Experimental Section

Materials

Plastic consumables for sample preparation were obtained from Applied Biosystems by Life Technologies (MicroAmp 8-tube Strip, REFN8010580 and MicroAmp 8-Cap-strip, REFN8010535). Sera-Mag SpeedBeads used in a 1:1 combination mix for SP3 protein clean-up were obtained from GE Healthcare, UK and prepared and maintained as described elsewhere.¹⁶ The DynaMag-96 Side magnet was purchased from Life Technologies, Oslo. Reagents for TMT labeling were obtained from Thermo Scientific (TMTsixplex Label Reagent Set, LOT#QL226165). Restriction enzymes were obtained from Thermo Scientific. Polymeric strong cation-exchange columns were obtained from Phenomenex, Part# 8B-S029-TAK. Unless specified otherwise, all other reagents were obtained from Sigma Aldrich.

Cell Culture

HCT116 GFP human epithelial colon carcinoma cells, constitutively expressing GFP (Anticancer 2009), were maintained in McCoy’s 5A modified medium (Sigma-Aldrich) supplemented with 10% fetal calf serum, 50 μg/mL streptomycin, 60 μg/mL penicillin, and 2 mM l-glutamine at 37 °C in 5% CO₂.

Monolayer Cultures

HCT116 GFP cells were seeded into 6-well plates (Nunc) at 300000 cells/well and cultivated in McCoy 5A medium for 24 h before drug treatment.

Spheroid Formation

Spheroids were formed from 5000 HCTS116 cells in 60 μL of fresh medium, plated into 384-well U-bottom ultralow attachment plates (Corning) using a Biomek 4000 liquid handling system (Beckman Coulter), and subsequently centrifuged at 200g for 5 min. Plates were covered with humidified MicroClime Environmental Microplate Lids (Labcyte) and cultured for 7 days at 37 °C in 5% CO₂. P-MCT culturing involved approximately 97% medium exchange with fresh medium on day 4 and 7. Q-MCT culturing did not involve any medium changes.

Drug Treatment

Monolayers, 24 h postseeding, and spheroids, 7 days postseeding, were treated with the OXPHOS inhibitors or simvastatin (at two different concentrations) or DMSO for 6 and 24 h (Figure 7A), respectively. Each treatment, time point, and concentration were performed as 28 replicates (single spheroids), which were pooled and frozen at −70 °C until preparation for MS analysis. Monolayer cells were washed with PBS after 6 or 24 h treatment, detached using Accutase, pelleted by centrifugation at 200g for 5 min, and frozen at −70 °C.

ATP Measurements in Spheroid Cultures (CellTiter Glo 3D)

Compounds were administered to spheroid cultures using Echo liquid handler 550. After 6 h incubation, the majority of the drug-containing medium was aspirated using ELx405 (BioTek). Then, 50 μL of AccuMax (Sigma) was added into each well, and plates were incubated at 37 °C for 30 min to allow spheroid dissociation. Subsequently, plates were centrifuged for 5 min at 200g, and the supernatant was aspirated using ELx405. The plates were then left at room temperature for 30 min in order to cool down. This was followed by the addition of 25 μL CellTiter-Glo 3D solution (Promega) into each well. Plates were then incubated at room temperature on an orbital shaker set to 900 rpm for 10 min to allow even disruption of cell pellets. Subsequently, plates were incubated at room temperature for 20 min, and luminescence was measured using automated plate reader FLUOstar OMEGA (BMG Labtech).

Immunological Staining

Spheroids were washed with PBS, fixed with 4% formalin in PBS, embedded in paraffin, and sectioned. Stainings for Ki67 performed using the Autostainer 480 (Thermo Fisher Scientific). Mouse anti-Ki67 clone MIB1 (#M7240) antibody was purchased from Dako AB. Slides were first incubated with Ultra V block (TA-125-UB, Thermo Fisher Scientific) for 5 min and thereafter incubated with primary antibodies at 1:1000 dilution for 30 min. The slides were then incubated with labeled horseradish peroxidase polymer for 30 min, followed by 3,3′-diaminobenzi-dine (DAB Quanto) solution for 5 min. Slides were counterstained in Mayer’s hematoxylin (01820; Histolab) for 5 min using Autostainer XL (Leica) and then rinsed in lithium carbonate water (diluted 1:5 from saturated solution) for 1 min. The slides were dehydrated in graded ethanol and coverslipped (PERTEX; Histolab) using an automated glass coverslipper (CV5030; Leica). The slides were scanned using the automated scanning system Aperio AT2 (Aperio Technologies).

Sample Preparation for MS Analysis

Cell Lysis, Digestion, and TMT Labeling

Cell pellets were resuspended in 200 μL of lysis buffer [50 mM triethylammonium bicarbonate (TEAB), 1% sodium dodecyl sulfate, and protease inhibitor] and subsequently incubated at 95 °C for 5 min. Samples were tip-sonicated at high frequency on ice with a tip-sonifier (Bandelin, Sonopuls UW 2070, Berlin) for 1.5 min in three 30 s increments until complete cell lysis was achieved. Samples were centrifuged in an Eppendorf benchtop centrifuge at 4 °C, 14,000g for 20 min. Protein concentration of the supernatants was measured by the bicinchoninic acid (Thermo Fisher) assay according to manufacturer’s instructions. Complete supernatants were reduced and alkylated with 10 mM dithiothreitol (DTT) at 45 °C for 30 min and 40 mM iodoacetamide at room temperature for 30 min in the dark. Following this, the alkylation reaction was quenched by a second reduction step resulting in a final DTT concentration of 20 mM for 20 min at 37 °C.

For TMT labeling and sample clean-up, paramagnetic beads were used, and sample processing was based on the method described by Hughes et al., 2014,¹⁶ as illustrated in Figure 7B. Briefly, samples were transferred to PCR strips, and 30 μg of protein was evaporated to dryness in a SpeedVac evaporator (SCANVAC, labogene). Proteins were resuspended in 8 μL of 50 mM TEAB buffer supplemented with 2 μL of magnetic bead mix and 20 μL of acetonitrile (ACN)/formic acid (FA) solution to obtain a final concentration of 66% ACN and 0.25% FA. After 8 min incubation at room temperature, followed by 2 min incubation on a magnetic rack, supernatants were removed. The bead-bound proteins were washed twice with 70% ethanol, once with ACN, and subsequently air-dried for 30 s. On-bead digestion was carried out with trypsin and Lys-C enzymes in 1:25 ratio in 25 μL of 50 mM TEAB buffer overnight at 37 °C. Supernatants were collected, and peptides were labeled with the TMT 6-plex reagent for 3 h at room temperature according to manufacturer’s specification. Digestion and labeling efficiency was tested on 16 randomly selected samples whereby approximately 95% zero-missed cleavages and 99% labeling efficiency were found. The labeling scheme is depicted in Figure 7C. The labeling reaction was quenched with Tris-HCL-buffer, pH 8 for 30 min at 37 °C, and samples were pooled resulting in 12 6-plex pools (five treatments in two concentrations and one DMSO control, three cell types, and two time points) for the treatment experiment. In order to compare baseline proteomes of the three cell types, one additional 6-plex pool containing vehicle-control samples of each cell type and time point was prepared, which will be further referred to as the control set. ACN excess was evaporated, and remaining reagents and detergents were removed by mixed-mode (reversed-phase cation exchange) solid-phase extraction (StrataXC, 8B-S029-TAK-TN). Samples were evaporated and stored at −20 °C until high-pH reversed-phase prefractionation.

Peptide Prefractionation by High-pH Reversed-Phase Liquid Chromatography

Peptide prefractionation was performed for labeled samples by high-pH reversed-phase liquid chromatography using an Agilent 1200 HPLC system. Samples were reconstituted in 80 μL of mobile phase A (20 mM HPLC-grade ammonia) and loaded on an XBridge BEH300 C18, 3.5 μm, 1 × 150 mm column (Waters, part number: 186003606). Over a 70 min gradient with gradually increasing mobile phase B (10 mM ammonia, 80% ACN) up to 100% at a flow rate of 0.12 mL/min, 96 fractions per sample were collected and subpooled into 21 fractions per sample and subsequently dried.

Nano LC–MS/MS Analysis

The labeled peptides dissolved in 20 and 4 μL were injected and separated using an Ultimate 3000 RSLCnano system (Dionex Ultimate-3000, Thermo Scientific, Sunnyvale CA, USA) set up in a trap and elute configuration. The peptides were trapped on a precolumn (75 μm × 20 mm 3 μm 100 Å C18 particles, Acclaim PepMap 100, Thermo Scientific) before eluting the peptides onto a 50 cm EasySpray 50 cm PepMap RSLC C-18 column (ES803, Thermo Scientific, Bremen, Germany) using LC gradient buffers consisting of A (3% ACN, 0.1% FA in LC–MS grade water) and B (96% ACN, 0.1% FA in LC–MS grade water). The buffer B gradually increased from 3 to 40% in 98 min and after that to 99% in the next 14 min. The column was equilibrated for an additional 11 min before the next sample was injected. Peptides were eluted at a flow rate of 250 nL/min and equilibrated at 350 nL/min. LC–MS data were acquired using a high-resolution quadrupole orbitrap mass spectrometer (Q-Exactive, Thermo Scientific, Bremen, Germany) in a data-dependent mode, where the top 10 most intense precursor ions were selected and fragmented using a normalized collision energy of 30. Full MS spectra were acquired at 70k resolution and fragmentation (MS/MS) spectra at 17.5k resolution.

Data Analysis

Raw Data Processing

Proteome Discoverer software (Thermo Fisher Scientific, version 2.1.0.81) was used for data analysis. UniProt human database (2015-09-24_uniprot-all_Human_42136_Can_Iso.fasta) was used in database searching. The RAW data file MS/MS spectra were filtered for the top 12 peaks per 100 Da mass windows and then subjected to SequestHT search and Percolator validation. Trypsin with full cleavage specificity and two maximum missed cleavage sites was specified for the search. Precursor and fragment mass tolerances were 15 ppm and 0.03 Da, respectively. Carbamidomethylation (C) and TMT 6plex (any N terminus, K) were set as static modifications. Methionine oxidation, protein N-terminal acetylation, and deamidation (NQ) were set as dynamic modifications. The database search protein identification acceptance criteria were, peptide-level FDR less than 1% based on Percolator scoring, a minimal number of peptide sequence per protein of 1 (rank 1 peptide only), and peptides were counted only in top scored proteins. The protein grouping was enabled, and the strict maximum parsimony principle was applied. For reporter ion-based quantification, Quan value corrections were enabled, and the coisolation threshold and average reporter S/N threshold were set to 50 and 10, respectively. The confidence threshold for protein FDR validation was set to 1%, total peptide amount was used for normalization, and obtained normalized abundances were used to determine alterations in expression based on TMT ratios: treatment/DMSO control (127-131)/126 or one entire control set, where the DMSO control of all cell types and time points were compared (127-131)/126.

Data Filtering

The data were loaded into the statistical software environment R v1.0.153. The protein list was filtered for high-protein FDR confidence, and a minimum of two unique peptides per identified protein was allowed, and isoforms and invalid values were removed. Finally, contaminants were removed using a contaminant list obtained from (http://lotus1.gwdg.de/mpg/mmbc/maxquant_input.nsf/7994124a4298328fc125748d0048fee2/$FILE/contaminants.fasta). For more information, see the Results and Discussion section, and protein lists are found in the Supporting Information.

Gene Set Enrichment Analysis

Normalized abundances of protein-coding genes were extracted for each DMSO-treated control, and drug-treated sample ratios were calculated, and rows containing invalid values were removed. Gene set enrichment analysis was performed using the default settings for pre-ranked enrichment analysis (GSEA 3.0) using Hallmarks (h.all.v6.1.symbols.gmt) as the database, NES scores were generated after 1000 permutations, statistics = classic.^17,18 For the identification of enriched pathways upon exclusively OXPHOS inhibitor treatment at the early 6 h time point, simvastatin treatment was removed from the data set, and differentially upregulated proteins with a fold change >0.6 were inserted to the online tool EnrichR.^19,20

Data and Software Availability

The MS proteomic data have been deposited to the ProteomeXchange Consortium via the PRIDE²¹ partner repository (http://www.ebi.ac.uk/pride/archive/) with the data set identifier PXD016495. Username: reviewer43746@ebi.ac.uk, Password: qRsMXAAd.

Glossary

Abbreviations

OXPHOS

oxidative phosphorylation

P-MCTS

proliferative multicellular tumor spheroids

Q-MCTS

quiescent multicellular tumor spheroids

SP3

single-pot solid phase-enhanced sample preparation

TMT

tandem mass tag

GFP

green fluorescence protein

DMSO

dimethyl sulfoxide

DTT

dithiothreitol

TEAB

triethylammonium bicarbonate

HPLC

high pH liquid chromatography

FDR

false discovery rate

LC–MS/MS

liquid chromatography tandem mass spectrometry

GSEA

gene set enrichment analysis

HMGCR

3-hydroxyl-3-methylglutaryl-coenzym A reductase

HMGCS

hydroxymethylglutaryl-coenzym A-synthase

MS

mass spectrometry

PCA

principal component analysis

ACN

acetonitrile

FA

formic acid

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsomega.0c01419.

• Immunohistochemical staining of Ki-67, data quality analysis, differences between the three cell types upon DMSO treatment, and pathways upregulated upon OXPHOS inhibition after 24 h treatment (PDF)

• GSEA CTRL data set (XLSX)

• Whole proteome obtained after PD database search (XLSX)

• Proteins filtered for a high-protein FDR confidence, at least two unique peptides and unique sequences. The table contains normalized abundances for each sample and TMT ratios for each treatment to respective DMSO control. Protein accession = 6025 (XLSX)

• Proteins filtered for a high-protein FDR confidence, at least two unique peptides and unique sequences. The table contains normalized abundances for the control set (DMSO controls for 6 and 24 h of all three cell types) after removal of any invalid values. Protein accession = 5965 (XLSX)

• Proteins filtered for a high-protein FDR confidence, at least two unique peptides. The table contains TMT ratios for each treatment to the respective DMSO control. Protein accession, including isoforms = 16,938 (XLSX)

Author Contributions

J.L., K.K., W.S., M.F., and J.S. contributed to study conception and design. W.S. cultured monolayer and spheroid cells and performed the treatment experiment. W.S. performed the ATP measurements. J.S., J.L., and J.R. performed the sample preparation. J.S., J.L., K.K., and E.O. performed HPLC prefractionation and nano-LC–MS/MS analysis. J.L. and J.S. performed the PD-based database search. J.S. and S.H. analyzed and visualized the data. J.S., W.S., K.K., and M.F. drafted the manuscript. The manuscript was critically revised through contributions of J.S., W.S., K.K., M.F., S.H., J.L., J.R., E.O., P.J.J., and R.L. All the authors have given approval to the final version of the manuscript.

(R.L., M.F., and W.S.) This study was supported by the Swedish Cancer Society, Swedish Research Council, the Lions Cancer Research Fund, (J.S and P.J.J.) the Else Kröner-Fresenius-Stiftung (Else Kröner-Fresenius-Graduiertenkolleg), the Swedish Cancer Society (grant no. CAN 2016/739), and the Swedish Research Council (grant nos. 2015-03955, 2017-01391, and 2017-02577).

The authors declare no competing financial interest.