Comparative microsomal proteomics of a model lung cancer cell line NCI-H23 reveals distinct differences between molecular profiles of 3D and 2D cultured cells

Lung cancer is the leading cause of cancer-related deaths in the USA and worldwide. Yet, about 95% of new drug candidates validated in preclinical phase eventually fail in clinical trials. Such a high attrition rate is attributed mostly to the inability of conventional two-dimensionally (2D) cultured cancer cells to mimic native three-dimensional (3D) growth of malignant cells in human tumors. To ascertain phenotypical differences between these two distinct culture conditions, we carried out a comparative proteomic analysis of a membrane fraction obtained from 3D- and 2D-cultured NSCLC model cell line NCI-H23. This analysis revealed a map of 1,166 (24%) protein species regulated in culture dependent manner, including differential regulation of a subset of cell surface-based CD molecules. We confirmed exclusive expression of CD99, CD146 and CD239 in 3D culture. Furthermore, label-free quantitation, targeting KRas proteoform-specific peptides, revealed upregulation of both wild type and monoallelic KRas4BG12C mutant at the surface of 3D cultured cells. In order to reduce the high attrition rate of new drug candidates, the results of this study strongly suggests exploiting base-line molecular profiling of a large number of patient-derived NSCLC cell lines grown in 2D and 3D culture, prior to actual drug candidate testing.


INTRODUCTION
Lung carcinoma is the deadliest cancer in the United States and worldwide [1]. In 2019, the American Cancer Society reported 142,670 deaths from lung cancer in the United States. Lung cancer affects men and women equally, with an overall five-year survival rate of 19%. Histologically, 80-85% of all cases belong to non-smallcell lung cancer (NSCLC). The most common form of NSCLC is the adenocarcinoma of the lung [2]. Despite extensive research and economic investments, about 95% of new drugs against lung carcinoma eventually fail in clinical trials [3,4]. While there are many hypotheses and explanations for this poor clinical translation rate, it is well accepted that conventionally used preclinical in vitro testing models (e.g., lung cancer cell lines) are incapable of reproducing the growth of malignant cells in human tumors in vivo. Hence, after the initial discovery Research Paper www.oncotarget.com phase, preclinical testing is heavily burdened by excessive proportion of false positive results that drive the high lung cancer drug attrition rate [5]. Adherent two-dimensional (2D) in vitro cultures of human cancer cell line are the mainstay preclinical testing model used in lung cancer drug development and discovery [6]. In solid tumors, however, cancer cells grow in a three-dimensional (3D) environment. This makes the 2D cultured cells unable to truly reproduce the natural proliferation, migration, and/or drug permeation taking place in their innate 3D environment [7]. To circumvent this shortcoming, the 3D cancer cell culture have been proposed as a more accurate/relevant preclinical testing models, not only biomechanically but also at the genome, proteome, and metabolome level [7][8][9]. Typically, preclinical assessment of drug effectiveness relies on functional cell assays, incidental imaging procedures, and histopathological examinations [10]. To better understand and standardize results obtained in the preclinical phase, it is of prime interest to better characterize and compare molecular phenotypes of lung cancer cells grown in 2D-and 3D-culture. Thus, there is an outstanding and unmet need for better characterization of preclinical lung cancer models to facilitate development of more effective predictive therapeutic biomarkers involved in molecular pathways driving lung cancer tumorigenesis [11]. This approach also creates a pathway towards the development of a preclinical atlas depicting the molecular profiles of cell lines grown in 2D vs. 3D, information that is sorely needed to accurately monitor drug development. Contemporary MS-based proteomics [12][13][14][15][16] represents a potent technology capable of revealing alterations in protein level expression/regulation and changes in posttranslational modifications associated with preclinical drug evaluation [17,18]. Furthermore, the differential molecular phenotype of a NSCLC cell line membrane proteome grown in 2D-vs. 3D-culture is still missing.
Herein we report results of differential shotgun membrane proteomic analysis of the microsomal fraction obtained from a NSCLC model cancer cell line grown in both 2D and 3D culture. Comparative proteomics that relies on the off-line strong cation exchange (SCX)-based fractionation of tryptic digests [19] and HR/AM LC-MS [20] was employed for in-depth mapping of the membrane proteome of the NCI-H23 cancer cell line. Next, spectral counting-based and label-free quantitative proteomics was utilized to elucidate differences in protein expression between cells grown in 2D and 3D culture.

Mapping microsomal proteome of 2D-and 3D-grown NCI-H23 cells using comparative proteomic analysis
The aim of this investigation was to map differences in protein expression between microsomal fractions obtained from 2D-and 3D-grown NCI-H23 lung cancer model cell line bearing heterozygous KRas-G12C mutant https://www.cancer.gov/research/key-initiatives/ ras/outreach/reference-reagents/cell-lines and elucidate biologically relevant alterations between these two cell culture conditions. Towards this goal, we used MS-based proteomics which has been established as an efficient approach for profiling of complex membrane protein mixtures [21][22][23]. The workflow is depicted in Figure  1. Briefly, tip sonication was used to disrupt intact cells and generate a microsomal fraction, typically comprising of double membrane-bound vesicles (e.g., plasma membrane, mitochondria, nuclei) and single membranebound vesicles (e.g., endoplasmic reticulum, lysosomes, Golgi apparatus). Finally, this crude microsomal fraction containing heterogenous membranous vesicles was isolated using ultracentrifugation [19]. Following tryptic digestion, the resulting peptide mixture was fractionated, and analyzed in duplicates using HR/AM LC-MS as previously described [19,23]. The analysis resulted in the identification of a total of 4,180 and 4,444 protein groups depicted in Supplementary Tables 1-2, along with subcellular location annotations currently accessible in the Human Proteome atlas, from a total 30,986 and 35,360 tryptic peptides (Supplementary Tables 3-4) identified in microsomal fraction of NCI-H23 cells grown in 3D and 2D culture, respectively. A total of 1,345 (i.e., 32.2%) and 1,404 (i.e., 31.6%) were classified as integral membrane proteins in 3D-and 2D-cultured NCI-H23 using PSORT and TMHMM prediction algorithms (Supplementary Tables 5-6). The efficiency of the isolation protocol is exemplified by the enrichment and unambiguous identification of microsomal fractionspecific enzymes, NADPH-cytochrome P450 reductase (POR), lanosterol 14-alpha demethylase (CYP51A1) and cytochrome P450 2S1 (CYP2S1) found in microsomal fraction of both 3D and 2D cultured cells (Supplementary  Tables 1-2). The search against the human CSPA found that 415 (i.e., 30.8%) and 434 (i.e., 30.9%) proteins identified in NCI-H23 cells grown in 3D and 2D culture, respectively, were cell surface proteins (Supplementary  Tables 7-8), previously unambiguously identified by mass-spectrometry in more than 40 human cell types and catalogued in CSPA [24]. Of these, 53 (i.e., 12.7%) and 51 (i.e., 11.7%) were classified as cluster of designation (CD) molecules identified in NCI-H23 cells grown in 3D and 2D culture, respectively (Supplementary Tables 7-8).

Classification of protein species identified in NCI-H23 cells grown in 3D and 2D culture
We used PANTHER classification system to examine protein functions, protein classes and corresponding pathways in NCI-H23 cells grown in 3D and 2D culture. The PANTHER analysis of differentially expressed proteins exhibited similar distribution of molecular functions involving mainly catalytic activity, binding, molecular regulation, and transporter activity (Supplementary Figure 1A-1B). Significantly, the protein class analysis revealed enrichment of cell adhesion molecules, intercellular signaling molecules, cell junction proteins, and extracellular matrix proteins in 3D-grown cells. 2D-cultured cells showed a slight increase in the identification rate of transporters, cytoskeletal and scaffold proteins (Figure 2A). Comparative pathway analysis revealed significant activation of the integrin signaling pathway and of general transcription regulation pathway in 3D cultured cells, while TGF-beta signaling and EGF receptor signaling pathways were found increasingly activated in 2D-cultured cells ( Figure 2B). Furthermore, p53 glucose deprivation pathway, glycolysis and hypoxia response via HIF activation were found activated exclusively in 3D cultured cells ( Figure 2B). While the findings of the PANTHER molecular enrichment analysis are not highly specific, the enrichment of extracellular matrix proteins, cell junction proteins, intercellular signaling molecules and cell adhesion molecules are in agreement with the expected tissue-like phenotypical transition taking place in 3D culture [25]. Correspondingly, the unique enrichment of p53 glucose deprivation, hypoxia response via HIF activation and glycolysis pathways in the context of the 3D cell culture are in agreement with widespread hypoxia that is typical for 3D culture and solid tumors [26].

Pathway analysis revealed differential activation/ enrichment of metabolic pathways and protein networks in culture dependent manner
To explore the biological significance of differential proteomic analysis and to prioritize and select crossvalidation targets, a subset of membrane proteins detected exclusively and/or found dysregulated in NCI-H23 cells grown in 3D-and 2D-culture were subjected to the Ingenuity ® Pathway Analysis (IPA ® ) [QIAGEN Redwood City, https://www.qiagen.com/ingenuity]. Canonical    whereas Organismal Survival (p-value range 6.93E-09-3.99E-09) was the most enriched in 2D cultured NCI-H23 cells (Supplementary Table 15B). Cluster of differentiation (CD) molecules at the cell surface play critical role in cancer immunology (e.g., tumor antigen recognition) [27], tumor biology (e.g., signal transduction and cell migration) [28], tumor microenvironment (e.g., angiogenesis) [29] and cancer immunotherapy therapy (e.g., non-Hodgkin lymphoma) [30]. Therefore, we focused specifically on CD molecules found differentially regulated in a culturedependent manner as depicted in the IPA's biological function analysis (Supplementary Table 15). Of these, CD87 (PLAUR), CD91 (LRP1), CD109 antigen, and CD280 (MRC2) were found to be involved in connective tissue growth and fibroblast migration in 3D cell culture (Supplementary Table 15A), whereas CD73 (NT5E), CD118 (LIFR) and CD221 (IGIFR) were found to involved in morbidity and organismal death in NCI-H23 cells grown in 2D culture (Supplementary Table 15B). The enrichment/activation of connective tissue development function and exclusive expression of CD87, CD91, CD109, and CD280 by NCI-H23 cells in 3D culture are in direct agreement with the ability of 3D cultured cells to phenotypically mimic and begin to recapitulate the tumor microenvironment [8].
Next, we explored results of the network analysis by aiming at the networks containing CD molecules. Unlike canonical pathway and/or disease/function analyses that group molecules based on known commonalities or what is accepted in the field, the network analysis groups clusters molecules into networks based on any direct or indirect biological relationship described in the literature [11,23,31]. Top identified networks are ranked by statistical significance and number of interacting proteins using Fisher's exact test [32].
Moreover, the network analysis also revealed that cancer and angiogenesis, network #2, cellular growth and proliferation, network #6, and connective tissue disorder and angiogenesis, network #17 represent overlapping networks in 3D cultured cells, indicating their interaction at functional, signaling and biological level. The merged network analysis ( Figure 4A) depicts PI3K complex, NFkB complex, P38 MAPK and IFNbeta as the most prominent signaling nodes, displaying direct and/or indirect interactions with all of the three CD molecules (i.e., CD99, CD146, CD239) we selected for validation. Taken together, these findings further corroborate the importance and distinctive culturedependent regulation of CD molecules in 3D-cultured NCI-H23 cancer cells.
Western blot analysis confirmed exclusive detection of CD146, CD99, and CD239 in 3D cultured NCI-H23 cells Based on results obtained using statistics, PANTHER bioinformatics, IPA analysis, their role in tumor biology (i.e., angiogenesis, proliferation, metastasis) and the availability of commercial WB antibodies, we selected CD146, CD99, and CD239, identified exclusively in 3D cultured cells, for orthogonal validation using WB.
The human CD146, also known as the melanoma cell adhesion molecule and/or the cell surface glycoprotein MUC18, is encoded by the MCAM gene [33]. CD146 plays a role in cell adhesion, particularly in vascular tissue. The human protein atlas (HPA) [34] showed that at the RNA level CD146 exhibits low specificity for normal/ healthy tissue and was detected in many organs including lung. However, at the protein level CD146 is detected highly expressed only in normal cerebral cortex, soft and adipose tissue. Medium expression was observed in normal kidney, adrenal gland placenta breast and tonsils. Interestingly, CD146 was not found expressed in normal lungs at the protein level [34].
In human cancers, CD146's RNA was detected in many malignancies with the highest levels found in melanoma and renal cancer, and lower level in lung cancer and colorectal cancer, according to the cancer genome atlas (TCGA). At the protein level, immunohistochemistry (IHC) revealed similar pattern depicted by distinct membranous CD146 staining in majority of patients with malignant melanomas and head and neck cancers, whereas lung cancer and endometrial cancer represent a smaller group with lower CD146 expression level [34].
In cell lines, in accordance to the human protein atlas (HPA) accessible consensus normalized expression ("NX") values, the highest level of CD146's RNA was observed in endothelial (i.e., HUVEC/TERT2, NXvalue: 125.3), skin (i.e., TIME, NX-value:122.3), and melanoma (i.e., WM-115, NX: 97.3) cells. Lower CD146 expression was found in sarcoma (i.e., U-2OS, NX: 17.9), breast cancer (i.e., EFO-21, NX: 10.7) and very low level in colon cancer (i.e., CACO-2, NX: 1.9) and lung cancer (i.e., SCLC-21H, NX 1.4; A569, NX:0.5). Interestingly, CD146 was found to play a critical role in the migration and proliferation in the human pulmonary large cell neuroendocrine carcinoma [35]. However, there is no record in HPA showing the CD146 expression in the NCI-H23 cells at the RNA and/or protein level [34]. The results of comparative WB analysis depicted in Figure 4B is concordant with the results of the LC-MS analysis showing exclusive identification of CD146 at the cell surface of 3D grown NCI-H23 cells. However, CD146 was previously detected by flow cytometry in A549 NSCLC cell line, harboring G12S homozygous KRas mutant, grown in 2D culture under oxidative stress [36]. For that reason, we wanted to ascertain the status of MCAM gene at the transcriptional level in 2D-cultured NCI-H23 cells. We carried out RNA ISH assay (i.e., RNAscope 2.5 HD) [37]. The analysis showed low to moderate level of MCAM expression at the RNA level in most of 2D cultured NCI-H23 cells (Supplementary Figure 13), suggesting that MCAM transcription is present in 2D culture.
The human CD99 antigen, also known as the protein MIC2, is a cell surface glycoprotein encoded by the pseudo-autosomal gene MIC2 [38]. At the RNA level, CD99 is detected in all tissues with highest in level brain, female tissues and muscles. However, at the protein level CD99 was detected highly expressed in normal bone marrow, lymphoid tissues, upper digestive tract, female tissues, and male tissues. Medium expression was observed in normal kidney, adrenal gland and skin, whereas lungs, adipose tissue and colon showed low expression level [34].
CD99's RNA has been detected in many malignancies in accordance to TCGA. CD99 showed the highest transcription level in malignant glioma and melanoma and lower level in lung cancer, pancreatic cancer, breast cancer and others. At the protein level, high CD99 expression has been observed in Ewing sarcoma, lymphoblastic lymphoma and malignant glioma whereas liver cancer and pancreatic cancer represent a smaller group with low CD99 expression [34].
In cell lines, the high level of CD99 at the transcriptional RNA level was observed in sarcoma (i.e., ASC diff, NX-value: 124.2), foreskin fibroblasts (i.e., BJ hTERT, NX-value:100.3), and malignant glioma (i.e., U138 MG, NX-value: 49.6) cells. Lower CD99 expression was found in lung cancers (i.e., A549, NXvalue: 27.4; HBEC3-KT, NX-value: 18.1), among others. Interestingly, there is no record in HPA showing the CD99 expression in the NCI-H23 cells at the RNA level [34]. The results of our WB analysis depicted in Figure  4B. are concordant with the results of the LC-MS analysis depicting exclusive identification of CD99 in 3D grown NCI-H23 cells.
The human CD239, also known as the basal cell adhesion molecule and/or the Lutheran antigen, is a cell surface glycoprotein which is encoded by the BCAM gene [39]. CD239 is a member of the immunoglobulin superfamily. It is a receptor for the extracellular matrix protein laminin, involved in tumor metastasis.
At the RNA level, CD239 has been detected in many tissues with highest expression in kidney, thyroid gland and female tissues. At the protein level CD239 has been found highly expressed in normal kidney, nasopharynx, fallopian tube and placenta. Medium expression was observed in normal lungs, colon, gallbladder, breast and heart muscle, whereas normal cerebellum, thyroid gland, duodenum, liver and pancreas show low CD239 expression levels [34].
CD239's RNA has been detected in many malignancies in accordance to TCGA. It showed the highest transcriptional level in ovarian cancer, endometrial and prostate cancer and lower level in lung cancer, breast cancer urothelial cancer and others. At the protein level, high CD239 expression has been observed in breast cancer, ovarian cancer prostate cancer whereas colorectal cancer, pancreatic cancer and urothelial cancer showed medium expression level. Lung, stomach and liver cancer represent a group with low CD239 expression [34].
Label-free quantitation confirmed upregulation of total Ras, K-Ras4B WT and K-Ras4B G12C mutant in 3D-cultured NCI-H23 cells Recently, CD239 was found to play a functional role in the metastatic spreading of monoallelic KRASmutant driven colorectal cancer by mediating interactions between the tumor and its microenvironment [40]. The expression and localization of CD239 was validated by immunohistochemistry in preclinical monoallelic KRAS mutant mouse model of hepatic metastasis [40]. In 3D-cultured NCI-H23 bearing the monoallelic KRAS mutant, we confirmed exclusive CD239 expression using WB. Notably, our spectral counting-based measurements utilizing total tryptic PSM counts showed significant upregulation of Ras proteoforms in a culture dependent manner (Supplementary Table 9). Therefore, we wanted to verify these measurements using label-free quantitation by the exclusive targeting of KRas proteoformsand allele specific tryptic peptides (i.e., KRas4B WT , QGVDDAFYTLVR; KRas4B G12C , LVVVGACGVGK), NRas (i.e., NRas WT , SFADINLYR) and a total Ras (i.e., Ras WT , LVVVGAVGVGK) shown in Supplementary Tables 3-4 [41]. Extracted ion chromatograms shown in Figure 5A-5C depict relative abundance of KRas4B WT , KRas4B G12C and total Ras. This analysis, unequivocally confirmed upregulation of KRas4B WT , KRas4B G12C and a total Ras WT in 3D-culture ( Figure 5D-5E). These results allow us to hypothesize that CD239 may have a role in metastatic spreading of NSCLC similar to the one in the context of colon cancer harboring monoallelic KRas mutants [40].

DISCUSSION
Lung cancer is the leading cause of cancer-related deaths in the USA and worldwide [1]. Yet, approximately 95% of new drug candidates validated in preclinical phase eventually fail in clinical trials [42]. This high failure rate is generally attributed to the inability of the commonly used two-dimensional (2D) cancer cell culture to recapitulate native three-dimensional (3D) growth of malignant cells in human tumors [5]. In particular, 2D culture does not mimic the natural 3D tumor microenvironment critical for tumor growth and survival [43]. Consequently, false positive compounds regularly qualify for clinical trials, leading to high attrition rates and a waste of time and money. To address these shortcomings, multiple preclinical testing models have been developed to better simulate the natural in vivo conditions. One such model is 3D cell culture that has been proposed as a more accurate and relevant preclinical testing model, not only biomechanically but also at the genomic, proteomic, and metabolomic level [25].
MS-based proteomics has been increasingly used in cell-surface profiling due to the limitation of genomic data to provide explicit information about cell surface protein expression level, actual post-translational modification status and/or explicit subcellular location of a given target [44]. Our laboratory has pioneered and applied proteomics for profiling of the cell-surface proteome in cell lines and tissue specimens [11, 19, 21-23, 45, 46]. Despite the importance of preclinical testing models for lung cancer drug target and biomarker discovery, there are no reports to date on indepth comparative proteomic profiling of any lung cancer cell lines grown in 3D vs. 2D culture. Recent advances in cancer immunotherapy involving the targeting of proteins at the surface of cancerous cells, further emphasize the need for comprehensive mapping of the lung cancer cell surface proteome and development of approaches which are capable of in-depth profiling of the surface of in vitro cultured cancer cells [23,47]. Our investigation begins to address

Figure 5: Label-free, gel-free, and antibody-free quantitation of Ras from complex mixture targeting exclusively proteoform-specific tryptic peptides detected in 3D-cultured and 2D-cultured NCI-H23 cells. (A) Extracted ion
chromatograms of the KRas4B WT proteoform-specific tryptic peptide QGVDDAFYTLVR identified in control (i.e., 2D cultured NCI-H23 cells) and the sample (i.e., 3D-cultured NCI-H23 cells). (B) Extracted ion chromatograms of the KRas4B G12C allele-specific N-terminal mutant tryptic peptide LVVVGACGVGK identified in control (i.e., 2D cultured NCI-H23 cells) and the sample (i.e., 3D-cultured NCI-H23 cells). (C) Extracted ion chromatogram of total Ras WT specific N-terminal LVVVGAVGVGK peptide identified in control (i.e., 2D cultured NCI-H23 cells) and the sample (i.e., 3D-cultured NCI-H23). (D) 3D and 2D quantitative channels readings for the total Ras WT specific N-terminal LVVVGAVGVGK peptide, the KRas4B WT proteoform-specific tryptic peptide QGVDDAFYTLVR and the KRas4B G12C allelespecific N-terminal mutant tryptic peptide LVVVGACGVGK, respectively. (E) Calculated 3D/2D ratios for total Ras WT , KRas4B WT , and KRas4B G12C proteoforms detected in membrane fraction of NCI-H23 cells grown in 3D and 2D culture, respectively. the basic shortage of differential molecular profiles/maps of preclinical lung cancer testing models with the ultimate goal to facilitate evidence-based drug target and biomarker testing and/or discovery.
To obtain a detailed map and differential molecular profile of the crude membrane proteome isolated from NSCLC preclinical model cell line, we applied shotgun proteomics to analyze microsomal fractions from the NCI-H23 cells grown in 3D-and 2D-culture. This analysis uncovered a total 334 protein species unique to 3D-cultured cells, 598 unique to 2D cultured cells and a total 234 common protein species found significantly dysregulated under both culture conditions, revealing a non-redundant map/catalogue containing a total 1,166 proteins regulated in a culture dependent manner (Supplementary Table 12), representing 24% of all nonredundant protein species identified in this study. This may explain why false positive compounds often enter clinical trials. Importantly, this molecular map provides detailed qualitative and distinct quantitative information on proteins differentially regulated in a culture-dependent manner. This catalogue enables reasonable selection and ranking of putative targets based on their detectability and relative abundance estimated by LC-MS.
Regardless of a cell culture type, cancer cells require a continuous supply of energy and structural components to support their growth and proliferation [48]. Due to persistent tumor hypoxia, cancer cells are forced to adapt their bioenergetic pathways to survive and proliferate in harsh hypoxic tumor microenvironment [49]. In comparisons to non-transformed healthy cells, transformed malignant cells possess remarkable metabolic flexibility characterized primarily by the ability upsurge glycolysis and repress oxidative phosphorylation (i.e., Warburg effect) [50], in order to secure cell survival and growth in hypoxia [49]. The principal regulator of intensified glycolysis is the hypoxia inducible factor 1 (HIF-1), a transcription factor responsible for increased transcription/ translation of glycolytic enzymes induced by hypoxia [51]. In comparison with 2D cultured cells, 3D-cultured cells are exposed to a much higher level of hypoxia and/or anoxia, featuring oxygen gradients which exist in the tumor microenvironment [52,53]. Accordingly, the PANTHER pathway analysis revealed activation of hypoxia response via HIF and glycolysis pathway in 3D cultured NCI-H23 cells ( Figure 2B). Correspondingly, the TGF-beta signaling was found highly activated in 2D culture, which is consistent with the role of this pathway in stimulating mitochondrial oxidative phosphorylation in the absence and/or low level of hypoxia present in 2D culture [52].
Also, it is widely recognized that tumors can switch to lipid-dependent metabolism for energy supply, employing adipogenesis/lipogenesis pathway for de novo synthesis of endogenous lipids and/or utilize the exogenous lipids from tumor microenvironment [54].
Significantly, lipogenesis-driven de novo synthesis of fatty acids (FA) is linked to oncogene signaling, responsible for endothelial cell recruitment and induction of tumor angiogenesis in response to hypoxia [55]. The IPA canonical pathway analysis, revealed the adipogenesis/lipogenesis pathway as the most enriched pathway in 3D cultured cells ( Figure 3A), suggesting that hypoxia-driven activation of de novo FA synthesis in 3D cultured NCI-H23 cells is taking place in order to initiate endothelial cell recruitment and induction of neo-angiogenesis [56]. In addition, canonical pathway analysis exposed the activation of HIF1-alpha signaling in 3D culture ( Figure 3A), indicative of hypoxia-driven glycolysis [51]. Significantly, it is now widely recognized that adipogenesis/lipogenesis pathway is closely linked to glycolytic pathway [57]. In 2D-cultured cells, the IPA canonical pathway analysis exposed the oxidative phosphorylation pathway as a top significantly enriched pathway ( Figure 3B). This is in agreement with findings showing that oxidative phosphorylation is predominant type of energy metabolism in lung cancer cell lines grown in 2D-culture [49]. Taken together, the canonical pathway analysis is indicative of the ability of shotgun proteomics to capture broad phenotypical changes, depicting differential activation of principal metabolic pathways, in a culture-dependent manner, indicating that 3D-cultured cells may be better than 2D-cultured cells in mirroring hypoxia, an important tumor feature that drives drug resistance in vivo [52].
The IPA disease and biological function analysis revealed connective tissue development as the top biological function (Supplementary Table 15), exemplified by the enrichment of multiple fibroblast functions and connective tissue growth, indicative of 3D cultured cells ability to sense 3D milieu and initiate phenotypical changes necessary for the creation of a tumor microenvironment that is critical for tumor survival [43]. Importantly, this analysis revealed a subset of CD molecules involved in this phenotypical transformation, showing exclusive expression of CD109 and CD146 in 3D culture, and upregulation of CD87, CD91 and CD280 in a culture dependent-manner (Table 1). This remarkable phenotypical transformation captured by shotgun membrane proteomics is equivalent to hypoxia-induced vascular mimicry in melanoma [58,59] that we previously investigated using similar approach [45].
Next, we focused specifically on differentially regulated CD molecules/antigens, primarily due to their importance in cancer diagnostics (i.e., cancer biomarkers) [60,61], cancer treatment (i.e., cancer immunotherapy) [62,63] and our interest in the cell surface proteomics of cancer cell lines [23] and tumor xenografts [11] harboring KRas mutants in the context of biomarker and drug target discovery [47]. We were the first show colocalization of CD147 (BSG) and CD318 (CDCP1) at the surface of cancer cell line expressing KRas4B mutant [23]. Importantly, subsequent investigations established CD318 as a viable target for radiological staging and treatment of pancreatic cancer [47,64].
In this study, we confirmed exclusive expression of CD146, CD99 and CD239 on the surface of 3D-grown NCI-H23 cells ( Figure 4B) and showed their potential interaction through the activation of protein networks facilitated by PI3K, NFkB, and IFN-beta signaling ( Figure  4A). A recent study focused on the evaluation of the KRas G12C direct covalent inhibitor ARS1620 using patientderived xenograft (PDX) models and found that intrinsic resistance of NSCLC to ARS1620 monotherapy, driven by nongenetic adaptive mechanisms, can be alleviated by concomitant application of PI3K inhibitors [65]. This study also showed that the PI3K pathway is not under the exclusive control of the KRas G12C mutant in NSCLC PDX models. Based on the IPA network analysis ( Figure  4A), and WB validation, it is tempting to speculate that CD146, CD99 and CD239 are downstream PI3K effectors involved in pro-angiogenic [66] malignant stroma [67] regulated metastasis [40] and may be explored as potential immunotherapeutic cell surface targets in the context of KRas G12C driven NSCLC.
In this investigation, both the LC-MS and the WB analysis unambiguously established the presence/ expression of CD146 in 3D-culture, whereas both LC-MS and WB were negative for the CD146 expression in 2D-cultured NCI-H23 cells ( Figure 4B). However, the RNAscope ISH assay was clearly positive for the presence of MCAM transcripts in 2D-cultured NCI-H23 cells (Supplementary Figure 12). It is well accepted that, with the exception for highly abundant housekeeping proteins, mRNA levels correlate poorly with protein level in cultured cell lines and tissues [68]. Furthermore, cell surface proteins show the worse correlation between protein and RNA abundance primarily because of complex posttranscriptional mechanisms controlling their expression (e.g., translation rate, degradation rate, transport) [69]. Importantly, our LC-MS and WB analyses were carried out using the membrane fraction, and it is possible that a very low level of ubiquitinated CD146 might exist in the cytosol, undergoing rapid degradation [68]. These results underscore the importance of direct proteomic analyses to identify/quantify deferentially expressed cell surface proteins in cancer cell lines or tumor tissue [44]. Significantly, a tissue based IHC study targeting CD146 showed that CD146 expression was clearly predictive of lymph node metastasis in patients with NSCLC (n = 118) [70]. Another tumor tissue based study, relying on well-characterized clinical cohorts, reveled CD99 as a novel clinically applicable NSCLC prognostic stromal marker [67]. Similarly, gene expression profiles obtained from 1106 NSCLC tissue specimens, revealed BCAM (CD239) as a part of stromal biomarker panel for early-stage NSCLC survival stratification [71].
Taken together, the results of these investigations are indicative of the ability of 3D-cultured cells to faithfully recapitulate tumor-like phenotype, and the efficacy of comparative shotgun proteomics in capturing these changes. Phenotypically these changes at the proteome level are consistent with stromalike transformation of epithelial NCI-H23 cells in 3D culture, where uniquely expressed CD molecules play important role. Specifically, CD146, CD99 and CD239 which may be considered as therapeutic targets and would not have been discovered employing conventional 2D-culture.
The NCI-H23 cells harbors KRas4B G12C homozygous mutant and could be used as a preclinical NSCLC model for testing allele-specific covalent inhibitors that bind to the cysteine at position 12 of the G12C KRas4B mutant [72]. In this study, we unambiguously showed that 3D culture induces upregulation of both wild type and mutant KRas4B allele in NCI-H23 cells ( Figure 5). Importantly, the present approach allows for direct proteoform-and allele-specific quantitation of changes in KRas WT and KRas G12C mutant regulation, and can be easily employed in quantifying responses to allele-specific covalent inhibitors or any other therapeutic compounds affecting Ras expression [73]. To our knowledge, this is the first report of gel-free and antibody-free bottom-up quantitative LC-MS analysis of KRas targeting exclusively proteoformspecific tryptic peptides. The shotgun proteomics is also amenable to absolute antibody-free quantitation of Ras isoforms using synthetic heavy isoform-specific tryptic peptide standards for LC-MS-based parallel reaction monitoring (PRM), we previously described targeting xenotropic and polytropic retrovirus receptor 1 (XPR1) in a complex membrane protein mixture [23]. Importantly, this approach allows for concomitant hypothesis-free proteomic profiling and the PRM-based quantitation applied on the same sample, as previously described [23]. This would greatly facilitates reduction in the variability associated with multiple sample preparations.

Cell culture
NCI-H23 cells were obtained from the American Type Culture Collection (ATCC, Rockville, MD). 2D culture was carried out in RPMI 1640 medium supplemented with 8 mM l-glutamine (GIBCO-BRL, Basel, Switzerland) and 5% fetal calf serum (FCS) (Fakola, Basel, Switzerland). 3D culture was carried out using global eukaryotic microcarriers platform from Global Cell Solutions (GCS, Charlottesville, Virginia) as previously described [74]. At the end of the culture process, both 2D and 3D grown cells were detached from the surface using trypsin. About 10 7 cells were collected and washed with PBS and stored at -80°C.

Microsomal fraction preparation, digestion, and fractionation
Microsomal fraction was isolated as previously described [23]. Briefly, each cell pellet was thawed in 50 mM NH 4 HCO 3 containing 1 mM PMSF. Cells were homogenized using tip sonication. After debris removal, the lysate was reduced using 3 mM TCEP and alkylated using 5 mM IAA. Following the ultracentrifugation at 100,000 × g for 1.5 h, pellets containing enriched microsomal fraction were resuspended using tip sonication in 25 mM NH 4 HCO 3 . Next, protein concentration was determined using the BCA Protein Assay Kit (Pierce). Microsomal fraction aliquots, 500 μg each, were lyophilized and then solubilized in 500 μL of 60% CH 3 OH/40% 25 mM NH 4 HCO 3 (v/v), containing 0.1% PPS, as previously described [21]. After tryptic digestion, samples were desalted, lyophilized and reconstituted in 200 μL of 45% acetonitrile containing 0.1% formic acid. Each digest was then fractionated using SCX chromatography (Supplementary Figure 13) and a total of 12 peptide fractions were collated, as previously described [19].

LC-MS and raw data analysis
SCX fractions were analyzed in duplicates using an EASY-nLC 1200 System (ThermoFisher Scientific) coupled on-line to an Orbitrap Elite mass spectrometer (ThermoFisher Scientific). Each SCX fraction was reconstituted in 0.1% TFA and 1 µg of peptides loaded onto an EASY-Spray, 25 cm long reversed-phase C18 column (ThermoFisher Scientific). After injection, peptides were eluted using a linear gradient starting with 2% mobile phase B (0.1% formic acid in ACN) to 40% solvent B (0.1% formic acid in 80% ACN). The mass spectrometer was operated in a data-dependent mode, using the peptide m/z range of 400−1800, monitored at the resolution of 60,000. Each MS scan was followed by 15 MS/MS scans, wherein the 15 most abundant precursor ions were dynamically selected for collisioninduced dissociation using normalized collision energy of 35%. Protein identification was carried out using the SEQUEST-based search against the non-redundant human proteome database (SwissProt release v57.15) utilizing the Proteome Discoverer 2.2 (ThermoFisher Scientific). For the monoisotopic peptide precursor ions (i.e., MS spectra), mass tolerance was set at 5 ppm; and for the fragment ions (i.e., MS/MS spectra), mass tolerance was set at 0.6 Da for fully tryptic peptides, allowing for up to two missed cleavages. Dynamic amino acid modifications were added for the detection of the following: +57.021 Da for carboxyamidomethylated cysteines, +15.994 Da for oxidized methionines and +0.984 deamidated asparagine/glutamine. A strict peptide false discovery rate (FDR) of ≤ 0.01 was set using Percolator-based statistical evaluation [75]. To increase the quality and reliability of protein identifications and enforce economy in the number of identified proteins, protein grouping was employed. Proteome Discoverer 2.2 was used for spectral counting based relative quantitation of changes in protein regulation between 2D and 3D cultured cells utilizing peptide-spectral matching (PSM) readouts computed by the Percolator algorithm [76]. The same software was used for the label-free quantitation of Ras targeting exclusively proteoform-specific peptides.

Statistics and bioinformatics
Significantly dysregulated proteins were revealed using binomial probability and FDR calculation [77]. PSORT (Horton P) and TMHMM (Krogh, A) algorithms were used to classify and characterize membrane proteins. To map cell surface proteins, we used the mass spectrometric-derived Cell Surface Protein Atlas (CSPA) [24] accessible at https://wlab.ethz.ch/cspa/. The PANTHER (protein annotation through evolutionary relationship) classification system (http://www.pantherdb. org/) was employed to classify significantly enriched protein species and corresponding pathways [78]. Finally, data were analyzed using literature-based Ingenuity Pathway Analysis IPA ® (QIAGEN Inc., https://www. qiagenbioinformatics.com/products/ingenuity-pathwayanalysis) to generate significantly enriched protein networks and select/prioritize cross-validation targets.

Western blot analysis
Cells were lysed in 25 mM of NH 4 HCO 3 buffer supplemented with 1 mM PMSF and homogenized by five cycles of 10-second sonication (20% intensity) using a Bronson microprobe tip-sonicator. The homogenate was centrifuged at 1000 × g for 10 min to remove unbroken cells and cellular debris. The supernatant was ultracentrifuged at (100,000 × g) for 1.5 hrs. using a Beckman MLS50 rotor (Brea, CA, USA). The membrane pellet was resolubilized in 25 mM of NH4HCO3 buffer, and the protein concentration of the solution was determined with the BCA Protein Assay Kit (Pierce). An equal amount of protein was run on SDS-PAGE (Life Technologies). Resolved proteins were transferred onto a 0.2 µm nitrocellulose membrane (Bio-Rad, Hercules, CA, USA), blocked with 5% non-fat dried milk in PBST (PBS with 0.05% Tween), incubated with primary Ab at 4 °C overnight, washed with PBST, and probed with HRPconjugated secondary Ab (Jackson ImmunoResearch, West Grove, PA). Immunoreactive bands were visualized by colorimetric detection using the Opti-4CN Substrate Kit (Bio-Rad). We used CD97 as WB control as well as membrane protein isolation reproducibility control since the LC-MS analysis showed no changes in CD97 expression between compared cell culture conditions.

Availability of data and materials
The datasets analyzed in the current study are accessible at the https://vmsshare.nist.gov.

CONCLUSIONS
In this study we generated a cell surface resource/ atlas of a preclinical testing model NCI-H23 cell line, providing insight into phenotypical changes at the proteome level unique to each culture type. This study provides for the first time an antibody-free and gel-free, proteomic method to directly quantify proteoformand allele-specific changes in KRas WT and KRas G12C mutant expression in complex protein mixture using comparative bottom-up proteomics. This resource describes protein species that are found significantly dysregulated in culture-dependent manner as well as proteins unique to 3D-and 2D-culture, which would not have been observed utilizing solely conventional 2D-culture. Capturing culture-dependent changes in protein expression is essential for providing a baseline to which cells can be accurately compared, followed by the application/testing a given drug candidate. In keeping with these concepts, we captured remarkable metabolic changes taking place in culture-dependent manner. Our proteomic resource also contains information on the difference in enrichment of biological functions and protein networks that can be mined to understand the transformation of NCI-H23 cells in 3D-culture consistent with activation of malignant tumor stroma. In particular, interactions between CD molecules mediating networks are key to regulating phenotypical changes under 3D culture. Among the candidate proteins selected for cross validation, unique expression of CD99, CD146 and CD239 in 3D-cultured cells are indicative of the development of malignant stroma (i.e., tumor microenvironment) induced angiogenesis and metastasis, likely triggered by increased hypoxia present in 3D-culture. Hence, our proteomic resource will provide a valuable set of protein targets for future studies and substantiate the advantage of using 3D-culture to bridge the gap between conventional 2D in vitro cultures and in vivo animal testing models.
The present proof of the principle proteomic strategy should be widely applicable to other patientderived cancer cell lines. Evidently, each patient-derived cancer cell line represents a phenotype of a single individual tumor/cancer. Therefore, it is critical to carry out comparative proteomic profiling of a large number of patient-derived cancer cell lines (e.g., NCI-60) grown in 2D-and 3D-culture in order to alleviate the heterogeneity issue and obtain comprehensive molecular profile/atlas of preclinical testing models, followed up by investigations focused on the differences ate the proteome level observed after application of selected drugs. Finally, phenotypical differences at the molecular level between 3D-and 2D-cultured NCI-H23 cells described in this investigation, underscores the importance of characterizing the properties of 3D cell models utilized in preclinical testing, in order to reduce high drug attrition rates. In fact, the use of 3D-cultured cells might be a necessary precursor, and perhaps an alternative, to animal models.

CONFLICTS OF INTEREST
Authors have no conflicts of interest to declare.