Proteomic Analysis of Chinese Hamster Ovary Cells

Abstract

In order to complement the recent genomic sequencing of Chinese hamster ovary (CHO) cells, proteomic analysis was performed on CHO including the cellular proteome, secretome, and glycoproteome using tandem mass spectrometry (MS/MS) of multiple fractions obtained from gel electrophoresis, multi-dimensional liquid chromatography, and solid phase extraction of glycopeptides (SPEG). From the 120 different mass spectrometry analyses generating 682,097 MS/MS spectra, 93,548 unique peptide sequences were identified with at most a 0.02 false discovery rate (FDR). A total of 6164 grouped proteins were identified from both glycoproteome and proteome analysis, representing an 8-fold increase in the number of proteins currently identified in the CHO proteome. Furthermore, this is the first proteomic study done using CHO genome exclusively which provides for more accurate identification of proteins. From this analysis, the CHO codon frequency was determined and found to be distinct from humans, which will facilitate expression of human proteins in CHO cells. Analysis of the combined proteomic and mRNA data sets indicated the enrichment of a number of pathways including protein processing and apoptosis but depletion of proteins involved in steroid hormone and glycosphingolipid metabolism. 504 of the detected proteins included N-acetylation modifications and 1292 different proteins were observed to be N-glycosylated. This first large-scale proteomic analysis will enhance the knowledge base about CHO capabilities for recombinant expression and provide information useful in cell engineering efforts aimed at modifying CHO cellular functions.

Introduction

Chinese Hamster Ovary (CHO) cells are the primary hosts for the production of a large number of recombinant therapeutics and antibodies with over $99 billion in market value ¹. The CHO expression system is often preferred in bio-processing because of its manufacturing adaptability and post-translational modification capabilities. CHO cells can be adapted to suspension culture and scaled up to produce 5-10 g/L biologics ². Furthermore, the presence of post-translational modifications in CHO cells such as glycosylation compatible with humans increases the quality of the therapeutics and the biological life-time of pharmaceuticals in the circulatory system. For all these reasons, CHO cells are the most widely used hosts for industrial production of biopharmaceuticals ^2-3.

In order to enhance the production capabilities and efficiency of the host, an increased understanding of cellular physiology is desirable. Unfortunately, over the past decade, advances in CHO biology and biotechnology have been hampered by the lack of genomic and proteomic data. CHO proteomic studies have relied on finding homologous peptides in other organisms or using expressed sequence tags (ESTs), which have limited the proteome sizes to only a few hundred proteins ^4-5. However, the availability of high-throughput technologies has enabled more studies at the genome, transcriptome and proteome levels ^{2, 6}. With the recently published draft genome sequence of CHO-K1, complementary proteomic and transcriptomic studies can be deployed to provide further insights into its 24,383 genes ⁷ and 29,291 transcripts. Indeed, transcriptome sequencing of this cell line has demonstrated the mRNA expression of 11,099 genes, almost half of the predicted genes ⁷.

Given that only a fraction of the CHO genes were expressed, information on the translation of genomic information would be useful to complement the transcriptomic information and enhance our knowledge concerning cellular activities in CHO at the protein level. To accomplish this goal, a detailed proteomic study was undertaken here. Proteomics can be used to directly monitor the thousands of proteins that can play key roles in a CHO production host including those involved in growth, cell death, protein processing, glycosylation, and cell metabolism. Identification of the proteome will enable users to better understand the characteristics of this important production host as well as its capabilities for performing specific protein processing functions such as glycosylation, which can alter the half-life activity of the biologics in the circulatory system and impacts the protein function ⁸.

Together, the genomic, transcriptomic and proteomic data can elucidate key properties about this important biotechnology system including amino acid utilization patterns, highly expressed proteins, and depleted and enriched pathways ⁹. Indeed, since the correlation between the transcriptome and proteome is not always one-to-one, a combined approach is often useful to better understand the cellular physiology ^10-11.

Recent improvements in mass spectrometry technologies have enabled qualitative and quantitative identifications of proteins present in human, Drosophila melanogaster, Anopheles gambiae, mouse and many other organisms ^12-14. While there have been a few protein profiling studies of CHO cell lines ^15-17, the proteome of CHO cells has not been well characterized due to the lack of genomic sequence information. With recent genomic and transcriptional studies in CHO cells ⁷, it will be equally useful to identify the protein components of this production host.

In the current study, the proteome of CHO cells has been elucidated using multiple proteomics technologies and using two different search engines, TagRecon and MyriMatch^18-19. In order to identify as many proteins as possible in the proteome, different fractionation strategies and enrichment methods were used in combination with mass spectrometry (MS) analysis. To increase the proteome coverage, cellular proteins (cell lysate) and secreted proteins (media) were first separated using two dimensional liquid chromatography (2-D LC) techniques followed by MS/MS analysis of fractions as shown in Figure 1. Furthermore, the glycoproteome of CHO cell lines were identified by coupling solid phase extraction of glycosylated peptides with MS/MS ⁸.

Figure 1.

Flow diagram of the CHO proteome sample preparation and analysis. Spent media and cells were separated followed by cell lysis. Proteins were extracted from both sources and subjected to gel separation and in gel proteolysis (lysates) or in-solution digestion and bRPLC separation (lysates and secretome). These fractions were then subjected to LC/MS/MS and bioinformatics analysis. The glycosylated peptides were also analyzed by using SPEG coupled LC/MS/MS analysis.

The identification of expressed proteins has enabled assessment of the codon frequency among expressed proteins, an examination of the abundance levels of different protein categories, and an evaluation of the depleted and enriched pathways in CHO cells. While increasing our knowledge about the general physiology of CHO in culture, we also anticipate that this expanded proteomics information will be used to help understand CHO capabilities for recombinant protein production and lay the groundwork for cell and metabolic engineering efforts in CHO for years to come.

Experimental Procedures

Materials

The CHO-K1 (CCL-61) cell line was obtained from ATCC (Manassas, VA). Fetal bovine serum (FBS), L-Glutamine, nonessential amino acids, DPBS, F-12K and DMEM media were purchased from Gibco (Grand Island, NY). Sequencing grade trypsin was purchased from Promega (Madison, WI). The BCA protein assay kit was bought from Thermo Scientific Pierce (Rockford, IL). Affi-prep hydrazide resin beads and sodium periodate were from Bio-Rad (Hercules, CA). The other reagents were Tris (2-carboxyethyl) phosphine (TCEP) (Pierce, Rockford, IL), PNGaseF (New England Biolabs, Ipswich, MA), trifluoroethanol (TFE) (Sigma-Aldrich, Milwaukee, WI), 1 mL C18 Sep-Pak cartridges (Millipore, Billerica, MA), 4%-12% SDS-PAGE gel (Invitrogen, Grand Island, NY) and ultrafilters (Waters, Milford, MA). All other chemicals used in this study were purchased from Sigma-Aldrich (St. Louis, MO).

Cell Culture and Preparation Protein Lysate

CHO-K1 was cultured in F-12K media, supplemented with 10 % FBS, 1% non-essential amino acids and 2 mM L-glutamine. The cells at 70-80% confluence were used for proteomic analysis. The cells were washed with ice-cold phosphate buffered saline six times prior to cell lysis to obtain intracellular proteins. For the collection of secreted proteins, the cells were grown up to 80 % confluence, then the media was decanted and cells were carefully washed six times with DPBS. Subsequently, the cells were starved for 12 hours with serum free media. After 12 hours, media was collected and concentrated with 10 kDa molecular weight cut-offs to decrease the volume.

In-gel Digestion

The cells were lysed with 2 % SDS and sonicated on ice for three times for 20 seconds with a probe sonicator. The lysed sample was then heated at 90 °C for 5 minute after mixing with Laemmli Buffer. Approximately 400 μg of lysate was loaded into three different wells and separated using 1D SDS-PAGE gel according to molecular weight differences. An Invitrogen 1D gel runner was used with a Novex 4-12% gradient Bis-tris gel (1.5 mm thickness). The gel was then stained with Coomassie blue and rinsed with distilled water for about 15 minutes. The adjacent protein bands were merged and then cut into 27 pieces of 1×1mm². These gel pieces were cut into smaller pieces and put into an eppondorf tube followed by destaining, reduction and alkylation as described before ²⁰. Trypsin was added into each sample at a 1: 50 ratio and the trypsin digestion was carried out at 37°C overnight. Digestion efficiency was high as we observed few undigested trypsin cleavage sites (<10%). The digested peptides were then dried for mass spectrometry analysis ²⁰.

In-solution digestion

The cells were lysed in 9 M urea containing 20 mM NaHEPES buffer pH 8.0 and sonicated on ice for three times for 20 seconds with a probe sonicator. The protein concentration of the lysate was determined with BCA assay. The lysate was then reduced with DTT and alkylated with iodeacetamide using the filter aided sample preparation (FASP) method as described previously ²¹. Ultracel 30 kDa Amicon filters were used and 95% ultrafiltration efficiency was observed.

Trypsin was added into the solution at ratio of 1 to 50 and digestion was carried out overnight at 37 °C. After the removal of non-digested proteins with ultrafiltration, the sample was dried and analyzed by LC/MS or fractionated with basic pH reversed-phase liquid chromatography (bRPLC) into 96 fractions before injecting to the LC/MS.

The secreted proteins were also denatured, reduced and alkylated with FASP method as previously explained ²¹. Then the proteins were digested by trypsin enzyme (1:50 ratio) at 37 °C overnight. The non-digested peptides were removed with ultrafiltration and the sample was separated into 75 fractions with bRPLC. The fractions were combined into 12 fractions and analyzed in LC/MS.

Glycopeptide capture with SPEG

Cells were lysed with RIPA buffer (5% Triton-X-100, 5% Sodium Deoxycholate, 0.05% SDS, 790mM NaCl, 50mM Tris) and sonicated. The lysates (800 μg) were denatured, reduced, alkylated and digested as previously explained ²². Following tryptic digestion, the samples were desalted through C18 columns and were oxidized with 10 mM sodium periodate at room temperature in dark for 1 hour. The oxidized samples were mixed with 50 μL of (50% slurry) hydrazide resins overnight at room temperature for coupling reaction. Then, the non-glycosylated peptides were removed by washing the immobilized resins with 800 μL of 1.5 M NaCl and water. In order to release the N-glycosylated peptides from the immobilized support, the resins were suspended in 50 μL of G7 reaction buffer (50 mM sodium phosphate at pH 7.5) with 2 μL PNGaseF at 37 °C overnight shaking. The released formerly glycosylated peptides were transferred into a glass vial and the resins were washed twice with 100 μL of water. The washes and the supernatant were then combined in a glass vial and applied to a C18 SepPak column and the eluted samples were dried ²³. The samples were separated in strong cation exchange (SCX) column into 8 fractions and each fraction was analyzed in LC/MS.

Fractionation of peptides using basic pH reversed-phase liquid chromatography (bRPLC)

A bRPLC method was adapted from a protocol described by Wang et al. (2011) ²⁴ with minor modifications. The bRPLC fractionation of peptides was carried out using an XBridge C₁₈ analytical column (4.6 mm × 250 mm) and a guard column (4.6 × 20 mm), with a 5μm particle size from Waters (Milford, MA). The binding and equilibrating mobile phase (Solvent A) consisted of 10 mM triethyl ammonium bicarbonate and the elution mobile phase (Solvent B) consisted of 10 mM triethyl ammonium bicarbonate in 90% acetonitrile (ACN) with the flow rate of 1 mL per minute. The digested peptides were loaded on to the column after diluting (3-fold) in Solvent A. The gradient of acetonitrile used for separation of peptides was as follows: 10% gradient was reached in 10 minutes and 35% gradient was achieved between 10-50 minutes eluting majority of the peptides. This was followed by a sharp increase to 70% between 50-60 minutes and to 100% B between 60-70 minutes. Ninety six fractions were collected and concatenated into 48 fractions by merging 1, 48; 2, 49; 3, 50 and so on (Figure 1). On basic RPLC, peak width was less than 30 sec which allowed us to collect more fractions with the expectation of identifying unique peptides in each fraction. This also minimized co-isolation of peptides, so the MS/MS spectrum will be more unbiased. The samples were dried in a Speed Vac (Farmingdale, NY) equipped with refrigerated condensation trap.

LC-MS/MS Analysis

We carried out a total of 120 LC-MS/MS analyses of different fractions from CHO cell protein digests on an LTQ-Orbitrap Velos (Thermo Electron, Bremen, Germany) mass spectrometer interfaced with an Eksigent 2D nanoflow LC system. The reversed phase-LC system consisted of a peptide trap column (75 μm × 2 cm) and an analytical column (75 μm × 10 cm) both packed with Magic AQ C₁₈ material (5 μm, 120Å, www.michrom.com). Eluting peptides were sprayed directly into an LTQ Orbitrap Velos at 2.0 kV using an electrospray (i.d. 8 μm) emitter tip (New Objective, Woburn, MA) at a flow rate of 300 nl / min and the capillary temperature was set at 200°C. The entire data dependent tandem MS analysis was carried out in orbitrap in which precursor and the fragment ions were analyzed at 60,000 and 7,500 resolution (measured at m/z 400), respectively. FTMS full MS and MSn AGC target were set to 1 million and 50,000 ions respectively. Survey scans (full ms) were acquired from m/z 350-1,800 with up to 15 peptide masses (precursor ions) individually isolated with a 1.9 Da window and fragmented (MS/MS) using a collision energy of 35% in a HCD (higher collision dissociation) cell and 30 s dynamic exclusion. Minimum signal requirement for triggering an MS2 scan was set to 2,000 and the first mass value was fixed at m/z 140. An ambient air lock mass was set at m/z 371.10123 for real time calibration (ref: PMID: 16249172). For MS/MS analysis, monoisotopic precursor mass selection and rejection of singly charged ion criteria were enabled.

MS/MS Data Analysis

The MS/MS data were searched against the RefSeq annotation (downloaded October 28, 2011) from the newly established CHO genomic sequence, with each sequence appearing in both normal and reversed orientation. Searches described in this manuscript include MyriMatch 2.1.103 and TagRecon 1.4.57 ¹⁹, search engines that use the same final scoring system but that differ in their selection of candidate peptides for each spectrum (precursor mass versus sequence tag-based). Both tools were configured for semi-tryptic enzyme specificity allowing maximum 2 missed cleavages, with precursor ions required to fall within 10ppm of projected m/z values and the mass tolerance for fragments ions was 0.5 m/z. The variable modifications for proteome inventories included both oxidation (M +15.996) and pyroglutamine (N-terminal Q -17.027), while glycoproteome searches added deamidation (N +0.984). In all cases, carbamidomethylation (C +57.021) was set as a fixed modification. Identifications in pepXML format were processed in IDPicker 2.6 for organization and viewing. The majority of the fractions showed a similar number of peptide spectrum matches (∼10,000 PSMs in each fraction). A 2% FDR was applied to peptide-spectrum matches, computing FDR by doubling the count of reversed hits above a threshold and dividing by the total number of passing IDs. At the protein level, parsimony filtering was applied, reporting only the proteins that were supported by two distinct peptide sequences and requiring at least six spectra to be observed for each protein with 4.2 % FDR.

KEGG pathway analysis

For all identified proteins, KEGG ortholog identifiers were obtained from the CHO-K1 genome annotation, and their associated KEGG pathway assignments were obtained from the KEGG database. Enrichment and depletion of expressed proteins and mRNAs were assessed using the hypergeometric test.

Results

Identification of the whole cell CHO proteome is a key step in the detailed characterization of this important production host. Obtaining proteins from diverse sources using multiple separation and analytical methodologies will expand the number of peptides analyzed and proteins identified. For this reason, proteins from whole cell lysate and media and also enriched glycosylated proteins were separated using multiple chromatographic techniques. Strong cation exchange chromatography (SCX), reversed phase liquid chromatography (RPLC) and 1-D SDS gel electrophoresis were used as a first dimension fractionation method followed by RPLC as a second dimension because of its compatibility with MS ^24-25 as shown in Figure 1. Fractionation prior to LC/MS/MS increased the identification of the proteins since there were a reduced number of proteins in each fraction.

For analysis of proteins in the cell lysate, both protein and peptide separation techniques were used to increase the number of identified intracellular proteins. 400 μg of cell lysate was run on an SDS gel to separate proteins, after which the gel was cut into 27 pieces and subjected to in-gel digestion prior to LC/MS/MS analysis. For peptide fractionation, 1 mg of protein was proteolyzed in solution and then the digested peptides were separated into 96 fractions using bRPLC and pairs of fractions were then pooled to yield 48 fractions for LC/MS/MS analysis. For the secretome, the serum free media was collected and concentrated prior to trypsin digestion. The digested peptides were separated into 72 fractions using bRPLC and fractions were combined into 12 pools to be analyzed by LC/MS/MS. Since glycosylated proteins are in low abundance and often localized to the membrane, they may be difficult to analyze using the traditional proteomic strategies. For this reason, the glycosylated proteins were enriched with solid phase extraction of glycosylated peptides (SPEG) and then separated into 8 fractions with SCX before LC/MS/MS analysis. Coupling SPEG to mass spectrometry enabled both identification of the N-glycosylated proteins and their N-glycosylation sites. Including replicates, a total of 120 separate mass spectrometry analyses were performed on the CHO lysate, spent media and glycoproteome.

To map the identified spectra to the correct peptides and proteins, TagRecon ¹⁹ and MyriMatch ¹⁸ software were used. These algorithms both enabled identification of individual proteins to increase proteome coverage and confirmed the positive identifications obtained with different programs. The results obtained using TagRecon and MyriMatch software are discussed in more detail in the following sections and the proteins and the glycoproteins identified are listed in Supplementary Table 1 and Supplementary Table 2 with their accession IDs. These two search engines use the same final scoring system but differ in their selection of candidate peptides for each spectrum (precursor mass versus sequence tag-based).

The secretome and proteome experiments spanned 94 LC-MS/MS experiments from which 682,097 MS/MS spectra were identified, corresponding to 93,548 peptide sequences. Peptide-spectrum matches were required to meet thresholds producing a 2% FDR based on reversed sequence decoys. A parsimonious protein list was produced from the cell lysaste and secretome analysis, where two distinct peptides and at least six MS/MS scans were required to support each protein identified. These peptides enabled identification of 6084 proteins, forming 5782 discernible protein groups, with an empirical 4.2% FDR based on reversed hits. The experiments from the glycoproteome yielded an additional 29,525 identified MS/MS scans, matching to 5690 distinct peptide sequences. Filtering in this set applied thresholds to yield a 2% FDR at the peptide-spectrum match level and required two spectra to be identified to each protein, with a resulting empirical FDR of 0.6% for protein sequences distributed across 1292 discernible protein groups.

The number of distinct peptides used to identify the proteins is shown in Figure 2, in which a median value of 10 peptides were used to identify a unique protein. All catalogued proteins contained at least 2 distinct peptides while some of them, such as actin, included a sufficient number of unique peptides to cover 97% of their sequence. Out of the 682,097 identified tandem mass spectra from the proteome inventories, 345,140 were produced with bRPLC, 231,045 were produced using Gel/LC/MS, and the rest were split between single-dimension RPLC chromatographic runs and the secretome. 5,694 out of 5,782 total proteins were identified from bRPLC runs and 5006 were identified with Gel/LC/MS. Applying gel separation and chromatographic fractionation separately to the cell lysate proteins increased the number of proteins identified and helped to confirm identifications. Separate secretome analysis identified 1977 proteins. In addition to the global proteomic analysis of the cell lysate and media proteins, SPEG coupled SCX and LC/MS identified 5690 N-glycosylated peptides belonging to 1292 unique glycosylated proteins. All together with the glycoproteome and proteome analysis, 6164 unique proteins were identified as listed in Supplementary Table 1. Supplementary Table 1 includes all the accession IDs of identified proteins together with their FDR values, sequence coverage, identified peptides for each proteins, number of peptides and spectra belonging to each protein, GO annotations, KEGG pathway annotations and swiss-protein annotations. In addition, known proteins with the greatest homology at the genetic level were also noted in this Table. The CHO proteins were then grouped according to the species with greatest homology including mouse, rat, human or other mammalian such as sheep, orangutan or bovine. As shown in Figure 3, 51 % of the identified proteins were found with closest homologs in mouse followed by 20% having closest homologs in rat and 12 % close homologs in human. In addition, eleven percent of the proteins identified did not show homology to any mammalian species.

Figure 2.

Number of peptides identified for each protein.

Figure 3.

Percentage of identified CHO protein with greatest genetic sequence homology to Mouse, Rat, Human, None (do not have any homology) and other Mammalian such as Sheep, Orangutan or Bovine.

The raw mass spectrometry data is deposited to Tranche and can be reached at ProteomeCommons.org²⁶ Tranche using the hash “O0m+rpvfYIBhq6GI36JaAzaOclnXlGjPZ45tmEEiaUK+q4999kI43JM6Kmg9vTMhDTfN13D FXbnZStS5kOzCCIq+LqYAAAAAAAACYg==.” The identified proteins will also be available at the CHO genome website (www.chogenome.org)²⁷.

Two different search engines TagRecon and MyriMatch were used to assign spectra emerging from mass spectrometry using the recently accessible genomic sequence of CHO ^{7, 28}. IDPicker 2.6 software filtered the identifications and organized the TagRecon and Myrimatch results in a hierarchy ²⁹. In our study, 93% of the proteins from Gel/LC/MS and 95% of the proteins from RPLC/LC/MS were identified with both TagRecon and MyriMatch search engines. While TagRecon and MyriMatch share the same PSM scoring and database sequence processing functions, they differ in the strategies by which they determine which peptide sequences get compared to spectra. TagRecon required database peptides to match the sequence and both flanking masses of at least one of the 30 sequence tags inferred from each spectrum as described in detail by Dasari et al³⁰. MyriMatch required database peptides to match the mass of the precursor for a spectrum within 10 ppm as described in detail by Tabb et al³¹. As a result, TagRecon compared far fewer sequences to each spectrum, potentially achieving better sensitivity for spectra that contained reasonable ladders of fragments.

Codon Frequency

All amino acids other than methionine and tryptophan are encoded by at least 2 different synonymous codons. The usage frequency of these codons is different for each organism which is due to the characteristics of organism-specific isoaccepting tRNAs ³². The most frequently used codon for each amino acid is generally read by the most abundant isoaccepting tRNAs which facilitates rapid and high protein expression ³³. Since the population of isoaccepting tRNAs positively correlates with the codon choice, this leads to a codon frequency for each genome. In addition to the tRNAs, other factors also contribute to codon bias in mammals ³⁴. For instance, expression of GC rich genes is relatively higher due to efficient transcription and mRNA processing in mammals ³⁵ whereas AU richness facilitates degradation ³⁶.

In order to increase the expression of human proteins of interest in CHO cell lines, the codons of human proteins should be optimized to the proper codon frequency for CHO cells. Thus, the codon frequency of all experimentally detected proteins in CHO was found for each amino acid and compared to the human proteome. Shown in Figure 4 is the relative frequency for each codon together with a heat map showing the ratio of each codon's frequency in CHO cells to human. Although there are similarities in the codon frequency for CHO and human for some amino acids such as glutamine and lysine, the codon usage is also quite different for many others such as proline, threonine, alanine, aspartate, and cysteine. When generating human recombinant proteins, it may be preferable to replace codons that are sub-optimal for these amino acids in order to increase efficiency of expression.

Figure 4.

Codon frequency for CHO cells. The ratio of codon frequency in CHO cells to humans is indicated by the color of the codon boxes using the heat map scale.

Gene Ontology Analysis

Gene ontology (GO) annotation represents one way to standardize the biological functions of the genes ³⁷. As a result, both predicted genes from CHO genome annotations and proteins experimentally identified by mass spectrometry were functionally annotated with GO-slim terms. Statistical analysis of the experimentally identified proteins was performed based on all the predicted gene models and p-values for each GO-slim term calculated using Fisher's exact test ³⁸. These p-values are listed in Supplementary Table 3 for each GO-slim term. Supplementary Table 3 also includes the list of proteins categorized by their GO annotation terms as a table.

The significantly enriched groups were categorized into 17 main GO groups as shown in Figure 5. In addition to the whole proteome, the glycoproteome was also statistically analyzed and categorized into the functional GO groups. While the largest percentage of proteins are allocated to binding category in the proteome such as protein, RNA, DNA and chaperone binding molecules, the highest percentage of proteins in the glycoproteome are allocated to the cellular components such as N-acetylglucosaminyltransferases in the secretory compartment of the endoplasmic reticulum and golgi apparatus. Also, the glycoproteome has a larger percentage of proteins dedicated to functions such as signal transduction and cell signaling, and cell communication while the proteome has a greater percentage of proteins devoted to catalytic activity, and metabolic or biosynthetic processes.

Figure 5.

Significantly enriched functional groups of a) proteome, and b) glycoproteome. Identified proteins both in proteome and glycoproteome were functionally annotated with GO-slim terms and statistically enriched GO functions were categorized into 17 main groups.

Correlation between protein and mRNA abundance

Very little analysis has been performed concerning the relationship between the mRNA and protein abundance in CHO cells due to the lack of corresponding data sets. Mass spectrometry profiling provides information on the relative levels of proteins by providing the number of spectra assigned to each protein (spectral count) ³⁹. The abundance of the identified proteins is then estimated by normalizing the total spectral count for each protein with the length of the protein in order to define the spectral abundance factor (SAF). In order to normalize the variability between sample runs, each SAF value is divided by total SAF values in order to determine the normalized spectral abundance factor (NSAF) ⁴⁰. Using this approach, a relationship between the mRNA levels obtained by previous transcriptomic analysis and protein abundance was estimated for each gene ⁷. The used mRNA expression levels are standard normalized FPKM values which are fragments of reads mapped per kilobase of exon model, given in Supplementary Table 4. From a biological standpoint, less abundant mRNAs are expected to encode less abundant proteins ⁴¹. Therefore one may expect to have a linear correlation between the mRNA and protein levels. However this is often not the case because modifications and degradation rates of mRNA and proteins can be regulated at the post-transcriptional or post-translational level, respectively ^41-42. For example, host cells may shut off transcription of some genes and degrade the mRNAs while maintaining a stable protein to function for extended periods ⁴¹. In order to identify the stable and unstable proteins in CHO cell lines, the relative protein abundance (or intensity) of each protein identified was plotted against its corresponding mRNA intensity level as shown in the log₂-log₂ plot of Figure 6.

Figure 6.

A) mRNA and protein intensities for each identified protein. The normalized mRNA values, FPKM values versus normalized protein values NSAF values are plotted to figure out the significantly unstable and stable proteins.

The relation between mRNA and protein intensities was fitted using a least-square regression and the p-value was calculated based on a fitting error ⁴³. Statistical analysis of mRNA and protein intensities showed a high degree of correlation of 0.48 (p-value <1e-16) with a linear slope of 0.4.

Although the majority of the protein intensities exhibited a linear correlation with the mRNA levels on a log₂-log₂ scale, a significant number of proteins were also detected with high protein intensity and low mRNA levels, representing stable proteins. Conversely, unstable proteins showed relatively high mRNA and low protein intensity. In order to determine the significance of stability for each protein, the p-value of each protein based on its relative mRNA and protein levels was examined. Stable and unstable proteins with p-value <0.05 are listed in Supplementary Table 5.

Annotating the stable and unstable proteins with GO terms indicated which functional categories of proteins are more stable or unstable in CHO cells. Transport, signal transduction and cell signaling molecules such as protein kinases, and transcription related proteins including transcription factors, zinc fingers and cell cycle checkpoint control proteins were enriched in the unstable group; whereas, tubulin, myosin and binding proteins such as interleukin binding factors and ribonuclear proteins were enriched in the stable protein category and depleted in the unstable protein group.

KEGG Pathway Analysis

The identified proteins were annotated with the KEGG ids and these are tabulated Supplementary Table 1. When this data was combined with previous mRNA analysis, many metabolic pathways accounted for in the CHO genome were observed according to proteome, transcriptome, or both sources. However, this analysis showed significant enrichment or depletion in protein or mRNAs relative to the genome for various pathways. Particular pathways significantly depleted in the proteome and transciptome include glycosphingolipid biosynthesis, steroid hormone biosynthesis and primary bile acid biosynthesis.

Other pathways including protein processing and apoptosis shown in Figure 7 were enriched in proteins identified in the current proteome study. These pathways are of particular relevance to viability of CHO and its use as a host for the production of complex secreted recombinant proteins.

Figure 7.

Detailed mapping of A) Protein processing in endoplasmic reticulum B) Apoptosis pathways (Yellow: Genome, Red: mRNA and Protein, Orange : mRNA, Green: Protein).

Some of the genes such as Sec6263 and ERGIC53 in the protein processing pathway were detected at the protein level whereas they did not show any expression at the mRNA level. On the other hand, genes such as IAP, Bcl2 and FLIP in the apoptosis pathway were not observed at the proteome level but expression was detected at the mRNA level. Some additional genes such as CASP9, ASK1 and ERGL were also found to be silenced at both mRNA and protein levels.

Post translational Modifications: N-Glycosylation and N-acetylation

The glycopeptide capture technique provided further enrichment of N-linked glycosylated proteins and identification of 1292 glycoproteins. In particular, glycosyltransferases, membrane proteins, growth factors and apoptosis related proteins were found to be glycosylated as listed in Supplementary Table 2.

Another abundant post-translational modifications in eukaryotes is N-terminal acetylation which is carried out by N-acetyl transferases ⁴⁴. The N-terminal acetylation was set as a variable modification while searching the mass spectrometry data within the CHO database. 736 of the 1010 peptides identified with N-termini were found to be N-acetylated as listed in Supplementary Table 6. Alanine (38%) and methionine (38%) were the two major amino acids including the N-terminal acetylation modifications whereas 15% of the peptides had an acetylated serine residue.

Discussion

CHO cells are the most widely used hosts for expression of commercial proteins since they can perform post-translational modifications compatible with human patients and are readily adapted to suspension and fed-batch cultures for scale-up ⁴⁵. Although well-characterized technologies such as transfection, clone selection, media formulation and bioprocess optimization have been used to improve productivity, the cellular machinery itself is poorly characterized ⁴⁶. Cellular processes including gene transcription, translation, post-translational modifications, transport, metabolism, degradation, signaling, apoptosis, and many other cellular events all play important roles in cell growth and protein synthesis, processing, and secretion. Knowledge concerning cellular processes at the molecular level will increase our understanding of CHO as an important production host and enhance our potential capabilities for improving cell specific productivity of recombinant proteins ⁴⁶. This knowledge will also be helpful in future targeted cell engineering strategies aimed at altering functions involving protein processing, cell growth, glycosylation, apoptosis, and metabolic pathways.

Previously, Xu et al. (2011) identified 24,383 predicted genes and 11,099 transcripts in their genomic and transcriptomic studies of a CHO-K1 cell line ⁷. In the current study, multiple fractionation and enrichment methods were used to generate over 682,097 mass spectra that were mapped to the genome and transcriptome of CHO cell lines in order to identify a total of 6,164 grouped proteins. Integration of the proteome with the transcriptome and genome confirmed many of the sequences of predicted genes. For instance, 3689 of the predicted genes were shown to be expressed at both mRNA and protein levels. However, the method also identified genes currently detected at the protein level only. Indeed, 2,475 of the proteins identified in the mass spectrometry analysis were not detected at the mRNA level in the previous study ⁷. Furthermore, the average number of spectra used in the assignment of these 2,475 proteins was 6 and all of these proteins were identified using at least 2 unique peptides with a protein false discovery rate of 4.2%. These undetected mRNAs might be present in low abundance due to regulation of transcription or mRNA stability. Thus, this study represents the most comprehensive proteome study of CHO cell lines to date since it extends the number of identified proteins by more than a factor of 8 times compared to the previous studies ^47-49. Furthermore, this study represents the first large-scale high-throughput proteomics study to use the CHO genome to identify these 6,164 proteins. Previous efforts typically used rodent or mammalian databases to match the CHO mass spectra due to the lack of CHO genome data.

Three important protein groups from the cellular domain, secretome and glycoproteome were characterized in this study. Due to its complexity, two different fractionation techniques, 1D gel and bRPLC were used to separate cellular proteins and peptides, respectively. These two methods were able to identify 5006 and 5694 proteins, respectively. For the current study, in-solution digestion of proteins and peptides separation together with the bRPLC technique appears to be more efficient for protein identification as compared to separating proteins using a 1D gel followed by in-gel digestion ²⁴.

One of the biggest challenges in proteomics is the data analysis, which can be a limiting step for protein identification. For this reason, 2 different search engines, Myrimatch and TagRecon, were used here. In a previous study, MyriMatch was observed to be highly discriminative compared to other search engines ³¹. Nonetheless, more than 90% of the proteins were identified with both of the search engines. Together, MyriMatch and TagRecon algorithms were able to increase the proteome coverage and provide confirmation and consistency for the catalogued proteins.

Annotating the proteins with GO terms and statistical analysis provided information on which functional groups are enriched in the CHO proteome. In the CHO proteome, binding proteins such as protein binding, RNA binding, ion binding, DNA binding proteins and proteins involved in catalytic activity dominated over the other functional groups. Enrichment of glycosylated proteins provided for the identification of other functional groups including cell communication, signal transduction, cell signaling, and response to biotic stimulus. Many of the proteins in these secreted and glycosylated groups are cell adhesion molecules, receptors, ion transporters and other proteins often localized in membrane or extracellular matrix. Thus, the inclusion of the SPEG method has provided for the enrichment of these glycoproteins of the proteome ⁵⁰.

This study also allowed us to categorize proteins and mRNA according to abundance by tracking the number of mass spectra associated with a particular peptide and protein. There are multiple factors affecting mRNA and protein abundance. In additional to transcriptional control of gene expression by promoter, transcriptional regulatory elements, and transcriptional factors, a number of factors also regulate the stability of the mRNA. The factors that affect the level of the mRNA can be grouped into structural and sequential determinants. The structural determinants include the 3′ terminal stem loop ⁵¹, iron responsive elements ⁵², and the long range stem loop in the 3′ untranslated region ⁵³. Sequential determinants such as a Poly-A-tail ⁵⁴ stabilize the mRNA whereas an AU rich sequence in 3′ region ⁵⁵ causes degradation of mRNA. As a result, the half-life of mRNA can change from 20 minutes to over 100 hours ⁵⁶. On the other hand, protein intensity is affected by translational control and the stability of mRNA, folding, localization, post-translational modifications and other cellular processes ⁵⁷. Due to these reasons, it is difficult to provide a direct correlation between the selected mRNAs and the protein levels for a small set of data ⁵⁸. Nonetheless, Greenbaum et al. (2003) observed correlations between the mRNA and protein abundances in yeast although the significance of the correlations depended on the function and location of the proteins ⁴². In this study, we have also found a significant correlation between mRNA and protein levels with a p-value of <1e-16.

Some proteins such as binding and structural proteins including tubulin were found to have high protein abundance with low mRNA levels whereas others such as transcription factors were found have relatively high mRNA levels and low protein abundance. Certain classes of proteins are functionally and structurally important and should not be turned over rapidly for cellular processes. Furthermore, autoregulation mechanisms can help to destabilize of some of the RNAs. For example, when tubulin is synthesized above a threshold level, it triggers ribosome associated RNase to degrade its own mRNA. On the other hand, transcription factors have been found to be unstable ⁵⁶ and were also found in low quantities in our study. Although their mRNA levels were high, their protein levels were significantly lower. Indeed, previous studies have found that transcription factors are unstable due to the need for the cell to turnover these factors rapidly ⁵⁶. These proteins also contain relatively higher amounts of proline, aspartate, glutamate, threonine and serine which tend to be less stable ⁵⁶.

Previous studies have shown that codon bias is another important feature affecting protein expression levels in humans ^{34, 59}. Given that it is desirable to obtain the highest possible expression levels of human genes in CHO cells for biopharmaceutical applications, factors such as codon bias that can have a negative effect on the protein expression must be considered. ³⁴. Having the largest known collection of CHO proteins and their gene sequences available enabled a comprehensive assessment of CHO codon usage along with a comparison between the CHO and human codon frequency. This comparison did indeed indicate differences especially in the codon usage of proline, threonine, aspartate and cysteine. Furthermore, since aspartate, threonine and proline containing proteins are relatively unstable ⁵⁶, the codons for these amino acids should be changed for optimal synthesis of human proteins in CHO.

Codon bias is thought to arise from ancient evolutionary mutations, in which some tRNAs for individual codons in an organism became overexpressed. This overexpression led to a greater abundance of proteins made from mRNA with these preferred codons. Over time evolution favored these codons more than others, until they became the codon bias of the organism ³². Recent studies have also shown that other factors affecting mRNA stability such as GC content also play a major role in the codon bias of mammals ³⁴. Codon bias can also have adverse effects on heterelogous expression such as ribosomal frameshifting or pausing and cleavage of mRNA during co-translation ³⁴. Since human proteins are often being expressed in CHO cells, codon optimization of the genes before transfection into CHO cell lines will be highly relevant to enhancing its translational efficiency ⁶⁰.

KEGG pathway analysis indicated that mucin type or other types of glycosphingolipid biosynthesis, steroid hormone biosynthesis and primary bile acid biosynthesis steps were depleted in CHO cell lines relative to some other pathways. Although most of the genes responsible for these metabolic pathways were identified in the genome, their expression was not validated with mRNA or mass spectrometry analysis to suggest that some of these genes may be silenced in CHO cells. On the other hand, statistical analysis indicated that some pathways such as fatty acid metabolism, amino sugar and nucleotide sugar metabolism, which provide important precursors to recombinant protein synthesis, as well as protein processing and apoptosis, were enriched in CHO cell lines. Even for these enriched pathways, some important proteins were not detected. For example, Bcl-2 was not evident at the protein level to suggest its protein expression may be low in CHO-K1 cells. Interestingly, previous studies have shown that transfection of heterologous Bcl-2 can help to prevent apoptosis in CHO cells ^61-63, perhaps by overcoming limitations in the endogenous levels of this anti-apoptotic protein.

One major reason for using CHO cells as hosts for the recombinant protein expression is their capability to perform N-glycosylation. Indeed 1292 glycosylated proteins including neural cell adhesion molecules, growth factors, GPCRs, interleukin and transferrin receptors and glycosyltransferases were identified using the SPEG method. The glycoproteomic technique also identified highly abundant sialidases and glycosidases which can lead to glycan heterogeneity. These glycosidases and sialidases genes represent potential targets of opportunity for cell engineering in order to alter the glycan content of expressed glycoproteins from CHO cells. Examining our proteomic data also revealed that 87% of these N-terminally identified peptides contained N-acetylation. Previous studies have also shown that 80 to 90% of proteins are N-acetylated in mammals supporting our results ⁴⁴.

Conclusion

High-throughput proteomic analysis has been successfully applied to CHO cells using two dimensional fractionation coupled with RPLC-MS/MS. From more than 682,097 MS/MS spectra, 93,548 unique peptides were identified representing 5782 grouped proteins. Together with the glycoproteomic analysis, a total of 6164 proteins were identified. In addition, the codon frequency of CHO cells and its comparison to human were specified. Furthermore, statistical analysis on the mRNA and protein intensity levels identified proteins both over and under expressed in CHO cells relative to their mRNA level. Analysis on this proteomic data revealed enriched and depleted functional gene categories in CHO expression systems. The analysis of the pathways associated with these protein categories such as protein processing and apoptosis that were enriched or depleted in the proteome also indicated overexpressed or underexpressed genes which represent targets of opportunity for future cell engineering efforts aimed at altering CHO cell growth, metabolism, protein expression, glycosylation or other protein modifications for improving production of biopharmaceuticals of interest.

Supplementary Material

1_si_001. List of Supplementary Tables.

Supplementary Table 1: Identified proteins with their accession IDs, descriptive names, swissprot, GO and KEGG pathway annotations, FDR values, number of identified peptides, number of spectra belonging to the identified protein, coverage of the protein, cluster and group ID. Cluster and group IDs provide to find the protein of interest in the raw data deposited to Tranche.

Supplementary Table 2: Listing of glycosylated proteins together with their accession IDs and descriptive names.

Supplementary Table 3: GO analysis of all proteins and GO analysis of glycosylated proteins together with the proteins list according to their GO annotation terms.

Supplementary Table 4: Identified proteins with their descriptive names and their FPKM values.

Supplementary Table 5: Proteins specified as unstable and stable according to correlation between mRNA and protein intensities. The protein accession IDs together with their descriptive names and their SAF, NSAF and FPKM values are listed. A table showing the unstable and stable proteins is also included.

Supplementary Table 6: Listing of N-acetylated proteins and specific N-acetylation sites on these specific proteins.

Nomenclature

CHO

Chinese Hamster Ovary

MS

Mass Spectrometry

SPEG

Solid Phase Extraction of Glycopeptides

FDR

False Discovery Rate

EST

Expressed Sequence Tag

FASP

Filter Aided Sample Preparation

bRPLC

Basic Reversed-Phase Liquid Chromatography

LTQ

Linear Trap Quadrupole

FT

Fourier Transform

SCX

Strong Cation Exchange

GO

Gene Ontology

SAF

Spectral Abundance Factor

NSAF

Normalized Spectral Abundance Factor