Dynamic Cellular Proteome Remodeling during SARS-CoV-2 Infection. Identification of Plasma Protein Readouts

Abstract

The outbreak of COVID-19, led to an ongoing pandemic with devastating consequences for the global economy and human health. With the global spread of SARS-CoV-2, multidisciplinary initiatives were launched to explore new diagnostic, therapeutic, and vaccination strategies. From this perspective, proteomics could help to understand the mechanisms associated with SARS-CoV-2 infection and to identify new therapeutic options. A TMT-based quantitative proteomics and phosphoproteomics analysis was performed to study the proteome remodeling of human lung alveolar cells expressing human ACE2 (A549-ACE2) after infection with SARS-CoV-2. Detectability and the prognostic value of selected proteins was analyzed by targeted PRM. A total of 6802 proteins and 6428 phospho-sites were identified in A549-ACE2 cells after infection with SARS-CoV-2. The differential proteins here identified revealed that A549-ACE2 cells undergo a time-dependent regulation of essential processes, delineating the precise intervention of the cellular machinery by the viral proteins. From this mechanistic background and by applying machine learning modeling, 29 differential proteins were selected and detected in the serum of COVID-19 patients, 14 of which showed promising prognostic capacity. Targeting these proteins and the protein kinases responsible for the reported phosphorylation changes may provide efficient alternative strategies for the clinical management of COVID-19.

Introduction

Coronaviruses (CoVs) are enveloped positive single-stranded RNA viruses that infect humans as well as other species, which makes these pathogens a health, veterinary and economic problem. Four genera can be distinguished among the Coronaviridae family, alpha and betacoronavirus that exclusively infect mammals and gamma and deltacoronavirus with a wider host tropism. CoVs are zoonotic agents infection by which usually correlates with respiratory and enteric diseases that may develop into a severe syndrome with respiratory complications and lung injury for which scarce therapeutic options are yet available.¹ Since the discovery of the first human coronavirus in mid-60s^2,3 large efforts have been dedicated to understanding the mechanistic principles of CoVs biology and pathogenesis. In particular, the SARS-CoV-2 pandemic urged the scientific community to look for new strategies to combat the severe respiratory syndrome frequently associated with COVID-19 disease. International initiatives, such as the COVID-19 MS Coalition⁴ or the “Global Health” Interdisciplinary Platform (PTI-CSIC) were among the collaborative efforts launched to fight COVID-19 pandemic.

Understanding the host-SARS-CoV-2 interaction appears as a priority to discover the mechanisms underlying the viral pathogenesis and to deduce new ways to combat COVID-19.⁴ SARS-CoV-2 enters the host cell upon the interaction of the Spike protein (S) with the ACE2 receptor at the cell surface through its receptor binding domain (RBD).^5,6 The S protein is then processed by the serine protease TMPRSS2 to allow fusion with the cell membrane followed by the penetration of the viral genomic RNA,¹ which is translated into polyproteins that are subsequently processed into smaller products (nonstructural proteins) by virus-encoded proteases. Extensive remodeling of the cell endoplasmic reticulum (ER) leads to the formation of double-membrane vesicles that host the synthesis of the viral RNAs that are translated into the structural and accessory proteins. Structural proteins and viral genomes assemble into new viral particles in the ER and are transported into the cell surface for exocytosis.⁷ In all these endeavors proteomics has emerged as a masterpiece providing unprecedented technical resources to study proteomes in their whole complexity, either biological fluids, cells, or tissues that complement other traditional approaches commonly used in biomedical research. Proteins are the main effectors of most cellular functions and constitute the principal intermediate module to transfer the information encoded in the genome into individual phenotypes. Therefore, understanding the dynamic reprogramming of the proteome upon SARS-CoV-2 infection is essential to understanding the cellular response, defining early diagnostic and prognostic methods, and developing effective therapeutic interventions. Since 2020, different studies were conducted to investigate the proteome rewiring induced by SARS-CoV-2 in different cell types. Despite the cellular diversity and the specific methods used across studies, central proteins driving the infection cycle and the host cell response were identified⁸⁻¹⁴

Changes in the serum proteome induced by SARS-CoV-2 infection have been widely investigated as they could provide a valuable readout of driver biological processes of COVID-19 progression and offer new opportunities to monitor and hopefully predict the severity of the disease. Studies using different cohorts and platforms have been reported including mass spectrometry and affinity reagents-based methods such as OLINK.¹⁵⁻¹⁹ Despite the expected biological heterogeneity and methodological diversity, a remarkable overlapping of differential proteins was reported, leading to the proposal of severity biomarkers that highlighted regulation of acute-phase response and inflammation, blood coagulation, lung and kidney damage, immune response, and complement cascade. The value of these proteins for the management of COVID-19 patients was further supported by the observation that these up or downregulated proteins returned to normal values 100 days after patient discharge.¹⁵

In this study, we have investigated the dynamic rewiring of the proteome and phosphoproteome induced by SARS-CoV-2 in the host cell to identify driver proteins of the cellular response to the infection. Ultimately, a selected panel of regulated target proteins has been verified for their suitability to stratify COVID-19 patients according to their disease severity.

Materials and Methods

Cells

Vero-E6 were obtained from ATCC (ATCC CRL-1586). Human lung adenocarcinoma cells were generated by transducing A549 cells (ATCC CCL-185) with a retroviral vector expressing the full-length human ACE2 cDNA²⁰ (ref A). Successful retroviral transduction confers resistance to blasticidine. Blasticidine resistant cell populations were expanded, aliquoted and frozen in liquid nitrogen.

Human lung adenocarcinoma cells (A549) expressing human ACE2 and Vero-E6 cells have been previously described. Cells were maintained complete media (Dulbecco-s Modified Eaglés medium; DMEM) supplemented with 10 mM HEPES, 1X nonessential amino acids (GIBCO), 100 U/mL penicillin–streptomycin (GIBCO) and 10% fetal bovine serum (FBS; heat-inactivated at 56 °C for 30 min). Unless otherwise stated, all infection experiments were performed at 37 °C in a CO2 incubator (5% CO2) the presence of 2% FBS in the absence of selection antibiotics.

Viruses

SARS-CoV-2 (Coronaviridae; Orthocoronavirinae; Betacoronavirus; Sarbecovirus; strain NL/2020) was provided by Dr. R. Molenkamp, Erasmus University Medical Center Rotterdam. SARS-CoV-2 stocks were produced in Vero-E6 cells by inoculation at a multiplicity of infection (MOI) of 0.001 TCID50/cell. Cell supernatants were harvested at 48 hpi, cleared by centrifugation, aliquoted, and stored at −80 °C. SARS-CoV-2 virus titers were determined by end point dilutions and immunofluorescence microscopy using an antibody against N protein (nucleoprotein), as previously described.²¹

Single Cycle Infection Experiments

A549-ACE2 cells were inoculated at a multiplicity of infection of 1FFU/cell (MOI 1) by incubation with infectious supernatants for 1 h at 37 °C. Inoculum was removed and cells were washed with warm PBS before replenishing the cells with complete media containing 2% FBS. Samples of the cells were collected at the indicated time points by scraping the cells in ice-cold 1X PBS after media removal and washing with ice-cold PBS.

Cell pellets were produced by centrifugation at 4500g/5 min/4 °C. Cell pellets were resuspended in lysis buffer (see below). All infection procedures were carried out following international guidelines in an authorized BSL3 facility under the supervision of the local Biosafety Committee. All activities were authorized by CSIC Ethics Committee and the Comisión Interministerial de Bioseguridad from the Spanish Government.

Cell Sample Preparation

A549-ACE2 cells (2 × 10⁶ cells per biological replicate) were lysed with 5% sodium dodecyl sulfate (SDS) and 25 mM triethylammonium bicarbonate (TEAB) supplemented with 10 mM tris(2-carboxyethyl)phosphine (TCEP) and 10 mM chloroacetamide (CAA). The samples were incubated at 60 °C for 1 h, followed by sonication using an ultrasonic processor UP50H (Hielscher Ultrasonics) for 1 min on ice (0.5 cycles, 100% amplitude). The protein extracts were centrifuged at 18,400g for 10 min and the supernatant was transferred to a new tube and quantified by PIERCE 660 nm reagent (Thermo Scientific) supplemented with ionic Detergent Compatibility Reagent (Thermo Scientific). Protein digestion on S-Trap columns (Protifi) was performed following the manufacturer’s instructions with minor changes.^22,23 Briefly, 80 μg of each sample were digested at 37 °C overnight using trypsin/protein ratio of 1:15. After digestion, tryptic peptides were quantified by fluorimetry using QuBit (Thermo Fisher Scientific), according to the manufacturer’s instructions, and 27 μg were labeled using Tandem Mass Tags (TMT)pro 16plex kit. For this purpose, tryptic peptides of each sample were resuspended with 25 mM TEAB buffer and 50% anhydrous acetonitrile (ACN) and labeled with distinct TMT label reagents (Figure S1). A pool of all digested samples (1.8 μg/each sample) was labeled with 134N to be included in the experiment as an internal standard (IS). After 2 h, TMT reaction was quenched with 0.3% hydroxylamine for 15 min at room temperature and, after that, the individual samples were combined in equal parts. For phosphopeptides enrichment, 270 μg were transferred to a new tube, dried in a speed vacuum, and frozen until further processing. In turn, 80 μg were kept for the fractionation using styrene Divinylbenzene reverse phase sulfonate (SDB-RPS) STAGE Tips.

Basic-pH Fractionation Using SDB-RPS STAGE Tips

The fractionation of TMT-labeled peptides was performed using in-homemade STAGE tips prepared from SDB-RPS solid-phase extraction disks (Empore) as previously reported.²⁴ SDB-RPS solid-phase extraction disks were packed into a 200 μL tip according to the STAGE tip protocol described by our research group.²³ The STAGE tip was inserted onto the top of a 2 mL tube using an in-homemade adapter, activated with 100 μL MeOH, and centrifuged at 900g for 3 min. The tip was then conditioned with steps of 100 μL 50% ACN and 0.1% formic acid (FA) and centrifuge, followed by three equilibration steps with 0.1% FA. After that, TMT-labeled peptides (80 μg) were reconstituted in 100 μL 1% FA (pH < 3) and loaded onto the STAGE tip. The sample was centrifuged at 900g for 5 min and the collected flow-through was loaded again to improve the peptide recovery yield. The STAGE tip was washed with 100 μL 0.1% FA, followed by 100 μL H2O. The elution was carried out using a 10-stepwise elution with 100 μL of 5 mM ammonium formate buffer and increasing acetonitrile concentrations (0, 5.0%, 7.5%, 10.0%, 12.5%, 15.0%, 17.5%, 20%, 25%, and 45%). Fractions were dried in a speed vacuum and frozen until further processing.

TiO2 Phosphopeptide Enrichment

Phosphopeptide enrichment was carried out as described previously.²⁵ The TiO2 slurry was prepared by mixing 10 mg Titansphere beads (10 μm, GL Sciences) with 80 μL of 1 M glycolic acid in 80% acetonitrile and 1% Trifluoroacetic acid (TFA) for 30 min at 25 °C. After being resuspended with 600 μL 60% ACN and 1% TFA, 250 μg TMT-labeled peptides were incubated at room temperature with a volume of TiO2 slurry corresponding to a TiO2:protein ratio of 10:1. Following a 25 min incubation period, the beads were spined down in a benchtop centrifuge and the flow-through was removed (nonphosphorylated peptides). The phosphopeptides bound to the titansphere beads were resuspended in 150 μL of 60% ACN and 1% TFA and transferred to a 200 μL tip with a 10 μm filter (MoBiTec). The wash step was repeated and the phosphopeptides were subsequently eluted using 5% ammonium hydroxide (NH4OH) (2x) and 25% ACN in 10% NH4OH (2x). The eluted fractions were pooled and acidified with TFA. Phosphopeptide enriched fraction was subsequently desalted using an in-house packed Oligo R3 reversed-phase microcolumn, dried and stored until further processing.

Analysis by Liquid Chromatography Coupled to Mass Spectrometry

Each sample (10 fractions or phosphopeptide enriched-fraction) was quantified by fluorimetry (QuBit) and 1 μg was individually analyzed by nano-Liquid Chromatography coupled to Electrospray Ionization Tandem Mass Spectrometry (nanoLC-ESI-MS/MS) analysis using an Ultimate 3000 nano HPLC system (Thermo Fisher Scientific) coupled online to an Orbitrap Exploris 240 mass spectrometer (Thermo Fisher Scientific). The fractions (1 μg in 5 μL of injection volume) were loaded on a 50 cm × 75 μm Easy-spray PepMap C18 analytical column at 45 °C and were separated at a flow rate of 300 nL/min using a 120 min gradient ranging from 2–95% mobile phase B (mobile phase A: 0.1% formic acid (FA); mobile phase B: 80% acetonitrile (ACN) in 0.1% FA). To avoid carry-over, two 40 min blank samples (mobile phase A) were systematically run between samples. Data acquisition was performed using a data-dependent top method, in full scan positive mode, scanning 375 to 1200 m/z. MS1scans were acquired at an Orbitrap resolution of 60,000 at m/z 200, with a normalized automatic gain control (AGC) target of 300%, a radio frequency (RF) lens of 80%, and an automatic maximum injection time (IT). The top 20 most intense ions from each MS1 scan were selected and fragmented with a Higher-energy collisional dissociation (HCD) of 30%. Resolution for HCD spectra was set to 45,000 at m/z 200, with an AGC target of 100% and an automatic maximum IT. Isolation of precursors was performed with an isolation window of 0.7 m/z and 45 s of exclusion duration. Precursor ions with single, unassigned, or six and higher charge states from fragmentation selection were excluded. For phosphopeptide analysis, HCD was set to 32% and the exclusion duration was decreased to 30 s.

Shotgun Data Analysis

Data obtained by mass spectrometry were analyzed with Proteome Discoverer (v2.5.0.400) using four search engines (Mascot (v2.7.0), MsAmanda (v2.4.0), MsFragger (v3.1.1), and Sequest HT) using a target/decoy Homo sapiens + SARS-CoV2 Uniprot Knowledgebase database (25th February 2021, 20,462 sequences) with the most common laboratory contaminants (cRAP database with 69 sequences). Search parameters were set as follows: cysteine carbamidomethyl (+57.021464 Da) and TMT6plex (+229.162932 Da) on lysine and N-term as fixed modifications; methionine oxidation (M) (+15.994915 Da), N-term acetylation (+42.010565 Da), and Gln → pyro-Glu (−17.026549 Da) as variable modifications. Precursor and fragment mass tolerances were set at 10 ppm and 0.02 Da respectively, and trypsin/P was selected as a protease with a maximum of 2 missed cleavage sites. The false discovery rate (FDR) for proteins, peptides, and peptide spectral matches (PSMs) peptides was kept at 1%. The quantitation was also performed in Proteome Discoverer using the “Reporter Ions Quantifier” feature in the quantification workflow using the following parameters: unique + razor peptides were used for quantitation, coisolation threshold was set at 50%, signal-to-noise of reporter ions was 10, and the normalization and scaling were performed considering the total peptide amount and the control (IS) average, respectively. The protein ratio was calculated considering the protein abundance and the hypothesis test was based on a t-test (background-based). Protein groups (master proteins) with an FDR lower than 1% and with abundance values in both IS were considered for quantitation. A p-value ≤0.05 adjusted using Benjamini-Hochberg was set to determine the proteins found differentially expressed at 3, 6, 9, and 16 h compared with the Mock-infected cells. Volcano plot and Principal Component Analysis (PCA) were performed in Proteome Discover considering the differentially expressed proteins for each comparison. Functional analysis of differentially expressed and phosphorylated proteins (peptides) was performed using Metascape and Coronascape.²⁶ Transcription Factor Enrichment Analysis (TFEA) and Kinase Enrichment Analysis (KEA) were performed using the Expression2Kinases webtool (X2Kweb; https://maayanlab.cloud/X2K/).^27,28

Feature Selection for Patient Stratification

Proteins whose presence in plasma has been already reported were retrieved from the total list of identified proteins according to the abundance information from paxdb (URL: https://pax-db.org/dataset/9606/1394854118/).²⁹ A classifier was then generated with random forest considering the 48 proteins to classify the different cell groups exposed to SARS-CoV-2, according to the following pipeline. Intensity data were normalized by Proteome Discoverer and then, log2 was calculated. All data were fed into a recursive random forest algorithm to rank the proteins by importance. Random forest was run for 30 rounds to eliminate iteratively those proteins with the lower important value, until only one protein was retained. Then, the data set was regenerated and the whole process of iterative variable elimination started again up to 50 times, accumulating the importance value from each iteration. Proteins were ranked according to the accumulated importance and the top 30 proteins were selected for further experimental testing in human samples (Table S5).

Serum Processing Using Automatic SP3-Based Protein Digestion

Serum samples were diluted with stock buffer to a final concentration of 2.5% SDS, 25 mM Triethylammonium bicarbonate (TEAB), 5 mM TCEP, and 10 mM chloroacetamide (CAA) and aliquoted into 8-well strips. Samples were incubated at 60 °C for 30 min in a Jitterbug 4 incubator (Fisher Scientific) for protein reduction and alkylation. For automatic SP3-based protein digestion, 50 μg of protein was processed in the OT-2 platform as previously reported.³⁰ Briefly, 100 μg of MagReSyn Amine beads (20 μg/μL suspension in 20% ethanol) were mixed with 30 μL of serum and 70 μL ACN. The protein-bead aggregation was carried out by two steps of 10 min incubation at room temperature without agitation. Protein-bead aggregates were washed three times with 100 μL ACN, followed by three washes with 70% ACN. For protein digestion, samples were incubated for 16 h at 37 °C with a mix of 1.5 μg Trypsin and 0.10 μg Lys-C (125–05061, WAKO) prepared in 25 mM TEAB. Following digestion, supernatants containing peptides were transferred to a new tube and acidified at a final concentration of 1% (v/v) formic acid. Peptides were dried in a speed vacuum and conserved at −20 °C until further processing.

Optimization of the Parallel Reaction Monitoring Method

Of the 30 proteins selected by random forest, a list of twenty-nine proteins to be monitored by parallel reaction monitoring method (PRM) was imported into Skyline v.22.2.0.255. The selection of 3 proteotypic peptides per protein was made considering the information on MS/MS spectral libraries in SRM Atlas and previous DDA experiments with plasma/serum according to the following criteria: peptide length between 7 and 25 residues, no missed cleavages, and exclusion of peptides containing methionine, histidine, and other amino acids susceptible to undergoing modifications (except for cysteine). To generate the experimental peptide library within Skyline, a serum pool containing synthetic peptides to be monitored (external standard) was analyzed by 3 PRM unscheduled submethods exported from Skyline v.22.2.0.255. For this purpose, 5 μL of sample (corresponding to 1 μg of peptide) were loaded on a 50 cm × 75 μm Easy-spray PepMap C18 analytical column at 45 °C using an Ultimate 3000 nano HPLC system (Thermo Fisher Scientific) coupled online to an Orbitrap Exploris 240 mass spectrometer (Thermo Fisher Scientific). Peptides were separated at a flow rate of 300 nL/min using a 60 min gradient ranging from 2–95% mobile phase B (mobile phase A: 0.1% formic acid (FA); mobile phase B: 80% acetonitrile (ACN) in 0.1% FA). Data acquisition was performed in PRM mode for monitoring the 86 peptides (approximately 29 peptides per method). The selected peptides were fragmented with a Higher-energy collisional dissociation (HCD) of 30%. The isolation window was set to 0.8 and the resolution to 30,000. The obtained data were searched against a target/decoy H. sapiens + SARS-CoV2 Uniprot Knowledgebase database (25th February 2021, 20,462 sequences) using Mascot (v2.7.0) within Proteome Discoverer (v2.5.0.400). The generated.msf and.pd files were imported into Skyline software for library generation. To optimize the PRM method, serum samples were analyzed using the same procedure to select the peptides that can be experimentally detected and quantified. The raw data was imported into Skyline software. Undetected peptides and proteins (not identified in proteome discoverer) or/and peptides with wide or unquantifiable peaks were removed from the analysis. The final method consisted of a scheduled method of 90 min for monitoring 61 peptides and 23 Proteins (Table S5).

PRM Analysis

Tryptic peptides were dried in a speed-vacuum system and resuspended at 200 ng/μL, according to QUBIT quantification (Thermofisher Scientific). Each sample (1 μg of peptides in 5 μL of injection volume) was loaded online on a C18 PepMap 300 μm I.D. 0.3 × 5 mm trapping column (5 μm, 100 Å, Thermo Scientific) and analyzed by nanoLC-ESI-MS/MS analysis using a Thermo Ultimate 3000 RSLC nanoUPLC coupled online to an Orbitrap Exploris 240 mass spectrometer (Thermo Fisher Scientific). Peptides were then separated on a 15 cm × 75 μm Easy-spray PepMap C18 analytical column at 45 °C with a flow rate of 300 nL/min using a 90 min gradient ranging from 2–95% mobile phase B (mobile phase A: 0.1% formic acid (FA); mobile phase B: 80% acetonitrile (ACN) in 0.1% FA). A pool (external standard) was run for each batch of 8 samples to monitor oscillations in the MS signal and in the retention time. The selected peptides (Table S2) were fragmented with an HCD of 30%. The isolation window was set to 0.8 and the resolution to 30,000.

PRM Data Analysis

Raw MS data were imported into Skyline and the automatically selected transitions were manually revised considering the intensity distribution of peaks from the fragmentation spectrum contained in the experimental peptide library. Peptides not detected in the most of samples (not identified in proteome discoverer) or/and with wide peaks were removed from the analysis. The peptide area was normalized by dividing the peptide area quantified in each sample by the area of the same peptide in the external standard analyzed in the same batch. The total normalized area of each peptide was calculated by summing all the normalized areas of all associated peptides (each one resulting from the sum of all transitions). Statistical analysis by one-way ANOVA (Tukey’s HSD) & posthoc tests.

Machine Learning Modeling of PRM Results

All calculations were performed in the R environment, version 4.1.3. All random forest-dependent results were obtained through the “randomForest” library v4.7–1.1. The “caret” library v6.0–93 was used to create the machine learning models. The “MASS” package v7.3–58.1 was used to generate the linear discriminant analysis (LDA) models. The “e1071” library v1.7–13 was used to build the support vector machine (SVM) classifiers. All SVMs created use a radial kernel. The hyperparameters of all classification models were selected by grid search of C and sigma values. The verification package v1.42, was used to create the ROC (receiver operating characteristic) curves. These were generated based on the probability values of the samples assigned to each class as a vector, plus a vector of information about the actual class of each sample. These vectors were entered into the “roc.plot” function to generate the ROC curve.

Results

Identification of Regulated Proteins in A549-ACE2 after Exposure to SARS-CoV-2

To elucidate the mechanisms underlying the response of A549-ACE2 cells to SARS-CoV-2 infection, we have performed a differential proteomic and phosphoproteomic analysis of A549-ACE2 at different times after infection with SARS-CoV-2 (Figure S1). A549-ACE cells were inoculated with SARS-CoV-2 in a single cycle infection setup (MOI 1). Total RNA samples were collected to determine the viral load at different times postinfection by RT-qPCR. Figure 1A shows that viral RNA steadily accumulates over time, reaching a plateau between 6 to 9 h postinoculation. Consistent with these results, infectivity titers in the supernatants reached maximum values at 16 h (Figure 1B), coinciding with the time points at which maximal RNA accumulation is achieved. Viral load correlates with a progressive accumulation of N protein in the majority of the cells, reaching over 95% positive cells at 16 hpi, as determined by immunofluorescence microscopy (Figure 1C).

Figure 1.

SARS-CoV-2 infection kinetics in A549-ACE2 cells: A549-ACE2 cells were inoculated with SARS-CoV-2 (strain NL/2020; MOI 1). Samples of cells and supernatants were collected at 3, 6, 9, and 16 h post infection (h.p.i.) to determine relative intracellular genomic RNA levels (panel A) normalized to a cellular housekeeping RNA (28S), extracellular infectivity titers in focus forming units per mL (panel B) and intracellular viral antigen staining (panel C) as described in the Methods section. Data in panels A and B are presented as mean and standard deviation of three biological replicates (n = 3). Dotted lines in panels A and B indicate the limit of detection (LOD) determined using mock-infected cell samples as controls (0 hpi). The inset in panel C shows the percentage of N-protein positive cells at each time point.

Altogether, 6766 protein groups were identified with an FDR ≤1%, from which 6712 were quantified (Table S1), resulting in 91, 115, 261, and 417 regulated proteins (adjusted p-value ≤5%) after 3, 6, 9, and 16 h of exposure to SARS-CoV-2, respectively (Figure 2A and Table S2). Moreover, 10 viral protein groups were also identified, which accumulate over time, except for nonstructural protein 8, which reaches a plateau at 9 h postinfection (Figure 2C). Upon TiO2 enrichment, 8223 peptides were identified, corresponding to 2762 protein groups (FDR ≤ 1%), revealing 6428 phosphorylation sites with 100% confidence. Among them, 136, 324, 687, and 902 peptides (adjusted p-value ≤5%) displayed a differential pattern at 3, 6, 9, and 16 h postinfection, respectively (Figure 2A and Table S2). The overlapping between the panels of regulated proteins by changes in the abundance or phosphorylation is negligible, indicating that the differential phosphorylation events cannot be explained by protein abundance changes (Figure S2). Both differential proteins and phosphoproteins efficiently segregated the cell groups at different viral cycle stages, likely reflecting the dynamics of the cellular process involved (Figure 2B). Besides cellular proteins, three SARS-CoV-2 proteins were phosphorylated peaking 6–9 h after cell exposure to the virus, coinciding with maximum replication periods (Figure 1), namely SARS-CoV-2 replicase, membrane protein, and nucleoprotein (Figure 2D). Interestingly, two new phosphorylation sites (S410 and S416) located at the C-terminal peptide of the SARS-CoV-2 nucleoprotein (QLQQSMSSADSTQA) are described here for the first time (Figures 2E and S3 and S4).

Figure 2.

Changes in the host and SARS-CoV-2 proteome and phosphoproteome detected in the analysis of A549-ACE2 cells (n = 3) harvested at 3, 6, 9, and 16 h postinfection compared to mock-infected cells (3 h). (A) Number of host (human) and SARS-CoV-2 proteins and phosphopeptides showing a differential pattern at 3, 6, 9, and 16 h post infection. (B) Principal component analysis (PCA) shows that these proteins/phosphopeptides clearly discriminate between different stages of the viral cycle, reflecting host cell dynamics during viral infection. (C) Changes in the levels of SARS-CoV-2 proteins after infection, showing an accumulation of all these proteins over time, except for nonstructural protein 8 (D) Phosphorylation dynamics of SARS-CoV-2 proteins including replicase, membrane protein and nucleoprotein during viral infection. (E) Annotated fragmentation spectrum representing a novel phosphorylation site (S416) located on the C-terminal peptide (QLQQSMSSADSTQA) of the SARS-CoV-2 nucleoprotein (P0DTC9 [406–419]).

Functional Analysis of A549-ACE2 Differential Proteins along SARS-CoV-2 Infection

A functional analysis was performed to infer the cellular processes in which the differential proteins and phosphoproteins are involved (Figure 3 and Table S3). Overall, formation of the cornified envelope, hemostasis, RNA metabolism and processing, transcription and translation regulation, protein transport and folding, posttranslational protein phosphorylation, ribosome biogenesis, cell cycle, signaling (e.g., Rho GTPases, in good agremment with previous observations,³¹ and microtubule/extracellular matrix organization, are the top-rank regulated cellular processes (Figure 3A,B). These observations fit well with previous studies that describe how SARS-CoV-2 controls the host cell biology to ensure a successful infection cycle^10,12,13,32 (Figures S5 and S6). It is also interesting to note that, although not many processes are enriched at the proteomic level, several cellular processes associated with changes in phosphorylation are activated 3 h postinfection.

Figure 3.

Functional enrichment of the host proteins and phosphoproteins (phosphopeptides) showing a differential pattern upon SARS-CoV-2 infection compared to mock-infected cells (3 h). According to the Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), Reactome, and WikiPathways databases, heatmaps generated using Metascape display the common and/or unique biological pathways associated with the (A) proteins or (B) phosphopeptides found to be differentially expressed at each infection time point compared to mock-infected cells. (C) Transcription Factor Enrichment Analysis (TFEA) and (D) Kinase Enrichment Analysis (KEA) were performed using the Expression2Kinases webtool of the peptides with differential phosphorylation at 3 h postinfection compared to mock-infected cells.

To define upstream regulators that might drive the phosphoproteome changes induced by SARS-CoV-2 in A549-ACE2 cells, a Transcription Factor Enrichment Analysis (TFEA) was used employing the X2Kweb pipeline (Figure S7). The analysis suggested an early activation of the transcriptional machinery as expected from the viral requirements for protein synthesis (ATF2, TAF1, YY1, MYC, PML), pointing to the viral control of the cellular apoptotic response through BCLAF1 at 3 h postinfection (Figure 3C). Similarly, a Kinase Enrichment Analysis (KEA) of differential phosphopeptides at each time was performed to identify potential kinases whose regulation might lead to the observed changes in protein phosphorylation (Figures S7 and 3D). In agreement with previous reports, multiple kinases are activated upon SARS-CoV-2 infection of A549-ACE2 cells^12,33 which highlights a significant reprogramming of the cellular signaling landscape. Among them, CK2 (CSNK2A1) and CDKs may contribute to the regulation of viral replication (Figures 3D and S7). It is worth noting that there is no mock for the late time points when more cell growth would have occurred so these observations may also reflect a change in cellular confluence.

In order to investigate the dynamics of the cellular processes regulated along the viral infection, the differential proteins and phosphoproteins were first grouped into clusters according to their variation over time. The number of clusters was selected based on their specific time-dependent profile, which was estimated by the minimum distance between cluster centroids (Figure S8). According to this criterium, four and six clusters were considered for the differential proteomics (DP clusters) and phosphoproteomics (DPP clusters) analyses, respectively (Figures 4A and 5A). Then, the regulated pathways within each cluster were inferred by functional enrichment analysis (Table S4). The top 10 cellular processes regulated by proteins and phosphoproteins of each cluster are shown in Figures 4B and 5B respectively and the specific proteins involved in each case are listed in the Table S4. DP3 cluster represents proteins that increase transiently at 3 h postinfection. Functional analysis revealed an early inflammatory reaction (α2 macroglobulin), deep remodeling of keratin filaments (several keratin species and filaggrin), and response to cellular damage by cell cycle arrest (HSU1). Clusters DP1 and DPP3/6 include proteins showing a steady accumulation or phosphorylation along the infection. As might be expected viral proteins are included in these clusters since they accumulate as the infection cycle progresses until virions are ensembled and released. This has also been verified for the phosphorylation of viral proteins, although the functional outcome of these modifications is not yet clear. Functional enrichment analysis of upregulated cellular proteins suggests suppression of IFN signaling, as well as cell division and mitosis (NCAPH, CDK1, CDK7, KIF11, BRSK2) and microtubule movement (Kinesines, dynein).

Figure 4.

Functional enrichment based on protein expression dynamics in proteome analysis: (A) Mfuzz clustering was applied to define 4 distinct protein clusters with common expression dynamics over the time course of SARS-CoV-2 infection. (B) Functional analysis was performed to assess the top biological processes associated with these Mfuzz-based clusters using the Gene Ontology (GO) database.

Figure 5.

Functional enrichment based on phosphopeptide expression dynamics in phosphoproteome analysis: (A) Mfuzz clustering was applied to define 6 distinct phosphoprotein clusters with common expression dynamics over the time course of SARS-CoV-2 infection. (B) Functional analysis was performed to assess the top biological processes associated with these Mfuzz-based clusters using the Gene Ontology (GO) database.

Other proteins are involved in cellular process that are known to be regulated by SARS-CoV-2, including autophagy (SQSTM1), histone remodeling (H2A, H2B, H1.2), regulation of Ca²⁺ metabolism (CALM3, CARHSP1), ER stress (KDELR1, ATP13A1, EMC4), and protein folding (PFDN1). Moreover, functional analysis indicates that phosphorylated proteins are involved in the activation of protein import to the nucleus (NUP35, RANBP2, KPNA5, KPNA4, KPNA3, RGPD5), nuclear pore complex assembly, mRNA transport and export from the nucleus (NUP98, NUP35, AHCTF1, RANBP2, RANGAP1, NUP214, NUP153, NUP50, NUP188, ANP32A, NUP107, POM121), activation of translation initiation (EIF4G1, EIF5, EIF5B, CTIF, EIF4B, NCBP1, CSDE1, HSPB1), and microtubule cytoskeleton organization (SON, MAP1B, MAP1S, EML4, DYNC1LI2, MAP4, TACC2, EML1, PHLDB2, CLASP1, TACC3, HAUS6, DYNC1LI1, DST). Clusters DP2 and DP4 integrate downregulated host cell proteins along the viral cycle. Since the expression dynamic profile is quite similar for proteins in these two clusters, a combined functional analysis was performed considering all downregulated proteins. The most significantly regulated processes include cell adhesion, antigen processing and presentation, response to vitamin D, regulation of IGFR signaling, IL6 signaling, and H2B ubiquitination, which fits well with the phosphorylation of several proteasome proteins (PSMA5, PSMD1, PSMD4, PSMD11, PSMF1). Besides, clusters DPP2 and DPP5 recapitulate proteins whose phosphorylation is reduced as the viral cycle progresses. Similarly, to the analysis of downregulated proteins, the functional enrichment analysis was done by grouping the proteins from both clusters. It revealed that these proteins participate in the regulation of mRNA splicing, immediate filament organization, mRNA processing, adherens junctions organization, rRNA transcription, RNA transport, and regulation of stress fiber assembly. Finally, DDP1 and DDP4 include proteins that transiently decrease or increase respectively their phosphorylation up to 9 h postinfection to then return to basal levels. Regulated pathways by DDP1 cluster proteins included mRNA splicing and processing, H2B ubiquitination, and apoptotic chromosome condensation. Regarding the DDP4 cluster, the principal enriched pathways were the regulation of protein localization and stabilization, microtubule anchoring, actin filament depolarization, and cell–cell adhesion.

Selection of a functionally relevant protein panel and investigation of its value as a plasma readout of COVID-19 severity.

Once the mechanisms underlying SARS-CoV-2 infection in lung cells were delineated, it was wondered if the identified driver proteins could be used as a readout of the infection process in humans. To address this question, we focused on proteins among those identified whose presence in serum/plasma has already been demonstrated. The rationale was that these biofluids can be accessed by relatively noninvasive methods, which may facilitate the implementation of the optimized method in future clinical applications. Overall, 826 proteins were selected considering their previous detection in plasma according to Paxdb. Of these, 48 have showed differential expression in cells after exposure to SARS-CoV-2 could be detected in serum/plasma. The top 30 proteins in the importance ranking were then selected by recursive random forest (Table S1 and Figure 6). The selected features displayed a time-dependent expression pattern (Figure 6B) that allowed the segregation of the cells along the infection process in the Multidimensional (MDS) plot (3D in Figure S9 and 2D in Figure 6A). These proteins represent drivers of the infection that can be detected in plasma/serum and are therefore candidates for the follow-up of COVID-19 patients. Upon optimization and removal of nondetected proteins, 23 proteins were initially quantified by monitoring their proteotypic peptides as indicated in Table S5. Desmogelin 1 was not further considered since it could not be detected in most serum samples from COVID-19 patients and, therefore, its quantification was not reliable. The verification of these potential biomarkers was carried out in a cohort of COVID-19 patients with different disease severity, previously characterized in our laboratory.¹⁵ This cohort was composed of 76 patients, including 16 nonhospitalized patients (NHOSP), 30 hospitalized patients (HOSP), 10 patients hospitalized in the intensive care unit (ICU), and 20 deceased patients hospitalized in ICU (EXI). Of the 22 proteins successfully quantified (Figure 6C), 14 were found differentially expressed among the different COVID-19 groups (Table S6 and Figure S10), showing an expression pattern dependent not only on disease severity but also on the age of the patients (Figure 6D).

Figure 6.

Selection of a functionally relevant protein panel and verification of its value as a plasma readout in the cohort of patients with varying COVID-19 severity. According to the results of the TMT-based proteomics, 30 proteins were selected based on a recursive random forest algorithm and their abundance in human serum/plasma for further validation in serum samples from patients with COVID-19. The selected features (proteins) showed a time-dependent expression pattern in the (B) heatmap, which allowed the segregation of cells along the infection process in the (A) multidimensional (MDS) plot. (C) A list of 22 proteins was successfully quantified by parallel reaction monitoring (PRM). (D) The heatmap shows the abundance of the 14 differentially expressed proteins between the different COVID-19 groups in the PRM analyses, showing an expression pattern that depends not only on the severity of the disease but also on the age of the patients.

Identification of Circulating Proteins as Readouts of COVID-19 Severity Using Machine Learning Modeling

Aiming to define a classifier that may help sample stratification, the intensity values extracted from the PRM experiment of the remaining 22 proteins across the 76 COVID-19 serum samples (Table S6) were normalized (log2 transformed and scaled by Z-score calculation). The full data set was divided into a training data set consisting of 70% of the samples, and the remaining 30% of the samples were allocated to the test data set, distributed as follows: 54 samples for the training data set, 12 NHOSP, 21 HOSP, 7 ICU and 14 EXI, and 22 samples for the test data set 4 samples from NHOSP, 9 HOSP, 3 ICU and 6 EXI. The partitioning of the data was done by computing the randomization of the samples and ensuring that the randomization results were reproducible after setting a seed in the programming environment.

To optimize the performance of the classifier, three different machine learning pipelines were initially explored that integrate multiple classification algorithms, and different combinations of features and cross-validation methods. The best results were obtained with the sequential combination of two classification methods: LDA and SVM (Figure S11). The total number of classifiers during the optimization of a combined classifier, considering all proteins, was 253. The goal was that the accuracy associated with the cross-validation data set should be higher than that of the test data set and that the accuracy should be relatively high in both, indicating that there was no overfitting in any data subset. The resulting protein combination was: A2MG, RET4, CYTC, CO5, CLUS, FETUA, FIBG, FIBB, GNPTG, FINC, K1C16, ALBU, K1C9, and IBP2 (Figure S12). Starting from this protein panel, the next step allowed the improvement of the classification accuracy of the SVM model. This was achieved by optimizing a combination of the hyperparameters C (0.31) and sigma (0.1601) through a holistic random search. Overall, this is a heuristic method as testing all possible combinations of proteins and hyperparameters would not be computationally feasible. The area under the ROC curve was 0.7355 (Figure 7A) an acceptable performance, considering the challenging classification of the hospitalized group that includes patients of very different severity, as can be assessed in the confusion matrices (Figure 7B). The three LDA dimensions of the model contribute asymmetrically to discriminate the different classes, i.e., each dimension (LD1, LD2, or LD3) can improve the classification accuracy of specific classes (Figure 7C). Similarly, the features contribute asymmetrically to the classification accuracy, according to their corresponding importance value (Figure 7D).

Figure 7.

Combined linear discriminant analysis (LDA) and support vector machine (SVM) classifier. (A) ROC curve of the classification efficiency of the combined LDA and SVM model on test set samples. The ROC curves and their associated AUC value and p-value are calculated from the number of samples correctly classified from a data set, taking into account the total number of classes and the number of samples proportionally classified in each group. (B) Confusion matrix of the cross-validation set (top) and test set (bottom) classification by the combined LDA and SVM model. (C) Feature importance bar chart of the random forest LDA dimensions from the combined LDA and SVM classification model. The overall importance of each dimension in classifying the samples is represented as the sum of the random forest mean decrease accuracy measure associated with each class. (D) Importance of all features measured by random forest on the entire data set. This plot illustrates how proteins contribute to the performance of the classification model. Color code: green: nonhospitalized (NHOSP); yellow: hospitalised (HOSP); red: Intensive Care Unit (ICU); gray: deceased (EXI).

LDA is a dimensional reduction method that depends on the calculation of the covariance matrices per class, having as final result a set of eigenvalues and eigenvectors (LDA biplots). The eigenvectors are a set of factors, each of which is associated with a feature (i.e., protein), which has a specific weight on the final value of the Linear Discriminant. Therefore, they make possible to represent the performance of the LDA facilitating the interpretation of the model by plotting two-dimensionally each paired combination of Linear Discriminants (Figure 8). This allows the discrimination between the different disease groups, considering a combination of different dimensions. Overall, longitudinal proteomic studies of SARS-CoV-2 infected human cells in culture provided a comprehensive description of the cellular proteome remodeling along the infection. In addition, a panel of differential proteins was identified as candidate biomarkers that allowed the establishment of a method to classify disease severity in a cohort of infected patients, providing new opportunities for their clinical management.

Figure 8.

Two-dimensional representation of the results of the combined LDA and SVM classification model. On the left, representations of how the eigenvector factors contribute as proportions to the linear LDA combinations. Given that LDA produces three dimensions or linear discriminants, there are three possible combinations without repetition of two features out of a total of three features. Each of the two-dimensional plots represents one of the combinations of LDA dimensions. The weight associated with each protein varies between the eigenvectors of each linear discriminant. Right, two-dimensional plots of each combination of linear discriminants corresponding to the arrow plots on their left. Samples are represented as points and the color of the point indicates the corresponding class. The colored area is the space detected by the SVM model where a sample is more likely to belong to a particular class by its coordinates in the LDA. The percentage of variance of the data retained by each linear discriminant after training the LDA model is shown in the axis names between brackets. Color code: green: nonhospitalised (NHOSP); yellow: hospitalised (HOSP); red: Intensive Care Unit (ICU); gray: deceased (EXI).

Discussion

In this study, we aimed to identify proteins regulated by SARS-CoV-2 in human lung epithelial cells that are associated with cellular processes driving the viral infection cycle. Based on the defined mechanistic background, we selected a panel of relevant proteins during infection that can be detected in plasma, using machine learning tools. These proteins were tested as severity markers of COVID-19 in serum samples from a cohort previously characterized in our laboratory.¹⁵ The combination of functional studies to understand the host cell response with statistical estimation to hypothesize potential biomarkers is emerging as a promising strategy to define robust disease-associated proteins with potential clinical applications.³⁴

To fully understand a complex biological process, it is essential to unravel the dynamics of the molecular events involved in order to deduce the hierarchical interplay between them. We describe here a deep time-dependent proteome rewiring that drives a sequential reprogramming of cell biology to allow the progress of SARS-CoV-2 infection, which is in very good agreement with previous reports (Table S7). Evasion of the cellular innate immunity, control of the translation machinery, regulation of the endocytic and secretory pathways, control of the apoptotic response and autophagy, induction ER stress and UPR, interference with the control of cell polarity and epithelial cell–cell junction integrity, lipid metabolism are the masterpieces of the host cell reprogramming by the virus, that is mediated by key viral proteins.^7,35−43 In addition to the expected accumulation of viral proteins over time, we also detected a time-dependent phosphorylation increase regulatory impact of which is unclear. Phosphorylation, as well as other posttranslational modifications, might influence the protein–protein interaction landscape of viral proteins, playing an essential role in coordinating the virus–cell interaction (reviewed by Wu et al.⁴⁴). Phosphorylation regulates the activity of N protein at different levels, including RNA binding, interaction with RNA helicase DDX1, viral NSP3 proteins and host proteins such as glycogen synthase 3, CDK-1, and 14–3–3 proteins. The phosphorylated residues involved are located in LKR and SR regions.⁴⁴ Among the phosphorylation sites previously identified in the nucleoprotein (Figure S4), there is no previous reference, to the best of our knowledge, to the S210 and S216 phosphosites described here. These residues are located at a highly disordered region in the C-terminal domain of the protein that is neither involved in RNA binding nor in dimerization. Although phosphorylation of T391 has been associated with disruption of the interaction capacity of the N protein,⁴⁵ the functional implications of these new modifications will need additional studies. It is worth noting that while a substantial portion of the proteomic alterations discussed in the manuscript may be ascribed to a direct impact of SARS-CoV-2 on the infected cells, the fact that the maximum achievable MOI reached 60% of the cells, suggests that a fraction of bystander, noninfected cells may contribute with specific alterations to the differential proteomic profiles, particularly at early time points, much as how infections proceed in nature.

The identification of circulating biomarkers to diagnose and predict the severity of COVID-19 and long-term complications has been a priority for the scientific community since the breakthrough of the SARS-CoV-2 pandemic. Different proteome-wide studies have been conducted in this endeavor on different biological fluids including gargle, saliva, CSF, and serum samples^46,47 (reviewed in 5). Although systematic studies involving larger cohorts are needed, several differential proteins have been commonly reported and, therefore, might be considered as potential biomarkers (Table S7). Most of these candidates emerge from the statistical assessment of heterogeneous cohorts and might reflect a systemic condition, as biological fluids collect information from the whole organism. Proteome analyses of target cells exposed to SARS-CoV-2 provide a functional framework for the selection of infection-associated proteins and therefore, enhance their specificity as biomarkers whose discriminatory capacity could be then validated in patient samples accessible by minimally invasive methods. In the past few years, machine learning strategies have been increasingly employed for unbiased feature selection from proteome-wide data sets, in particular for SARS-CoV-2 studies.^48,49 In this study, we selected 30 differential proteins in A549-ACE2 cells exposed to SARS-CoV-2, which have been previously detected in plasma/serum according to PAXdb. The selection process was done by recursive random forest based on the accumulated importance value of variables after 50 iterations. PRM analysis allowed the quantification of 22 of these proteins in serum samples from patients with increasing COVID-19 severity that was previously studied in our laboratory (5).

Starting from the PRM-derived protein abundance data, LDA and SVM were sequentially used to optimize a classifier to stratify COVID-19 serum patients according to the severity of the disease. LDA classification provided higher accuracy associated with the cross-validation and test set simultaneously than other classifiers tested in this analysis. Therefore, LDA was applied as a dimensional reduction technique and feature extraction method to select relevant proteins that are used in a second classification algorithm, SVM, which uses a radial kernel. Though the overall important value decreased from LDA1 to 3, each specific LDA is particularly important for classifying samples belonging to certain classes. The final model included 14 proteins with an area under the ROC curve of 0.7355 whose association with COVID-19 progression can be postulated based on their differential expression in A549-ACE2 cells exposed to SARS-CoV-2. Of the 14 proteins, some (A2GM, RET4, CYTC, FETUA, IBP2, FIBB, FIBG, FINC, CO5, and ALBU) have already been proposed as indicators of COVID-19 severity and associated comorbidities (5). Previous studies have demonstrated that acute renal failure is a common complication of COVID-19, affecting up to 46%.⁵⁰ The increase of A2GM and the concomitant decrease of ALBU suggest the progression of renal damage, a hypothesis that is further supported by the upregulation of CYTC and IBP2.⁵¹ From the early stages of the pandemic, coagulation abnormalities were recognized as one of the complications of COVID-19, with an estimate of thrombotic events in up to 60% of the patients, with the highest incidence in those individuals with severe disease and requiring an ICU admission.^52,53 According to this evidence, the observed accumulation of FIBB and FIBG may serve as a measure of coagulation changes associated with the progression and severity of COVID-19. Interestingly, FIBB and FINC peak their abundances at 3 h post infection on the analyzed cellular model and therefore, they might be considered as early markers of COVID-19 severity. To the best of our knowledge, the additional 4 proteins of the classifier (CLUS, K1C16, K1C9, and GNPTG) have not been associated so far with COVID-19 progression. Keratin intermediate filaments are essential elements of epithelial cells and tissue architecture. Downregulation of K1C16 and K1C9 in A549-ACE2 cells during SRS-CoV-2 infection, as well as the severity-dependent drop observed in the serum of COVID-19 patients, might be associated with the lung epithelial dysfunction resulting from the inflammatory reaction that leads to the acute respiratory distress syndrome.⁵⁴ The progressive decrease of these proteins in serum might also explain the skin manifestations that have been observed in some COVID-19 patients.⁵⁵ CO5 upregulation is likely part of the well-known activation of the complement cascade in COVID-19 5⁵⁶ as well as the decrease of CLUS. CLUS is an extracellular chaperone that can alter the membrane attack complex (MAC) formation preventing complement-mediated cytolysis.⁵⁷ Finally, as for other coronaviruses, SARS-CoV-2 virion assembly occurs into the ER-Golgi intermediate compartment (ERGIC).^58,59 However, it has been recently shown that, differently from other enveloped viruses, SARS-CoV-2 and other betacoronaviruses follow a nonlytic lysosomal exocytotic pathway once assembled in the ERGIC compartment.⁶⁰ GNPTG plays an essential role in the trafficking of proteins from the Golgi to the lysosome⁶¹ and, therefore, it is tempting to speculate that it may play a central role in the assembly and exocytosis of SARS-CoV-2 viral particles.

Overall, we have identified proteins and phosphoproteins in A549-ACE2 lung epithelial cells that are regulated by SARS-CoV-2 and provide molecular evidence to understand the host cell response. A panel of the differentially regulated proteins that are measurable in blood was verified in human serum samples from COVID-19 patients by PRM-targeted mass spectrometry. Changes in the abundance of 22 proteins parallel our findings in the cellular model, indicating their association with the host response to SARS-CoV-2 infection. PRM data were then analyzed with machine learning tools resulting in an optimized classifier of 14 proteins to stratify patients according to their severity. Remarkably, these proteins summarize some of the well–known processes associated with SARS-CoV-2 infection.

Data Availability Statement

Data have been uploaded to PRIDE and Peptide Atlas and can be accessed for review using the following credentials: Project accession: PXD053240. Servername: ftp.peptideatlas.org. Full URL: ftp://PASS05872:YZ3954qow@ftp.peptideatlas.org/.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jproteome.4c00566.

• Figure S1. Experimental Design applied for sample labeling using Tandem Mass Tags (TMT)pro 16plex kit. Figure S2. Overlapping of proteins found differentially expressed in the proteome and phosphoproteome analysis. Figure S3. Table representing the phosphorylation sites previously detected in SARS-CoV-2 proteins Figure S4. Phosphorylation sites previously detected in SARS-CoV-2 nucleoprotein (P0DTC9) and newly detected phosphorylation sites located at the C-terminal peptide (QLQQSMSSADSTQA) Figure S5. Functional analysis performed by Coronaspace comparing the proteins found differentially expressed in the A549-ACE2 cells proteome and respective enriched terms with other studies in the literature. Figure S6. Functional analysis performed by Coronaspace comparing the proteins found differentially phosphorylated in the A549-ACE2 cells and respective enriched terms with other studies in the literature. Figure S7. Transcription Factor Enrichment Analysis (TFEA) and Kinase Enrichment Analysis (KEA) performed using the Expression2Kinases webtool. Figure S8. Selection of the number of clusters based on their specific time-dependent profile estimated by the minimum distance between cluster centroid. Figure S9. Protein selected for parallel reaction monitoring (PRM) based on Random Forest algorithm. Figure S10. One-way ANOVA & posthoc Tests performed in MetaboAnalyst 5.0 of the proteins quantified by parallel reaction monitoring (PRM). Figure S11. Machine learning analysis graphical workflow. The best results were obtained with the sequential combination of two classification methods: LDA and SVM. Figure S12. Boxplots of the proteins used by the best built classifier (PDF)

• Table S1. Proteins and phosphoproteins identified by proteomics (XLSX)

• Table S2. Protein and phosphoprotein quantification (XLSX)

• Table S3. Functional analysis of the differential proteins and phosphoproteins (XLSX)

• Table S4. Functional analysis of the differential protein clusters grouped according to their time-dependent expression pattern (XLSX)

• Table S5. Peptides monitored by PRM in the validation analysis (XLSX)

• Table S6. Results from the PRM quantitation of the selected protein panel (XLSX)

• Table S7. Summary of previous studies reporting COVID-19 biomarkers in blood (XLSX)

Author Contributions

Experiments and procedures: FM, JV, SC, VC, IO; concept and design: FJC, PG, UG; supervision: FJC; writing of article: FJC.

The CNB was supported by Grant CEX2023–001386 S funded by MICIU/AEI/10.13,039/501100011033. Comunidad de Madrid Grants B2017/BMD-3817 and 2022/BMD-7232. Intramural CSIC PIE/COVID-19 projects 202020 × 10⁷⁹ and 202020E108. MICIN PID2021–127496NB-100. This research work was also funded by the European Commission—NextGenerationEU (Regulation EU 2020/2094), through CSIC’s Global Health Platform (PTI Salud Global) and Conexión Cancer.

The authors declare no competing financial interest.