Application of Proteomics in Sarcoidosis

Abstract

Sarcoidosis is a multisystem disease with heterogeneity in manifestations and outcomes. System-level studies leveraging “omics” technologies are expected to define mechanisms contributing to sarcoidosis heterogeneous manifestations and course. With improvements in mass spectrometry (MS) and bioinformatics, it is possible to study protein abundance for a large number of proteins simultaneously. Contemporary fast-scanning MS enables the acquisition of spectral data for deep coverage of the proteins with data-dependent or data-independent acquisition MS modes. Studies leveraging MS-based proteomics in sarcoidosis have characterized BAL fluid (BALF), alveolar macrophages, plasma, and exosomes. These studies identified several differentially expressed proteins, including protocadherin-2 precursor, annexin A2, pulmonary surfactant A2, complement factors C3, vitamin-D–binding protein, cystatin B, and amyloid P, comparing subjects with sarcoidosis with control subjects. Other studies identified ceruloplasmin, complement factors B, C3, and 1, and others with differential abundance in sarcoidosis compared with other interstitial lung diseases. Using quantitative proteomics, most recent studies found differences in PI3K/Akt/mTOR, MAP kinase, pluripotency-associated transcriptional factor, and hypoxia response pathways. Other studies identified increased clathrin-mediated endocytosis and Fcγ receptor–mediated phagocytosis pathways in sarcoidosis alveolar macrophages. Although studies in mixed BAL and blood cells or plasma are limited, some of the changes in lung compartment are detected in the blood cells and plasma. We review proteomics for sarcoidosis with a focus on the existing MS data acquisition strategies, bioinformatics for spectral data analysis to infer protein identity and quantity, unique aspects about biospecimen collection and processing for lung-related proteomics, and proteomics studies conducted to date in sarcoidosis.

Sarcoidosis is a rare multisystem disease of unknown cause characterized by granulomatous inflammation with variable severity and outcomes. The pathologic hallmark of sarcoidosis is epithelioid granuloma associated with the infiltration of CD4⁺ T cells, scattered macrophages, giant cells, CD8⁺ T cells, and B cells around the granuloma. Despite a greater understanding of sarcoidosis pathogenesis, the mechanisms contributing to the heterogeneity of disease manifestations and predictors of disease outcomes are poorly defined (1). A crucial step to narrow the gaps in the understanding of sarcoidosis includes the use of “omics” and systems biology research for the molecular characterization of sarcoidosis phenotypes and to uncover novel disease mechanisms that could represent new disease biomarkers or therapeutic targets.

Powerful high-throughput technologies make it possible to investigate large numbers of biological functions at multiple levels, which can be particularly useful in a polygenic condition such as sarcoidosis (2). Transcriptional studies in sarcoidosis have found genes that regulate the IFN-α, β, and γ signaling (STAT1), consistent with the Th1-type inflammatory response (3). In a genome-wide analysis of lung tissue, networks corresponding with cell movement and immune function, including STAT1, IL-7, IL-15, and specific chemokine genes, were overexpressed in subjects with sarcoidosis (4). However, proteins are the main effectors of cellular function, and their alterations result in the disruption of biologic systems. To gain a comprehensive understanding in sarcoidosis, systematic testing of the protein changes is essential.

Several platforms identify proteins by detecting antigenic epitopes. Meso scale discovery (5), proximity extension assays (6) and aptamer-based tools (7) are available to quantify candidate proteins. A limitation of methodologies that detect antigens is that they are limited to an a priori set of proteins. With the improvement in the mass spectrometers and bioinformatics, it is now possible to quantify thousands of proteins in complex protein mixtures. For this review, we will focus on workflows for mass spectrometry (MS)-based proteomics, which beneficially provides comprehensive identification and quantification of unknown proteins, and their application to sarcoidosis.

Biospecimens for Lung Proteomics

In proteomics research, sample handling and processing are crucial for robust results. The lung has inflammatory cells as well as epithelial lining fluid (ELF) consisting of mucins and surfactants. These “matrix” biomolecules in ELF make the MS analysis challenging, as they interfere with spectral data acquisition. The wide dynamic range also limits the detection of less abundant proteins (8), and high abundant protein depletion increases protein detection (9). The recent American Thoracic Soceity document provides guidance for the preparation of samples for proteomic studies (10). A critical issue is to prevent ex vivo changes. For this reason, standard operating procedures are necessary. These standard operating procedures should include sample transportation on ice, aliquot in usable sample volumes to avoid future freeze and thaw cycles, immediate processing for freezing (−80°C) until the time of assay, and long-term storage in liquid nitrogen (greater than 6 mo) (10).

For sarcoidosis, it would be ideal to have affected tissue for analysis; however, the patchy nature of involvement and limited tissue availability is a barrier to examining changes in the areas of granulomatous inflammation. Using blood or urine offers the benefit of repeated sampling, but the lung-specific signal is diluted. Consequently, sputum (11), ELF (12), lung edema fluid (13), exhaled breath condensate (EBC) (14), BAL fluid (BALF), and BALF-derived exosomes (15) have been investigated (Table 1). BAL cells provide insights into the immune mechanisms of inflammatory lung diseases. In sarcoidosis, although gene expression studies have identified the activation of alveolar resident immune processes such as mechanisms involved in T-cell, IL-12, IL-23, and IL-17 signaling (16) and lower concentrations of steroid receptor coactivator-3 in cases needing treatment (17), the investigation of BAL cells is underused, with only a few reports implementing comprehensive protein evaluation (18, 19) (described later). Examining EBC is challenging because water vapor constitutes more than 99% of its content (20). Higher EBC TNF-α concentrations, as a marker of alveolar macrophage activity, were observed by ELISA in sarcoidosis patients with sarcoidosis (21). In chronic obstructive pulmonary disease (COPD), differences in EBC cytokines, cytokeratin (22), and surfactant proteins indicated distinct COPD variants (23). Such studies could identify subtypes in sarcoidosis. Sputum offers ease of collection but is a challenging biospecimen because of variable viscosity; components such as inflammatory cells, microbes and their peptides, cellular debris, and most importantly, mucins; and contamination by salivary proteins. The mucins and highly charged proteins form crosslinks with other sputum proteins (24) and make detection by MS challenging. Specific sputum sample preparation identified the protein changes characteristic of current smoking status and thirteen differentially expressed proteins between asymptomatic smokers and COPD cases. Blood-based proteomics is attractive, and several studies in sarcoidosis have used multiplexed platforms that detected a priori identified antigens (25, 26), and only a few studies have used MS (27). The high dynamic range of the plasma/serum has made it challenging to examine the global protein changes.

Table 1.

Biospecimens for Proteomic Investigations in Sarcoidosis

	Advantages	Limitations	Sample-Specific Recommendation for Sarcoidosis	Role in Sarcoidosis
Lung/other tissue	Samples the site of injury Highest likelihood to capture biological changes	Granulomatous changes are patchy and prone to missed sampling Limited amount of tissue available Need invasive procedures for biopsy Contamination of tissue samples by blood and plasma proteins	Snap-freeze lung tissue promptly Washing before sample preparation for proteomics	Determine biological mechanism of granulomatous inflammation Promising for integration with other omics, including genomics
BALF	Easy to obtain via bronchoscopy Samples epithelial lining fluid (most proximate to the site of injury) Samples proteins from diverse cellular sources Used with standardized protocols by several investigators	High salt content needs removal (10) Variable dilution because of unpredictable volume of return High dynamic range because of albumin and other proteins Surfactant protein may impact spectral data acquisition	Enrichment of specific proteins of interest or removal of higher abundance proteins Standardized protocols for sample collection and storage Salt removal before analysis Avoid freeze-thaw	Better suited for developing biomarkers for diagnosis and prognosis Promising for integrative omics, such as metabolomics, but not genomics because of limited availability of RNA
BAL cells	Easy to obtain via bronchoscopy Identifies immune mechanisms by sampling inflammatory cells	Large volume lavage to ensure adequate cellular yield Contamination by red blood cells Variable proportion of cells depending on disease states and environmental exposures	Account for cell proportion during assay and analysis Cryopreservation may alter cellular proteins Avoid freeze-thaw	Determine biological mechanism of granulomatous inflammation Promising for integration with other omics, including genomics
EBC	Can be obtaining without any procedures Could be repeated without difficulty	Because EBC is >99% condensed water vapor (20), most analytes are close to the detection limit (10, 91) Variable collection methods alter EBC analytes and concentrations	Standard procedures to minimize the variability in collection (20)	Limited role unless standardized sample preparation and improved resolution of MS instruments
Sputum	Easy to obtain	Variable viscosity from sampling methods and sputum components Large amounts of mucins Contamination by oral microbes and microbial peptides	Adding reducing agents such as DTT (20), size-exclusion with or without dialysis (24), or gel-cleanup (92) to remove mucins Standardized approach to sample collection and processing Acetone precipitation for protein isolation	Unexplored but challenging for sarcoidosis pathogenesis and biomarker discovery
Blood-based proteomics	Easy to obtain without special procedures Better to capture systemic changes beyond organ-specific changes	High dynamic range of plasma Low-abundance proteins or those with a small change may be difficult to quantify reliably High-abundance protein removal techniques may remove other proteins Semple preparation could decrease throughput for discovery studies	Plasma is preferred over serum with EDTA (over heparin) (93) Removal of high-abundance proteins to improve depth of coverage (94) Fractionation before MS analysis (95) Removal of disease-relevant proteins bound to high-abundance proteins remains a challenge	Biomarker in blood would be idea for diagnosis and prognosis Targeted proteomics approaches offer promise to develop diagnostic tools and clinical assay

Definition of abbreviations: BALF = BAL fluid; EBC = exhaled breath condensate; MS = mass spectrometry.

The official report of the American Thoracic Society workgroup on metabolomics and proteomics offers guidance on techniques and challenges and recommendations for sample handling and processing (10).

BALF is performed with normal saline, and therefore, removal of high salt content is necessary before MS analysis. During BAL, ELF is diluted up to 100-fold in an unpredictable manner that is difficult to account for. Although some recommend using BALF urea (10, 28), others conclude that no method can sufficiently account for the variability in the dilution (29). Standardizing the instilled volume (29), the number of input aliquots (29), and the site of lavage (29) and using a single-cycle lavage (30) could decrease the variability in ELF dilution. ELF is rich in plasma-derived proteins (albumin, transferrin, etc.), especially in conditions with increased epithelial permeability. The enrichment of specific proteins of interest or removal of higher abundance proteins are commonly used strategies when examining BALF (31–33). Despite challenges of sample processing, global or targeted protein studies have provided valuable insights into the biology of lung diseases such as idiopathic pulmonary fibrosis (IPF) (34, 35), acute respiratory distress syndrome (13, 36), asthma (37, 38), COPD (39), and lung cancer (40, 41). The studies in sarcoidosis are described later.

Disease models provide excitement for the identification of biological systems that participate in the formation of sarcoid granuloma. Recently, in an innovative human granuloma model, the incubation of peripheral blood mononuclear cells from sarcoidosis cases with bacillus Calmette-Guérin–coated beads resulted in the formation of cell aggregates with centralized macrophage aggregation and peripheral T-cell clusters, as is observed in sarcoidosis human tissue (42). The genomic profile of the in vitro sarcoidosis granulomas conforms to human sarcoidosis tissue granulomas, including the mTORc1 signaling pathway, IL-13, and STAT6 as important upstream regulators of genes and pathways active in human lung and lymph node sarcoidosis tissues (43). Because of limited access to tissue with granulomatous inflammation, examining the landscape of protein changes in the in vitro granuloma models offers promise for the identification of both biomarkers for diagnosis and novel mechanisms that dictate sarcoidosis development and course.

Mass Spectrometry–based Proteomics

With advances in MS, MS instruments are ever increasing in sensitivity, speed, and selectivity. MS instrumentation, coupled with upfront ion funneling technologies, such as gas-phase ion mobility fractionation, substantially increases the ability to comprehensively detect and quantify peptides across the dynamic range of the proteome (44–46). In combination with evolving sample preparation workflows for organelle enrichment and/or membrane structure enrichment (47, 48), these techniques now provide opportunities to specifically investigate membrane or mitochondrial proteins. There are also powerful in vivo labeling methods emerging to target specific membrane structures and enrich these for proteomics analysis (49, 50). In sarcoidosis, MS proteomics could be performed for discovery studies that analyze a large number of proteins or for the targeted measurement of specific proteins. Discovery proteomics analyze either intact proteins (top-down) (51) or peptides from proteolytic digests (bottom-up). Bottom-up proteomics offers protein inference through tandem MS (MS/MS) peptide spectral matching using traditional data-dependent acquisition (DDA), which has failed to identify low-abundance proteins such as transcription factors, cell receptors, and lipid-bound or mitochondrial proteins. Novel data-independent acquisition (DIA) MS provides an alternative method that comprehensively detects and robustly quantifies medium- and low-abundance proteins (52, 53).

Three top-down MS applications exist: matrix-assisted laser desorption Ionization-time of flight (MALDI-TOF), surface-enhanced laser desorption ionization (SELDI)-TOF, and electrospray (ESI)-MS. Both MALDI-TOF and SELDI-TOF can screen and compare protein profiles, but neither method can afford protein identification. Protein identification from MS/MS is feasible with ESI-MS/MS or liquid chromatography (LC)-ESI-MS/MS, but this technique is available in limited laboratories because specialized MS instrumentation is needed. However, high-throughput abilities of top-down MS make it appealing for discovery applications despite protein interference that occurs because of dynamic ranges in the proteome. Therefore, additional sample preparation, such as solid-phase extraction and chip-based techniques, is used to decrease sample complexity (54, 55). In contrast, bottom-up proteomics exploits two sample preparation methods (in-gel digestion and in-solution digestion) and permits relative and absolute protein quantitation. Such experiments are executed in the following steps: 1) proteolytic digestion, 2) chromatographic separation, 3) peptide MS/MS, 4) database search for peptide identification, and 5) protein assembly and quantification. The peptide MS/MS data acquisition can be acquired via DDA-MS or DIA-MS techniques. During DDA-MS, MS1 scans, known as “survey scans,” are acquired for all real-time precursor ions eluting during a time-course LC gradient. Subsequently, MS/MS scans (or MS2 scans) are triggered on the most intense precursor ions from the survey scan, using either top-speed or top-N mode (Figure 1A). This cycle is repeated over the duration of the LC gradient. Upstream sample separation methods, such as two-dimensional LC orthogonal separation is employed to reduce sample complexity and improve protein sequence coverage. Two-dimensional LC separation techniques often include offline high pH reverse-phase (RP) fractionation followed by low pH RP separation in line to ESI-MS (56). Discovery-based proteomics employs both label-free (57) and labeled-based (58, 59) approaches for relative quantification. Label-free quantitation (Figure 2A) can be determined in one of the following two ways: 1) integration of peak area of MS1 ions or 2) peptide spectral counting (60). Both techniques require replicate measurements of each sample to validate the results statistically. Label-free methods experience drawbacks because of chromatographic reproducibility, spectral variability, sample complexity, and throughput.

Figure 1.

Data-dependent and data-independent mass spectrometry (MS). (A) Data-dependent acquisition (DDA) MS. During DDA-MS, protein samples are proteolytically digested into peptides. The peptides are then subjected to complementary offline liquid chromatography separation and subsequently concatenated into fractions. Fractions are then each separated using low pH liquid chromatography coupled to MS. Each duty cycle over a time-course gradient will include one precursor scan over the entire mass range. The precursor spectral intensity will then dictate which MS1 ions are selected for tandem MS (MS/MS). A dynamic exclusion time parameter can be used to prohibit duplication of MS/MS for a predefined time, allowing new MS/MS to be acquired. Effectively, DDA will acquire MS/MS spectra on the most abundant peptides in a mixture. Raw data are then processed using traditional peptide assignment software using the theoretical peptide values expected from genomic databases. Lastly, protein inference is assigned based principle of parsimony, in which the lowest number of proteins accounting for the peptides identified. (B) Data-independent acquisition (DIA) MS. In contrast to DDA, DIA omits real-time ion selection but obtains MS/MS spectra on all precursor ions irrespective of MS1 spectral intensity. DIA-MS uses small precursor windows ranging from 10 to 25 m/z to produce MS/MS spectra. The duty cycle will encompass successive windows of precursor isolation and fragmentation until the complete m/z range is completed. This process will repeat for the entire time course. The MS/MS spectra of wider isolations windows will result in a convolution of MS/MS spectra, which results in multiple peptides being fragmented. For bioinformatics analysis of DIA-MS spectral dataset, an upstream spectral library generation via DDA methods and specialized computational algorithms will deconvolute spectra and extract peptide extracted ion chromatogram (XIC) at MS1 and MS/MS levels for quantification. DIA-MS workflows have a high computational complexity. AUC = area under the curve; LC = liquid chromatography.

Figure 2.

Quantitative proteomics approaches. Techniques for label-free quantitation (A) Peptide peak AUC or peptide spectral counting. The AUC for peptides can be calculated using the analyte intensity in combination with retention time profiles. The liquid chromatography elution profiles for peptide features within defined retention time window and mass accuracy settings are integrated. Peptide features are aligned and matched to proteins by their corresponding MS/MS and finally, peptide signals intensities between sample types are compared for protein abundance differences. Relative peptide amounts present with healthy and disease state are proportional to the peak AUC. Peptide spectral counting can use spectral counting of precursor (MS1) ions. The frequency of MS/MS spectra is proportional to the relative abundance of the peptide being surveyed. Targeted peptide identification is confirmed with MS/MS on a subsequent sample injection. (B) Label-based quantification involves the incorporation of metabolic or chemical tags for estimating protein abundance. Here, we demonstrate the use of chemical tags with two tandem mass tag labels. Healthy and diseased states are lysed and digested in similar ways as label-free approaches. The digested peptide from each sample is labeled with a tandem mass tag and fractionated offline. The fractionated samples are then individually subjected to liquid chromatography MS/MS. Under MS/MS conditions, each tandem mass tag label will cleave and generate a reporter signal. The ratio of these reported signals is indicative of peptide abundance within a labeled sample.

The principal of label-based quantification is the incorporation of metabolic or chemical tags for estimating protein abundance. Stable isotope labeling by amino acids in cell culture (SILAC) is metabolic labeling that incorporates stable heavy isotopes into proteins via labeled amino acids to compare two or three conditions of interest. During culture, cells are exposed to growth media containing heavy- or light-labeled amino acids that are globally incorporated into the proteome (61). In the case of a dual comparison, one sample condition is prepared with “light” media, which does not contain heavy-labeled amino acids. In the second condition, cells are cultured from “heavy” media. After protein extraction, equal amounts of light and heavy proteins are combined, proteolytically digested, and analyzed by ESI-MS. The resulting MS peak intensities for light and heavy are used for relative protein quantification. Isobaric labeling with tandem mass tags (TMT) or isobaric tags for relative and absolute quantification (iTRAQ) (Figure 2B) are used when biological applications may warrant multiple conditions be analyzed in an experiment. Isobaric tag labeling technologies use a differential chemical tag (reporter ion) to label the free amine group of peptides. Each tag comprises a heavy isotope–labeled functional group that contains a variation in the composition of heavy isotopes, resulting in the labeling of up to 16 samples simultaneously. Because the chemical tag is isobaric, peptides common to each sample type have identical precursor m/z values. Relative intensities of the reporter ions coincide with peptide abundance in each condition. Isobaric labeling offers major advantages in reducing MS variability, improving sample throughput, and decreasing laborious data processing.

In contrast to DDA, DIA omits using real-time ion selection dependent on precursor scans to obtain MS/MS spectra (Figure 1B) (62–64). Thus, the fragmentation of peptide ions is independent of survey scan intensity. Instead, all ionized species of a sample that fall within a specified mass range (e.g., m/z windows; i.e., 25 m/z) are fragmented systematically and impartially. Subsequently, MS/MS acquisition will represent all fragment ions generated from ionizable analytes in the defined window. The acquisition of the next MS/MS is a result of a new precursor isolation window. A spectral library is used to match the MS/MS^all spectral dataset and infer protein identification.

Bioinformatics for DDA-MS and DIA-MS

The MS/MS spectra are assigned to peptide spectral match via peptide spectral matching. Publicly available species-specific repositories, which contain in silico translated genomic sequences, are used as databases in computational algorithms for determining peptide spectral matches. These theoretical in silico fragments ions are compared with experimental data, which is scored based on mass accuracy, consecutive ion matches, relative intensity, and number of unmatched fragment ions. The scored PSMs are then ranked, and the highest scoring PSMs are used to determine inferred proteins. Typically, one or two high-scoring peptide matches are sufficient evidence for the detection of a protein (65). Statistical validation of peptide and protein identification is determined by calculating false discovery rates when protein sequence databases are concatenated with reversed or scrambled databases (66). In comparison with DDA-MS bioinformatic algorithms, DIA-MS workflows have a high computational complexity because of wide isolation windows that generate fragment ions from multiple precursor ions (62, 67). Computational tools are available in which analyses can be performed in a peptide-centric or data-centric approach. Peptide-centric approaches (OpenSWATH, SWATH2.0, Skyline, and Spectronaut) require MS/MS spectral libraries to extract groups of information that reliably represent a particular peptide and use statistical methods to determine true matches (68). A data-centric software package, DIA-Umpire, constructs pseudo-MS/MS spectra that are subsequently identified and quantified with traditional database-searching and protein inference tools without an assay library (69). DIA offers promise as a reliable quantification method that can be scaled to clinical diagnostics. Although no published studies employing DIA-MS are available, it is a promising MS platform for improveing the understanding of sarcoidosis pathogenesis.

Discovery-based proteomics approaches are limited by their sensitivity, specificity, throughput, and cost-effectiveness for clinical applications. When specific signature features are known, targeted MS can be employed to provide absolute and/or relative quantification. Targeted assays improve the limits of detection and quantification and expand the working dynamic range when compared with DDA and DIA methods. Targeted assays require considerable method development but if used for absolute quantification, can be highly reproducible (70), and they are ideal for clinical assay development. Targeted MS employs multiple reaction monitoring (MRM) and selective reaction monitoring (SRM) or parallel reaction monitoring (PRM)–MS. For MRM or SRM, both precursor and fragment ions must be determined during the experimental design to ensure the appropriate precursor charge states and fragment ions are chosen. During analysis, precursor ions are selected and isolated in the first quadrupole. These ions are fragmented in a collision cell to generate fragment ions, which are selectively transmitted to the mass analyzer. MRM/SRM can monitor thousands of peptide transitions with appropriate chromatographic window scheduling for signature peptides SRM/MRM can monitor thousands of peptide transitions with appropriate chromatographic window scheduling for signature peptides. This targeted application excels in analyte sensitivity, sample throughput, and short acquisition times. MRM/SRM is well suited for validating selected protein and peptides form discovery studies in biomarker development and for clinical assay development. The MRM-based clinical classifier was developed to differentiate malignant lung nodules from benign lung nodules (71), but such studies have not been undertaken in sarcoidosis. The advantage of PRM over SRM/MRM is that it eliminates the need to preselect an optimum precursor to fragment PRM-MS requires the high-resolution MS/MS acquisition of all fragment ions after precursor ion isolation and fragmentation, typically on an orbitrap MS instrument. With full MS/MS spectra, PRM data can be searched with traditional DDA software algorithms, unlike SRM/MRM data. Ultimately, PRM assays have many uses, including biomarker discovery and validation in sarcoidosis.

Inferences from Proteomic Studies

Studies employing global proteomics provide high-dimensional data that could be leveraged to identify subtypes, to develop models (classifiers), and to gain insights into mechanisms contributing to sarcoidosis development and progression. Coupling contemporary proteomic platforms that offer deep coverage with appropriately powered studies and machine-learning algorithms will provide an opportunity to develop in silico disease models. The machine-learning principles that could be used for such studies include classification methods (Bayesian classifier, decision trees, random forest, rule-based learners, support vector machines, and artificial neural networks) and feature selection methods. The stability of the inferences could be tested by using partitioning tools, including cross-validation. These approaches are reviewed in prior publications (72, 73).

To gain novel insights into the mechanisms implicated in sarcoidosis development and progression, computational biology algorithms can be applied on proteomics datasets to provide insights into protein pathways and networks. Pathway analysis can organize proteins into existing canonical modules (signaling, gene-regulatory, or metabolic) to infer function. Networks use principles of graph theory to visualize and infer protein interactions using prior knowledge or de novo construction based on shared behaviors (such as regulation by transcription factors). Several currently available tools provide varying degrees of information about pathways and networks. These include tools that offer basic pathway functional information and basic network topological features. Such tools focus on the functional annotation of an unordered list of proteins by identifying statistically significant proteins first and functionally annotating these proteins, thus organizing interesting proteins into biological functions (most commonly, gene ontology terms). Several existing tools have this capability such as the Database for Annotation, Visualization and Integrated Discovery (DAVID), BioCarta, Reactome, GoMiner and others. Other tools provide rich pathway information by integrating statistical significance into functional integration by analyzing a ranked list of proteins in functional context, as done by our group previously (74). These tools (e.g., Gene Set Enrichment Analysis) provide information regarding the enrichment of the dataset proteins among eight curated gene sets, such as those predicted by targets for microRNA (regulatory target gene sets) or immunological gene sets. In comparison with basic pathway information tools, rich pathway information tools add information regarding shared behaviors, e.g., regulation or function. Bioinformatics algorithms can also provide basic or rich network information, including topological connections of proteins and events (protein interaction/regulation), that are carefully assembled into a graph. A number of tools including Cystoscope and Ingenuity Pathway Analysis are capable of identifying network topology and high-connectivity “hub” features that may be critical for network function. Previous publications summarize tools that may provide principles and available tools for pathway and network analysis of proteomics datasets that could be used for future studies in sarcoidosis (75, 76). To keep up with the ever evolving high-throughput platforms, novel bioinformatic algorithms can integrate multiomics datasets, which have positioned the scientific community to gain deeper insights into disease mechanisms that are not possible with reductionist approaches.

Proteomic Studies in Sarcoidosis

Several studies have examined system-level changes using genome-wide analysis, RNA sequencing, and protein changes and are summarized in prior reviews (77, 78). In lung tissue, the genes involved in Th1 (4, 79) and Th17 (80) immune responses were implicated in sarcoidosis and were drivers of fibrosis (81). These studies identified the crucial role of STAT1- and STAT3-related chemokines (CXCL9, CXCL10, and CXCL11) regulated by INF-γ and other genes, such as CXR5, GBP5, AIM2, ICAM1, JAK2, IL-5, IL-7, IL-12, and IL-15. Although significant overlap was observed in whole blood and lung tissue (3), some changes of pulmonary sarcoidosis are compartmentalized to the lung. In blood, the genes participating in IFN signaling (STAT1, STAT2, and GBPI) and pattern recognition (TLR2, TLR4, and TLR8), TCRs (CD28, ITK, and LEF1), and CXCL9 were linked to sarcoidosis, and CXCL9 and TCR genes were associated with progressive sarcoidosis (82). Some gene expression changes have also been detected at the proteome level. Table 2 lists proteomic studies in sarcoidosis that characterized BALF (26, 55, 83–88) and a few studies that examined serum/plasma (25–27), alveolar macrophages (18, 19), and BALF exosomes (15).

Table 2.

Summary of Studies Investigating Protein Changes in Sarcoidosis

Study	Tissue Type Studied	Sarcoidosis	Control Subjects	Proteomics Method	Findings
Magi 2002 (83)	BALF	Active sarcoidosis (n = 6)	IPF (n = 6)	2DE	Thirty-two protein spots with differential expression, of which 15 were more abundant in sarcoidosis
Sabounchi-Schütt 2003 (84)	BALF	Acute onset LS (n = 6)	Healthy (n = 4)	2DE-MALDI-TOF	Seventeen protein spots different in sarcoidosis; seven proteins not previously reported in BALF were also identified
Sabounchi-Schütt 2004 (27)	Serum	Sarcoidosis (n = 6)	Healthy (n = 4)	2DE and MALDI-TOF	Twenty-two protein spots were different between subjects with sarcoidosis and control subjects, 14 were more abundant and eight were less abundant in sarcoidosis. Nineteen spots had a statistically significant difference. Eleven proteins were identified from thee protein spots
Kriegova 2006 (55)	BALF	Stage I (n = 32 [LS = 14; NLS = 18]), stage II (n = 22), and stage III (n = 11)	Healthy (n = 23)	SELDI-TOF MS	Forty differentially expressed protein peaks, of which three were identified to be protococadherein-2 precursor, HSA, and alpha-1 antitrypsin
Rottoli 2005 (88)	BALF	Active sarcoidosis (n = 11)	IPF (n = 10) and SSc pulmonary fibrosis (n = 10)	2DE and cellular profile by flow cytometry.	An increase in plasma protein in BALF from sarcoidosis and SSc and more locally produced lower molecular weight proteins in IPF
Rottoli 2005 (87)	BALF	Active sarcoidosis	IPF (n = 10) and SSc pulmonary fibrosis (n = 10)	Carbonylated proteins assessed by ELISA	Increase in carbonylated proteins in the subjects with diffuse lung diseases compared with control subjects. Patients with IPF had more carbonylation protein targets, including plasmatic proteins and transferrin immunoglobulin light chain, and IgA. Albumin and immunoglobulin heavy chain target carbonylated in all cases and control subjects and may act as a buffer and prevent other proteins from oxidant damage
Silva 2007 (86)	BALF	Sarcoidosis (n = 4)	Chronic beryllium disease (n = 4) and control subjects (n = 5)	DIGE coupled with MALDI-TOF	Sixteen protein spots and five proteins were different in subjects with sarcoidosis versus control subjects. Seventeen protein spots and eight proteins were different in subjects with CBD and control subjects, and one protein was downregulated in sarcoidosis compared with CBD. One protein was more abundant in sarcoidosis (compared with CBD) and could not be identified
Landi 2013 (89)	BALF	Sarcoidosis (n = 9)	IPF (n = 7), PLCH (n = 9), and SSc pulmonary fibrosis, n = 7	2DE with MALDI-TOF MS/MS	An average of 1,000 spots were identified in each gel, and 154 spots were differentially expressed in various comparisons corresponding with 77 proteins. The proteins mapped to biological pathways involved in defense, inflammation, and response to stress
Silva 2013 (19)	Soluble protein fraction from alveolar macrophages	Sarcoidosis (n = 7)	Healthy (n = 7)	DIGE followed by MALDI-TOF	Sixty-nine protein spots were significantly different, corresponding with 25 unique proteins (17 upregulated and eight downregulated) in sarcoidosis. Clathrin mediated endocytosis and Fcχ receptor mediated phagocytosis
Du 2015 (25)	Serum	Sarcoidosis (n = 20)	Tuberculosis (n = 20) and healthy control subjects (n = 20)	Human cytokine antibody array	ICAM-1 and leptin discriminate cases with sarcoidosis from those with tuberculosis.
Landi 2015 (85)	BALF	Stage II/III active pulmonary sarcoidosis (n = 9)	Nonsmokers (n = 10) and smokers (n = 8)	2DE and MALDI-TOF MS or LC MS/MS	Differentially expressed proteins map to PI3K/Akt/mTOR, MAP kinase, pluripotency-associated transcriptional factor, and hypoxia response pathway
Häggmark 2015 (26)	BALF and plasma	LS (n = 139) and NLS (n = 140)	Asthma (n = 17) and healthy subjects (n = 49)	Planar antigen microarray	Sarcoidosis cases have higher reactivity to zinc finger protein 688 and mitochondrial protein L43
Kjellin 2016 (18)	Alveolar macrophages	Sarcoidosis (n = 8)	Control subjects (n = 6)	iTRAQ LC MS/MS	Eighty differentially expressed proteins were identified. There was upregulation of two phagocytotic pathways (Fcγ receptor–mediated phagocytosis and clathrin-mediated endocytosis signaling). Pyruvate metabolism was upregulated in sarcoidosis cases. Oxidative phosphorylation pathway was upregulated in patients with LS and downregulated in patients without LS
Martinez-Bravo 2017 (15)	BALF exosomes	Sarcoidosis (n = 15)	Healthy (n = 5)	iTRAQ LC MS/MS on BALF exosomes	Six-hundred and ninety proteins were quantified consistently in exosomes. Proteins more abundant in patients with sarcoidosis are involved in innate and adaptive immune response, protein maturation, and homeostasis. There were indications of downregulation of endocytosis, catabolism (such as peptidase and endopeptidase activity and ubiquitination), and metabolic processes

Definition of abbreviations: 2DE = two-dimensional electrophoresis; CBD = chronic beryllium disease; DIGE = difference gel electrophoresis; ICAM-1 = intercellular adhesion molecule-1; IPF = idiopathic pulmonary fibrosis; iTRAQ = isobaric tags for relative and absolute quantification; LC = liquid chromatography; LS = Lofgren’s syndrome; MALDI = matrix-assisted laser desorption ionization; MAP = mitogen-actived protein; MS = mass spectrometry; MS/MS = tandem MS; NLS = non–Lofgren’s syndrome; PLCH = pulmonary Langerhans cell histiocytosis; SELDI = surface-enhanced laser desorption ionization; SSc = systemic sclerosis; TOF = time of flight.

Early studies implemented either two-dimensional electrophoresis or top-down proteomics in sarcoidosis. Magi and colleagues (83) identified 32 differentially abundant (DA) proteins between sarcoidosis and IPF. Sabounchi-Schut and colleagues (27) observed 21 DA spots comparing subjects with Lofgren’s syndrome (LS) with healthy control subjects. Seventeen proteins were identified; many were nonplasmatic and participate in the inflammatory and redox processes. Twelve had lower intensity in patients with sarcoidosis, including some antioxidant proteins. In a larger study, Kriegova and colleagues (55) compared subjects with pulmonary sarcoidosis with control subjects using SELDI-TOF and detected 40 DA protein peaks, including 26 that were upregulated in sarcoidosis. The protein pattern among patients with chest X-ray (CXR) stage I disease was dependent on a diagnosis of LS, suggesting that the CXR stage influences protein abundance. Rottoli and colleagues assessed the differences in oxidant stress in diffuse lung diseases (DLDs [IPF, sarcoidosis, and systemic sclerosis (SSc)]) by quantifying BALF carbonylation (87). A higher oxidant burden was noted in subjects with DLDs compared with control subjects. A limited number of carbonylated proteins were present in SSc and sarcoidosis compared with IPF, suggesting differences in oxidant stress in DLDs. Landi and colleagues (89) compared BALF from patients with sarcoidosis with IPF and SSc as well as pulmonary langerhans cell histiocytosis (PLCH). Seventy-seven unique DA protein spots were selected for identification. These proteins included those involved in immune response, inflammation, coagulation, oxidative stress, and antiprotease activity. Silva and colleagues (86) compared BALF from patients with sarcoidosis and chronic beryllium disease (CBD) and control subjects. Five unique proteins were DA between subjects with sarcoidosis and control subjects, three were DA between subjects with sarcoidosis and those with CBD. These studies demonstrate the ability of MS proteomics to detect clinically meaningful differences in sarcoidosis.

With the improvement in the MS instrumentation, several studies have obtained deeper proteome coverage to provide insights into disease biology. Leveraging MALDI-TOF MS, Landi and colleagues (85) identified 34 DA proteins between nine patients without LS and 18 healthy control subjects in BALF. DA proteins map to PI3K/Akt/mTOR, MAP kinase, pluripotency-associated transcriptional factor, and hypoxia response pathways. Kjellin and colleagues (18) examined alveolar macrophage membrane-associated proteins employing iTRAQ LC MS/MS. In contrast to the limited coverage observed in prior studies, they detected 1,656 proteins, 423 of which were identified with at least two high-confidence peptides. Eighty proteins were DA between eight patients with sarcoidosis (four with LS) and six control subjects. There was upregulation of two phagocytotic pathways and pyruvate metabolism in sarcoidosis. Oxidative phosphorylation was downregulated in patients with sarcoidosis (mostly in the group without LS). Martinez-Bravo and colleagues (15) compared BALF exosomes from 15 patients with sarcoidosis with those from five healthy control subjects using three iTRAQ LC MS/MS and detected an average of 1,400 proteins in each run. Of the 690 consistent proteins, 144 proteins were DA. The proteins more abundant in sarcoidosis included those mapping to cell-to-cell interaction, LPS binding, and vitamin D binding. One specific protein, vitamin D–binding protein, was more abundant in sarcoidosis and was validated in plasma. These studies demonstrate the value of contemporary tools to gain novel insights into sarcoidosis biology as well as to develop biomarkers that could have utility in diagnosis, prognosis, and monitoring disease activity.

Conclusions and Future Directions

The field of MS proteomics has improved substantially because of advances in the instrumentation and sophistication of bioinformatics algorithms. Substantial knowledge gaps remain in the understanding of mechanisms that explain the heterogeneity of sarcoidosis manifestation and course (1). The implementation of system-level studies using “omics” tools is expected to narrow the knowledge gaps by identifying biomarkers and mechanisms that contribute to high-risk manifestations in sarcoidosis (90). MS proteomic studies conducted in sarcoidosis have provided insights into disease mechanisms (15, 85) and clues to biomarkers, despite being limited by small sample sizes. Systematic studies characterizing the proteins in well-phenotyped patients are needed. Ongoing efforts by the Foundation of Sarcoidosis Research and other organizations to develop sarcoidosis cohorts offer promise for undertaking proteomic studies. However, close attention will be needed for the collection and storage of biospecimens from these cohorts. Examining the post-translational modification will be crucial to gaining insights into the mechanisms contributing to sarcoidosis progression and identifying precise therapeutic targets. The integration of gene–protein data will identify genetic contributions to the heterogenicity. Applying network analysis to proteomic data can elucidate connections between pathways not previously recognized. The current state of proteomic technology is well positioned to address knowledge gaps that exist in the pathobiology of sarcoidosis.