Comprehensive proteomic classifier for molecular characterisation of pulmonary sarcoidosis: protocol for a longitudinal multi-centre study to evaluate bronchoalveolar fluid and cell diagnostic and prognostic biomarkers of pulmonary sarcoidosis

Abstract

Introduction

Sarcoidosis is a multisystem disorder with variable presentation and disease course. Diagnosis requires the exclusion of other causes of granulomatous inflammation. Current clinical management is often fraught with diagnostic uncertainy and the lack of tools to predict pulmonary disease progression. To address these challenges, we designed a study using data from bronchoalveolar lavage (BAL) fluid and cells to develop diagnostic and prognostic tools in patients with pulmonary sarcoidosis.

Methods and Analysis

This multicentre study will include discovery and validation cohorts of healthy controls, interstitial lung disease controls and pulmonary sarcoidosis cases from three study sites. Sarcoidosis participants will be grouped into progressive and non-progressive pulmonary disease based on changes in pulmonary function testing, chest radiographs or treatment requirements. The discovery cohort consists of participants with existing BAL fluid, BAL cells, and clinical datasets; the validation cohort will be prospectively enrolled and participants will consent for BAL collection from either a clinical or research bronchoscopy. Untargeted proteomic profiling of BALF along with statistical modelling with variable selection techniques will generate a classifier for diagnosis and prognosis. Targeted proteomics using parallel reaction monitoring–mass spectrometry will be used for internal and external validation. Additionally, BAL cell single-cell gene-expression analysis using Cellular Indexing of Transcriptomes and Epitopes by Sequencing (CITE-seq) will be integrated with proteome-wide data to elucidate cell-specific pathways implicated in the development and progression of sarcoidosis.

Ethics and Dissemination

The study will be conducted in accordance with Good Clinical Practice and the Declaration of Helsinki. The protocol has been approved by the Biomedical Research Alliance of New York Institutional Review Board (IRB), which serves as the single IRB across all study sites. The findings of this study will be presented as abstracts at scientific meetings and summarised in peer-reviewed journal manuscripts.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• Rigorous study design with discovery and validation cohorts, combining machine learning and biostatistical methods aimed at preventing underfitting and overfitting of models.

• An enrichment strategy focusing on a subset of sarcoidosis cases with primarily pulmonary involvement to account for the variability in sarcoidosis clinical menifesttaions. The comparison group includes both healthy controls and diseased controls to increase the generalisability of the findings.

• Use of integrative contemporary RNA-sequencing at a single cell level with proteome-wide changes is likely to define novel and known molecular pathways and signatures of disease and clinical phenotypes.

• A limitation is that the current sample size may constrain the power to detect small differences among sarcoidosis subtypes, given the heterogeneity in sarcoidosis course and clinical manifestations.

Introduction

Sarcoidosis is a multisystem immune-mediated disease of unknown cause^{1 2} with variable severity and outcomes.^{3 4} Sarcoidosis affects at least 45–300/100 000 individuals in the USA,⁵ across all ages, sexes and races.6,9 Pulmonary involvement, present in over 90% of cases, is the most frequent manifestation in Western countries and is associated with substantial morbidity, mortality, poor quality of life and increased healthcare costs.¹⁰ Respiratory failure remains the leading cause of sarcoidosis-related death in the USA.^{11 12} The annual mortality is approximately 2.8 per million people^{4 13} and continues to rise.¹¹ Outcomes differ by race and sex: women and Black individuals experience worse prognosis, with Black women having the highest mortality rate, although absolute numbers of deaths are similar across races.¹⁴ Notably, mortality rates are increasing among White cases but declining among Black cases, suggesting evolving disease patterns and manifestations.¹³ The clinical course is heterogeneous—while some patients achieve resolution, others progress¹⁵ to end organ failure despite treatment—resulting in significantly impaired health-related quality of life.

Diagnosing sarcoidosis is challenging because it requires histological confirmation and exclusion of alternative causes of granulomatous inflammation.^{1 6 16} No single test can definitively establish the diagnosis, and standardised criteria remain lacking. The absence of reliable biomarkers further complicates diagnosis and assessment of disease activity. Current blood-based markers, such as angiotensin-converting enzyme, soluble interleukin-2 receptor, chitotriosidase and lysozyme, lack sufficient sensitivity and specificity for routine clinical use, while there are no bronchoscopic markers to aid in the diagnosis of sarcoidosis. Consequently, diagnosis often involves extensive evaluation and may be delayed; one study reported an average delay of 8 months,¹⁷ particularly given that up to 13.6% of pulmonary sarcoidosis cases are asymptomatic.¹⁸

Despite advances in elucidating sarcoidosis pathogenesis,^{19 20} the mechanisms underlying the heterogeneity of disease outcomes remain poorly defined.^{3 12 20} Genome-wide association studies have identified both well-known human leucocyte antigen (HLA) class II and unexpected HLA variants,²¹ including BTNL2²² and ANXA11,²³ while omic profiling of transcripts,24,29 proteins,30,37 and metabolites^{38 39} implicates INFγ (Interferon‐gamma),²⁸ STAT1 (signal transducer and activator of transcription 1),²⁸ Wnt-β catenin,²⁴ NF-κB (Nuclear factor kappa-light-chain-enhancer of activated B cells)³⁵ and PI3K/AKT/mTOR (phosphoinositide 3 kinase/AKT kinase/mechanistic target of rapamycin) signalling in sarcoidosis. Progressive disease appears to involve distinct immune programmes, with lung gene expression studies showing heightened immune responses in progressive cases^{40 41} and multiple studies confirming roles for JAK–STAT (Janus kinase/signal transducers and activators of transcription) signalling, cytokines/chemokines, T-cell receptor (TCR) signalling, phagocytosis and metabolic/redox pathways.26,2940 Longitudinal blood transcriptomics using time-to-event modelling demonstrated differential CXCL9 (Chemokine (C-X-C motif) ligand 9) and TCR gene expression associated with progressive versus non-progressive pulmonary sarcoidosis⁴¹ and baseline protein expression in the same cohort predicted the progressive phenotype.⁴² Although emerging proteomic studies of bronchoalveolar lavage (BAL) fluid and cells in sarcoidosis show promise,^{30 32 36} their usefulness has been limited by phenotypic heterogeneity and incomplete proteome coverage. Analogous mass-spectrometry (MS) approaches integrating clinical variables and protein measurements (eg, Xpresys Lung 2) have improved classification of lung cancer versus benign nodules.43,45 To date, lung-compartment-specific proteomic profiling in sarcoidosis, with diagnostic and prognostic correlations, has not been performed, motivating the present study. There are no predictive models to determine whether pulmonary sarcoidosis will remit or progress, leading to non-targeted immunosuppressive therapy that often causes disabling side effects, particularly with corticosteroids, nor validated integrated biomarkers that include clinical and patient-reported outcomes (PROs) with biological pathways for diagnosis or prognosis.⁴⁶ Hence, an unmet need in the sarcoidosis field addressed ihere is to develop classifiers for rapid diagnosis of pulmonary sarcoidosis and prediction of progressive disease, as well as to define critical cellular mechanisms. Using high-throughput, state-of-the-art platforms and bioinformatics tools, our systems-level study aligns with the recommendations of a recent National Heart, Lung, and Blood Institute workshop on sarcoidosis.^{3 47}

Methods and analysis

Objective

Characterise proteome-wide lung compartment changes in sarcoidosis to identify differences between diseased and healthy controls (HC) and develop models that can serve as a foundation for improving diagnosis and prognosis in pulmonary sarcoidosis.

1. Primary objective: develop and validate a multicomponent BAL classifier for diagnosing and predicting the progression of pulmonary sarcoidosis.

2. Secondary objective: conduct pilot studies to integrate gene expression and protein levels for identifying new cell-specific pathways in pulmonary sarcoidosis and its progression.

Study design and participants

This cohort study will be conducted across three sites: National Jewish Health (NJH) in Denver, Colorado, USA; the University of Minnesota (UMN) in Minneapolis, Minnesota, USA, and HealthPartners Institute (HPI) in Saint Paul, Minnesota, USA, a teaching hospital affiliated with UMN (figure 1). These sites were selected to capture a diverse patient population with respect to demographic characteristics, disease severity and health system factors. The study aims to examine two independent cohorts: the discovery cohort, which was retrospectively enrolled based on availability of biospecimens and clinical data, and the validation cohort, which will be prospectively enrolled. The discovery cohort was available primarily at NJH with participants who had leftover BALF samples and data available from prior research studies and/or the NJH biorepository. The participants in the validation cohort was enrolled as follows: at NJH (n=99), UMN (n=81) and HPI (n=25). Study visits and data collection will be integrated into routine clinical operations at each site. Research activities—including bronchoscopies, blood draws, questionnaires, imaging, chart reviews and follow-up assessments—are to be conducted by trained research personnel in accordance with a standardised protocol to ensure consistency across all centres. The first study patient was enrolled on 2 December 2021, and the last patient was enrolled on 10 July 2025, with the anticipated completion of follow-up in July 2027.

Figure 1. All three sites will enrol study participants. NJH will coordinate enrolment, uniform data collection, case phenotyping, and the CITE-seq studies. The UMN site will conduct the mass-spectrometric studies, including untargeted MS, targeted MS, analysis of global phosphorylation events, and conduct bioinformatics and biostatistical analysis. CITE-seq, Cellular Indexing of Transcriptomes and Epitopes by Sequencing; HPI: Health Partners Institute; MS, mass spectrometry; NJH, National Jewish Health; UMN, University of Minnesota.

Eligibility criteria

We enrolled participants with sarcoidosis (n=120), HC (n=25), and diseased interstitial lung disease (ILD) controls (n=60) with stringent entry criteria. Sarcoidosis cases will have non-Löfgren’s or non-acute onset, be within 4 years of initial visit to the study sites and meet the case definition based on the American Thoracic Society (ATS) Statement¹ (box 1), or have highly probable clinical diagnosis.⁴⁸ HC are healthy volunteers without systemic or respiratory disease on no medications (box 2). Diseased ILD controls are ILD cases without exacerbation. Patients with sarcoidosis or ILD on systemic anti-inflammatory therapy at screening were excluded, except for patients with sarcoidosis on hydroxychloroquine (box 2). Participants with sarcoidosis will be followed for at least 2 years and grouped as non-progressive or progressive based on changes in pulmonary function tests (PFTs), chest X-ray (CXR), and/or the need for treatment (box 3).⁴¹ The study aims to enrol 120 sarcoidosis cases to account for potential dropouts and a small percentage of subjects who may not be classifiable as progressive or NP sarcoidosis, with an expected distribution of 50 each for progressive and NP sarcoidosis.

Box 1. Study eligibility criteria for sarcoidosis cases.

Inclusion criteria

• Biopsy proven or highly probable granulomatous disease, with exclusion of other causes of granulomatous inflammation (such as CBD, common variable immunodeficiency, mycobacterial and/or fungal infection or suspected hypersensitivity pneumonitis, etc) as per the ATS criteria. In case of extra-pulmonary site of biopsy providing evidence of granulomatous inflammation, abnormal chest imaging (either CXR or CT) is needed for evidence of pulmonary involvement.

• For the validation cohort we will enroll subjects at the time of diagnosis or first evaluation at one of our centres, with pulmonary involvement including those with radiographic Scadding stage 0 (no pulmonary disease on CXR but with lung biopsy and/or CT confirmation of lung involvement), stage I (lymphadenopathy on CXR but no parenchymal disease), stage II (lymphadenopathy and parenchymal disease on CXR) or stage III (no lymphadenopathy but parenchymal disease on CXR) disease.

• On no immunosuppressive therapy for at least the last three months (with the exception of inhaled steroids), to include corticosteroids (eg, prednisone, prednisolone, dexamethasone, etc), steroid-sparing agents (eg, methotrexate, mycophenolate mofetil, leflunomide, azathioprine, etc), biologics, including anti-TNF, anti-B cell therapy, and others.

• Former smokers (quit >6 months ago) or never smokers, including marijuana smoking and e-cigarettes/vaping.

Exclusion criteria

• Löfgren’s syndrome or acute sarcoidosis: with sudden onset (<1 month) of prominent constitutional symptoms that include fever, arthritis/arthralgias, and fatigue.

• Current use or within three months of systemic anti-inflammatory therapy for pulmonary sarcoidosis. Use of hydroxychloroquine is not exclusionary.

• Stage IV sarcoidosis.

• Current cigarette, e-cigarette or marijuana smoking or smoking within six months prior to the study.

• Pregnancy.

• Weight less than 110 pounds (for venipuncture).

• Positive lung washing or biopsy cultures for fungi, mycobacteria or other respiratory pathogen consistent with an acute or chronic infection.

• Severe room air hypoxemia (eg, pulse oximetry saturation <85% on room air or need for >4 L oxygen to maintain saturations above 90%) and or hypercapnia (precluding BAL), for example, resting PaCO₂ >45 mmHg.

• An FEV₁<1 L.

• Presence of another disease that may be expected to significantly affect patient mortality and or the immune response or increase risks of complications with bronchoscopy (eg, HIV, HCV, cancer, uncorrected bleeding diathesis, another autoimmune disease or other interstitial lung disease, chronic liver disease, chronic kidney disease needing haemodialysis, serious cardiac arrhythmia, recent myocardial infarction within 6 weeks, unstable coronary artery disease or decompensated heart failure, uncontrolled hypertension and/or diabetes).

• On anticoagulation with the exception of aspirin.

• Inability to participate in the study, such as inability to undergo venipuncture and BAL procedures that form part of the inclusion/exclusion criteria or part of the outcome measure.

• Age >80 years.

ATS, American Thoracic Society; BAL, bronchoalveolar lavage; CBD, Chronic Beryllium Disease CXR, chest X-ray; FEV₁, forced expiratory volume in 1 s; HCV, Hepatitis C Virus; PaCO2, Partial Pressure of Carbon Dioxide; TNF, Tumor Necrosis Factor.

Box 2. Study eligibility criteria for controls.

Healthy controls

Inclusion criteria

• Healthy volunteers without systemic or respiratory disease.

• Former smokers (quit >6 months ago) or never smokers, including marijuana smoking.

Exclusion criteria

• Known underlying systemic or lung disease.

• Immunosuppressive therapy or other medication for a systemic disease process in the last 6 months.

• Current cigarette, e-cigarette or marijuana smokers or smoking within 6 months of study.

• Pregnancy.

• Weight less than 110 pounds (for venipuncture).

• Inability to undergo bronchoscopy with BAL or any procedures.

Diseased ILD controls

Inclusion criteria

• Bilateral infiltrates on CXR or chest CT undergoing bronchoscopy for suspicion of ILD such as hypersensitive pneumonitis, idiopathic pulmonary fibrosis, cellular or fibrotic non-specific interstitial pneumonitis, cryptogenic organising pneumonia, connective tissue disease-related ILD, interstitial pneumonia with autoimmune features, pneumoconioses and others.

• On no immunosuppressive therapy for at least the last three months with the exception of inhaled steroids, to include corticosteroids (eg, prednisone, prednisolone, dexamethasone, etc), steroid-sparing agents (eg, methotrexate, mycophenolate mofetil, leflunomide, azathioprine, etc), biologics, including anti-TNF, anti-B cell therapy and others.

• Former smokers (quit >6 months ago) or never smokers, including marijuana smoking and e-cigarettes/vaping.

Exclusion criteria

• Acute exacerbation of ILD as the clinical symptomatology is different and sarcoidosis is not a competing diagnosis.

• Current use or within three months of systemic anti-inflammatory therapy for ILD. Use of hydroxychloroquine is not exclusionary.

• Positive lung washing, BAL or biopsy cultures for fungi, mycobacteria or other respiratory pathogens consistent with an acute or chronic infection.

• Pregnancy.

• Weight less than 110 pounds (for venipuncture).

• Severe room air hypoxemia (eg, pulse oximetry saturation <85% on room air or need for >4 L oxygen to maintain saturations above 90%) and/or hypercapnia (precluding BAL), for example, resting PaCO₂>45 mmHg.

• An FEV₁<1 L.

• Presence of another disease that may be expected to significantly affect patient mortality and or the immune response or increase risks of complications with bronchoscopy (eg, HIV, HCV, cancer, uncorrected bleeding diathesis, another autoimmune disease, chronic liver disease, chronic kidney disease needing haemodialysis, serious cardiac arrhythmia, recent myocardial infarction within 6 weeks, unstable coronary artery disease or decompensated heart failure, uncontrolled hypertension and/or diabetes).

• On anticoagulation except for aspirin.

• Inability to participate in the study, such as inability to undergo venipuncture and BAL procedures that form part of the inclusion/exclusion criteria or part of the outcome measure.

• Age >80 years.

BAL, bronchoalveolar lavage; CXR, chest X-ray; FEV₁, forced expiratory volume in the 1 s; HCV, Hepatitis C Virus; ILD, interstitial lung disease; PaCO2, Partial Pressure of Carbon Dioxide; TNF, Tumor Necrosis Factor.

Box 3. Pulmonary sarcoidosis phenotypes.

Progressive pulmonary disease

• A ≥10% decline in FVC and/or FEV₁ and/or ≥15% decline in DLCO.

• Worsening chest radiography as determined by the interpreting radiologist/investigator.

• Need for immunosuppressive therapy for treatment of pulmonary sarcoidosis, including corticosteroids (eg, prednisone, prednisolone, dexamethasone, etc), steroid-sparing agents (eg, methotrexate, mycophenolate mofetil, leflunomide, azathioprine, etc), biologics, including anti-TNF (eg, infliximab, adalimumab, etc.) and/or anti-B cell therapy (rituximab) and others used for sarcoidosis.

• No resolution of disease.

Non-progressive pulmonary disease

• A ≤10% decline in FVC or FEV₁ and ≤15% decline in DLCO (stable) and/or ≥10% improvement in FVC or ≥15% improvement in DLCO (improved).

• A stable or improved chest radiograph as determined by the interpreting radiologist/investigator.

• No indication for immunosuppressive therapy as outlined above.

DLCO, diffusion capacity for carbon monoxide; FEV₁, forced expiratory volume in 1 s; FVC, forced vital capacity; TNF, Tumor Necrosis Factor.

Study procedures/assessments

Validation cohort study participants will undergo procedures outlined in table 1, figure 2. Discovery Cohort data and samples were available from prior research studies or the NJH biorepository.

Table 1. Timeline of study procedures.

Procedures		Enrolment Visit 1	6 months Visit 2	12 months Visit 3	18 months Visit 4	24 months Visit 5
Informed consent		X
Bronchoscopy		X
Medication and disease update		X	X	X	X	X
Patient-reported outcomes		X	X	X	X	X
Spirometry+diffusing capacity		X		X		X
6 min walk test		X		X		X
Chest X-ray		X		X		X
Patient-reported outcome measures	FAS	X	X	X	x	x
	SF-12	X	X	X	x	x
	SGRQ	X	X	x	x	x

FAS, Fatigue Assesment Score; SF-12, 12-item Short Form Survey; SGRQ, St. George’s Respiratory Questionnaire.

Figure 2. The study personnel will screen cases and controls to enroll participants who meet the entry criteria. Bronchoscopy for collection of BAL will occur at first research visit (V1) along with exposure assessment, medical health and PRO measure questionnaires. Subsequent follow-up visits to determine if sarcoidosis cases will have progressive or non-progressive pulmonary phenotypes. 6MWT, 6-min walk test; BAL, bronchoalveolar lavage; CXR, chest X-ray; DLCO, diffusion capacity for carbon monoxide; PRO, patient-reported outcome; spiro, spirometry.

Biospecimen collection

The study will examine BALF and BAL cells obtained via bronchoscopy performed at the time of enrollment either (1) as part of a clinical bronchoscopy with leftover BAL, or (2) as a research bronchoscopy. Participants were asked to undergo a research bronchoscopy if they were HC or sarcoidosis cases whose diagnosis was made outside of the healthcare system. BAL collection was standardised across all study sites. Research BAL was performed using a sterile, room-temperature, isotonic NaCl solution without additives for intravenous use, instilled into two adjacent segments of the anterior lung lobe, as described in prior studies.⁴⁹ BAL samples were processed promptly to separate cells and fluid for storage and subsequent analyses, in accordance with standard operating procedures across all study sites.

Patient-reported outcome measures and questionnaires

PROs will be assessed using multiple validated instruments, including respiratory symptoms through the St. George’s Respiratory Questionnaire and quality of life will be assessed using the 12-item Short Form Survey. Fatigue will be assessed using the Fatigue Assessment Scale. PROs will be assessed at enrolment and follow-up visits. Additionally, we will record occupational and environmental exposures, including smoking status, via Genomic Research in Alpha-1 Antitrypsin Deficiency and Sarcoidosis (GRADS) questionnaire⁵⁰ at the time of study enrolment.

Physiological assessments

PFTs—including forced vital capacity, forced expiratory volume in 1 s and single-breath diffusion capacity for carbon monoxide—will be performed at baseline (V1) and at 12-month intervals (V3, V5). PFTs will be obtained if not performed clinically and will be optional at the 6-month (V2) and 18-month (V4) visits.

Radiology

An anterior–posterior CXR will be obtained at enrolment (V1), 12 months (V3) and 24 months (V5), unless performed clinically. Radiographs will be interpreted by a radiologist at the clinical centre. If a chest CT is obtained for clinical purposes, a research CXR will not be performed. Information related to stability or change in radiography as outlined by the interpreting radiologist will be extracted from the medical record.

Data abstraction for phenotyping

Information relevant to the study will be obtained from the medical record of each participant, including duration of disease, PFTs, CXR Scadding stage/CT imaging reports and change in radiography, biopsy results and past medication use at the time of enrolment and current medications over the course of the study. As part of the institutional review board (IRB)-approved consent process, volunteers will sign a Health Insurance Portability and Accountability Act (HIPAA) authorisation form authorising the investigator to review and abstract appropriate data from their medical records. Any data recorded by the investigators will be de-identified and given a subject number. This information, along with consents, will be stored in a locked filing cabinet in the office of the investigator at the study sites.

Data management and biorepository

Clinical site personnel will enter data into the secure, encrypted REDCap (Research Electronic Data Capture, Vanderbilt University, Nashville, TN, USA) system hosted by NJH, with password authentication. Participants may complete PROs via REDCap web-based surveys or paper forms.⁵¹ Questionnaire links will be sent directly to participants with authentication and verification. Instruments include screening checklists with alerts for ineligibility, as well as demographic and clinical data, medication and medical history, lung function tests and radiology results. Data integrity will be maintained through validation rules (eg, range checks for outliers and alerts for missing data), regular queries for inconsistencies and periodic audits to verify accuracy and completeness across sites.

Molecular characterisation of BALF and BAL cells

We will characterise BALF proteins to build a classifier that distinguishes sarcoidosis from progressive sarcoidosis. Untargeted proteomics will be followed by targeted analysis of selected classifier proteins. BALF will be processed using an optimised protocol to enrich medium-abundance and low-abundance proteins, minimise loss and yield high quality tryptic peptides free of contaminants.⁵² tandem mass tag (TMT)-based liquid chromatography–tandem MS will be performed on a high-resolution Orbitrap Eclipse (Thermo Fisher Scientific, Rockford, IL, USA). Bioinformatics analysis will use tools within the Galaxy for proteomics (Galaxy-P, University of Minneota, Minneapolis, MN) software platform⁵³ with stringent thresholds to identify high-confidence peptides and proteins. For targeted validation, parallel reaction monitoring–MS (PRM-MS) will quantify classifier proteins via internal and external validation.

This study will integrate cell-specific gene expression with global protein changes to elucidate mechanisms driving sarcoidosis and pulmonary disease progression. We will perform Cellular Indexing of Transcriptomes and Epitopes by Sequencing (CITE-seq) using a panel of antibodies to identify macrophages/monocytes, T cells, B cells and NK cells, with protein epitopes tagged via antibody-directed barcoding prior to single cell RNA-seq.⁵⁴ RNA-seq FASTQ files will be processed using 10X Cell Ranger (10x Genomics, Pleasanton, CA, USA) for mapping and counting, while antibody barcode counts will be obtained via CITE-seq-Count. Data will be analysed in R using Seurat (V3),⁵⁵ applying SCTransform for normalisation and reciprocal principal component analysis (PCA) for sample alignment. Clusters will be defined from CITE-seq data, and differential gene expression will be assessed using EdgeR⁵⁶ after generating pseudo-bulk gene expression tables by cluster and sample.

BAL cells from the same participants will also undergo phosphoproteomic analysis. Cell lysates will be processed using dual-stage enrichment with Sequential Metal Oxide Affinity Chromatography (SMOAC)⁵⁷ to capture phosphorylation events. For advanced multiplexing, 16-plex TMTPro tags will be applied, and TMT-labelled tryptic peptides will be resolved on a high-resolution Orbitrap Eclipse instrument. Spectral data will be analysed in Galaxy-P, following our published workflows,⁵⁸ complemented by the use of the FragPipe software suite,⁵⁹ which contains tools customised for characterisation of quantitative phosphoproteomics data and results.^{60 61}

Analysis plan

Our primary goal is to (a) build and validate a classifier (figure 3) that distinguishes sarcoidosis from healthy and diseased controls (Aim 1) and a classifier that predicts sarcoidosis progression (Aim 2) and (b) integrate broad gene expression and protein level data to identify the intricate interplay underlying the development and progression of sarcoidosis in Aim 3.

Figure 3. The approach to building classifiers for sarcoidosis diagnosis and disease progression. Untargeted MS, coupled with statistical modelling, will identify classifiers that balance performance and parsimony. Proteins featured in the classification model will subsequently be internally validated using targeted PRM-MS, followed by external validation in an independent cohort of study participants. ILD, interstitial lung disease; NP, non-progressive; P, progressive; PRM-MS, parallel reaction monitoring–mass spectrometry.

1. To build the classifier, the top differentially abundant (DA) proteins are of interest, as is their collective ability to distinguish the comparison groups. The classifier will be developed in the discovery cohort and be confirmed in the validation cohort. We describe the process for the HC versus sarcoidosis classifier; the same approach will be used for the P versus NP.

   • To identify DA proteins in the discovery cohort, we will perform single-protein analysis using regression to model the association between each protein and the categorical outcome (sarcoidosis, diseased ILD and healthy), adjusting for demographic and clinical covariates. For each protein, t-tests for the corresponding coefficient in the linear model will assess predictive value for sarcoidosis versus diseased ILD and sarcoidosis versus healthy. Proteins with substantial missing data will be analysed using a dichotomised presence/absence variable. For each pairwise comparison, the false discovery rate (FDR) will be controlled using the Benjamini-Hochberg method with threshold of ≤5%. Validation will involve testing all DA proteins in an independent cohort, with primary analyses focusing on the top ~25 proteins using Bonferroni correction (family-wise error rate (FWER) ≤0.05) and secondary analyses applying an FDR ≤5%.

   • Development of classifiers based on multi-protein profiles: we will develop a sarcoidosis classifier using a modelling approach that integrates protein biomarker profiles with clinical variables. In the discovery cohort, protein feature selection will be performed by partitioning data into training and test sets, applying cross-validation to identify the optimal number of DA proteins for the classifier. Multinomial logistic regression will be used to model sarcoidosis risk from the top DA proteins and leave-one-out cross-validation will be used to select the model with the highest area under the curve (AUC). The top DA proteins will be re-computed for each cross-validation fold to avoid overfitting bias. Final models will include: (i) clinical predictors alone, (ii) data from the top K proteins alone (with K chosen by cross-validation) and (iii) clinical predictors with the top proteins. Predictive performance will be evaluated in the validation cohort using receiver operating characteristic (ROC) curves to select probability thresholds that maximise the AUC. These thresholds will classify individuals as having sarcoidosis or not, enabling the calculation of sensitivity, specificity, positive predictive value and negative predictive value. AUCs and 95% CIs will be computed for training and test sets using the pROC package in R.

   • We will internally validate the multiprotein classifiers using targeted quantification by parallel reaction monitoring (PRM)-MS in the combined discovery and validation cohorts. Three models will be fit using PRM-MS data: (i) clinical predictors alone, (ii) data from the top proteins and (iii) clinical predictors with the top proteins. The predictive performance of these models will be qualitatively compared with those developed using discovery proteomics and formally assessed by comparing ROC curves using bootstrap methods to estimate sampling error. Additionally, we will evaluate whether incorporating disease-related variables—such as PROs, 6-min walk test, organ involvement patterns (pulmonary-only vs pulmonary-plus-extrapulmonary or multi-organ involvement), and smoking status—improved classifier performance in the validation cohort.

   • Power: in our preliminary data, DA proteins differed by ~0.5 SD between comparison groups, corresponding to an approximate OR of 2.48. Using GPower software,⁶² we have >0.80 power to detect OR of 1.82 and 1.63 for a given protein, comparing sarcoidosis (n=100) to HC (n=25) and diseased ILD (n=60), respectively. We assume a very conservative Bonferroni correction for the approximately 1000 tests in the discovery that would result in at most one false positive (α=0.00005) for our power calculations but in practice will use FDR adjustments. For our validation set, we have >0.80 power to detect OR of 1.64 and 1.43 comparing sarcoidosis (n=100) to HC (n=25) and diseased ILD (n=60), respectively, assuming a conservative Bonferroni correction for ~25 tests that would result in at most one false positive (α=0.002). For each of the discovery and validation classification, with 100 sarcoidosis cases and 85 unaffected cases each, we have 90% power to bound the sensitivity of our sarcoidosis classification with a margin of error of 9% if the sensitivity is 70%, and 4% if the sensitivity is 95%.

2. To determine differences in BAL cell programming and protein pathways, the statistical analysis will be performed to identify differentially expressed (DE) transcripts in cases versus controls and progressive versus NP, and to integrate this information with DA cellular proteins. We will process single cell RNA and protein data using CITE-seq, align reads with Cell Ranger, and quantify antibody barcodes with CITE-seq-Count. Data will be analysed in R using Seurat (V3) with SCTransform normalisation and reciprocal PCA for sample alignment, followed by clustering based on combined transcriptomic and surface marker protein profiles for cell type annotation. Differential expression of genes and differential abundance of proteins will be assessed across clusters and between sarcoidosis phenotypes (cases vs controls; progressive vs NP) using t-tests or negative binomial models in EdgeR,⁵⁶ applying a two-step FDR correction (≤5%). Functional interpretation will involve mapping significant genes and proteins to Gene Ontology and canonical pathways using IPA, EnrichR (Qiagen, Germantown, Maryland, USA), and proteomics-specific tools (eg, PSEA-Quant,⁶³ with integrated pathway enrichment analyses at global and cluster levels. Kinase signalling perturbations will be inferred using Kinase Substrate Enrichment Analysis (KSEA)^{64 65} to identify dysregulated phosphorylation networks and candidate kinase inhibitors for future validation. Power calculations indicate >80% power to detect moderate-to-large effect sizes under FDR control. Finally, multi-omics integration of gene, CITE-Seq surface marker profiles, and bulk protein data will be performed using SmCCNet⁶⁶ and Bidimensional Integrative Factorization (BIDIFAC) method,⁶⁷ which extends the joint and indvidual variability (JIVE)⁶⁸ to identify shared and cluster-specific molecular modules associated with sarcoidosis.

Study timeline

Recruitment began in December 2021, with completion projected for August 2026. In years 1–3, we aim to focus on enrolling sarcoidosis cases, and in years 2–4, controls, for a total of n=120 sarcoidosis cases, n=60 diseased ILD controls and n=25 HC. We recruited 120 sarcoidosis cases in the validation cohort to ensure that follow-up was completed and that phenotyping was unambiguous for at least 100 subjects, as our power calculations for Aims 1 and 2 are based on a sample size of 100 sarcoidosis cases for Aim 1 and 100 each of NP or progressive phenotypes for Aim 2. Sarcoidosis cases will be followed for a total of 2 years to determine their disease course/phenotype as progressive or NP pulmonary disease, as described above.

Patient and public involvement

While it was not possible to involve patients or the public in the design, or conduct, or reporting or dissemination plans of our research, we reviewed research and white papers with patient preferences indicated to help ensure patient needs were being considered.

Discussion

This study represents a uniquely ambitious effort to develop both diagnostic and prognostic models in sarcoidosis using a rigorously designed, multi-site cohort. By enrolling participants across three independent centres, the study enhances generalisability and reduces site-specific bias—an essential strength in a heterogeneous disease such as sarcoidosis. Importantly, the use of BALF and BAL cells analysed through contemporary, unbiased MS provides a significant methodological advantage. Unlike targeted approaches that rely on predefined protein panels, many of which are unlikely to be present in the lung, this discovery-based strategy does not require an a priori list of candidate proteins which is particularly relevant, as BALF contains many proteins not represented in standard human lung proteome reference sets. This approach enables the identification of previously unrecognised features and pathways, ultimately enabling a more comprehensive understanding of disease mechanisms and providing a stronger foundation for predictive modelling in sarcoidosis.

Despite its strengths, this study has several limitations that need to be considered. Although enrolment across three independent sites enhances generalisability, under-representation of minority populations—who are disproportionately affected by sarcoidosis—may limit the applicability of the findings if these groups do not participate in the proposed work. Additionally, despite standardised operating procedures, subtle differences in BAL collection, processing, and storage across centres may introduce technical heterogeneity, which will need to be rigorously tested for and controlled. The reliance on discovery-based MS, while methodologically advantageous, also presents inherent challenges: BALF is a dilute, low-protein biospecimen with a wide dynamic range, which can lead to incomplete proteome coverage and under-detection of low-abundance features. Although depletion strategies can mitigate these issues, their use may complicate downstream translation because widespread clinical implementation is more feasible when a model is developed on ’neat’, undepleted matrix. Furthermore, MS studies remain susceptible to batch effects and instrument-specific biases, potentially limiting reproducibility across laboratories and underscoring the need for rigorous evaluation of batch effects as a driver of variability and for appropriate controls. However, we anticipate that our proposed work offers promise to fulfil an area of critical need.

Ethics and dissemination

This multicentre study is being conducted in accordance with globally accepted standards of good practice, in agreement with the Declaration of Helsinki and with local regulations. The study protocol has been approved by the Biomedical Research Alliance of New York (IRB# 21–08-248), the Institutional Review Boards at NJH (21-02-248 – 528), UMN (IRB SITE00001283) and HPI (21–02-248-1065). All participants are required to provide written informed consent. All manuscripts resulting from the study will be submitted to peer-reviewed journals. Findings will be disseminated via journal publications, scientific conferences, patient advocacy group online content and social media platforms.