Novel biomarker discovery through comprehensive proteomic analysis of lupus mouse serum

Abstract

Objectives:

The difficulty of monitoring organ-specific pathology in systemic lupus erythematosus (SLE) often complicates disease prognostication and treatment. Improved non-invasive biomarkers of active organ pathology, particularly lupus nephritis, would improve patient care. We sought to validate and apply a novel strategy to generate the first comprehensive serum proteome of a lupus mouse model and identify mechanism-linked lupus biomarker candidates for subsequent clinical investigation.

Methods:

Serum levels of 1308 diverse proteins were measured in eight adult female MRL/lpr lupus mice and eight control MRL/mpj mice. ELISA validation confirmed fold increases. Protein enrichment analysis provided biological relevance to findings. Individual protein levels were correlated with measures of lymphoproliferative, humoral, and renal disease.

Results:

Four hundred and six proteins were increased in MRL/lpr serum, including proteins increased in human SLE such as VCAM-1, L-selectin, TNFRI/II, TWEAK, CXCL13, MCP-1, IP-10, IL-10, and TARC. Newly validated proteins included IL-6, IL-17, and MDC. Results of pathway enrichment analysis, which revealed enhancement of cytokine signaling and immune cell migration, reinforced the similarity of the MRL/lpr disease to human pathology. Fifty-two proteins positively correlated with at least one measure of lupus-like disease. TECK, TSLP, PDGFR-alpha, and MDC were identified as novel candidate biomarkers of renal disease.

Conclusions:

We successfully validated a novel serum proteomic screening strategy in a spontaneous murine lupus model that highlighted potential new biomarkers. Importantly, we generated a comprehensive snapshot of the serum proteome which will enable identification of other candidates and serve as a reference for future mechanistic and therapeutic studies in lupus.

1. INTRODUCTION

Systemic lupus erythematosus (SLE; lupus) primarily affects women of reproductive age[1]. Unpredictable flares of inflammatory activity followed by relative quiescence complicate monitoring and managing lupus[2]. A physician's ability to accurately track individual organ pathology continues to be limited by the heterogeneity of clinical presentations[3] and conflicting evidence regarding the utility of current biomarkers[4]. While elevated serum anti-nuclear antibodies are required for diagnosis, they are not lupus specific[5] and do not consistently correlate with disease activity[4, 6]. Lupus implicated cytokines such as B-cell activating factor and interleukin 6 (IL-6) show promising flare prediction and can be targeted therapeutically[7]; however, their serum levels appear too variable or do not reliably correspond to specific manifestations[8, 9]. SLE can affect nearly all organ systems, including the skin, joints, heart, and brain[10]. However, among those lupus-attributable complications, severe kidney inflammation, a.k.a lupus nephritis, poses the greatest risk of patient mortality[11].

Biomarkers corresponding to components of systemic disease, active kidney pathology, or both would enable clinicians to tailor therapy to each patient and intervene early to prevent loss of organ function[6]. The list of putative biomarkers in SLE is quite long. One way to categorize the mediators of lupus would be to consider the cellular location of each protein. Extracellular signals and plasma membrane receptors, such as cytokines and their receptors, can drive inflammatory signaling[8]. Alternatively, cytoplasmic and nuclear proteins, such as second messengers and transcription factors, mediate cellular changes that facilitate long-lasting impacts of autoimmunity[12]. Many prior biomarker studies focused primarily on those extracellular markers[4, 7], but few have looked at the potential of intracellular markers.

The present study applied high-throughput array analysis of lupus mouse serum to detect novel protein biomarker candidates relevant to pathogenic mechanisms in SLE. Performing these exploratory analyses in an animal model that replicates many lupus features provides unrivaled access to the tissues needed for analysis. Simultaneously, animal studies limit treatment confounders and environmental variations which often complicate the interpretation of clinical exploratory studies[4]. Moreover, to our knowledge, no prior studies have simultaneously assessed the expression of hundreds of proteins, ranging from inflammatory cytokines to homeostatic nuclear signals, in the serum of lupus mice.

These analyses utilize the MRL/lpr mouse, a widely studied lupus model exhibiting female-dominance, lymphoproliferation, humoral immunity, kidney pathology, and many other features of SLE[13]. Decades of MRL/lpr research have not yet generated a comprehensive serum proteome of this strain (or, for that matter, any other lupus mouse model). The degree to which the MRL/lpr proteome overlaps with the human serum inflammatory profile remains unknown. Moreover, no prior investigations in any lupus mouse model have combined high-throughput and unbiased serum microarray analysis with measurements of disease features. This knowledge gap could be obscuring potential targets for future inhibition or knockout studies. By validating a novel array-based screening approach in lupus mice, generating an MRL/lpr serum proteome, and identifying new biomarker candidates, the present study illuminates many avenues for future basic and clinical research which could eventually improve care for SLE patients.

2. MATERIAL AND METHODS

Mice.

All animal protocols were approved by the IACUC either at Albert Einstein College of Medicine (New York) or University of Houston (Texas). MRL/lpr mice served as our experimental model, and the congenic strain MRL/mpj was the control group.

Microarray detection of 1308 serum proteins.

Mouse sera from 18-19 week-old female MRL/lpr (n = 8; Table 1) and MRL/mpj (n = 8) were interrogated for 1308 antigens using the RayBio L-Series Mouse Antibody Array (RayBiotech, Catalog# AAM-BLG-1308-8). The assay was carried out and slides quantified according to the manufacturer's instructions (Figure 1A). Initial power analysis (d = 5.1; 1-β = 0.95; GPower 3.1) of known serum anti-dsDNA levels in MRL/lpr and MRL/mpj mice showed that groups of eight were sufficient for means-based comparisons with an alpha of 10⁻⁵. This sample size mirrors those used in prior MRL/lpr phenotyping studies from our group and others[14, 15].

Table 1. Description of MRL/lpr mice used on the screening array.

Serum was collected from eight female MRL/lpr and eight age-and-sex-matched MRL/mpj mice. Key measures of systemic, lymphoproliferative, and renal disease are provided for the MRL/lpr strain (mean +/− standard error).

	Age (wks)	Body weight (g)	Serum total IgG (mg/mL)	Serum anti- dsDNA (OD)	Lymph node weight (mg)	Serum BUN (mg/dL)	Proteinuria (mg/dL)	Renal path (score)
MRL/lpr (n = 8)	18.4 +/− 0.2	42.8 +/− 1.4	31.2 +/− 2.4	44.2 +/− 9.9	148.4 +/− 23.3	72.9 +/− 8.5	161 +/− 53	8.1 +/− 1.5

Figure 1. Schematic of experimental design.

A) Protein microarray analyses. Eight MRL/lpr (age 18.1-19.4 weeks) and eight MRL/mpj (age 18.0-19.4 weeks) were sacrificed for tissue and serum collection following urine retrieval. Serum from each mouse was measured in duplicate on RayBiotech L1.3k mouse protein microarrays. Fluorescence intensity data was then processed according to manufacturer recommendations. Individual protein levels were analyzed for MRL/lpr vs MRL/mpj and MRL/lpr vs MRL/lpr comparisons. B) ELISA validation. Following array analyses, proteins were chosen for further ELISA measurement. Among those proteins significantly increased in MRL/lpr (fold change > 1.25, multiple test corrected significance < 0.05), ten were chosen which collectively ranged from low to high increase, were located extracellularly, and had readily available and reliable ELISA kits. Each protein was measured in the sera of 7 to 15 MRL/lpr or MRL/mpj from the array cohort or other age-matched cohorts or both. The measured fold change was compared to the array finding to determine validity of the high-throughput screening array.

Assessment of lupus-like disease in MRL/lpr mice.

We assessed four domains of disease according to published protocols. Lymphoproliferative disease: Spleen and cervical lymph node weights: splenomegaly and lymphadenopathy are established indicators of disease progression in MRL/lpr mice[16]. Humoral disease: Serum levels of total and dsDNA specific IgG. Renal dysfunction: Serum BUN and urinary albumin-creatinine ratios. Renal pathology: Histologic measures of kidney glomerular and tubular inflammation performed by an experienced nephropathologist (L.H.) blinded to the experimental groups.

Validation of microarray screening results.

Adiponectin, erythropoietin (EPO), fibroblast growth factor 21 (FGF-21), granulocyte colony stimulating factor (G-CSF), granulocyte-macrophage colony stimulating factor (GM-CSF), IL-6, interleukin-17 (IL-17), L-selectin, macrophage-derived chemokine (MDC), and vascular cell adhesion molecule 1 (VCAM-1) were measured using ELISA (Figure 1B).

For each marker, fold changes (FC) were generated by dividing MRL/lpr by MRL/mpj average levels. We compared array-measured FC (aFC) to ELISA-measured FC (eFC). aFC and eFC values of similar direction and magnitude were considered concordant, emphasizing eFCs with significant p-values (student's two-tail T-test; p < 0.05). Receiver operating characteristics (ROC) curves were generated for all candidate and validation markers to assess area-under-the-curve (AUC), sensitivity, and specificity metrics in discerning lupus from non-lupus mice.

Protein enrichment analysis.

DAVID Bioinformatics Resources 2021 was used to generate annotations for gene ontology (GO) biological process, and Kyoto Encyclopedia of Genes and Genomes (KEGG) and Ingenuity Pathway Analysis (IPA) were used to generate top canonical pathways.

Statistical analyses.

All data were analyzed and plotted using GraphPad Prism 9, Microsoft Excel, Cytoscape, and R. Mann-Whitney U test was used for the antibody array assessment of MRL/lpr FC. Q-values (false discovery rate; q < 0.05) were computed for each protein assayed. The relationship of each significantly different marker with MRL/lpr disease features was assessed using Spearman correlation analysis (p < 0.05).

A detailed description of Material and Methods can be found in the supplementary materials.

3. RESULTS

3.1. Serum proteomic differences in lupus mice

One goal of this study was to generate the first comprehensive serum proteome of the MRL/lpr mouse. Of the 1308 proteins measured in the sera of MRL/lpr and MRL/mpj mice, 406 were significantly increased (FC > 1.1, q < 0.05; Supplementary Table 1), and 397 were significantly decreased (FC < −1.1, q < 0.05) in MRL/lpr mice (Figure 2A). FC values ranged from −5.7 to 3.96. Figure 2B reports the top ten increased and decreased proteins, based on FC magnitude. Serum levels of the remaining 505 proteins did not significantly differ between lupus mice and controls.

Figure 2. Protein biomarkers and protein enrichment in MRL/lpr vs MRL/mpj mice.

A) Volcano plot depicting the number of significantly different proteins in MRL/lpr compared to MRL/mpj at false-discovery rate (q-value) less than 0.05. Fold changes greater than 1.1 were termed "increased", and those below −1.1 termed "decreased". B) Tables reporting the top 10 increased and decreased markers in MRL/lpr serum. Protein enrichment analysis performed using DAVID software shows top gene ontology biological pathways (C) and KEGG pathways (D) associated with all significantly increased proteins in MRL/lpr. E) Protein enrichment using Ingenuity pathway analysis (IPA) which considers fold changes of all increased and decreased markers to produce weighted canonical pathways.

To better characterize lupus-related differences in our subsequent analyses, we grouped all significant proteins by their cellular location. 103 nuclear (32 increased, 71 decreased), 296 cytoplasmic (141 increased, 155 decreased), 177 plasma membrane (104 increased, 73 decreased), and 215 extracellular (124 increased, 91 decreased) proteins were significantly different in MRL/lpr mice. Among the 505 proteins unaffected by lupus-like pathology, a higher proportion were nuclear or cytoplasmic compared to the 803 significantly different proteins (57% vs 50%).

3.2. Protein enrichment and network analysis of lupus mouse proteome

Using the list of 406 significantly increased proteins in MRL/lpr mice, we used DAVID software to identify the top over-represented GO biological processes (Figure 2C) and KEGG pathways (Figure 2D). Significant terms included JAK-STAT signaling, immune cell chemotaxis, T-cell proliferation, cytokine signaling, inflammatory response, and other immune functions. (We note that while enrichment terms such as African trypanosomiasis, Inflammatory bowel disease, and Malaria (Figure 2D) may seem surprising at first in a lupus model, the SLE proteome includes many innate and adaptive inflammatory mediators which are upregulated in other immunologic conditions). Figure 2E includes the results of IPA canonical pathway analysis, which differs from DAVID; IPA used all 803 proteins and considered each protein's fold change and significance as it generated a collection of canonical pathway terms. Migration of immune cells, cellular infiltration, and necrosis were key pathways upregulated in MRL/lpr serum.

To identify potential regulators of the significantly expressed proteins, the Cytoscape plugin iRegulon was used, and the results are shown in Figure 3A-B. NFκb1 (encoding NF-kappa B p105) was recognized as the top transcription factor controlling the significantly expressed proteins in lupus mice (Figure 3A). It was also identified as the top signaling motif regulating the differentially expressed proteins (Figure 3B). Protein-protein interaction networks were also generated using Cytoscape. MCODE was used to identify clusters, which represent highly interconnected proteins within a network. The first cluster is shown in Figure 3C. The top five reactome pathways to which this cluster mapped included cytokine signaling in the immune system, chemokine receptors binding chemokines, immune system, signaling by interleukins, and interleukin-2 family signaling. Additionally, the full network of the top 549 proteins altered in lupus mice (p<0.01), as determined by Cytoscape, is shown in Supplementary Figure 1A. The most interconnected cluster within this global network is represented in Supplementary Figure 1B

Figure 3. Network analysis of proteins increased in MRL/lpr serum.

Analysis of interactions within the 214 proteins with a p-value < 0.01 and fold change > 1.5. iRegulon was used to determine the top transcription factor (A) and signaling motif (B) in this network. iRegulon is a computational method to reverse-engineer the transcriptional regulatory network and signaling network underlying a co-expressed gene set using cis-regulatory sequence analysis. It implements a genome-wide ranking-and-recovery approach to detect enriched transcription factor motifs and their optimal sets of direct targets. C) MCODE was used to find clusters, which are highly interconnected regions within a network. The top five reactome pathways to which this cluster mapped include cytokine signaling in the immune system, chemokine receptors binding chemokines, immune system, signaling by interleukins, and interleukin-2 family signaling. Each circle represents a protein, and the color of the protein is based on the fold change value of the protein. The minimum value is 1.5 and the maximum value is 3.96.

3.3. Correlates of disease features in lupus mice

For this analysis, we grouped systemic disease and kidney manifestations into four major categories: lymphoproliferative disease (spleen and lymph node weights), humoral disease (serum titers of IgG and anti-dsDNA antibodies), renal dysfunction (urine A/C ratio and serum BUN), and renal pathology (glomerular and tubular inflammation). Figure 4A depicts the Spearman correlation coefficients for all 803 significantly different proteins with each of the eight disease features, with the serum proteins being parsed by intra-cellular location. Figure 4B lists the top correlates under each category of features.

Figure 4. Relationship between individual proteins and MRL/lpr disease features.

Spearman correlations were used to assess the relationship between individual serum protein levels in MRL/lpr mice and several disease metrics. These features were quantified for each mouse as follows. Lymphoproliferative disease was assessed by weighing the spleen and a cervical lymph node. Humoral disease was determined by measuring serum levels of total IgG and anti-dsDNA antibodies via ELISA. Serum levels of BUN and urinary albumin-creatinine ratio, both measured by ELISA, were quantified to reflect renal dysfunction. Histologic analysis scoring of kidney sections by an experienced nephropathologist resulted in glomerular and tubular scores which reflect renal pathology. A) R coefficients from all 1308 measured proteins (y axis) were plotted to visualize correlation between disease features (x axis) and protein cellular location (row headings). B) Top correlates of lymphoproliferative and humoral disease metrics as well as renal dysfunction and pathology are reported. Asterisks represent fold change q-value or correlation p-value (* < 0.05, ** < 0.01). Data for all significant correlates is included in Supplementary Table 2.

Of the 803 significantly different proteins, 161 correlated with at least one disease feature (Supplementary Table 2). Ninety-two of these 161 proteins were increased in MRL/lpr serum, of which 52 showed a positive correlation. Among those exhibiting increased expression and positive correlations, 21 proteins were extracellular, 13 were membrane-bound, 14 were cytoplasmic, and four were nuclear. When considering the specific disease manifestations in Figure 4, we saw positive correlations between extracellular and membrane proteins with lymphoproliferative and renal disease; intracellular markers did not appear to have as numerous positive correlations with disease.

In order to better understand the origin of these 92 proteins, we consulted established proteomic atlases (i.e., Mouse Genome Database[17] and the Human Protein Atlas[18]) to link each protein to the cell type that most likely secreted it. Priority was given to cell types for which the protein is an established transcriptomic marker. Twenty-seven proteins were either non-specific or likely released by remote cell types (i.e., neurons). Considering the remaining 65, the majority (43 of 65; 66%) were likely produced by myeloid cells, while lymphoid and stromal cells may have contributed 14 (22%) and 8 (12%) proteins, respectively (Supplementary Figure 2A). Looking at specific cell types, neutrophils and other granulocytes account for 19 (29%), macrophages and monocytes for 15 (23%), dendritic cells for 9 (14%), and T-cells for 8 (12%). Fibroblasts, endothelial cells, natural killer cells, B-cells, and plasma cells combined to produce the remaining 14 (22%) proteins (Supplementary Figure 2B). Proteins that correlated with renal dysfunction or pathology (Supplementary Figure 2C) appear to come primarily from macrophages and monocytes (12 of 50; 24%), neutrophils and granulocytes (12 of 50; 24%), or T-cells (9 of 50; 18%).

Figure 5 is a focused depiction of the extracellular and membrane proteins implicated, showing their respective Spearman correlation coefficients and significance. Among these top marker candidates, TECK, TSLP, PDGFR-alpha, and MDC belonged to several top terms in our protein enrichment analyses. Though not depicted, BLC, TARC, and TNF RII had highly significant fold changes, correlated with disease, and also belonged to top enrichment terms.

Figure 5. Correlation coefficients and significance for the top extracellular and membrane-bound proteins.

The top 15 extracellular and membrane bound proteins elevated in MRL/lpr over MRL/mpj mice are shown on the y axis. Shown are each protein's Spearman r-coefficients (circle color and size) and corresponding p-values (square color) for its correlation with each of the eight disease metrics: spleen weight (grams), lymph node weight (grams), serum total IgG (ng/mL), serum anti-dsDNA (relative optical density), serum BUN (mg/dL), urine albumin—creatinine ratio (AC ratio; μg/L:mg/L), glomerular pathology score, and tubular pathology score. The data for correlates is included in Supplementary Table 2. Red boxes indicate proteins common to top protein enrichment terms. EBP50 (Ezrin-radixin-moesin-binding phosphoprotein 50); TECK (Thymus-expressed chemokine, CCL25); TGF-beta 1 (transforming growth factor beta); TLR4 (Toll-like receptor 4); TSLP (thymic stromal lymphopoietin); VEGF-D (vascular endothelial growth factor type D); EDAR (Ectodermal dysplasia receptor); PCDH8 (Protocadherin-8); MDC (macrophage-derived chemokine, CCL22); Reg3A (Regenerating islet-derived protein 3-alpha); PDGF R alpha (platelet-derived growth factor receptor alpha).

3.4. Protein enrichment analysis of proteins that correlate with disease

To investigate the pathways associated with disease manifestations in MRL/lpr mice, we again used DAVID and IPA software to perform protein enrichment analysis of the 161 proteins correlated with at least one disease feature (Figure 6A-C). We further focused our analyses on those 104 correlates of renal disease (Figure 6D-F).

Figure 6. Protein enrichment analysis of serum biomarker disease correlates in MRL/lpr mice.

All significantly different proteins (q-value < 0.05) regardless of fold change direction were included in these enrichment studies. Spearman correlations between individual marker levels in the MRL/lpr mice and disease metrics were performed. Using DAVID and IPA software, top terms were generated for all proteins which correlated with at least one feature of systemic disease (A-C; n = 161) and for the subset of proteins which correlated with at least one renal feature (D-F; n = 104). IPA results in (F) did not yield enough terms with activation z-scores, so number of molecules is plotted along the x-axis.

Top terms related to overall lupus severity belonged primarily to two categories: cell migration and inflammatory cytokine signaling. Specifically, the term cell chemotaxis (Figure 6A) encompassed 6Ckine, CXCR3, Endoglin, MCP-2, MDC, PDGFR alpha, and TECK, all of which were elevated in MRL/lpr serum. Cytokine-cytokine receptor interaction (Figure 6B) and pathogen induced cytokine storm signaling (Figure 6C) involved 6Ckine, Activin A, BLC, CD40-L, CXCR3, EDAR, G-CSF, IL-1 R2, Leptin R, Lymphotactin, MCP-2, MDC, TARC, TECK, TLR1, TLR4, TNF RII, TRAIL, TSLP, and VEGF D, all of which were elevated in MRL/lpr serum.

While cytokine signaling remained linked to renal disease (Figure 6E-F), immune cell migration was also a dominant theme distinguishing lupus nephritis from systemic disease, particularly by IPA. Specifically, BLC, Lymphotactin, MDC, TARC, and VEGF R3 related to the diapedesis of both granulocytes and non-granulocytes, while G-CSF and TNF RII related only to granulocyte movement (Figure 6F), all representing serum proteins elevated in MRL/lpr mice.

One interesting finding was the relationship of systemic disease with positive regulation of the ERK1 and ERK2 cascade (Figure 6A). Proteins involved in this cluster include 6Ckine, Lymphotactin, MDC, PDGF R alpha, TARC, TECK, TLR2, and TLR4. A relationship with renal correlates was also noted for the ERK1 and ERK2 cascade (Figure 6D), which additionally included GPNMB and ICAM-1. Upon further investigation, ERK1/2's relationship with renal disease was attributable to the proteins correlated with serum BUN and tubular pathology.

3.5. Selection of candidate proteins for further validation

From the 1308 proteins initially interrogated, 406 proteins were noted to be significantly increased in MRL/lpr serum (Figure 2A). Next, we performed correlation analysis and found that 52 of the increased proteins positively correlated with at least one disease feature. Of these 52 proteins, 34 were extracellular or membrane bound, and 28 of 34 correlated with lymphoproliferative or renal disease or both. We then used protein enrichment analyses and prior publications to further winnow down this list, resulting in seven promising candidates: BLC, TARC, TNF RII, TECK, TSLP, PDGFR-alpha, and MDC. Each candidate showed strong biomarker potential based on ROC analysis (Table 2).

Table 2. Comparison of array identified biomarkers to published findings.

Through our biomarker analyses, we identified seven promising proteins that exhibit serum levels that correlated with disease features. Receiver operating characteristic curve analyses included area-under-the-curve (AUC), sensitivity, and specificity. AUC^p-val provides the AUC value and its associated p-value. Results, if available, from prior human lupus and mouse model studies are included in the table. P-values: * < 0.05, ** < 0.01

Top biomarker candidates	Array Findings						Literature Review
	Fold change (lpr : mpj)	q- value	AUCp-val (%)	Sensitivity (%)	Specificity (%)	Disease correlates	Human SLE	Murine Lupus
C-C motif chemokine 22 (MDC)	1.70	0.003	95**	87.5	100	spleen weight, serum BUN	Inconclusive evidence50, 51
Platelet-derived growth factor receptor alpha (PDGFR-alpha)	1.59	0.004	91**	75	100	serum BUN, tubular score	Kidney expression ↑ in lupus nephritis47
Thymic stromal lymphopoietin (TSLP)	1.67	0.003	94**	87.5	100	spleen weight, serum BUN	Pathway upregulated in SLE45
Thymus-expressed chemokine (CCL25; TECK)	1.68	0.001	97**	100	87.5	spleen weight
TNF receptor superfamily member 1B (TNF-RII)	1.96	0.024	94**	75	100	A/C ratio	SLE > control20, linked to renal disease36,37
Thymus and activation regulated chemokine (CCL17; TARC)	1.54	0.019	81*	75	87.5	serum BUN	SLE > control, ↓ with treatment34
C-X-C motif chemokine 13 (CXCL13; BLC)	1.86	0.005	91**	100	75	A/C ratio	SLE > control20, ↑ in renal disease35	Blockade improved nephritis in MRL/lpr mice15

3.6. Validation of array results using ELISA

To assess the validity of the array results and the broader applicability of the array platform to experimental lupus research, we measured serum levels of ten proteins using ELISA (RayBiotech, R&D): adiponectin, EPO, FGF-21, G-CSF, GM-CSF, IL-6, IL-17, L-selectin, MDC, and VCAM-1. These markers were chosen based on screening array fold increases, extracellular location, and ELISA kit availability (Figure 1B). When selecting these validation markers, we assumed ELISA would reliably measure proteins strongly increased in MRL/lpr serum and located outside cells. Moreover, several of these validation markers, including IL-6, IL-17, L-selectin, and VCAM-1, also have established roles in human SLE.

When evaluating the fold change on the screening array (aFC) and the ELISA assay (eFC), IL-6 (aFC: 2.05; eFC: 18.7; p < 0.01), IL-17 (aFC: 2.23; eFC: 1.58; p < 0.05), L-selectin (aFC: 3.96; eFC: 3.02; p < 0.001), MDC (aFC: 1.70; eFC: 2.37; p < 0.001), and VCAM-1 (aFC: 2.84; eFC: 6.78; p < 0.001) showed significant increases in MRL/lpr sera by both platforms. ELISA-measured serum levels of anti-dsDNA antibodies, a standard lupus serum marker, served as our reference for the ability to distinguish between lupus and non-lupus mice (Table 3).

Table 3. ELISA validation of serum biomarkers identified by the screening array.

RayBiotech and R&D ELISA kits were used to measure individual marker levels in mouse sera. When serum availability limited the number of validation candidates that could be tested on samples from the screening array mice, serum from other age-matched cohorts of female mice was substituted. The inclusion of additional cohorts was required to yield enough serum to assess the levels of each marker. Student's 2-tail t-test was used for ELISA comparison. Serum anti-dsDNA titers were used as a reference benchmark for comparison against the area-under-the-curve (AUC), sensitivity, and specificity values of the other markers. These ten markers were chosen for their range of array-measured increases and extracellular location. P-values: * < 0.05, ** < 0.01, *** < 0.001

Proteins used for validation	Array		ELISA
	Fold change (lpr : mpj)	q- value	n (lpr + mpj)	Fold change (lpr : mpj)	AUCp-val (%)	Sensitivity (%)	Specificity (%)
Fibroblast growth factor 21 (FGF-21)	1.78	0.007	16 (8 + 8)	1.54	63	75	57
Interleukin-6 (IL-6)	2.05	0.005	30 (15 + 15)	18.70**	85***	84	78
Interleukin-17 (IL-17)	2.23	0.001	14 (8 + 6)	1.58*	69*	50	84
L-selectin	3.96	0.009	16 (8 + 8)	3.02***	100***	100	100
Macrophage-derived chemokine (MDC)	1.70	0.003	15 (7 + 8)	2.37***	98**	100	87.5
Vascular cell adhesion protein 1 (VCAM-1)	2.84	0.005	16 (8 + 8)	6.78***	100***	100	100
Anti-dsDNA	-	-	16 (8 + 8)	2.23*	80*	75	87.5

FGF-21 (aFC: 1.78; eFC: 1.54; p: 0.347) showed a similar but non-significant eFC. Two proteins, EPO (aFC: 2.29; eFC: 1.08; p: 0.898) and G-CSF (aFC: 3.25; eFC: −1.01; roles in human SLE. p 0.448), showed relative equivalence of serum levels in lupus mice compared to non-lupus controls on ELISA. GM-CSF levels in both genotypes of mice were below the detection threshold of ELISA. Finally, adiponectin (aFC: 1.55; eFC: −1.75; p < 0.01) levels were significantly decreased in lupus mice on ELISA. Additionally, we compared array findings to published results[14, 19-23]. We found ten additional proteins increased in MRL/lpr sera on the array, including CD40, CXCL10, CXCL16, IFN-gamma, IL-5, IL-10, MCP-1, MCP-3, MCP-5, and MIP-3b, which were consistent with data in human lupus patients, MRL/lpr mice, or both (Supplementary Table 3).

4. DISCUSSION

This study's primary goal was to generate a comprehensive and high-resolution snapshot of the circulating MRL/lpr proteome, as this would constitute a valuable resource of potential lupus biomarkers and novel therapeutic targets. Most prior studies investigating serum protein levels in this model focused on inflammatory cytokines and chemokines with established roles in human lupus (i.e., interleukins, interferons, TNF family members), leaving less obvious but potentially relevant mediators (i.e., second messengers and other intracellular mediators of inflammation) underexplored. Importantly, both cytoplasmic and nuclear proteins were included in our study.

4.1. Use of a novel array technique to generate a comprehensive view of the MRL/lpr serum proteome

Existing high-throughput proteomic studies of lupus mice largely measured protein levels within organs such as the lung[24] or kidney[25, 26]. Urinary protein screens, such as those performed by our group[14], have shown a convincing overlap between mouse model and patient findings. Proteomic profiling of the serum in lupus mice has been limited to studies of pre-selected cytokines and antibodies. We found only one prior study that measured levels of many circulating proteins in MRL/lpr and control MRL/mpj mice[27]; however, this work detected few novel markers and did not attempt to relate findings to disease or pathology. Prior to our study, there were no reports of a large-scale, unbiased protein microarray screen of lupus mouse serum paired with measures of disease manifestations.

MRL/lpr serum increases detected using this novel array platform were confirmed for 16 of the 21 proteins, as assessed by subsequent ELISA analyses or previous studies or both. Of note, while similar in magnitude, the ELISA-measured fold change in FGF-21 was not significant. A wider dispersion of measured FGF-21 concentrations on ELISA (normalized range = 2.71) compared to array (normalized range = 0.81) within the MRL/lpr group likely accounts for the reduced statistical significance. Due to the amount of serum needed, ELISAs were generally performed on different cohorts of mice than those used for the array. Therefore, we consider the high concordance between array and ELISA findings to be especially noteworthy and to reflect the array results' general applicability to the MRL/lpr mouse strain. Furthermore, our protein enrichment analyses indicated that the proteomic methodology detected known aspects of the immunopathology of SLE and disease-related pathways, offering further validation.

Among the proteins with the largest fold increases in lupus, L-selectin[28], VCAM-1[29], and IL-10[20] have documented immune effects and were previously shown to be elevated in the serum of lupus patients. Many other top proteins increased in these lupus mice, such as interferon-gamma and IL-17, also have essential immune functions and can promote tissue damage[7, 10]. On the other hand, some of the most reduced serum proteins, including VEGF-C[30], have anti-inflammatory properties.

4.2. Human SLE findings are replicated in MRL/lpr mice with high fidelity

Our results demonstrated striking similarities in serum markers and disease pathways between the MRL/lpr lupus model and patient disease. When we assessed the likely cell types producing those proteins correlated with disease features, leukocytes were the candidate sources for most of these correlates. The involvement of both myeloid and lymphoid cells mirrors known SLE pathogenic mechanisms[31]. Macrophages and T-cells represented the specific cell types of each category most linked to renal disease correlates. Both types of immune cells comprise large fractions of the kidney infiltrates seen in SLE[32], and their increased levels in the infiltrates correlate with worse lupus nephritis[33, 34].

Several elevated proteins we identified as potential biomarkers of renal disease, including BLC[20], TARC[35], and TNF RII[20], are also increased in the serum of SLE patients. We found that BLC correlated with renal dysfunction, while prior work from our group showed that BLC blockade attenuated renal disease in the same MRL/lpr mouse strain[15]. Both findings further corroborate patient data showing that BLC is elevated with flares of renal disease[36]. Interestingly, studies of serum levels of TNF RII highlight its complex relationship with renal dysfunction[20, 37, 38]. Focused biomarker studies should be carried out to confirm or clarify the sensitivity, specificity, and accuracy of serum BLC, TARC, and TNF RII as biomarkers of renal dysfunction in SLE.

In addition to specific proteins, our protein enrichment findings corroborated known cellular pathways participating in lupus pathophysiology[10, 31]. In our study, we found drivers of murine lupus; these included increased NFkB, JAK-STAT signaling, IL-6 production, IL-17 signaling, and responses to IFN-gamma and TNF. Each of these components has a well-established role in human lupus[31], and these cytokines have been previously explored as biomarkers and even therapeutic targets[7].

From our enrichment analyses, positive regulation of ERK1 and ERK2 signaling was related to renal disease. ERK1 and ERK2 are two downstream kinases of the mitogen activated protein kinase (MAPK) pathway, which facilitates cellular responses to extracellular signals[39]. TNF activation of NF-kB was previously shown to involve ERK1/2[40]. Interestingly, our network analyses identified NF-kB as the top transcription factor related to proteins differentially regulated in MRL/lpr serum. While ERK signaling has been implicated in multiple sclerosis and rheumatoid arthritis patients[41], few studies have investigated the role of ERK1/2 in lupus nephritis. These pathways were increased in lupus B-cells in our group's previous studies[42]. Increased ERK1/2 could propagate the vicious cycle of inflammatory cytokine signaling in lupus and may be amenable for targeting by FDA-approved ERK1/2 inhibitors used to treat oncologic indications.

4.3. Discovery of novel biomarker candidates

Here, we sought to utilize a unique high-throughput proteomic platform to identify previously underexplored biomarker candidates. However, study of the MRL/lpr model alone is insufficient to establish the clinical utility of potential lupus biomarkers. While our study is a necessary first step, future studies in SLE patients must establish the translational significance and clinical utility of our findings.

For translational reasons[6], we chose in our study to prioritize those markers increased in the serum of MRL/lpr mice and those that positively correlated with disease manifestations. By categorizing protein correlates according to their cellular location, we noted the importance of extracellular and membrane-bound proteins in lymphoproliferative or renal disease. Through our protein enrichment analyses, we further identified a core group of promising candidates: BLC, TARC, TNF RII, TECK, TSLP, PDGFR-alpha, and MDC. With this list of seven candidates, the elevated levels of which have not been described previously in MRL/lpr mice, we compared their biomarker performance using ROC analysis. Each of these seven proteins yielded ROC AUCs, sensitivities, and specificities equal to or greater than those of anti-dsDNA, a conventional yardstick. Among these seven, the top-performing candidates were serum TECK, TSLP, and MDC.

As previously discussed, BLC, TARC, and TNF RII replicated results from clinical lupus studies. TECK, a.k.a. CCL25, is secreted by thymic medullary dendritic cells and recruits marrow-derived thymocytes to the thymus for T-cell development[43]. Serum levels of TECK were increased in MRL/lpr mice and strongly correlated with splenomegaly. TSLP, a member of the interleukin-2 cytokine family, mediates the transition from an innate immune response to a type 2 humoral response[44]. Components of its downstream signaling pathway were upregulated in lupus patients[45]. PDGFR-alpha, or CD140a, primarily receives signals which regulate connective tissue. This receptor mediates the development of glomerular mesangial cells, alters fibroblast characteristics, and has been linked to kidney pathology[46]. Indeed, kidney levels of PDGFR-alpha are increased in lupus patients with active renal disease[47], and soluble forms of the receptor circulate in human serum[48]. Moreover, this represents a gene that confers increased susceptibility to lupus nephritis[47]. Thus, TECK, TSLP, and PDGFR-alpha are new biomarker candidates with plausible pathophysiologic contributions to SLE and its organ manifestations. However, we were unable at this time to perform validation ELISA's for all candidate biomarkers. This technical limitation is due to minimal residual serum volume from the mice used to screen the initial proteome and small dilution factors required to measure these serum proteins.

Our array analyses, confirmed by ELISA testing, identified that MDC was significantly associated with lymphoproliferation and renal dysfunction. MDC, a.k.a. CCL22, derives from dendritic cells and macrophages to promote the infiltration of immune cells, particularly activated T-cells, into local sites of inflammation[49]. Some studies point to higher MDC levels in lupus patient serum[50]. Other studies not replicating these results were small and not well-controlled [51]. Nevertheless, in rat models of kidney disease, MDC mediates immune cell glomerular infiltration in anti-GBM and crescentic glomerulonephritis[49, 52]. This mechanism could explain our findings which pointed to MDC's relationship with both lymphoproliferation and renal dysfunction. These findings necessitate well-designed clinical biomarker studies to interrogate the role of MDC as a non-invasive marker of ongoing renal pathology.

Finally, leptin and the soluble leptin receptor, both the focus of clinical studies[53], were increased in the MRL/lpr mice serum on array screening. In our study, leptin did not correlate with disease features. The leptin receptor, however, met many of our criteria and correlated strongly with both renal pathology and dysfunction; however, it did not contribute to disease-relevant terms in our protein enrichment analyses. Though we excluded it from our downstream analyses, our results along with conflicting patient data support the study of circulating leptin receptor as a marker of lupus nephritis.

5. CONCLUSION

Significant progress in our understanding of SLE has been made over the last decade. Despite these advances, prognostic biomarkers for specific disease manifestations are limited, and current options for therapy can be toxic and too often yield suboptimal results. To ultimately improve patient outcomes, we validated a novel 1308-plex screen of MRL/lpr and control mouse serum with two main goals. First, we sought to generate the first comprehensive proteome of the MRL/lpr lupus mouse. Second, we aimed to identify novel biomarkers of specific SLE manifestations and identify potential new therapeutic targets.

We successfully detected 406 increased and 397 decreased proteins in the serum of MRL/lpr mice. Our array results and correlation analyses identified several viable biomarker candidates. BLC, TARC, and TNF RII are well-studied but require dedicated clinical assessments of biomarker characteristics. TECK, TSLP, PDGFR-alpha, and MDC are under-explored and represent potentially novel biomarkers of renal disease in SLE. In addition, the role of ERK1 and ERK2 signaling in renal pathogenesis appears to be an exciting future direction. These results can also guide future mechanistic studies in lupus animal models and may eventually provide new directions for clinical investigation.

Supplementary Material

Supplementary Figure 1

Supplementary Figure 1. Network analysis of proteins increased in MRL/lpr serum. Analysis of interactions between the 549 proteins with a p-value < 0.01 and fold change > 1.0, elevated in the lupus serum. A) Shown is the overall Cytoscape network depicting all interactions and the top cluster within it (B). The top five reactome pathways for this cluster are SRP-dependent cotranslational protein targeting to membrane, nonsense mediated decay independent of the exon junction complex, formation of a pool of free 40S subunits, L13a-mediated translational silencing of ceruloplasmin expression, and GTP hydrolysis and joining of the 60S ribosomal subunit. Each circle represents a protein. Protein color is based on the fold change value of the protein. The minimum value is 0.18 and the maximum value is 3.96. The gradient turns to red at fold change of 2.

Supplementary Figure 2

Supplementary Figure 2. Likely cellular origins of biomarkers. Considering only those 92 proteins which were both increased in the serum of MRL/lpr mice and correlated with a feature of lupus-like disease, we consulted established protein atlases (i.e. Mouse Genome Database, The Human Protein Atlas) to determine the likely cell type secreting each protein. Data is presented in pie-chart form as percentage of all 92 proteins (A, B) or percentage of the 50 proteins which related to renal disease (C).

HIGHLIGHTS.

• Better biomarkers of ongoing organ pathology in SLE and lupus nephritis are needed

• 1308 mouse serum proteins assessed for correlation with organ pathology

• First comprehensive serum proteome of a lupus mouse model

• Disease-relevant candidate biomarkers, such as MDC, ready for clinical assessment

Abbreviations

6Ckine

C-C motif chemokine ligand 21 (CCL21)

aFC

array fold change

BLC

B lymphocyte chemoattractant (CXCL13)

BUN

blood urea nitrogen

CCL28

C-C motif chemokine ligand 28

COL9A3

collagen type 9 alpha 3 chain

CTACK

cutaneous T-cell-attracting chemokine (CCL27)

CXCL10

C-X-C motif chemokine ligand 10

CXCR3

C-X-C chemokine receptor type 3

EDAR

ectodermal dysplasia receptor

eFC

ELISA fold change

EPO

erythropoietin

ERK

extracellular regulated MAP kinase

FC

fold change

FDR

false discovery rate

FGF-21

fibroblast growth factor 21

G-CSF

granulocyte colony stimulating factor

GDF7

growth and differentiation factor 7

GM-CSF

granulocyte-monocyte colony stimulating factor

GO

gene ontology

ICAM-1

intercellular adhesion molecule 1

IFN

interferon

IL1 R4

interleukin-1 receptor type 4

IL-1

interleukin-1

IL-10

interleukin-10

IL-15

interleukin-15

IL-17

interleukin-17

IL-27

interleukin-27

IL-5

interleukin-5

IL-6

interleukin-6

IPA

Ingenuity Pathway Analysis

JAK

janus kinase

KEGG

Kyoto Encyclopedia of Genes and Genomes

MCP-1

monocyte chemoattractant protein 1 (CCL2)

MCP-2

monocyte chemoattractant protein 2 (CCL8)

MCP-3

monocyte chemoattractant protein 3 (CCL7)

MCP-5

monocyte chemotactic protein 5 (CCL12)

MDC

macrophage-derived chemokine (CCL22)

MIP-3b

macrophage inflammatory protein 3 beta (CCL20)

PDGFR-alpha

platelet-derived growth factor receptor alpha

SLE

systemic lupus erythematosus (lupus)

STAT

signal transducer and activator of transcription

TARC

thymus and activation regulated chemokine (CCL17)

TECK

thymus-expressed chemokine (CCL25) TLR: toll-like receptor

TNF RII

tumor necrosis factor receptor type 2

TNF

tumor necrosis factor

TRAIL

TNF superfamily member 10

TSLP

thymic stromal lymphopoietin

VCAM-1

vascular cell adhesion molecule 1

VEGF

vascular endothelial growth factor

Data availability statement:

Data are available on reasonable request.