Article
Dynamic Response Genes in CD4+ T Cells Reveal a
Network of Interactive Proteins that Classifies
Disease Activity in Multiple Sclerosis
Graphical Abstract
Highlights
d
Dynamic CD4+ T cell responses in vitro revealed a
dysregulated network module in MS
d
This module was highly enriched for GWAS genes and
translated into protein expression
d
A combined protein score readily predicted disease activity
and treatment response
d
Integrating GWASs with dynamic expression profiling
disclosed personalized biomarkers
Authors
Sandra Hellberg, Daniel Eklund,
Danuta R. Gawel, ..., Maria C. Jenmalm,
Mika Gustafsson, Jan Ernerudh
Correspondence
daniel.eklund@liu.se (D.E.),
mika.gustafsson@liu.se (M.G.)
In Brief
Hellberg et al. have constructed a highly
connected gene module of dysregulated
genes in multiple sclerosis patients in
which the gene products can collectively
be used to classify patients and predict
disease activity and response to
treatment. This set of key proteins holds
promise as clinically useful biomarkers in
personalized MS treatment.
Accession Numbers
GSE78244
Hellberg et al., 2016, Cell Reports16, 2928–2939 September 13, 2016ª 2016 The Author(s). http://dx.doi.org/10.1016/j.celrep.2016.08.036
Cell Reports
Article
Dynamic Response Genes in CD4+ T Cells Reveal
a Network of Interactive Proteins that Classifies
Disease Activity in Multiple Sclerosis
Sandra Hellberg,1,8Daniel Eklund,1,8,*Danuta R. Gawel,2Mattias Ko¨pse´n,2,3Huan Zhang,2Colm E. Nestor,2
Ingrid Kockum,4Tomas Olsson,4Thomas Skogh,5Alf Kastbom,5Christopher Sjo¨wall,5Magnus Vrethem,6
Irene Ha˚kansson,6Mikael Benson,2Maria C. Jenmalm,1Mika Gustafsson,3,9,*and Jan Ernerudh7
1Department of Clinical and Experimental Medicine, Linko¨ping University, 581 83 Linko¨ping, Sweden
2The Centre for Individualised Medicine, Department of Clinical and Experimental Medicine, Linko¨ping University, 581 83 Linko¨ping, Sweden 3Bioinformatics, Department of Physics, Chemistry and Biology, Linko¨ping University, 581 83 Linko¨ping, Sweden
4Neuroimmunology Unit, Department of Clinical Neuroscience, Karolinska Institutet, Stockholm, 171 77 Linko¨ping, Sweden
5Department of Rheumatology and Department of Clinical and Experimental Medicine, Linko¨ping University, 581 83 Linko¨ping, Sweden 6Department of Neurology and Department of Clinical and Experimental Medicine, Linko¨ping University, 581 83 Linko¨ping, Sweden 7Department of Clinical Immunology and Transfusion Medicine and Department of Clinical and Experimental Medicine, Linko¨ping University,
581 83 Linko¨ping, Sweden
8Co-first author 9Lead Contact
*Correspondence:daniel.eklund@liu.se(D.E.),mika.gustafsson@liu.se(M.G.) http://dx.doi.org/10.1016/j.celrep.2016.08.036
SUMMARY
Multiple sclerosis (MS) is a chronic inflammatory
dis-ease of the CNS and has a varying disdis-ease course
as well as variable response to treatment.
Bio-markers may therefore aid personalized treatment.
We tested whether in vitro activation of MS
pa-tient-derived CD4+ T cells could reveal potential
bio-markers. The dynamic gene expression response to
activation was dysregulated in patient-derived CD4+
T cells. By integrating our findings with
genome-wide association studies, we constructed a highly
connected MS gene module, disclosing cell
acti-vation and chemotaxis as central components.
Changes in several module genes were associated
with differences in protein levels, which were
measurable in cerebrospinal fluid and were used to
classify patients from control individuals. In
addi-tion, these measurements could predict disease
activity after 2 years and distinguish low and high
re-sponders to treatment in two additional,
indepen-dent cohorts. While further validation is needed
in larger cohorts prior to clinical implementation,
we have uncovered a set of potentially promising
biomarkers.
INTRODUCTION
Multiple sclerosis (MS), a chronic inflammatory disease of the CNS, is a common cause of neurological disability in young persons (Dendrou et al., 2015). The introduction of new
treat-ments has improved the situation for many patients, yet some patients do not respond to a given treatment and treatments are costly and associated with side effects, sometimes severe. Thus, biomarkers for personalized treatment would be highly beneficial (Derfuss, 2012). The development of biomarkers is hampered by the heterogeneity and complexity of the disease process in MS, in which the underlying disease mechanisms reflect the interplay between a large number of genes and their downstream targets. Accordingly, genome-wide associa-tion studies (GWASs) have revealed a growing list of genes associated with MS (Beecham et al., 2013). While subse-quent genome-wide expression profiling of immune cells has increased our functional understanding of disease mechanisms in MS, these studies have so far not resulted in clinically useful biomarkers (Comabella and Montalban, 2014; Kemppinen et al., 2011).
Previous gene expression profiling studies in MS primarily examined unstimulated cells, i.e., a ‘‘snapshot’’ of how the cells appear in vivo, where most cells are not stimulated. We hypothesized that the dynamic response to activation would reveal disease-associated pathways that were not found during conventional profiling of blood cells. We there-fore examined gene expression in unstimulated as well as in stimulated cells, with the goal of identifying relevant genes that might translate into clinically valuable biomarkers. We focused our attention on CD4+ T cells because (1) they are early and important regulators of adaptive immunity (Sallusto, 2016), (2) substantial evidence implicates peripherally acti-vated CD4+ T cells in MS pathogenesis both in humans (Den-drou et al., 2015; Fletcher et al., 2010; Okuda et al., 2005; Putheti et al., 2003) and in animal models (Fletcher et al., 2010), and (3) the association of MS with the CD4+ T cell-spe-cific antigen-presenting molecule HLA Class II (DRB1*15:01) is well established, as well as the more recent identification of
several other T cell-associated genes in large GWASs of MS, such as the interleukin (IL)-2 receptor alpha chain, the IL-12 receptor, and IL-7 (Sawcer et al., 2011; Patsopoulos et al., 2011).
By combining traditional analysis of differential gene expres-sion in stimulated CD4+ T cells with a network-based modular approach, we found that aberrant dynamic response (DR) genes, i.e. genes behaving differently upon stimulation in pa-tients than in control individuals, were highly enriched for MS genes, as defined by GWAS data (Sawcer et al., 2011). The aberrantly expressed genes were associated with immune activation and chemotaxis and a set of them encoded secreted proteins (chemokine [C-X-C motif] ligand 1-3 [CXCL1-3], CXCL10, chemokine [C-C motif] ligand 2 [CCL2], and osteopon-tin [OPN]). A combination of these proteins, as measured in ce-rebrospinal fluid (CSF), readily classified patients and controls and, in two new cohorts, predicted 2-year disease activity in early MS as well as response to treatment, respectively.
RESULTS
Stimulation of CD4+ T Cells Reveals Differently Expressed Genes Enriched for MS-Associated GWAS Genes
The principal workflow of the study was as follows: identifying genes that were dysregulated in isolated and activated CD4+ T cells from MS patients (Figure 1A); investigating whether the identified genes were known to be associated with MS (Fig-ure 1B); identifying functional relationships of the dysregulated genes (Figures 1C and 1D); and testing whether this information might be useful for identifying biomarkers in new cohorts of MS patients (Figure 1E). To assess gene expression dynamics in CD4+ T cells, we used an in vitro model, in which isolated CD4+ T cells were cultured unstimulated or with commonly used combined activation through the TCR (anti-CD3) and co-stimulatory CD28 receptor (anti-CD28) (Sharpe and Freeman, 2002). We used a moderate level of stimulation, as measured
Figure 1. Overview of the Study
The figure shows the principle of the different analyses performed in the study.
(A) CD4+ T cells from both MS patients and controls were cultured in vitro for 24 hr in the presence or absence of 0.03 mg/ml anti-CD3 and anti-CD28 antibodies. After mRNA isolation and microarray analysis of gene expression levels, three different datasets of differentially expressed genes between patients (n = 14) and controls (n = 14) were generated, as follows: (1) in unstimulated CD4+ T cells, (2) in stimulated CD4+ T cells, and (3) the dynamic response dataset represented by the unstimulated expression subtracted from the stimulated expression.
(B) Datasets were overlapped with MS-associated genes identified from GWAS and analyzed for enrichment to validate relevance for MS.
(C) The average distance in the protein-protein interaction network between random genes and the MS-associated genes present in the dynamic response dataset was calculated.
(D) The dynamic response dataset was used to construct a module, which was further dissected into submodules with similar biological function.
(E) Four genes, selected from the module, and their corresponding proteins were measured in supernatants of stimulated and unstimulated CD4+ T cells as well as in cerebrospinal fluid (CSF) from MS patients and controls. The combined score of these proteins was used to separate patients from controls and predict evidence of disease activity within 2-year follow-up as well as response to treatment in three different cohorts of MS patients.
by CD69 surface expression (Figure S1), a marker of global T cell co-stimulation and activation. Gene expression profiling (22,610 genes) was performed on unstimulated and stimulated CD4+ T cells from a cohort of patients (n = 14;Table 1) with relaps-ing-remitting MS, without treatment prior to the sampling
occa-sion, as well as cells from matched control subjects (n = 14; Table 1). In addition, to assess the dynamic response to stimula-tion, we calculated the changes in gene expression. This was done by subtracting expression in unstimulated cells from expression in the stimulated cells in both patient and control cells (unlogged data were used). The groups were thereafter compared statistically, i.e. the change in gene expression upon activation was compared between patients and controls. We refer to these genes as dynamic response (DR) genes. When comparing patients and controls using an unpaired t test (unad-justed double-sided p < 0.05), we identified nominally differen-tially expressed genes in unstimulated CD4+ T cells (n = 2,075), stimulated CD4+ T cells (n = 1,250), and DR genes ((stim-ulated) (unstimulated), n = 932)). However, when adjusting for genome-wide testing (at 20% Benjamin Hochberg false discov-ery rate), no genes were differentially expressed in any of the three gene lists (unstimulated, stimulated, and DR). We thus dis-placed the single gene analysis and proceeded by analyzing groups of genes, starting by testing whether any of the lists were enriched for genes relevant to MS. By assessing original GWAS data from the International MS Genetic Consortium (Sawcer et al., 2011) based on25,000 patients and controls, and by using a moderate stringency cut off (p < 10 5), we identified 760 MS-associated SNPs that mapped, through the closest transcription start site, to 216 unique genes (hereafter referred to as MS genes;Table S1). To investigate whether our three gene lists with differently expressed genes (unstimulated, stimulated, and DR) actually contained genes that were associ-ated with MS (as defined by previous GWASs), we overlapped the three gene lists with the 216 MS genes and computed statis-tical enrichment using Fisher’s exact test (Figures 1B and2). The highest fold enrichment (FE) of MS genes was found for the DR genes (FE = 2.08, n = 19, p = 2.03 10 3;Figure 2). These 19 genes are referred to as MS-DR genes. Unstimulated CD4+ T cells showed a significant, but smaller, fold enrichment (FE = 1.65, n = 44, p = 5.3 3 10 4;Figure 2), while the stimulated gene set showed no enrichment (FE = 0.93, n = 14, p = 0.648; Figure 2) of GWAS genes (Table S1). While the unstimulated gene set also included a high proportion of GWAS genes, we decided to perform subsequent analyses using the DR genes, since the primary objective of the study was to investigate dys-regulated genes upon activation of CD4+ T cells from MS pa-tients. To rule out the possibility that the GWAS genes found among the DR genes were not generally associated with autoim-mune disease, but specific to MS, we calculated the FE of GWAS genes from two other autoimmune diseases, rheumatoid arthritis (RA) and type I diabetes (T1D). For RA, we analyzed around 2,500,000 SNPs (Stahl et al., 2010) and found 81 genes at a mod-erate stringency cut off (p < 10 5). Only four of these RA-associ-ated genes were found among the DR genes (FE = 1.16, p = 0.46). For T1D, about 500,000 SNPs were analyzed (Barrett et al., 2009), of which 109 genes were considered to be associ-ated with this disease (p < 10 5). Only two of these T1D-associ-ated genes were found among the DR genes (FE = 0.43, p = 0.95). Taken together, genes that were differentially expressed during in vitro activation of CD4+ T cells from MS patients, rep-resenting the dynamic response to stimulation, were highly asso-ciated with MS-relevant genes, as defined by GWASs.
Table 1. Characteristics of the MS Cohorts Controlsa MS
Study Cohort
n 16 16
Relapsing remitting (n) 16
Median age, years (range) 41.5 (27–51) 40.5 (26–52)
Sex (M/F) 0/16 0/16
Median disease duration, years (range)
9.5 (0–17)
Annual relapse rate (mean)b 0.4
Median EDSS (range) 2.5 (0–7)
Median MSSS (range) 2.37 (0.09–9.21) Previous treatmentc Natalizumab 8 Fingolimod 1 IFN-b 4 Intravenous immunoglobulins 2 No previous treatment 1 Early MS Cohortd n 11 41 Possible MS 19 Relapsing remitting (n) 22
Median age, years (range) 30 (22–47) 31 (21–62)
Sex (M/F) 2/9 9/32
Median disease duration, months (range)
3.5 (0.3–120)
No previous treatment 41
Response to Treatment Cohorte
n 15
Median age, years (range) 34 (25–46)
Sex (M/F) 12/3
EDSS, Expanded Disability Status Scale; MS, multiple sclerosis; MSSS, MS severity score. M, male; F, female.
a
Controls in both cohorts consisted of blood donors or employees at Linko¨ping University Hospital. All controls were free from disease, and medication and the controls in the early MS cohort had normal CSF find-ings and no evidence of neurological disease.
bCalculated over 2 years prior to sampling.
cNone of the patients had received immunomodulatory or
immunosup-pressive treatment for at least 2 months prior to sampling, with the excep-tion of one patient who received intravenous immunoglobulin that was stopped 15 days before sampling.
dConsecutively included when first seeking medical attention at the
Department of Neurology, Linko¨ping University Hospital, and fulfilling criteria of clinical isolated syndrome (CIS) or relapsing-remitting (RR) MS according to the revised MacDonald criteria from 2010 (Polman et al., 2011).
e
This cohort was described previously (Gustafsson et al., 2014; Meller-ga˚rd et al., 2010, 2013).
Dynamic Response Genes Co-localize in the PPI Network
It is known that proteins encoded by MS-associated genes are more likely to be connected in the protein-protein interac-tion (PPI) network (Internainterac-tional Multiple Sclerosis Genetics Consortium, 2013), thereby suggesting a functional relation-ship. We therefore examined whether the 19 MS-DR genes identified herein were also functionally related (Figure 1C). To test this we used all high confidence human protein inter-actions (n = 413,896) from among 14,543 proteins (STRING database v.10 [Szklarczyk et al., 2015]). Analyzing this network showed that most of the MS-DR genes were co-localized in a neighborhood with a significantly shorter dis-tance, compared to randomly equally sized sets of proteins (harmonic average < d > = 2.9 versus < drand> = 5.2; permu-tation test p = 2.03 10 3), where distance is defined as the smallest number of interacting proteins between two given genes. This analysis suggests a functional relationship be-tween the MS-DR genes.
A Module-Based Approach Reveals a Network of Dysregulated Dynamic Response Genes with Chemotaxis and Immune Regulation as Central Components
The observed co-localization of MS-DR genes, together with the fact that the DR gene list was enriched for MS-associated genes, motivated us to use a network-based modular approach to further search for functionally related genes among the 932 DR genes with the purpose of finding the most interconnected DR genes and their most interactive proteins (Figure 1D). A modular approach means that we include both gene products that are physically interconnected in the protein network, as well as other tightly interconnected gene products (seeExperimental Proced-ures). These module genes are likely to be functionally related in cellular processes and are thus considered to belong to a dis-ease module. The resulting MS module contained 75 protein-coding genes that differed significantly between patients and controls (Figures 3A and 3B) and corresponded to proteins involved in similar biological processes. To further elucidate whether the module was associated with MS, we tested whether these 75 genes overlapped with the MS genes that were identi-fied using GWAS data. As shown inFigure 3C, the module was highly enriched for MS genes (FE = 7.26, n = 7, Fisher’s exact test, p = 5.03 10 5), thereby confirming that the module was relevant for MS. To understand the biological function of this module, we performed a gene ontology (GO) analysis using DAVID (Huang et al., 2009) and defined five submodules based on their localization within the module. The two larger submod-ules contained significant enrichment of several GO terms related to cell activation and chemotaxis (Table S2). In addition to the MS-associated genes identified by GWASs, the module contained both well-known genes previously linked to MS, such asSPP1 (encoding for OPN) (Bo¨rnsen et al., 2011; Shimizu et al., 2013) andTNFRSF1A (Gregory et al., 2012), as well as less-studied genes of interest, such asNTF3 and several chemo-kine genes. Collectively, by integrating GWAS and expression profiling data and by using a modular approach to genes that differed in their dynamic response to stimulation, we identified a gene network of potential relevance to MS. The dominating modules contained genes involved in intracellular signaling and chemotaxis, and both well-known and less-explored biomarker candidates were represented.
Module Proteins in Cell Culture Supernatants Are Able to Classify Patients versus Controls
Since proteins measurable in fluids are most suitable as bio-markers, we investigated whether the observed differences in gene expression between patients and controls could be asso-ciated with differences in protein expression. A total of six pro-teins were selected from the two main submodules (Figure 3) based on whether the protein was (1) known to be a secreted product, (2) associated with CD4+ T cells, and (3) possible to measure by commercial ELISA or multiplex bead technology. Next, we measured the selected proteins (CCL2, CXCL1-3, CXCL10, OPN, neurotrophin [NT]-3, and soluble tumor necrosis factor receptor 1 [sTNFR1];Table S3) in supernatants collected from unstimulated and stimulated CD4+ T cells from MS pa-tients (n = 16) and healthy controls (n = 16). The supernatants
Figure 2. Dynamic Response Genes Are Highly Enriched for MS-Associated Genes
(A and B) Three datasets were generated by gene expression profiling of isolated CD4+ T cells from MS patients (n = 14) and controls (n = 14), as fol-lows: (1) differentially expressed genes in unstimulated cells, (2) stimulated cells, and (3) the dynamic response to stimulation, defined as the expression in stimulated cells subtracted by the expression in the unstimulated cells (unpaired t test, unadjusted p < 0.05). (A) The genes in these three datasets were overlapped with the 216 MS-associated genes (as defined by GWAS), and (B) fold enrichment was calculated using Fisher’s exact test. See also
were obtained from the same primary cohort in which gene expression was analyzed. As with the dynamic gene expression data, protein levels were defined as the difference in concentra-tion between stimulated and unstimulated cells (referred to as DR protein levels;Figure S2). In line with the gene expression data, there was a significant increase in the levels of OPN (Mann-Whitney test, p = 0.007) and a tendency toward an in-crease in CXCL1-3 (p = 0.09) and CXCL10 (p = 0.08) in patient samples, while no difference was observed in the levels of CCL2 (p = 0.39). NT-3 and sTNFR1 were undetectable in all samples.
To see if the secreted protein levels after stimulation of CD4+ T cells could distinguish patients from controls in our primary study cohort, we used the DR protein levels and tested the ability of a sin-gle protein or the combination of the four proteins to classify pa-tients and controls. As a combined disease score, the protein levels of OPN, CXCL1-3, and CXCL10 were added, while the CCL2 protein level was subtracted. We found that the sum of all four proteins could discriminate patients and controls with high ac-curacy (area under the curve [AUC] = 0.84, Mann-Whitney test p = 0.0013;Figure 4A), whereas using a single protein or combinations of two or three proteins resulted in poorer discrimination (Table
Figure 3. A Dynamic Response Module Reveals Chemotaxis and Cell Activation as Central Components
(A–C) The 932 dynamic response (DR) genes, which responded differently between patients and controls upon T cell activation, were used to construct a module of the most interconnected DR proteins, by combining it with the human PPI database STRING.
(A) The figure shows the 75 genes included in the module and their individual interactions, where GWAS genes are indicated with red rings. The module was further dissected into biologically relevant submodules using DAVID GO analysis, where the two main submodules (‘‘Chemotaxis’’ and ‘‘Cell activation’’) con-tained statistically significant GO terms.
(B) Heatmap showing the dynamic response (expression levels of stimulated cells expression levels of unstimulated cells) of individual genes in the submodules in patients and controls.
(C) The graph shows the fold enrichment (using Fisher’s exact test) of MS-associated GWAS genes among the 75 module genes as well as the fold enrichment in the original dynamic response (DR) dataset (shown previously inFigure 2).
S4), as did mRNA levels obtained from the expression data (AUC = 0.71, p = 0.063). As a control classifier we constructed a similar combined score by measuring eight other secreted proteins in the same supernatants (granulocyte macrophage colony-stimu-lating factor [GM-CSF], interferon [IFN]-g, IL-10, IL-13, IL-17, CCL17, macrophage colony-stimulating factor [M-CSF], and CCL20;Table S3) that were not part of the module, but are of immunological relevance and known to be affected by stimulation. The 1,000 random combined scores created from the DR protein levels from this set of 8 control proteins (mean AUC = 0.54) were outperformed by the combined score from our selected module proteins (p = 4.383 10 5, FE = 1.55). Taken together, the increase in gene expression of the selected module genes was associated with an increase in protein secretion, and by considering the secreted levels of a combination of proteins in the supernatants we were able to classify patients from healthy controls.
Module Proteins in Cerebrospinal Fluid Are Able to Predict Disease Activity after 2 Years
Encouraged by the possibility of classifying patients from con-trols based on protein levels after stimulation of CD4+ T cells,
we asked if a combined score of proteins could serve as a biomarker in MS (Figure 1E). To become a useful biomarker, it would be advantageous if the proteins could be measured in vivo, i.e. as soluble proteins in plasma or CSF. Of note, these levels would indirectly reflect the dynamic response in vivo, as an increase is probably caused by cell activation. This is particularly pertinent to CSF, where disease-associated, activated T cells are highly enriched. To test biomarker potential in a clinical setting, we used a second, independent cohort of early MS patients (Table 1) and measured the four proteins in CSF and plasma samples obtained when patients first sought medical attention. Since CSF cells could not be activated, as performed for CD4+ T cells, and a dynamic response could therefore not be formed, we inferred their relations in CSF. We performed logistic regression of the four proteins in CSF from patients (n = 41) versus healthy controls (n = 11) and estimated the predictive value by computing the probability by leave-one-out cross-vali-dation in order to assess the prediction error (Table S5). Interest-ingly, the relationships among the four proteins was again inferred as previously, i.e. higher levels of OPN, CXCL1-3, and CXCL10 added to the combined disease score, while higher
Figure 4. Secreted Proteins from the Module Can Be Used to Construct a Combined Protein Score, which Discriminates Patients from Controls and Predicts Disease Activity and Response to Treatment
(A–D) Each subplot represents discrimination of different cohorts, where the top panel depicts boxplots (boxes show the interquartile range and whiskers show the total range) of the combined score and the lower panel represents receiver operating characteristic (ROC) curves.
(A) The combined score from secreted proteins in supernatants defined by (CXCL1-3) + (CXCL10) + (OPN) (CCL2) discriminated patients (n = 16) from controls (n = 16) with high accuracy (AUC = 0.84, Wilcoxon rank test p = 0.0013). Each of the four selected proteins was defined by the difference between the stimulated and unstimulated levels in cultures of CD4+ T cells.
(B) A combined protein score in CSF, defined from the probability computed by leave-one-out cross-validation and logistic regression of the four proteins measured in CSF was able to discriminate early MS patients (n = 41) from controls (n = 11) in a new cohort with high accuracy (AUC = 0.79, p = 3.83 10 3
). (C) The combined score in CSF also discriminated between subgroups of MS patients who showed signs of disease activity (n = 27) from those with no evidence of disease activity (NEDA, n = 12) at 2-year follow-up (AUC = 0.84, p = 9.63 10 4).
(D) The CSF score discriminated also, in an independent cohort, between high (n = 6) and low (n = 9) responders to natalizumab treatment with high accuracy (AUC = 0.89, p = 0.012).
levels of CCL2 decreased the score. We found that the com-bined regression score for proteins in CSF separated patients and controls with a high accuracy (AUC = 0.79, p = 3.8 3 10 3;Figure 4B).
A critical factor in clinical management is prediction of disease activity. At 2-year follow-up, 12 patients had no evidence of dis-ease activity (NEDA) (Rotstein et al., 2015), while 27 had signs of it (non-NEDA). We then tested if our combined score of these four proteins, as measured in CSF (Table S5) at baseline, was able to distinguish patients with or without disease activity by displaying the disease probability scores. We found that the combined score in CSF at baseline was able to predict NEDA af-ter 2 years with high accuracy (AUC = 0.84, p = 9.63 10 4; Figure 4C).
Module Proteins Can Distinguish Low and High Responders to Treatment
Another aspect of personalized treatment is the possibility of predicting response to treatment. We therefore tested, in a third cohort of patients, whether our combined score in CSF (Table S5) could also be used to distinguish subsequent high (n = 6) and low (n = 9) responders to natalizumab, a commonly used drug for MS, which targets a4-integrin and blocks entry of lym-phocytes into the CNS (Mellerga˚rd et al., 2013). Since natalizu-mab is one of the most effective MS drugs in use (Tramacere et al., 2015), the low responders constitute a minor group (Gus-tafsson et al., 2014; Mellerga˚rd et al., 2010) that was defined by having one (n = 7) or two (n = 1) relapses during the first year of treatment or having two relapses during 3 years of treat-ment (n = 1). Although choosing this treattreat-ment is based on several considerations, we reasoned that this cohort (Gustafs-son et al., 2014; Mellerga˚rd et al., 2010), while very small, could indicate the potential for predicting responses to treatment. For these samples we lacked CXCL1-3 values so we re-trained our original model without this protein on the 41 patients with early MS. Remarkably, we found that the combined score obtained in CSF before start of treatment was able to separate high and low responders with high accuracy (AUC = 0.89, p = 0.012; Fig-ure 4D). AUC values for both the second cohort (disease activity) and the third cohort (response to treatment) were not affected by the difference in group size, which we tested by performing 1 million random balanced subsets. In contrast to CSF, the combined protein score in plasma (Table S5) did not distinguish any of the tested clinical groups (disease versus non-disease in the primary and the second cohort, NEDA versus non-NEDA in the second cohort, or high versus low response to treatment in the third cohort).
The Module Proteins Can Classify Patients and Controls in Other Autoimmune Diseases, but They Cannot Predict Disease Activity
Lastly, we asked if the ability of the four proteins to predict the presence of disease and disease activity was limited to MS or if it was relevant in other autoimmune diseases involving CD4+ T cells. To test this we measured the four proteins in serum from patients diagnosed with early RA (n = 40) and systemic lupus erythematosus (SLE, n = 43), as well as a new cohort of healthy donors (n = 40). Similarly to the previous analyses in
the MS cohorts, we performed logistic regression and assessed the predicted probabilities with cross-validation (combined scores are shown in Table S5). Interestingly, we found the same relationship among the proteins as for the MS patients (positive for CXCL1-3, CXCL10, and OPN, but negative for CCL2) and a significant separation for both RA (AUC = 0.89, p = 1.6 3 10 9) and SLE (AUC = 0.77, p = 2.0 3 10 5) with respect to controls. However, when we tested the ability of the combined proteins score to predict disease activity (defined as DAS28 in RA and mSLEDAI in SLE;Table S5) at 2-year follow-up, we found no significant correlation between the combined protein score and DAS28 at 2-year follow-up (Spearman cor-relation = 0.27, p = 0.13) for RA patients, nor between mSLEDAI scores at 2-year follow-up and combined protein score (Spearman correlation = 0.006, p = 0.97) in SLE patients. Furthermore, there was no correlation between the combined protein score and disease activity at inclusion in any of the two cohorts (data not shown). In summary, this analysis suggests the general relevance of using the identified combined biomarker score for autoimmune diseases, but shows that the correlation to disease activity seems to be limited to MS.
DISCUSSION
In the present study we demonstrated an aberrant response to activation in CD4+ T cells from MS patients and found that a set of differentially activated proteins can act as biomarkers in MS. This was achieved by transcriptomic profiling of the dy-namic response to activation in CD4+ T cells, which is in contrast to previous studies based on transcriptomic profiling of unstimu-lated cells. Using a combined protein score based on CSF levels, we were able, in two MS cohorts, to (1) classify patients and con-trols, (2) predict evidence or no evidence of disease activity after 2 years, and (3) classify low and high responders to treatment. Our results confirm that activation of CD4+ T cells is a central process in MS pathogenesis, as suggested by previous GWAS findings (Sawcer et al., 2011). Our findings also suggest that po-tential biomarkers might be identifiable for personalized treat-ment, which is highly needed given variable disease course and responses to given treatments for MS.
In order to validate our approach, we confirmed that each step increased the concordance with GWAS data, thereby perform-ing an a posteriori validation usperform-ing previous evidence supportperform-ing MS relevance. Indeed, MS-GWAS genes, in contrast to GWAS genes from two other autoimmune diseases, RA and T1D, were highly enriched in the DR gene set. Next, based on co-localization of the DR genes, we took a network-based approach and identified functionally interconnected genes in a disease module that was again highly enriched with MS genes. Our finding that chemotaxis forms a prominent submodule was not suggested by previous GWAS analyses, but could potentially be regarded as an event downstream of cell activation. The role of chemokines in MS is supported by several gene expres-sion profiling studies of T cells (Corvol et al., 2008; Mayne et al., 2004; Satoh et al., 2006). In addition, chemotaxis is a pre-requisite for entry of CD4+ T cells to the CNS. From a clinical perspective, the dysregulated chemokine response can be uti-lized for biomarker identification since chemokines are often
measurable in plasma and CSF, in general at higher concentra-tions than cytokines (Mellerga˚rd et al., 2010).
The general TCR-mediated stimulation, affecting all CD4+ T cells, was chosen because myelin-specific T cells are scarce in the circulation (L€unemann et al., 2004). Furthermore, the most relevant specificities of myelin-specific CD4+ T cells may still be unknown, although recent progress may address this issue (Ayoglu et al., 2016). We used both naive and memory CD4+ T cells, since the memory fraction may contain disease-associated memory cells and the naive fraction facilitates the finding of activation-induced dysregulation. In addition, we and others (Okuda et al., 2005) found no difference between patients and controls in the distribution of naive and memory cells.
Of the six proteins selected for biomarker studies, two (sTNFR-1 and NT-3) were not secreted by CD4+ T cells in vitro, probably because the model was designed for detection of gene expression and not protein secretion and levels may have been below the detection limit. However, four proteins could be measured as secreted products from CD4+ T cells, and when combined, their secreted levels could be used to distinguish patients from controls. Notably, three of the proteins were chemokines. CXCL1 is downstream of Th17 and involved in recruitment of granulocytes, a process that was recently high-lighted in experimental autoimmune encephalomyelitis (EAE) and MS, including a potential role as a biomarker (Rumble et al., 2015). CXCL10 is induced by IFN-g and involved in the recruitment of CXCR3+ Th1 cells. Although pertinent to Th1-associated mechanisms, as well as findings of increased CSF levels (Mellerga˚rd et al., 2010; Sørensen et al., 2002), the role of CXCL10 as a biomarker in MS has not been fully explored (Va-zirinejad et al., 2014). CCL2 contributed to the classifier, but was associated with lower levels in patients compared with controls, in patients with disease activity and in patients not responding to natalizumab treatment. The levels of CCL2 are in line with other reports in MS (Mahad et al., 2006; Sørensen et al., 1999) and have a suggested beneficial role in neuroinflammation (Kwon et al., 2015). Furthermore, a beneficial role of CCL2 is supported by its role in (1) the recruitment and activation of Th2 cells (Ip et al., 2006), (2) M2-associated fetal tolerance in pregnancy (Gustafsson et al., 2008; Svensson et al., 2011), and (3) reduced macrophage production of proinflammatory cytokines (Sierra-Filardi et al., 2014). The fourth protein, OPN (SPP1), is a well-known driver of Th1-and Th17-responses, which is associated with MS through increased levels in both CSF (Bo¨rnsen et al., 2011) and plasma (Shimizu et al., 2013).
A key finding is that a combined protein score of CXCL1-3, CXCL10, CCL2, and OPN, measured in CSF, was able not only to discriminate early MS from controls in a new cohort but also to predict evidence of disease activity after 2 years, a most valu-able finding when deciding treatment strategies. Other proteins in CSF, such as CXCL13, have been suggested as prognostic tools to predict disease activity (Khademi et al., 2011), but show lower AUC (0.64) when identifying patients with clinically isolated syndrome (CIS) developing definite MS within 2 years (Brettschneider et al., 2010). Of note, the combined score of pro-teins being able to predict disease activity was also able to predict low or high response to treatment, which suggests that proteins relevant to the disease process are also important for
the response to treatment, further strengthening the potential use of the combined score in personalized treatment. It should also be noted that the vast majority of MS patients do respond to natalizumab and that other considerations, such as JC-virus status, decide treatment preference. Combinations of bio-markers to predict response to treatment have been evaluated previously, for example the ability of certain gene triplets pre-dicted response to interferon beta in MS patients with an AUC ranging between 0.63 and 0.80 (Baranzini et al., 2015). Neverthe-less, the present results clearly show a potential of the combined disease score in predicting response to treatment. While all clear biomarker results were based on protein levels in CSF, there were no such findings for plasma levels of proteins. Returning to the dysregulated dynamic response in CD4+ T cells, it is logical that CSF displays an altered protein profile, since acti-vated disease-associated T cells travel to and are enriched in the CNS/CSF compartment, thus representing the dynamic response in vivo. It would have been of interest to study the dy-namic response in CD4+ T cells from both our response to treat-ment cohort and early MS cohort, but this lies beyond the scope of the current study. There has been an extensive search for bio-markers in MS (Comabella and Montalban, 2014; Raphael et al., 2015), although they were mainly tested to detect the presence of disease or ongoing disease activity rather than to predict dis-ease activity (Raphael et al., 2015). Our combination of proteins performed well in relation to previous studies of biomarkers, showing a unique potential to predict both disease activity and response to treatment. Although we tested performance in new independent cohorts, it is important to conduct further testing in prospective studies before clinical implementation.
One potential limitation of the protein module is that we selected proteins that could be measured by commercial prod-ucts. Thus, other gene products in the module could further contribute to the performance as a biomarker. Indeed, our study clearly shows the principle that aberrations in the dynamic response can be used to identify relevant pathways and bio-markers. The identified biomarkers showed some general rele-vance by distinguishing other autoimmune disease (RA and SLE) from controls, while disease activity was not reflected. Thus, the general principle of aberrant dynamic response should first be specifically applied to each disease. It should be empha-sized that although our combination of four proteins showed excellent performance, it cannot be excluded that other bio-markers, in particular reflecting the neurodegenerative compo-nent of the disease, might add to the biomarker potential (Modvig et al., 2015).
In summary, we showed a dysregulated dynamic gene response in the activation process of CD4+ T cells in MS. The dysregulated genes formed an interconnected gene module with cell activation and chemotaxis as central components. From this gene module, we could identify a set of proteins that can be measured in CSF, and by using the combined information from this protein set, we were able to classify patients from con-trols, predict disease activity after 2 years, and distinguish low and high responders to treatment. We conclude that our study shows a useful principle for detecting aberrant dynamic re-sponses to cell activation. The combined score of proteins holds promise as a clinically useful biomarker in personalized MS
treatment, and the proposed translational strategy is generally applicable and could be used to identify biomarkers for other diseases in which CD4+ T cells play an active role.
EXPERIMENTAL PROCEDURES Patients and Controls
CD4+ T cells were isolated from 16 women diagnosed with definite relaps-ing-remitting MS according to the McDonald criteria (Polman et al., 2011). None of the patients had experienced a relapse within 3 months prior to blood sampling or had received immunomodulatory or immunosuppressive treatment for at least 2 months (one exception was intravenous immu-noglobulin treatment given 15 days before treatment in one patient). 16 age-matched, healthy control women were recruited among blood donors (Table 1). To validate the clinical classifying capacity at the protein level, plasma and CSF from a second cohort of 41 patients with early MS was used (Table 1). These patients were followed for 2 years and classified into NEDA (Rotstein et al., 2015) or not. CSF was also collected from 11 matched healthy control subjects (Table 1). A third cohort was used to eval-uate if biomarkers were able to predict response to natalizumab; pre-treat-ment CSF samples were selected from nine patients that did not fully respond to treatment. The high responders (n = 6) were matched according to age and disability (Table 1). This study cohort has been described else-where (Gustafsson et al., 2014; Mellerga˚rd et al., 2010, 2013). To investigate the specificity of the findings to MS, patients diagnosed with two other autoimmune diseases (RA and SLE) were also included. 43 cases classified with SLE, according to the 1982 American College of Rheumatology classi-fication criteria (Tan et al., 1982) and the 2012 SLICC criteria (Petri et al., 2012), were selected from the cohort Clinical Lupus Register in North-eastern Gothia (KLURING), previously described in detail (Ighe et al., 2015), and were included and followed for 2 years. Disease activity was re-corded using the SLE Disease Activity Index 2000 (SLEDAI-2K), which was also modified (mSLEDAI) by the exclusion of laboratory items for hypocom-plementemia and anti-dsDNA antibody binding (Gladman et al., 2002). The RA cohort consisted of 40 patients with early RA (first joint swelling < 12 months ago), fulfilling the 1987 American College of Rheumatology clas-sification criteria for RA (Arnett et al., 1988), who were recruited to ‘‘the Swedish TIRA project’’ 2006–2009 (Sva¨rd et al., 2015). Serum samples were drawn at the inclusion visit (prior to institution of disease-modifying anti-rheumatic drugs) and stored at 70C. 29 (73%) tested positive for agglutinating rheumatoid factor, and 31 (78%) had antibodies against cyclic citrullinated peptides (anti-CCP2). Disease activity was assessed according to the original ‘‘28-joint disease activity score’’ (DAS28 [Prevoo et al., 1995]). At inclusion, DAS28 was available from 39 of the 40 cases. At the 2-year follow-up, DAS28-scores were reported in 31 cases. A cohort of 40 healthy blood donors was also included as control. All patients and controls were recruited at the Linko¨ping University Hospital, and the studies were approved by the regional ethics committee in Linko¨ping, Sweden. See Sup-plemental Experimental Proceduresfor further details.
Sample Preparation
Peripheral blood mononuclear cells (PBMCs) were isolated from whole blood by density centrifugation (Lymphoprep, Axis-Shield), placed in a freezing container at 70C overnight, and subsequently transferred and stored in liquid nitrogen until use. After thawing, CD4+ T cells were isolated by immuno-magnetic positive selection according to the instructions provided by the manufacturer (Miltenyi Biotec). The purity of the isolated CD4+ T cells was > 96%, based on flow cytometry (Figure S1). SeeSupplemental Experi-mental Proceduresfor details.
Cell Culture
Isolated CD4+ T cells were stimulated with 0.03 mg/ml anti-CD3 and anti-CD28 antibodies or cultured unstimulated at a density of 1 million cells/ml for 24 hr at 37C. Culture supernatants were collected and the cells were used for flow cytometry or homogenized and lysed in buffer RLT Plus (QIAGEN) and stored at 70C. SeeSupplemental Experimental Proceduresfor details.
Flow Cytometry Staining and Analysis
CD4+ T cells were labeled with mouse anti-human CD4, CD45RA, CCR7/ CD197, and CD69 (all from BD Biosciences) prior to flow cytometry analysis. To assess the viability, CD4+ T cells were stained with Annexin V and 7AAD (BD Biosciences). Patients and controls responded similarly to anti-CD3/ CD28 stimulation, based on CD69 expression (Figure S1). Further, the propor-tion of naive (CCR7+CD45RA+) and memory (CD45RA ) CD4+ T cells did not differ between patients and controls (Figure S1). SeeSupplemental Experi-mental Proceduresfor details.
RNA Extraction and Microarray Analysis
Total RNA from CD4+ T cells was extracted using the Allprep DNA/RNA Micro kit (QIAGEN), amplified and labeled using the Low Input Quick Amp Labeling Kit (Agilent Technologies), and purified with RNeasy Mini Kit (QIAGEN) before hybridized onto Agilent Sureprint G3 Human Gene Expression 8x60k arrays (Agilent Technologies) using the Gene Expression Hybridization Kit (Agilent Technologies). The arrays were scanned on an Agilent Microarray Scanner (Agilent Technologies). Data extraction and quality control fulfillment was performed with the Feature Extraction software v.10.7.3.1 (Agilent Technolo-gies). The accession number for the datasets reported in this paper is GEO: GSE78244. SeeSupplemental Experimental Proceduresfor details. Module Identification
The module was constructed using a modified, more robust and CD4+ cell-specific version of a previously described method (Barrena¨s et al., 2012; Gus-tafsson et al., 2014), by identifying a disease module from the median over an ensemble of modules. We used the human PPI database STRING (v.10 [(Szklarczyk et al., 2015]) and considered the interactions with confidence scores (ps
ij) above 0.7 between all included 14,543 proteins of the database.
We calculated an interaction weight wijby the geometric mean of psijand
(1 pC
ij), where pCijis the double-sided Pearson correlation p value between
gene i and gene j expressions in all samples. We then repeated the following steps 100 times. (1) Realize a perturbed network by drawing a random matrix rijwith matrix elements uniformly in [0,1] and let each interaction be included in
the network if wij> rij. Then, preceding with steps similar as inBarrena¨s et al.
(2012)to (2) identify all maximal cliques of the network with size 3 or more, (3) calculate enrichment of each cliques for differentially expressed genes (unad-justed limma t test p < 0.05) by Fisher’s exact test for each clique. (4) Select all cliques with p < 0.01 as differentially expressed, and (5) extract the set of genes and interactions present in these differentially expressed cliques as one per-turbed module. From the 100 modules we inferred the robust module as the genes present in more than half of the runs and removed the genes with less than twice as many interactions as a randomization module created with shuf-fled input data. Lastly, we removed the singletons from the disease module, i.e. genes with no connections.
Analysis of Cytokines and Chemokines
Quantification of NT-3 was performed using DuoSet ELISA reagents from R&D Systems (DY267), as described elsewhere (Boij et al., 2012). Multiplex Bead Technology (MILLIPLEX MAP Kit, Catalog No. HCYTOMAG-60K-11, HSCRMAG-32K, and HBNMAG-51K, Millipore) was used to measure all other cytokines and chemokines in the supernatant and CSF samples according to the manufacturer’s description. SeeSupplemental Experimental Procedures
for details.
Classification of Patients and Controls
To investigate if the measured mRNA and proteins, selected from the module, could be used to separate patients from controls in culture supernatants, we used the following combined score: (OPNStim OPNUnstim) +
(CXCL1-3Stim CXCL1-3Unstim) + (CXCL10Stim CXCL10Unstim) (CCL2Stim
CCL2Unstim). Area under ROC curve (AUC) value was calculated using the
perfcurve MATLAB function; p value was calculated using double-sided Wilcoxon test. Bootstrap p values were calculated by performing 1,000 random selections of 3 positive and 1 negative proteins among a set of 8 con-trol proteins (GM-CSF, IFNg, IL-10, IL-13, IL-17, CCL17, M-CSF, and CCL20). The p values were estimated from the bootstrap distribution of AUC values ob-tained for the random classifier. For the CSF we had no baseline to subtract,
and instead for each of the subjects we performed a leave-one-out regression score by performing logistic regression of patient probability estimation without that subject in the training. For the high and low responders to natali-zumab treatment, we lacked CXCL1-3 values and therefore retrained the logis-tic regression on the original 41 patients without this protein. Moreover, as three subjects also lacked CXCL10 values, we imputed their values using knnimpute MATLAB function with default settings.
Statistical Analysis
Gene expression microarray data were quantile normalized, and p values for differential expression were calculated using t test from the LIMMA package in R. The p values for protein measurements were calculated using Mann-Whitney test. For set enrichment analysis of GWAS genes, p values were calculated using one-sided Fisher’s exact test, using all genes measured on the gene expression array with at least one assayed SNP mapped as closest as background (n = 16,799). Analysis of correlation between disease activity and combined proteins score in SLE and RA patients was performed using Spearman rank correlation.
ACCESSION NUMBERS
The accession number for the datasets reported in this paper is GEO: GSE78244.
SUPPLEMENTAL INFORMATION
Supplemental Information includes Supplemental Experimental Procedures, two figures, and five tables and can be found with this article online athttp:// dx.doi.org/10.1016/j.celrep.2016.08.036.
AUTHOR CONTRIBUTIONS
M.G., D.R.G., and M.K. conceived and performed bioinformatics analyses; S.H., D.E., M.V, I.H., T.S., A.K. and C.S. collected and analyzed clinical ma-terials; D.E., S.H., H.Z., D.R.G., M.K., M.G. and C.E.N. performed and analyzed microarray experiments; D.E. and S.H. performed cell culture ex-periments, flow cytometry, and protein analyses; M.C.J., M.B., I.K., and T.O. contributed to the study design and bioinformatics analyses; and M.G. and J.E. jointly supervised the study and should be regarded as shared senior authors.
ACKNOWLEDGMENTS
We thank the International Multiple Sclerosis Genetics Consortium for providing GWAS data and Petra Cassel for excellent support in the multiple bead assay and with preparation and collection of samples. This work was supported by the Swedish Research Council (K2013-61X-22310-01-4), The Medical Research Council of Southeast Sweden (FORSS), The Swedish Foun-dation for MS Research, and NEURO Sweden. M.G. has received funding from the Swedish Research Council, the Centre for Industrial IT (CENIIT), the A˚ke Wiberg Foundation, and NEURO Sweden. T.O. has received MS research funding from the Swedish Research Council (K2013-61X-07488-28-5), the Knut and Alice Wallenberg Foundation, The Swedish Brain Foundation, The AFA foundation, and the Margareta af Ugglas foundation. C.E.N. has received research funding from the Swedish Research Council (2015-03495) and The A˚ke Wiberg Foundation. C.S. has received funding from The Swedish Society for Medical Research. T.O. has received lecture/advisory board compensa-tions and/or unrestricted MS research grants from Biogen, Novartis, Genzyme, Merck, and Allmiral. M.V. has received unrestricted research grants from Bio-gen Idec and Novartis and lecture honoraria from BioBio-gen Idec, Merck Serono, and Genzyme. Received: March 7, 2016 Revised: June 1, 2016 Accepted: August 11, 2016 Published: September 13, 2016 REFERENCES
Arnett, F.C., Edworthy, S.M., Bloch, D.A., McShane, D.J., Fries, J.F., Cooper, N.S., Healey, L.A., Kaplan, S.R., Liang, M.H., Luthra, H.S., et al. (1988). The American Rheumatism Association 1987 revised criteria for the classification of rheumatoid arthritis. Arthritis Rheum.31, 315–324.
Ayoglu, B., Mitsios, N., Kockum, I., Khademi, M., Zandian, A., Sjo¨berg, R., For-sstro¨m, B., Bredenberg, J., Lima Bomfim, I., Holmgren, E., et al. (2016). Anoc-tamin 2 identified as an autoimmune target in multiple sclerosis. Proc. Natl. Acad. Sci. USA113, 2188–2193.
Baranzini, S.E., Madireddy, L.R., Cromer, A., D’Antonio, M., Lehr, L., Beelke, M., Farmer, P., Battaglini, M., Caillier, S.J., Stromillo, M.L., et al. (2015). Prog-nostic biomarkers of IFNb therapy in multiple sclerosis patients. Mult. Scler.
21, 894–904.
Barrena¨s, F., Chavali, S., Alves, A.C., Coin, L., Jarvelin, M.R., Jo¨rnsten, R., Langston, M.A., Ramasamy, A., Rogers, G., Wang, H., and Benson, M. (2012). Highly interconnected genes in disease-specific networks are enriched for disease-associated polymorphisms. Genome Biol.13, R46.
Barrett, J.C., Clayton, D.G., Concannon, P., Akolkar, B., Cooper, J.D., Erlich, H.A., Julier, C., Morahan, G., Nerup, J., Nierras, C., et al.; Type 1 Diabetes Ge-netics Consortium (2009). Genome-wide association study and meta-analysis find that over 40 loci affect risk of type 1 diabetes. Nat. Genet.41, 703–707.
Beecham, A.H., Patsopoulos, N.A., Xifara, D.K., Davis, M.F., Kemppinen, A., Cotsapas, C., Shah, T.S., Spencer, C., Booth, D., Goris, A., et al.; International Multiple Sclerosis Genetics Consortium (IMSGC); Wellcome Trust Case Con-trol Consortium 2 (WTCCC2); International IBD Genetics Consortium (IIBDGC) (2013). Analysis of immune-related loci identifies 48 new susceptibility variants for multiple sclerosis. Nat. Genet.45, 1353–1360.
Boij, R., Svensson, J., Nilsson-Ekdahl, K., Sandholm, K., Lindahl, T.L., Palo-nek, E., Garle, M., Berg, G., Ernerudh, J., Jenmalm, M., and Matthiesen, L. (2012). Biomarkers of coagulation, inflammation, and angiogenesis are independently associated with preeclampsia. Am. J. Reprod. Immunol.68, 258–270.
Bo¨rnsen, L., Khademi, M., Olsson, T., Sørensen, P.S., and Sellebjerg, F. (2011). Osteopontin concentrations are increased in cerebrospinal fluid during attacks of multiple sclerosis. Mult. Scler.17, 32–42.
Brettschneider, J., Czerwoniak, A., Senel, M., Fang, L., Kassubek, J., Pink-hardt, E., Lauda, F., Kapfer, T., Jesse, S., Lehmensiek, V., et al. (2010). The chemokine CXCL13 is a prognostic marker in clinically isolated syndrome (CIS). PLoS ONE5, e11986.
Comabella, M., and Montalban, X. (2014). Body fluid biomarkers in multiple sclerosis. Lancet Neurol.13, 113–126.
Corvol, J.C., Pelletier, D., Henry, R.G., Caillier, S.J., Wang, J., Pappas, D., Ca-sazza, S., Okuda, D.T., Hauser, S.L., Oksenberg, J.R., and Baranzini, S.E. (2008). Abrogation of T cell quiescence characterizes patients at high risk for multiple sclerosis after the initial neurological event. Proc. Natl. Acad. Sci. USA105, 11839–11844.
Dendrou, C.A., Fugger, L., and Friese, M.A. (2015). Immunopathology of mul-tiple sclerosis. Nat. Rev. Immunol.15, 545–558.
Derfuss, T. (2012). Personalized medicine in multiple sclerosis: hope or reality? BMC Med.10, 116.
Fletcher, J.M., Lalor, S.J., Sweeney, C.M., Tubridy, N., and Mills, K.H. (2010). T cells in multiple sclerosis and experimental autoimmune encephalomyelitis. Clin. Exp. Immunol.162, 1–11.
Gladman, D.D., Iban˜ez, D., and Urowitz, M.B. (2002). Systemic lupus erythe-matosus disease activity index 2000. J. Rheumatol.29, 288–291.
Gregory, A.P., Dendrou, C.A., Attfield, K.E., Haghikia, A., Xifara, D.K., Butter, F., Poschmann, G., Kaur, G., Lambert, L., Leach, O.A., et al. (2012). TNF recep-tor 1 genetic risk mirrors outcome of anti-TNF therapy in multiple sclerosis. Nature488, 508–511.
Gustafsson, C., Mjo¨sberg, J., Matussek, A., Geffers, R., Matthiesen, L., Berg, G., Sharma, S., Buer, J., and Ernerudh, J. (2008). Gene expression profiling of
human decidual macrophages: evidence for immunosuppressive phenotype. PLoS ONE3, e2078.
Gustafsson, M., Edstro¨m, M., Gawel, D., Nestor, C.E., Wang, H., Zhang, H., Barrena¨s, F., Tojo, J., Kockum, I., Olsson, T., et al. (2014). Integrated genomic and prospective clinical studies show the importance of modular pleiotropy for disease susceptibility, diagnosis and treatment. Genome Med.6, 17.
Huang, W., Sherman, B.T., and Lempicki, R.A. (2009). Bioinformatics enrich-ment tools: paths toward the comprehensive functional analysis of large gene lists. Nucleic Acids Res.37, 1–13.
Ighe, A., Dahlstro¨m, O¨ ., Skogh, T., and Sjo¨wall, C. (2015). Application of the 2012 Systemic Lupus International Collaborating Clinics classification criteria to patients in a regional Swedish systemic lupus erythematosus register. Arthritis Res. Ther.17, 3.
International Multiple Sclerosis Genetics Consortium (2013). Network-based multiple sclerosis pathway analysis with GWAS data from 15,000 cases and 30,000 controls. Am. J. Hum. Genet.92, 854–865.
Ip, W.K., Wong, C.K., and Lam, C.W. (2006). Interleukin (IL)-4 and IL-13 up-regulate monocyte chemoattractant protein-1 expression in human bronchial epithelial cells: involvement of p38 mitogen-activated protein kinase, extracel-lular signal-regulated kinase 1/2 and Janus kinase-2 but not c-Jun NH2-termi-nal kinase 1/2 sigNH2-termi-nalling pathways. Clin. Exp. Immunol.145, 162–172.
Kemppinen, A.K., Kaprio, J., Palotie, A., and Saarela, J. (2011). Systematic re-view of genome-wide expression studies in multiple sclerosis. BMJ Open1,
e000053.
Khademi, M., Kockum, I., Andersson, M.L., Iacobaeus, E., Brundin, L., Selleb-jerg, F., Hillert, J., Piehl, F., and Olsson, T. (2011). Cerebrospinal fluid CXCL13 in multiple sclerosis: a suggestive prognostic marker for the disease course. Mult. Scler.17, 335–343.
Kwon, M.J., Shin, H.Y., Cui, Y., Kim, H., Thi, A.H., Choi, J.Y., Kim, E.Y., Hwang, D.H., and Kim, B.G. (2015). CCL2 mediates neuron-macrophage interactions to drive proregenerative macrophage activation following preconditioning injury. J. Neurosci.35, 15934–15947.
L€unemann, J.D., R€uckert, S., Kern, F., Wendling, U., van der Zee, R., Volk, H.D., and Zipp, F. (2004). Cross-sectional and longitudinal analysis of myelin-reactive T cells in patients with multiple sclerosis. J. Neurol.251,
1111–1120.
Mahad, D., Callahan, M.K., Williams, K.A., Ubogu, E.E., Kivisa¨kk, P., Tucky, B., Kidd, G., Kingsbury, G.A., Chang, A., Fox, R.J., et al. (2006). Modulating CCR2 and CCL2 at the blood-brain barrier: relevance for multiple sclerosis pathogen-esis. Brain129, 212–223.
Mayne, M., Moffatt, T., Kong, H., McLaren, P.J., Fowke, K.R., Becker, K.G., Namaka, M., Schenck, A., Bardoni, B., Bernstein, C.N., and Melanson, M. (2004). CYFIP2 is highly abundant in CD4+ cells from multiple sclerosis pa-tients and is involved in T cell adhesion. Eur. J. Immunol.34, 1217–1227.
Mellerga˚rd, J., Edstro¨m, M., Vrethem, M., Ernerudh, J., and Dahle, C. (2010). Natalizumab treatment in multiple sclerosis: marked decline of chemokines and cytokines in cerebrospinal fluid. Mult. Scler.16, 208–217.
Mellerga˚rd, J., Edstro¨m, M., Jenmalm, M.C., Dahle, C., Vrethem, M., and Er-nerudh, J. (2013). Increased B cell and cytotoxic NK cell proportions and increased T cell responsiveness in blood of natalizumab-treated multiple sclerosis patients. PLoS ONE8, e81685.
Modvig, S., Degn, M., Roed, H., Sørensen, T.L., Larsson, H.B., Langkilde, A.R., Frederiksen, J.L., and Sellebjerg, F. (2015). Cerebrospinal fluid levels of chiti-nase 3-like 1 and neurofilament light chain predict multiple sclerosis develop-ment and disability after optic neuritis. Mult. Scler.21, 1761–1770.
Okuda, Y., Okuda, M., Apatoff, B.R., and Posnett, D.N. (2005). The activation of memory CD4(+) T cells and CD8(+) T cells in patients with multiple sclerosis. J. Neurol. Sci.235, 11–17.
Patsopoulos, N.A., Esposito, F., Reischl, J., Lehr, S., Bauer, D., Heubach, J., Sandbrink, R., Pohl, C., Edan, G., Kappos, L., et al.; Bayer Pharma MS Ge-netics Working Group; Steering Committees of Studies Evaluating IFNb-1b and a CCR1-Antagonist; ANZgene Consortium; GeneMSA; International
Mul-tiple Sclerosis Genetics Consortium (2011). Genome-wide meta-analysis iden-tifies novel multiple sclerosis susceptibility loci. Ann. Neurol.70, 897–912.
Petri, M., Orbai, A.M., Alarco´n, G.S., Gordon, C., Merrill, J.T., Fortin, P.R., Bruce, I.N., Isenberg, D., Wallace, D.J., Nived, O., et al. (2012). Derivation and validation of the Systemic Lupus International Collaborating Clinics clas-sification criteria for systemic lupus erythematosus. Arthritis Rheum.64, 2677–
2686.
Polman, C.H., Reingold, S.C., Banwell, B., Clanet, M., Cohen, J.A., Filippi, M., Fujihara, K., Havrdova, E., Hutchinson, M., Kappos, L., et al. (2011). Diagnostic criteria for multiple sclerosis: 2010 revisions to the McDonald criteria. Ann. Neurol.69, 292–302.
Prevoo, M.L., van ’t Hof, M.A., Kuper, H.H., van Leeuwen, M.A., van de Putte, L.B., and van Riel, P.L. (1995). Modified disease activity scores that include twenty-eight-joint counts. Development and validation in a prospective longi-tudinal study of patients with rheumatoid arthritis. Arthritis Rheum.38, 44–48.
Putheti, P., Morris, M., Stawiarz, L., Teleshova, N., Kivisa¨kk, P., Pashenkov, M., Kouwenhoven, M., Wiberg, M.K., Bronge, L., Huang, Y.M., et al. (2003). Multiple sclerosis: a study of chemokine receptors and regulatory T cells in relation to MRI variables. Eur. J. Neurol.10, 529–535.
Raphael, I., Webb, J., Stuve, O., Haskins, W., and Forsthuber, T. (2015). Body fluid biomarkers in multiple sclerosis: how far we have come and how they could affect the clinic now and in the future. Expert Rev. Clin. Immunol.11, 69–91.
Rotstein, D.L., Healy, B.C., Malik, M.T., Chitnis, T., and Weiner, H.L. (2015). Evaluation of no evidence of disease activity in a 7-year longitudinal multiple sclerosis cohort. JAMA Neurol.72, 152–158.
Rumble, J.M., Huber, A.K., Krishnamoorthy, G., Srinivasan, A., Giles, D.A., Zhang, X., Wang, L., and Segal, B.M. (2015). Neutrophil-related factors as bio-markers in EAE and MS. J. Exp. Med.212, 23–35.
Sallusto, F. (2016). Heterogeneity of human CD4(+) T cells against microbes. Annu. Rev. Immunol.34, 317–334.
Satoh, J., Nakanishi, M., Koike, F., Onoue, H., Aranami, T., Yamamoto, T., Ka-wai, M., Kikuchi, S., Nomura, K., Yokoyama, K., et al. (2006). T cell gene expression profiling identifies distinct subgroups of Japanese multiple scle-rosis patients. J. Neuroimmunol.174, 108–118.
Sawcer, S., Hellenthal, G., Pirinen, M., Spencer, C.C., Patsopoulos, N.A., Moutsianas, L., Dilthey, A., Su, Z., Freeman, C., Hunt, S.E., et al.; International Multiple Sclerosis Genetics Consortium; Wellcome Trust Case Control Con-sortium 2 (2011). Genetic risk and a primary role for cell-mediated immune mechanisms in multiple sclerosis. Nature476, 214–219.
Sharpe, A.H., and Freeman, G.J. (2002). The B7-CD28 superfamily. Nat. Rev. Immunol.2, 116–126.
Shimizu, Y., Ota, K., Ikeguchi, R., Kubo, S., Kabasawa, C., and Uchiyama, S. (2013). Plasma osteopontin levels are associated with disease activity in the patients with multiple sclerosis and neuromyelitis optica. J. Neuroimmunol.
263, 148–151.
Sierra-Filardi, E., Nieto, C., Domı´nguez-Soto, A., Barroso, R., Sa´nchez-Ma-teos, P., Puig-Kroger, A., Lo´pez-Bravo, M., Joven, J., Ardavı´n, C., Rodrı´-guez-Ferna´ndez, J.L., et al. (2014). CCL2 shapes macrophage polarization by GM-CSF and M-CSF: identification of CCL2/CCR2-dependent gene expression profile. J. Immunol.192, 3858–3867.
Sørensen, T.L., Tani, M., Jensen, J., Pierce, V., Lucchinetti, C., Folcik, V.A., Qin, S., Rottman, J., Sellebjerg, F., Strieter, R.M., et al. (1999). Expression of specific chemokines and chemokine receptors in the central nervous system of multiple sclerosis patients. J. Clin. Invest.103, 807–815.
Sørensen, T.L., Trebst, C., Kivisa¨kk, P., Klaege, K.L., Majmudar, A., Ravid, R., Lassmann, H., Olsen, D.B., Strieter, R.M., Ransohoff, R.M., and Sellebjerg, F. (2002). Multiple sclerosis: a study of CXCL10 and CXCR3 co-localization in the inflamed central nervous system. J. Neuroimmunol.127, 59–68.
Stahl, E.A., Raychaudhuri, S., Remmers, E.F., Xie, G., Eyre, S., Thomson, B.P., Li, Y., Kurreeman, F.A., Zhernakova, A., Hinks, A., et al.; BIRAC Consortium; YEAR Consortium (2010). Genome-wide association study meta-analysis identifies seven new rheumatoid arthritis risk loci. Nat. Genet.42, 508–514.
Sva¨rd, A., Skogh, T., Alfredsson, L., Ilar, A., Klareskog, L., Bengtsson, C., and Kastbom, A. (2015). Associations with smoking and shared epitope differ be-tween IgA- and IgG-class antibodies to cyclic citrullinated peptides in early rheumatoid arthritis. Arthritis Rheumatol.67, 2032–2037.
Svensson, J., Jenmalm, M.C., Matussek, A., Geffers, R., Berg, G., and Erner-udh, J. (2011). Macrophages at the fetal-maternal interface express markers of alternative activation and are induced by M-CSF and IL-10. J. Immunol.187, 3671–3682.
Szklarczyk, D., Franceschini, A., Wyder, S., Forslund, K., Heller, D., Huerta-Cepas, J., Simonovic, M., Roth, A., Santos, A., Tsafou, K.P., et al. (2015). STRING v10: protein-protein interaction networks, integrated over the tree of life. Nucleic Acids Res.43, D447–D452.
Tan, E.M., Cohen, A.S., Fries, J.F., Masi, A.T., McShane, D.J., Rothfield, N.F., Schaller, J.G., Talal, N., and Winchester, R.J. (1982). The 1982 revised criteria for the classification of systemic lupus erythematosus. Arthritis Rheum.25, 1271–1277.
Tramacere, I., Del Giovane, C., Salanti, G., D’Amico, R., and Filippini, G. (2015). Immunomodulators and immunosuppressants for relapsing-remitting multi-ple sclerosis: a network meta-analysis. Cochrane Database Syst. Rev.9, CD011381.
Vazirinejad, R., Ahmadi, Z., Kazemi Arababadi, M., Hassanshahi, G., and Ken-nedy, D. (2014). The biological functions, structure and sources of CXCL10 and its outstanding part in the pathophysiology of multiple sclerosis. Neuroimmu-nomodulation21, 322–330.
