|Home | About | Journals | Submit | Contact Us | Français|
Skin biopsy gene expression was analyzed by DNA microarray from 13 dSSc patients enrolled in an open label study of rituximab, 9 dSSc patients not treated with rituximab, and 9 healthy controls. These data recapitulate the patient ‘intrinsic’ gene expression subsets described previously including proliferation, inflammatory, and normal-like groups. Serial skin biopsies showed consistent and non-progressing gene expression over time, and importantly, the patients in the inflammatory subset do not move to the fibroproliferative subset, and vice versa. We were unable to detect significant differences in gene expression before and after rituximab treatment, consistent with an apparent lack of clinical response. Serial biopsies from each patient stayed within the same gene expression subset regardless of treatment regimen or the time point at which they were taken. Collectively, these data emphasize the heterogeneous nature of SSc and demonstrate that the intrinsic subsets are an inherent, reproducible and stable feature of the disease that is independent of disease duration. Moreover, these data have fundamental importance for the future development of personalized therapy for SSc; drugs targeting inflammation are likely to benefit those patients with an inflammatory signature, whereas drugs targeting fibrosis are likely to benefit those with a fibroproliferative signature.
Systemic sclerosis (SSc) is a multi-system autoimmune disorder with a hallmark of skin fibrosis and thickening along with significant internal organ involvement (Mayes, 2003). SSc has historically been divided into limited and diffuse disease based on the extent of skin involvement, with limited cutaneous SSc (lSSc) involving skin restricted to the regions below the elbows, knees and face, and diffuse cutaneous SSc (dSSc) including more proximal skin. The degree of skin involvement has a direct correlation with SSc prognosis and internal organ complications (Barnett et al., 1988; Scussel-Lonzetti et al., 2002). However, within dSSc and lSSc, there is a heterogeneous range of skin and internal organ involvement. Approaches that objectively quantify disease heterogeneity and predict internal organ involvement are critically needed.
Previous genome-wide gene expression studies in SSc skin identified disease specific gene expression signatures in both lesional and non-lesional skin biopsies that are distinct from those found in healthy controls (Gardner et al., 2006; Milano et al., 2008; Whitfield et al., 2003). In addition, we have shown that distinct gene expression signatures divide SSc patients into ‘intrinsic subsets’, capturing the clinical heterogeneity of limited versus diffuse SSc, but extending this heterogeneity by revealing that patients with dSSc fall into several different subsets based on gene expression in skin (Milano et al., 2008). These results suggested that distinct pathogenic mechanisms may drive disease in different patients or at different stages of the disease. We previously identified four intrinsic gene expression subsets: a ‘diffuse-proliferation’ group comprised completely of patients with dSSc, showing increased expression of genes associated with the cell proliferation that could be further subdivided into two groups ‘diffuse1’ and ‘diffuse2’; an ‘inflammatory’ group comprised of dSSc, lSSc and morphea samples, showing increased expression of genes associated with inflammation; a ‘limited’ group comprised primarily of patients with lSSc; and the ‘normal-like’ group of dSSc and lSSc patients, showing gene expression similar to healthy controls. A weak relationship was found between disease duration and these intrinsic subsets, suggesting that they might reflect evolution of the disease process rather than biologically distinct pathogenic processes.
Here we recapitulate the intrinsic subsets, show these subsets are stable over time, and that treatment with rituximab fails to alter skin gene expression. These data illustrate that patients with an inflammatory signature do not go on to develop a fibroproliferative signature, suggesting a possible explanation why, in the past, broad spectrum anti-inflammatory agents may not have worked in SSc. It also indicates that different pathogenic mechanisms drive disease pathogenesis within phenotypically similar patients with dSSc and that this heterogeneity can be consistently and reproducibly detected by analyzing skin gene expression, having broad implications for the future development of therapies for SSc.
We analyzed skin biopsies from dSSc patients for gene expression changes indicative of patient specific heterogeneity. Gene expression was measured in 60 skin biopsies from 22 patients with dSSc and 9 healthy controls (Supplemental Table 1). A total of 89 microarrays were hybridized, which included 29 technical replicates. All patients were biopsied at a lesional forearm site, a subset were also biopsied at a non-lesional back site. Clinical data can be found in Supplemental Table 2.
Skin biopsies from dSSc patients were analyzed before and after treatment with rituximab for gene expression changes. Consistent with the lack of clinical response (Lafyatis et al., 2009), we did not find a significant change in gene expression associated with rituximab treatment. Instead, gene expression was nearly identical between serial biopsies of patients before and after treatment (see Supplemental Material; Supplemental Data File S1; Supplemental Figure S2).
We previously selected a set of ‘intrinsic’ genes that showed consistent non-changing gene expression between the lesional forearm and non-lesional back biopsies, but showed the largest changes between different patients (Milano et al. 2008), allowing us to compare differences between patients rather than between lesional/non-lesional biopsies. This resulted in the identification of patient subsets based on gene expression. These groups were labeled fibroproliferative, inflammatory, limited and normal-like based on the biological gene expression programs that predominated in each subset. We used the same strategy to classify patients in this independent cohort of patient. Hierarchical clustering using 2377 intrinsic genes (FDR 0.4%) recapitulated the major intrinsic subsets including the Fibroproliferative (Diffuse 1 and 2), Inflammatory, and Normal-like groups (Figure 1A and and22).
We carried out a second analysis to specifically explore whether patients showed significant alterations in gene expression subset over time (intrinsic-by-patient analysis). Genes were selected that showed the most consistent expression at a single time point for each patient, but had the most diverse expression between time points (1888 genes, FDR of 1.58%). Organizing the samples by hierarchical clustering shows that serial biopsies from 13 of 14 patients group together, even though this analysis emphasizes differences between time points (Figure 1B, Supplemental Figure S1), indicating that serial biopsies are more similar to each other than to any other samples over the 6 months to 2 years analyzed. The dendrograms for the intrinsic by time point and intrinsic by patient analysis are remarkably similar confirming that gene expression varies little across time (Figure 1).
Distinct sets of genes were associated with each subset that corresponded to specific biological processes in the skin, represented bv GO biological processes ((Milano et al.2008), Figure 2; Supplemental Data file S3). Genes associated with the inflammatory group are enriched for the GO biological processes of immune system response and inflammatory response (p ≤ 0.001, DAVID analysis; Figure 2D and E) and include interferon-induced genes such IFIT1, IFIT2, OAS3. This group of genes is also enriched for the GO biological processes of vasculature development (p ≤ .01) including the genes vascular endothelial growth factor C (VEGFC) and endoglin (ENG), as well as genes associated with fibrosis (COL6A3, COL6A1, COL5A2, COL5A1, COL1A1, COL1A2), Collagen Oligomeric Matrix Protein (COMP), matrix metalloproteinase 9, (MMP9) (Jimenez et al., 1996; Liu and Zhang, 2008; Ramirez et al., 2006; Varga and Jimenez, 1995).
Two groups of dSSc patients showed increased expression of the proliferation signature indicative of dividing cells (Figure 2G) and low expression of the inflammatory signature (Figure 2D), labeled diffuse 1 (blue) and diffuse 2 (red, showing a more prominent proliferation signature)(Whitfield et al., 2002). Genes associated with this subset are enriched for the GO biological processes of mitosis, m-phase of the mitotic cell cycle, chromosome segregation (p ≤ 0.001), and DNA metabolic process (p ≤ 0.05). These include the cell cycle-regulators CDCA8, CDC2, the kinesins KIF2C, KIF11, and cyclins CCNB2 and CCNB1.
Pathways more prominent in this study than seen previously include fatty acid metabolism (Milano et al.2008), with increased expression in the normal-like and diffuse 1 subsets (Figure 2C). Enriched GO Biological processes included lipid metabolism and fatty acid metabolism (p ≤ .001) include the peroxisome proliferation activated receptor gamma (PPAR-γ) coactivator alpha 1 gene. A group of genes with related function, but a different expression pattern centers on PPAR-γ gene expression (Figure 2F) (Wei et al.2010). Genes within this PPAR-γ pathway related group include several PPAR-γ target genes: CD36, (Huang et al., 2004), lipoprotein lipase (LPL) (Schoonjans et al., 1996), stearoyl-CoA desaturase (SCD) (Vondrichova et al., 2007), and Catalase (CAT) (Girnun et al., 2002).
We summarized the differentially regulated biological pathways by averaging the genes associated with each (Supplemental Material; Supplemental Figure S3-S4). We observe an increase in pathways associated with immune system activation (Supp. Fig. S3A) and an increase in gene expression associated with B cells, CD8+ T cells, leukocytes, macrophages and IFN treated keratinocytes and fibroblasts (Supp. Fig. S3B). The diffuse 2 subset showed an increase in cell cycle-related processes, decreases in cholesterol, steroid and fatty acid metabolism, as well as enrichment of genes associated with activated PBMCs, TNFalpha treated keratinocytes, T cells and dendritic cells (Supp. Fig. S3A–B). The normal-like and diffuse 1 group showed a decrease in immune activation and increases in lipid metabolism.
Two results from this study suggest the intrinsic subsets are not dependent on disease duration. The four major intrinsic subsets identified in our previous study, which had variable and longer disease duration (Milano et al.2008) were also found in this study cohort, all of these patients having early stage disease (Supplemental Figure S5A). If the subsets were dependent on disease duration then we would expect a skewing toward the early subset in these data. In addition, there is no significant difference in disease duration between subsets measured here (Supplemental Figure S5B). Collectively, these findings indicate that gene expression subsets are an inherent feature of the clinical dSSc phenotype and that this feature is independent of disease duration.
Despite the consistent gene expression, MRSS did change in patients that provide longitudinal biopsies (Supplemental Figure S6). Seven of the 15 patients showed increases in skin score, two patients (RIT7 and RIT14) showed a decrease and the remaining six showed little change. In all cases, patients maintained a stable pattern of gene expression. The range of MRSS for patients in this dataset has a slightly broader distribution than the skin scores in Milano et al. (Supplemental Figure S5C). When lSSc patients are excluded, the MRSS of proliferative and inflammatory groups though broader in distributions are otherwise similar to our past study (Milano et al.2008) (Supplemental Figure S5C-D). The inflammatory group has the widest range of MRSS scores, while the normal-like group consistently shows lower MRSS scores in both datasets. Autoantibodies did not show a significant association with intrinsic subset (Supplemental Table 2). One Diffuse 1 patient was anti-scl70 positive (1/4; p=0.52, Fisher’s Exact Test), two inflammatory patients were anti-RNA Polymerase III positive (2/7; p=0.12), and three unclassified patients were anti-scl-70 positive (2/3; p=0.051) and anti-centromere positive (1/3; p=0.16) positive. Diffuse 2 patients were negative for all three measured autoantibodies (p=0.52).
To validate the mRNA expression, we analyzed proteins for COMP and IFITM3 in representative biopsies spanning the intrinsic subsets. Both showed highest expression in the inflammatory subset consistent with gene expression data (Figure 2D, Supplemental Figure S7). IHC staining results paralleled and confirmed the gene expression findings. COMP showed highest expression in the inflammatory subset and lowest in the diffuse 2 subset (p<0.05) with slightly higher expression in the diffuse 1 subset (Supplemental Figure S7A); protein staining was most prevalent in dermal fibroblasts of SSc patients of the inflammatory subset (Supplemental Figure S7C and G) while absent in controls (Supplemental Figure S7E). IFITM3 showed highest expression in both the inflammatory and diffuse 2 subsets (Supplemental Figure S7B) and lowest expression in the diffuse 1 subset (p<0.05) with staining around the microvasculature in the skin (Supplemental Figure S7D and H). These data confirm and extend the gene expression findings at the protein level.
We next determined whether the 995 genes selected in our previous study could stratify the cohort of patients described here into the intrinsic subsets (Milano et al.2008) and thus could be developed into a classifier for subset stratification. 808 genes that passed basic quality filters were used to organize the samples by hierarchical clustering (Figure 3B), showing that 26 out of 31 dSSc skin biopsies from different anatomical sites or time points grouped together by patient (Figure 3A). The subsets identified previously are similar to those found here using the same set of genes ((Milano et al.2008), Figure 3C). The proliferation and inflammatory groups in the two datasets are indicated with the diffuse-proliferation groups in red and the inflammatory group in purple. Subsets of overlapping genes between the groups are indicated. The inflammatory signature is more prominent in the dataset presented here, whereas the proliferation signature is more prominent in our previous report (Milano et al.2008). Therefore, the original 995 genes can stratify an independent cohort into the intrinsic patient subsets.
We previously reported a 177-gene expression signature that could serve as surrogate biomarker of MRSS (Milano et al.2008). We refined this by identifying 44 genes that were present in the 177-gene signature but also found in the intrinsic by patient analysis (Figure 2). Organizing the samples using these 44 genes revealed two major subdivisions of samples. One included both dSSc patients and healthy controls (Figure 4A, group 1) while the other included only dSSc patients (Figure 4A, group 2). Subjects in these groups showed a significant difference in mean MRSS (group 1: mean 17.92, standard deviation 11.61; Group 2: mean 25.90, standard deviation 8.52; p = 0.005; Figure 4B).
To further power the analysis of SSc skin gene expression across time scales that cannot be easily captured with longitudinal biopsies, we combined the gene expression analyses from both this and our previous study (Milano et al.2008). The combined dataset includes skin biopsies from 39 patients with dSSc, 7 with lSSc, 3 with morphea, 1 patient with eosinophilic fasciitis, and skin biopsies from 15 healthy control subjects, totaling 121 biopsies. After Distance Weighted Discrimination (DWD) adjustment to remove systematic differences (Benito et al., 2004), intrinsic analysis was performed, 3551 probes selected (FDR 0.07%; Figure 5A), and the combined datsets clustered hierarchically. The two datasets recapitulated the major intrinsic subsets (Figure 5A). Gene expression in the ‘proliferative’ and ‘inflammatory’ groups both showed gradients of gene expression within their respective groups (Figure 5A). In the proliferation group the dendrogram split samples with high expression of the signature (‘Prolif 1’), and low expression (‘Prolif 2’). Similarly, the inflammatory group contained samples with high (‘Inflam 1’) and low (‘Inflam 2’) expression of the signature. To determine if the intensity of these signatures were associated with disease duration or MRSS, we compared the distributions between these two parameters within each group (Figure 6). The ‘Prolif 1’ group had a significantly longer disease duration (t-test, p = 0.0027) compared to ‘Prolif 2’ and a trend towards a higher MRSS (p < 0.05). In contrast, the ‘Inflam 2’ group, had a higher inflammatory gene expression signature, showed a significantly higher average MRSS (t-test, p = 0.016) with little difference in disease duration across the subgrouping. Thus, this analysis suggests that the gene expression within a group changes intensity with increased disease severity and/or duration.
SSc is a progressive disease, with skin going through various phases that can begin with edema, then progressive fibrosis, and in some cases a skin softening late in the disease. Past studies have suggested that SSc skin pathology evolves from inflammatory to fibrotic changes over time (Fleischmajer et al., 1978; Fleischmajer et al., 1977; Kraling et al., 1995; Roumm et al., 1984). Our molecular analyses suggest that inflammatory changes in the skin are not part of an evolving process leading to fibrosis, but rather represent a subset of patients, with other patients having significantly less skin inflammation as part of the pathologic process. The data presented here indicate that the gene expression subsets are stable over time. We show that measuring gene expression of skin biopsies from patients at different time points consistently classifies the patients into the same intrinsic subsets. In addition, this data taken from a cohort of patients with shorter disease duration compared to our previous studies of patients with longer disease duration show the same disease subsets.
Strikingly, although patients do not appear to move between subsets over time, patients within inflammatory and particularly proliferative subsets show changes in the intensity of the signature associated with disease duration. Similar changes in intensity of gene expression was found in a study of limited SSc patients with and without PAH, suggesting that as disease becomes more severe the intensity of the changes increases (Pendergrass et al. 2010). Our data is consistent with previous observations that the expression of pro-fibrotic genes in non-lesional SSc fibroblasts tends to be intermediate between that observed in healthy and lesional SSc fibroblasts (Chen et al., 2005). Thus our data suggest that different biological processes may underpin dSSc pathogenesis in different patient subsets, and that the intensity of these processes change over time. Although we cannot exclude the possibility that analysis of serial biopsies over a longer period of time would show patients changing subsets, the data here support the notion that a patient could be assigned to a specific intrinsic subset early in their disease course. These observations then have the potential of suggesting treatments that block the immune response if a patient displays an inflammatory signature, or treatments that block fibrosis, if they posses a fibroproliferative signature. Thus, as the pathogenic mechanisms underlying intrinsic subsets are identified, patient subset identification might permit specific targeting of pathogenic processes. A TGFβ–responsive signature appears to drive at least in part pathogenesis in the diffuse-proliferative subset of patients (Sargent et al. 2010), emphasizing the contribution of this cytokine to the phenotype of SSc (Leask, 2009; Varga and Pasche, 2009) while IL-13 appears to play a key role in the inflammatory subset (Greenblatt et al., 2011).
This dataset recapitulates and validates our originally defined inflammatory, diffuse-proliferation, and normal-like subsets of SSc patients. The disease duration of the cohort analyzed here (< 2 years) is more homogeneous and more typical of when the disease is most actively progressing compared to our previous report. Despite these differences, the results show reproducibility of patient subsets in a completely new patient population from a different clinical center. The reproducibility and stability of these gene expression-based subsets confirms that they reflect fundamental underlying pathogenic processes that differ between patients in the different subsets, rather than stages in the progressive disease. The addition of more serial biopsies covering a larger span of time, coupled with longitudinal clinical data, should provide prognostic information for patients in each group.
A subset of patients in this analysis were treated with rituximab, a monoclonal antibody that depletes mature B cells. In that open-label trial of rituximab, patients did not show a significant change in skin score, pulmonary function or other measures of organ involvement, although immunohistochemistry showed depletion of both peripheral and skin resident B-cells (Lafyatis et al., 2009). Several other unblinded clinical trials have suggested some potential beneficial effect of rituximab on skin and/or lung disease in patients with SSc (Bosello et al.; Daoussis et al.; Smith et al.; Wesson et al., 2008). Rituximab may prove to have value in SSc, and it is particularly intriguing to consider its potential efficacy in SSc-associated interstitial lung disease where B cell infiltration is often quite prominent (Lafyatis et al., 2007). However our studies here are consistent with our clinical trial results suggesting no or modest effect on skin disease. We therefore believe that rituximab is more likely to show efficacy for SSc-associated interstitial lung disease and that lung disease would be a better target for a large-scale trial of B cell depletion.
Comparing the results of Chung et al. who reported two patients with dSSc that showed a response to treatment with imatinib mesylate (Chung et al., 2009), with the results reported here in rituximab-treated patients suggests that when a clinical response is evident, one is also likely to find a gene expression response. Thus, gene expression may be useful as a surrogate outcome measure and for stratifying likely patient responders in clinical trials. Since the patients with an inflammatory signature do not go on to display a fibroproliferative signature, and vice versa, it is essential to target patients with a therapy that is appropriate to their gene expression signature. Therefore, it will be imperative to identify patient subsets in clinical trials as drugs may target only those in a single subset.
All study participants gave informed consent under Boston University Medical Center Institutional Review Board an approved protocol. Forearm lesional skin, and for a subset of patients, non-lesional back skin were collected by punch biopsies (3-6 mm) from 13 patients enrolled in the rituximab study (Lafyatis et al., 2009), as well as from 9 additional patients with early dSSc (less than 1.5 years from diagnosis). A total of 60 skin biopsies were collected from 22 patients with dSSc and 9 healthy controls (Supplemental Tables 1 and 2). RNA purification and microarray hybridization was carried out essentially as described (Milano et al.2008). Detailed methodology is provided in supplemental text.
Intrinsic gene analysis was carried out as previously described (Supplemental material; (Milano et al.2008)). Gene expression data was organized by average linkage hierarchically clustering using Cluster 3.0 and visualized using Java TreeView. Significance of clustering was determined by Statistical Significance of Clustering (SigClust)(Liu et al., 2008). Enriched GO biological processes were determined using The Database for Annotation, Visualization, and Integrated Discovery (DAVID) (Huang da et al., 2007)(Supplementary DataFile S5). Module maps were created with Genomica. Correlation between gene expression and clinical parameters were calculated using Matlab. Statistics were done using R. Distance Weighted Discrimination (DWD) was performed utilizing the java implementation of DWD analysis (Benito et al., 2004). Gene expression data from this study is available from GEO (accession number GSE32413).
This work was supported by NIH grant U01-AR055063 to RL and MLW, and by a grant from the Scleroderma Research Foundation to MLW. MLW is a Hulda Irene Duggan Arthritis Investigator. SAP was supported by the NIAMS Autoimmunity and Connective Tissue Training Grant T32 AR007575-11 and by funds from the Arthritis Foundation.
Conflict of Interest Disclosure Michael L. Whitfield, Robert Lafyatis and Sarah A. Pendergrass have filed patent applications for gene expression biomarkers in scleroderma. Michael L. Whitfield is a scientific founder and holds an interest in Celdara Medical, LLC, that is aiming to translate these discoveries into clinical use.