MicroRNA Expression Profiles as Biomarkers of Minor Salivary Gland Inflammation and Dysfunction in Sjögren's Syndrome

Abstract

Objective

MicroRNA reflect physiologic and pathologic processes and may be used as biomarkers of concurrent pathophysiologic events in complex settings such as autoimmune diseases. We generated microRNA microarray profiles from the minor salivary glands of control subjects without Sjögren's syndrome (SS) and patients with SS who had low-grade or high-grade inflammation and impaired or normal saliva production, to identify microRNA patterns specific to salivary gland inflammation or dysfunction.

Methods

MicroRNA expression profiles were generated by Agilent microRNA arrays. We developed a novel method for data normalization by identifying housekeeping microRNA. MicroRNA profiles were compared by unsupervised mathematical methods to test how well they distinguish between control subjects and various subsets of patients with SS. Several bioinformatics methods were used to predict the messenger RNA targets of the differentially expressed microRNA.

Results

MicroRNA expression patterns accurately distinguished salivary glands from control subjects and patients with SS who had low-degree or high-degree inflammation. Using real-time quantitative polymerase chain reaction, we validated 2 microRNA as markers of inflammation in an independent cohort. Comparing microRNA from patients with preserved or low salivary flow identified a set of differentially expressed microRNA, most of which were up-regulated in the group with decreased salivary gland function, suggesting that the targets of microRNA may have a protective effect on epithelial cells. The predicted biologic targets of microRNA associated with inflammation or salivary gland dysfunction identified both overlapping and distinct biologic pathways and processes.

Conclusion

Distinct microRNA expression patterns are associated with salivary gland inflammation and dysfunction in patients with SS, and microRNA represent a novel group of potential biomarkers.

Sjögren's syndrome (SS) is characterized by features of systemic autoimmunity and exocrine gland dysfunction and inflammation. The exact cause of exocrine gland dysfunction in SS has not been delineated, but it is thought that both immunologically mediated and nonimmune mechanisms contribute significantly (1). The diagnosis of SS is based on the combination of symptoms and signs of dry mouth and/or dry eyes, the presence of autoantibodies, and an inflammatory infiltrate in the minor salivary glands (MSGs). The intensity of the infiltrate varies considerably and is described by the focus score, which can range from 0 to 12, with the highest score representing diffuse lymphocytic infiltrates. There is also a great deal of variation in salivary flow, ranging from essentially normal production to no production of saliva. The correlation between the focus score and salivary flow is poor, suggesting that these may be 2 independent processes. Alternatively, the poor correlation between the focus score and salivary flow may reflect the limitations of our current methods, which rely solely on quantitative assessments of inflammation and dysfunction. More sophisticated biomarkers that reliably describe the different pathophysiologic aspects of SS are needed to establish the diagnosis, predict the prognosis, characterize disease activity, and develop effective therapies.

MicroRNA are a group of small RNAs, 21–24 nucleotides in length, involved in the regulation of a wide variety of cellular and physiologic processes (2–4). They exert their effects by 2 mechanisms: messenger RNA (mRNA) degradation and disruption of translation (5). A single mRNA is usually translated into a single protein; however, a single microRNA is capable of regulating the translation of a multitude of genes involved in a certain function. Changes in individual mRNA levels can be ultimately modulated or nullified by posttranscriptional regulation and thus may be less representative than microRNA of the physiologic status of the cell.

A biomarker is defined as a characteristic that can be measured and evaluated objectively and reproducibly and serves as an indicator of normal or pathogenic biologic processes or pharmacologic responses to therapeutic interventions (6). MicroRNA fulfill all of these requirements and therefore are appealing biomarker candidates. It has been shown in both experimental and clinical settings that microRNA reflect important biologic processes (7–11). In addition, microRNA are remarkably stable, a characteristic that is extremely important in clinical settings and makes microRNA superior to other classes of biomarkers; when using other classes of biomarkers, minor differences in sample processing can have a major impact on outcomes (8). Finally, microRNA can be objectively measured by several methods. Single microRNA are best measured by real-time quantitative polymerase chain reaction (qPCR), whereas microRNA microarrays can provide a more global assessment of microRNA expression patterns. MicroRNA have been shown to be valuable biomarkers for classifying cancers and their prognosis (12–14). Associations of selected biomarkers with autoimmune diseases have also been described (15–20), but to date there have been no reports of the use of microRNA as biomarkers of the diagnosis and pathophysiologic processes in a systemic autoimmune disease.

In this report, we present results of a proof-of-concept study of the feasibility of using microRNA as diagnostic and functional biomarkers in autoimmune diseases. More specifically, we used microRNA microarrays to profile MSGs from control subjects and patients with primary SS. To assess the biologic plausibility of differences in microRNA patterns among various groups, we identified the predicted target pathways of the differentially expressed microRNA, using the Ingenuity Pathways Analysis program.

PATIENTS AND METHODS

Sample acquisition and microRNA isolation

MSGs were obtained from female patients with primary SS (n = 16) and female control subjects (n = 8) enrolled in studies of the natural history of SS or studies evaluating the salivary gland function of healthy control subjects. All subjects had a standardized evaluation for SS that included oral and ophthalmologic examinations, laboratory testing, and a rheumatologic evaluation or a chart review by a rheumatologist at the National Institutes of Health. All patients with SS but none of the control subjects fulfilled the American–European Consensus Group criteria for primary SS (21). Among the control subjects, 5 were healthy volunteers, 1 was seen because of dry eye symptoms, 1 (with a diagnosis of myositis) was seen for symptoms of dry mouth, and 1 was seen as part of a workup for peripheral neuropathy. The study was approved by the Institutional Review Board of the National Institute of Dental and Craniofacial Research, and all subjects provided signed informed consent.

Half of the patients had low-grade inflammation in the MSGs (low focus score of 1 or 2), and half had extensive inflammation in the MSGs (high focus score of 12). Among the patients with SS, 6 had preserved unstimulated salivary flow (≥1.5 ml/15 minutes; “high flow”), and 10 had a low unstimulated salivary flow rate (<1.5 ml/15 minutes; “low flow”) (Figure 1), whereas 5 control subjects had a high flow rate, and 3 had a low flow rate. At the time of biopsy, the median age of subjects in the control group was 43.6 years (range 21.6–58.1), the median age of patients in the group with low focus scores was 57.7 years (range 22.6–66.9), and the median age of patients in the group with high focus scores was 51.9 years (range 37.9–70.6). MSG samples from the lower lip were excised and snap-frozen for microRNA isolation. The tissues were homogenized, and microRNA was isolated with the Qiagen miRNeasy Mini Kit. The quality of RNA was assessed with the Agilent Small RNA Kit on an Agilent 2100 Bioanalyzer and a NanoDrop 8000 spectrophotometer (NanoDrop Technologies).

Figure 1.

Minor salivary gland classes used for the microRNA microarray expression analysis. Each of the subgroups comprised biopsy specimens obtained from Sjögren's syndrome (SS) patients with high salivary flow and those with low salivary flow.

Microarrays

Agilent miRNA V1 microarrays with probes for 470 human and 64 viral microRNA were used to profile MSG microRNA. Only samples with an RNA integrity number >7 on the Agilent RNA 6000 Nano Chip, an optimum microRNA curve on the Agilent Small RNA Chip (determined according to the manufacturer's instructions), and 260:280 and 260:230 ratios <1.8 on the NanoDrop 8000 were used for microarray analysis. For each array, 100 ng of total RNA was used. Arrays were run according to the manufacturer's protocol. Microarrays were scanned using an Agilent Microarray Scanner (model G2505B), and data were extracted and array quality control was performed using Agilent Feature Extraction software.

Analysis

Normalization

MicroRNA array data were normalized using a set of (putative) housekeeper microRNA that were identified in our MSG arrays. (See Results and Supplementary material, available on the Arthritis & Rheumatism Web site at http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1529-0131 for details.)

Data analysis and software

All of the microarray analyses reported here were performed using BRB-ArrayTools (version 3.8.1) developed by Dr. Richard Simon and the BRB-ArrayTools Development Team, with the exception of the principal components analysis (PCA), which was carried out using GeneSpring version 10.0 (Agilent). Normalized microRNA array data were loaded into BRB-ArrayTools, where they were thresholded to 1.0, log₂-transformed, and filtered to remove microRNA with low variability (specifically those microRNA whose variance across all arrays was not significantly [P > 0.01] greater than that of the median of all such variances). Of the 534 microRNA on the arrays, 239 passed the filtering criteria. The resultant filtered data set of human microRNA was used for all subsequent analyses.

In order to increase the specificity for microRNA target prediction, we identified only mRNA that were predicted to be targeted using 2 different prediction algorithms, namely TargetScan (22) and MiRanda (23). Data were also analyzed using Ingenuity Pathways Analysis (www.ingenuity.com). A data set containing mRNA identifiers was uploaded into the application. Only mRNA targeted by at least 50% of the microRNA in each of the groups of interest were used for downstream analysis (see Supplementary material, available on the Arthritis & Rheumatism Web site at http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1529-0131).

Real-time qPCR

Real-time qPCR analysis was performed on 15 MSG samples with various inflammation scores, using the TaqMan MicroRNA Assay (5); these samples were independent of those used in the microRNA microarray analyses. Reverse transcription (RT) was performed according to the manufacturer's instructions, using 10 ng of starting material (determined based on optimized test assays to ensure that detection was within sensitivity limits). Specific microRNA primers were used for the detection of microRNA hsa-miR-768-3p and hsa-miR-574, both of which were acquired from Applied Biosystems (part no. 4395188 and part no. 4395460, respectively). Briefly, for the RT, a 15-μl RT reaction was run on a Veriti 96-Well Thermal Cycler (Applied Biosystems) for 30 minutes at 16°C, 30 minutes at 42°C, and 5 minutes at 85°C. Real-time PCR was performed using an ABI Prism 7500 (Applied Biosystems). Each PCR was run in triplicate. The 20-μl PCR was run with cycling conditions of 10 minutes at 95°C, followed by 40 cycles of denaturing for 15 seconds at 95°C, and annealing and extending for 60 seconds at 60°C.

RESULTS

Samples and array hybridization

Samples obtained from 24 MSGs were hybridized on microRNA microarrays. Eight of these glands were obtained from non-SS control subjects, and 16 were obtained from patients with primary SS (Table 1). Half of the primary SS samples had extensive inflammation (focus score = 12), and half had low-grade inflammation (focus score = 1 or 2). Within the control group and the groups of SS patients with high or low focus scores, some of the subjects had decreased salivary flow, and some had preserved salivary flow (Figure 1). We selected subjects with these clinical parameters to allow for the exploration of microRNA alterations not only between control subjects and patients with SS but also to test whether there is a difference in specific microRNA patterns between patients with different degrees of inflammation or between hypofunctional salivary glands and salivary glands with preserved function.

Table 1.

Characteristics of the control subjects and patients with Sjögren's syndrome (SS)^*

		SS patients
	Non-SS controls (n = 8)	Low focus score (n = 8)	High focus score (n = 8)
Age, median (range) years	43.6 (21.6–58.1)	57.7 (22.6–66.9)	51.9 (37.9–70.6)
Race, white/African American/Asian	5/2/1	7/0/1	5/2/1
Unstimulated salivary flow rate, median (range) ml/15 minutes	5.18 (0–12.9)	2.40 (0–4.735)	0 (0–2.716)
Focus score
0	8	0	0
1	0	5	0
2	0	3	0
12	0	0	8
Objective xerostomia†	3	3	7
Objective xerophthalmia‡	1	8	8
SSA/SSB positive	0	3	7
Hydroxychloroquine treatment	0	2	3
Immunosuppressive treatment	1	2	1
Extraglandular manifestation	Not applicable	3	3

^*Except where indicated otherwise, values are the number of subjects. All subjects were female. All patients had a diagnosis of SS according to the American–European classification criteria. See Patients and Methods for a description of the control subjects.

^†Unstimulated salivary flow rate ≤ 1.5 ml/15 minutes.

^‡Unanesthetized Schirmer test result ≤5 mm/5 minutes or van Bijsterveld score ≥4 in either eye.

Data normalization

Normalization is an important step in microarray data analysis, because it allows for the removal of systematic differences between samples that represent technical rather than biologic variations. A major limitation in the interpretation of microRNA array data is the lack of a clear consensus on the utility and appropriateness of various normalization methods. Many of the classic normalization methods developed for the analysis of mRNA microarrays may not be appropriate for microRNA profiling because of the significant differences between microRNA and mRNA data sets (24). For example, compared with mRNA, many microRNA are expressed at very low levels or not at all, and there are far fewer microRNA than mRNA per array (hundreds of microRNA versus tens of thousands of mRNA). Many normalization methods assume that mRNA intensity distributions are invariant over disease or experimental conditions, but it is not at all clear that this assumption is valid for microRNA.

A widely accepted and well validated method for normalizing mRNA microarrays is the use of housekeeping genes, i.e., genes that are expressed at constant levels in all samples. However, there are no well-established housekeeping microRNA that could be used in a similar way for microRNA arrays. For applications such as qPCR, other noncoding housekeeping genes such as nucleolar RNAs have been used, but their variability under different conditions is significant, and extensive testing must be performed before they can be used as “housekeepers.” To overcome this obstacle, we devised a method for the identification of microRNA that can serve as housekeepers in our data set. This method, which is described in detail in supplementary material (available on the Arthritis & Rheumatism Web site at http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1529-0131), essentially involves searching for microRNA with behavior that is consistent with a housekeeping role (specifically, microRNA whose expression levels vary in synchrony over all arrays, as would be expected if their expression levels were constant). Confining our search to those microRNA that were scored as being present on all arrays (n = 132), we identified a set of 27 such microRNA (Figure 2A). The existence of such a large set of microRNA behaving in this way is consistent with their role as housekeepers and is difficult to explain by any other mechanism.

Figure 2.

Data normalization and analysis. A, Housekeeping microRNA used to normalize data (see Patients and Methods). B, Distribution of Pearson's correlation coefficients for expression of all pairs of microRNA over all microarrays before and after normalization to the housekeeping microRNA. Only microRNA that were scored as present using Agilent Feature Extraction software on all microarrays were included in the calculations. Data are presented as box plots, where the boxes represent the 25th to 75th percentiles, the lines within the boxes represent the median, and the lines outside the boxes represent the 10th and 90th percentiles. Circles indicate outliers.

To confirm the efficacy of our normalization method, we carried out a recently proposed correlation test (25) to assess normalization methods for mRNA arrays. This test is based on the assumption that if 2 genes are chosen at random, it is highly unlikely that their expression levels will be correlated with each other. In other words, although some pairs of genes unquestionably are biologically related, the vast majority of pairs are not. This means that if one calculates the correlation coefficients for the expression levels of all pairs of genes over all microarrays, a properly normalized data set will yield a distribution of correlation coefficients centered near zero. The presence of poorly normalized arrays will introduce an artifactual correlation (because all signals will be increased or decreased in concert) that will skew this distribution. Figure 2B illustrates that normalization with the housekeeper microRNA clearly improves the symmetry and centeredness of this distribution near zero, consistent with the hypothesis that the microRNA listed in Figure 2A behave as housekeepers.

Classification of control and disease subsets

Principal components analysis

PCA is a mathematical method for reducing the dimensionality of a data set while retaining most of the variation. In the case of microarray data, the first principal component is a linear combination of expression patterns that accounts for the greatest amount of variability in the data. The second principal component is independent of the first and accounts for the greatest amount of remaining variability, and so on. Thus, PCA provides an unsupervised analysis that allows one to visualize a multidimensional data set in 2 or 3 dimensions that retains much of the experimental variability. In Figure 3A, each point (circle) represents an array plotted according to its coordinates along the first 3 principal components. This analysis shows a clear separation of all SS samples from control samples as well as a separation of low-focus-score and high-focus-score samples from one another. Therefore, this analysis provided strong evidence that these various groups can be distinguished according to their microRNA profiles.

Figure 3.

A, Principal components analysis (PCA) of all 24 hybridized samples. The samples are plotted along their first 3 principal components; 3 orientations of the PCA plot are shown. Non-Sjögren's syndrome (SS) minor salivary glands (MSGs) are shown in brown, MSGs with low focus scores are shown in blue, and MSGs with high focus scores are shown in red. The plots were exported from Gene-Spring. B, Hierarchical clustering of the microRNA arrays. There is a clear separation of control (C) samples from SS samples as well as a separation between SS samples with high focus scores (HF) and those with low focus scores (LF), with only 1 sample with a high focus score (HF-1) clustering with the samples with low focus scores. The dendro-gram was exported from BRB-ArrayTools. * = samples with high salivary flow.

Hierarchical clustering

Hierarchical clustering, another unsupervised classification method, identifies clusters by merging the samples determined by a defined measure of pairwise similarity of microRNA expression. We used average linkage clustering, in which the distance between 2 clusters is calculated as the average of the distances between all pairs of elements. Similar to the results observed with PCA, there was a distinct separation of MSG samples from control subjects and those from SS patients with high focus scores and SS patients with low focus scores (Figure 3B). The separation of normal tissue from samples with a high focus score was not surprising, because they represent tissues with very different cellular composition. However, the clear distinction between normal tissue and tissue with a low focus score was encouraging and suggests that microRNA profiles are sensitive enough to distinguish between MSGs with minimal histologic differences.

Class prediction using differentially expressed microRNA

Next, we tested whether differentially expressed microRNA could predict control versus SS class membership. Various prediction algorithms (compound covariate predictor, diagonal linear discriminant analysis, 1-nearest neighbor, 3-nearest neighbors, nearest centroid, support vector machines, and Bayesian compound covariate predictor) all correctly classified subjects as patients or controls in 100% of cases, yielding both sensitivity and specificity of 1.0 (see Supplementary Table 1, available on the Arthritis & Rheumatism Web site at http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1529-0131). The classifier was composed of 58 microRNA that were significantly different between the classes at the 0.001 significance level. The leave-one-out cross-validation method was used to compute the mis-classification rate.

Validation of potential biomarkers

To confirm the differences seen on microarrays and to assess the potential of using selected microRNA to distinguish between subgroups of patients with SS, we identified 2 microRNA that changed in opposite directions between the control group, the group with low focus scores, and the group with high focus scores. The expression of one of these microRNA (hsa-miR-768-3p) increased, and expression of the other (hsa-miR-574) decreased with an increasing focus score. In another experiment (Figure 4), we tested the feasibility of using the relative expression levels of these 2 microRNA as predictors of focus scores in a set of 15 samples independent of those used for our microarray studies. The relative expression of these 2 microRNA was determined by calculating the difference between their respective C_t values from the same TaqMan real-time qPCR. A 1-unit difference in the C_t value between the 2 microRNA represents a 2-fold difference in expression between the 2 microRNA. Using this method, we observed statistically significant differences among tissues with low (score = 0–2), medium (score = 5–7), and high (score = 12) focus scores (Figure 4).

Figure 4.

Correlation between relative expression of selected microRNA and minor salivary gland focus scores (FS). From the normalized data set, we identified a pair of microRNA with distinct and opposite expression patterns in control samples, Sjögren's syndrome (SS) samples with low focus scores, and SS samples with high focus scores (the expression of miR-768-3p increases, whereas the expression of miR-574 decreases with increasing focus scores). We validated the expression patterns of these 2 microRNA by real-time quantitative polymerase chain reaction (PCR) in an independent set of samples (n = 15) with various focus scores, determining the relative expression of these 2 microRNA by calculating the difference for their respective C_t values from the same TaqMan real-time quantitative PCR. A 1-unit difference in the C_t value between the 2 microRNA represents a 2-fold difference between the 2 microRNA. There was a statistically significant difference in the C_t values between samples with low focus scores (FS 0–2), those with medium focus scores (FS 5–7), and those with high focus scores (FS 12) (P = 0.0003 by one-way analysis of variance). Values are the mean ± SEM.

Biologic targets of differentially expressed microRNA discriminating among various degrees of inflammation or salivary gland dysfunction

In addition to showing a strong association with clinical characteristics, such as diagnosis, disease activity, or tissue damage, a reliable biomarker must reflect an important underlying physiologic process. Therefore, we assessed the potential biologic significance of our findings by identifying the physiologic processes that may be affected by the differentially expressed microRNA.

To achieve this, we identified microRNA that are associated with either inflammation or salivary gland dysfunction (Table 2). First, using the class comparison algorithm of BRB-ArrayTools set to allow a maximum of 1 false-positive result with a confidence level of 75%, we identified the microRNA that were most significantly differentially expressed among the control, low-focus-score SS, and high-focus-score SS groups (P < 0.01) (see Supplementary Table 2, available on the Arthritis & Rheumatism Web site at http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1529-0131). From these, we selected microRNA that were most strongly associated with an inflammation pattern. The expression of microRNA (n = 10) in this group increased >50% in the group with low focus scores compared with the control group and increased >2.5-fold in the group with high focus scores compared with the group with low focus scores; i.e., their expression levels consistently increased as inflammation increased (Table 2).

Table 2.

Class comparison between control samples, Sjögren's syndrome (SS) samples with low focus scores, and SS samples with high focus scores and between samples with normal salivary flow and samples with decreased salivary flow independent of the focus score and independent of being derived from SS patients or controls^*

		Focus score		Salivary flow
Unique ID	Control	Low	High	Normal	Decreased
hsa-miR-150	3.2	122.4	931.3	-	-
hsa-miR-650	4.6	7.2	49.3	-	-
hsa-miR-142-5p	221.6	442.5	2,067.8	-	-
hsa-miR-135b	1.0	8.0	31.8	-	-
hsa-miR-330	1.6	3.3	11.7	-	-
hsa-miR-513	3.7	60.4	189.5	-	-
hsa-miR-342	61.5	175.5	530.5	-	-
hsa-miR-501	2.9	6.9	20.6	-	-
hsa-miR-126*	1.0	55.9	162.0	-	-
ebv-miR-BART19	5.2	482.4	1,206.8	-	-
hsa-miR-765	-	-	-	37.0	91.1
hsa-miR-181a	-	-	-	113.6	210.5
hsa-miR-766	-	-	-	13.4	42.4
hsa-miR-335	-	-	-	407.4	231.5
hsa-miR-16	-	-	-	1,883.7	2,973.8
hsa-miR-671	-	-	-	32.0	108.0
hsa-miR-663	-	-	-	43.2	156.2
hsa-miR-340	-	-	-	10.6	2.7
hsa-miR-155	-	-	-	171.2	436.5

^*Values are the geometric mean for each group. Differentially expressed microRNA were identified using the BRB-ArrayTools class comparison algorithm set to allow a maximum of 1 false-positive hit with a confidence level for false discovery of 75%. The complete set of 94 differentially expressed microRNA with P values less than 0.01 is listed in Supplementary Table 1 (available on the Arthritis & Rheumatism Web site at http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1529-0131). See Patients and Methods for details.

To further explore whether specific microRNA are associated with decreased function independent of inflammation or the diagnosis of SS, we compared the microRNA expression between all samples with either preserved or low salivary flow. We identified 9 microRNA that were differentially expressed between patients with normal salivary flow and those with decreased salivary flow (Table 2). Interestingly, 7 of these 9 microRNA were up-regulated in the group with decreased gland function, suggesting that their targets may have a protective effect on epithelial cells.

We hypothesized that microRNA related to decreased salivary flow would target pathways that are not related to increased inflammation in SS but are related to other aspects of the disease process. To test this hypothesis, first we identified the predicted mRNA targets of these microRNA using the TargetScan (22) and MiRanda (23) target prediction algorithms and then used Ingenuity Pathways Analysis software to predict the physiologic networks and functions in which those predicted mRNA targets are involved. To reduce the possibility of identifying spurious pathways arising from multiple microRNA targets, we included only those mRNA that were predicted to be targeted by at least 50% of the microRNA in each group. The resulting most statistically significant biologic functions in the inflammation group identified complex biologic functions, such as neurogenesis, pathways involved in hepatic system disorder, neoplasia, and autoimmune diseases (see Supplementary Figure, available on the Arthritis & Rheumatism Web site at http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1529-0131), whereas use of Ingenuity Pathways Analysis software for those mRNA that were predicted to be targeted by at least 50% of the microRNA associated with salivary gland dysfunction identified neurogenesis, cellular and tissue growth, receptor signaling, and cellular adhesion as the most significant biologic functions targeted by these microRNA (see Supplementary Figure, available on the Arthritis & Rheumatism Web site at http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1529-0131).

DISCUSSION

This is the first study examining the utility of using MSG microRNA expression profiles as biomarkers in SS. Using microRNA microarrays and a novel method of data normalization, we demonstrated that microRNA profiling is capable of distinguishing between MSGs from control subjects and those from patients with SS, and that specific microRNA profiles are associated with 2 concurrent but distinct pathophysiologic processes, namely inflammation and exocrine gland dysfunction.

MicroRNA have several features that make them attractive as biomarkers: they play important regulatory roles, their expression patterns reflect specific physiologic processes, and importantly, microRNA are very stable and resistant to differences in specimen handling, thereby increasing their appeal as clinical biomarkers. MicroRNA expression profiles have been shown to have diagnostic and prognostic value in some malignancies. However, the feasibility of using microRNA as diagnostic and prognostic biomarkers has not been tested in autoimmune diseases, which arguably represent a more complex situation.

For this proof-of-concept study, we chose MSG biopsy specimens obtained from patients with SS, because they allowed us to study the relationship between microRNA and two physiologic processes, inflammation and exocrine gland dysfunction. By choosing patients representing the two extremes of the inflammation spectrum and including both patients with preserved salivary flow and those with decreased salivary flow, we were able to identify microRNA patterns associated with either inflammation or exocrine gland dysfunction.

Normalization is an important step in the analysis of microRNA microarrays, but at present there is no consensus on how this should be done. The applicability of many of the methods used to normalize mRNA arrays to microRNA arrays is still uncertain. The use of housekeeper microRNA that are expressed at constant levels in all samples is an ideal solution, but to date, no such housekeepers have been identified. We approached this problem from an operational perspective by searching for a group of microRNA that behaved as if they had such a housekeeping function. This approach enabled us to identify a set of 27 putative housekeeper microRNA, which we showed could be effectively used to normalize our microarray data (Figure 2).

MicroRNA expression patterns clearly separated non-SS control subjects from all patients with SS, using unsupervised clustering and classification analyses. This was not surprising for the samples with high focus scores, given their differences in cellular composition from control biopsy samples or those with low focus scores. The real challenge for a new putative biomarker is to distinguish between histologically similar but clinically and functionally distinct biopsy specimens. In this study, microRNA profiles achieved these goals to a great degree. They were able to separate SS biopsy specimens with low focus scores from both SS biopsy specimens with high focus scores and control specimens, including some from patients with xerostomia or systemic autoimmune disease. Because using microRNA arrays is not practical in everyday practice, we tested 2 selected microRNA in an independent cohort of patients with a wide range of inflammation-related conditions. We chose hsa-miR-768-3p and hsa-miR-574, based on their inverse relationship to focus scores and observed good correlation between the differential expression of these 2 microRNA and focus scores. If confirmed in a larger number of patients, this approach could significantly reduce the substantial subjectivity of scoring inflammatory infiltrates (26,27) and may aid in the diagnosis of SS.

Purely establishing the degree of inflammation is insufficient in many autoimmune diseases, in which, in addition to inflammation, end organ function is determined by the combination of other factors, such as irreversible damage, comorbidities, and end organ–specific pathophysiologic processes. Finding biomarkers that distinguish between the various pathologic processes could significantly improve the selection of optimal treatment strategies for these patients. The poor correlation between inflammation and decreased endocrine gland function in SS strongly suggests that both inflammatory and noninflammatory processes contribute to end organ dysfunction. This dichotomy makes SS salivary glands an excellent model with which to evaluate microRNA as biomarkers of different pathophysiologic processes. We were able to identify distinct sets of microRNA that were related to inflammation or salivary gland dysfunction.

To assess the biologic plausibility of these differences, we used Ingenuity Pathways Analysis software to predict the most likely targets of these microRNA. To reduce the probability of false-positive hits, we included only those mRNA in the Ingenuity Pathways Analysis program that were targeted by at least 50% of microRNA in each group. Comparing the predicted biologic targets of these two groups identified both overlapping and distinct biologic pathways and processes. Despite the differences in the specific pathways, it is intriguing that various neurologic functions were among the main targets for both groups, suggesting that neurologic regulation of salivary glands may be central to the pathogenesis of SS.

This is a pilot, proof-of-concept study with a limited number of patients. To reduce the inherent heterogeneity of SS, we included patients at the two ends of the inflammation spectrum and focused on salivary glands to specifically address the question of whether salivary gland microRNA could be used as biomarkers of salivary gland inflammation and dysfunction. The relationship between extraglandular manifestations and salivary gland microRNA expression was not evaluated because of concerns that the highly selected patient population used in this study would result in questionable validity of any conclusion going beyond these two issues. Further studies are needed to identify a smaller set of microRNA associated with salivary gland inflammation and dysfunction or extraglandular manifestations. These microRNA will have to be validated in larger cohorts including healthy control subjects, patients with the full spectrum of SS, and appropriate disease controls, such as patients with other autoimmune diseases or sicca symptoms without evidence of inflammation.

In summary, we have shown that microRNA are promising candidate biomarkers of inflammation and salivary gland dysfunction in patients with SS. Further exploration of the predicted pathways associated with decreased salivary flow in this study will provide insight into the pathophysiology of SS and may identify novel therapeutic targets.