Gene Expression Profiling and Heterogeneity of Nonspecific Orbital Inflammation Affecting the Lacrimal Gland

Key Points

Question  Is nonspecific orbital inflammation affecting the lacrimal gland a single disease or multiple different diseases, and is it a limited form of other diseases, such as sarcoidosis or granulomatosis with polyangiitis?

Findings  This cohort study comparing the expression of 40 genes from biopsy specimens of 48 patients having lacrimal nonspecific orbital inflammation with healthy controls indicates that such inflammation is a heterogeneous collection of diseases and suggests that it is often a limited form of known lacrimal inflammation.

Meaning  Gene expression patterns can subdivide lacrimal gland nonspecific orbital inflammation, having the potential to lead to more precise therapy and new insights into pathogenesis.

Importance  Although a variety of well-characterized diseases, such as sarcoidosis and granulomatosis with polyangiitis, affect the lacrimal gland, many patients with dacryoadenitis are diagnosed as having nonspecific orbital inflammation (NSOI) on the basis of histology and systemic disease evaluation. The ability to further classify the disease in these patients should facilitate selection of effective therapies.

Objective  To test the a priori hypothesis that gene expression profiles would complement clinical and histopathologic evaluations in identifying well-characterized diseases and in subdividing NSOI into clinically relevant groups.

Design, Setting, and Participants  In this cohort study, gene expression levels in biopsy specimens of inflamed and control lacrimal glands were measured with microarrays. Stained sections of the same biopsy specimens were used for evaluation of histopathology. Tissue samples of patients were obtained from oculoplastic surgeons at 7 international centers representing 4 countries (United States, Saudi Arabia, Canada, and Taiwan). Gene expression analysis was done at Oregon Health & Science University. Participants were 48 patients, including 3 with granulomatosis with polyangiitis, 28 with NSOI, 7 with sarcoidosis, 4 with thyroid eye disease, and 6 healthy controls. The study dates were March 2012 to April 2017.

Main Outcomes and Measures  The primary outcome was subdivision of biopsy specimens based on gene expression of a published list of approximately 40 differentially expressed transcripts in blood, lacrimal gland, and orbital adipose tissue from patients with sarcoidosis. Stained sections were evaluated for inflammation (none, mild, moderate, or marked), granulomas, nodules, or fibrosis by 2 independent ocular pathologists masked to the clinical diagnosis.

Results  Among 48 patients (mean [SD] age, 41.6 [19.0] years; 32 [67%] female), the mclust algorithm segregated the biopsy specimens into 4 subsets, with the differences illustrated by a heat map and multidimensional scaling plots. Most of the sarcoidosis biopsy specimens were in subset 1, which had the highest granuloma score. Three NSOI biopsy specimens in subset 1 had no apparent granulomas. Thirty-two percent (9 of 28) of the NSOI biopsy specimens could not be distinguished from biopsy specimens of healthy controls in subset 4, while other examples of NSOI tended to group with gene expression resembling granulomatosis with polyangiitis or thyroid eye disease. The 4 subsets could also be partially differentiated by their fibrosis, granulomas, and inflammation pathology scores but not their lymphoid nodule scores.

Conclusions and Relevance  Gene expression profiling discloses clear heterogeneity among patients with lacrimal inflammatory disease. Comparison of the expression profiles suggests that a subset of patients with nonspecific dacryoadenitis might have a limited form of sarcoidosis, while other patients with NSOI cannot be distinguished from healthy controls.

Dacryoadenitis has a broad differential diagnosis that includes sarcoidosis, nonspecific orbital inflammation (NSOI), Sjögren syndrome, and, less commonly, entities like granulomatosis with polyangiitis (GPA), thyroid eye disease (TED), or IgG4-related disease.^1-5 Other considerations in the differential diagnosis are infections, tumors (including lymphoma), and lymphoid hyperplasia. Among these diagnoses, nonspecific inflammation is the most common biopsy diagnosis.²

Although histopathology is certainly a component of the criterion standard for diagnosing diseases of the lacrimal gland, the approach has limitations. For example, GPA is a medium-sized vessel vasculitis, but vessels of that size are rarely present within a lacrimal specimen. Although the pathology of sarcoidosis includes granulomas, which are readily appreciated by histology, a sampling error might exclude granulomas from the tissue that is microscopically examined. The presence of IgG4-positive plasma cells is emerging as an important factor in lacrimal inflammation,⁴ but the specificity of these cells has been questioned.⁶ Clearly, there is opportunity to improve the diagnostic yield from lacrimal gland biopsy. Furthermore, an unresolved issue is whether nonspecific inflammation represents a single diagnostic entity or a variety of different inflammations.

Implementation of molecular techniques has the potential to increase the accuracy and specificity of diagnosing the different forms of dacryoadenitis. For example, gene expression profiling can distinguish different causes of synovitis,⁷ esophagitis,⁸ or myocarditis.⁹ Our group previously used gene expression profiling to characterize different causes of uveitis^10,11 and orbital adipose tissue inflammation.^12-15 In another study,¹⁶ our group analyzed gene expression in blood, lacrimal gland, and orbital adipose tissue from individuals affected by sarcoidosis. In that study, a set of 40 messenger RNA transcripts was selected that were differentially expressed in all 3 diseased tissue samples of patients compared with the same tissue samples from healthy controls.

We now have compared gene expression profiles of lacrimal gland from individuals having GPA, NSOI, sarcoidosis, or TED with profiles from healthy controls. Based on the 40 transcripts identified from our group’s prior sarcoidosis study,¹⁶ we used the mclust algorithm¹⁷ to test the hypothesis that nonspecific lacrimal gland inflammation is a heterogeneous collection of diseases that sometimes resembles sarcoidosis. The study dates were March 2012 to April 2017.

We obtained 49 lacrimal gland biopsy specimens from 48 individuals (1 lacrimal gland from a patient with sarcoidosis was sampled from 2 separate locations) at the following 7 international centers: Oregon Health & Science University, Emory University, King Khaled Eye Specialist Hospital, University of British Columbia, Ophthalmic Surgeons and Consultants of Ohio, Medical College of Wisconsin, and Kaohsiung Veteran’s General Hospital. All tissue had been formalin fixed and paraffin embedded. All tissue samples had been reviewed by an ophthalmic pathologist from the contributing center and then further independently reviewed by two of us who are ocular pathologists (D.J.W. and H.E.G.) and who collaborated in the preparation of this report. These centers and the ocular pathologists have previously used a similar method of tissue collection to analyze gene expression from a variety of orbital diseases, including TED,¹³ GPA,¹² sarcoidosis,¹⁶ and NSOI.¹⁴ Control tissue from healthy individuals was obtained at the time of cosmetic surgery or blepharoplasty. In some instances, surgeons removed portions of normal lacrimal gland if the gland prolapsed during the course of orbital surgery. The study was approved by the institutional review board at Oregon Health & Science University and by the institutional review board at each participating center. Written informed consent was obtained when required by the local review board.

Two of us (D.J.W. and H.E.G.) were tasked with confirming the diagnosis from the institution where the biopsy was obtained and also independently scored each tissue on the basis of fibrosis, granulomas, lymphoid nodules, and inflammation. In each case, an ordinal scale from 0 to 3 was used, with the pathology scores reflecting none, mild, moderate, or marked change for each given descriptor. The scores of 2 of us (D.J.W. and H.E.G.) were in agreement two-thirds of the time. The scores were averaged for analyses.

All tissue samples were sent to Oregon Health & Science University for RNA extraction and microarray as previously described.^12,13,16 In brief, complementary DNA was synthesized from purified RNA, amplified, labeled, and hybridized to arrays (U133 Plus 2.0; Affymetrix), which include probe sets for approximately 45 000 transcripts. Furthermore, our group has reported on the RNA quality and the correlation between our array data and quantitative polymerase chain reaction.¹⁸

Files (CEL; Affymetrix) were imported into R statistical language,¹⁹ and expression levels were calculated by the robust multiarray analysis.²⁰ The mclust algorithm¹⁷ was used to cluster the normalized gene expression values of selected probe sets. A clustering analysis is a type of unsupervised machine learning that groups samples into clusters such that those within a cluster are more closely related to one another than those assigned to different clusters based on some dissimilarity measure.²¹ Most of the conventional clustering algorithms, such as hierarchical clustering or K-means clustering,²² require that the number of clusters be prespecified, which can be a subjective decision and disadvantage in applications. In contrast, the mclust algorithm assumes that samples are collected from a number of gaussian distributions (or normal distributions). First, the algorithm fits a range of different gaussian mixture models, and then it chooses an optimal number of gaussian distributions or clusters based on the Bayesian information criterion.²³ Therefore, the mclust algorithm can provide a more objective number of clusters than the conventional clustering algorithms.

Heat maps and multidimensional scaling (MDS) plots were used to visualize the cluster analysis results. All computations were done using affy, limma, and mclust packages of R statistical language.

In our group’s previous study¹⁶ investigating gene expression differences due to sarcoidosis in blood, lacrimal gland, or orbital adipose tissue, we used linear models and empirical Bayes methods²⁴ while adjusting for potential confounding effects of age and sex. The linear models were fitted to each tissue type separately, and race was not included in the models due to too many missing values. Our group reported 159 Affymetrix probe sets indicating differential expression common to all 3 of these tissue samples (at least 1.5-fold change compared with healthy controls, with an adjusted false discovery rate of P < .05).¹⁶ Among them, 45 probe sets had a false discovery rate of P < .005 and were used for the analyses shown in Figure 1 and Figure 2. Because of redundancy in the probe sets, the 45 probe sets represent 40 different genes.

Among 48 patients, the mean (SD) age was 41.6 (19.0) years, and 32 (67%) were female. We analyzed 49 lacrimal biopsy specimens from 48 patients, including 3 with GPA, 28 with NSOI, 7 with sarcoidosis, 4 with TED, and 6 healthy controls. The mean age and sex for each of these 5 groups are listed in the Table. Information on race is also provided. Although the groups differ in some comparisons on the basis of age, sex, or race, the differences are not statistically significant. Because of the numbers of specimens in each of our disease groups, we focused our initial analysis on 45 core probe sets that our group previously reported as discriminating between gene expression profiles of blood, lacrimal gland, and orbital adipose tissue from individuals having sarcoidosis compared with profiles from healthy controls.¹⁶ Application of the mclust algorithm with the selected probe sets used unsupervised machine learning to separate the individuals into an optimal number of clusters or subsets with 5 different gene expression patterns based on the Bayesian information criterion.²³ Inclusion of additional probe sets that our group previously found to be differentially expressed in orbital adipose tissue samples from individuals with these diseases^12,13,15,16 decreased the consistency of clustering as indicated by the mclust algorithm with the disease diagnosis.

The mclust algorithm results based on the core probe sets are shown in the heat map (Figure 1) and the MDS plot (Figure 2 and Video). These probe sets discriminate among the healthy controls and the 3 known diseases (GPA, sarcoidosis, and TED). As seen in both the heat map and the MDS plot, 5 of the 6 healthy control biopsy specimens, 2 of the 3 GPA specimens, 6 of the 8 sarcoidosis specimens, and 3 of the 4 TED specimens cluster together. These same genes subdivide the NSOI biopsy specimens into the 4 clusters. Three of the 28 NSOI biopsy specimens have a gene expression pattern resembling sarcoidosis even though granulomas were not detected in the biopsy specimens by either pathologist. Clinical information was available on 2 of the 3 individuals with NSOI whose gene expression profile resembled sarcoidosis. Both individuals had received observation only without therapy, and neither had chest computed tomography imaging. Other NSOI biopsy specimens group with other diagnoses, including GPA or TED, and with healthy controls. In both Figures 1 and 2, underlining is used to identify the 2 biopsy specimens from different portions of the lacrimal gland from the same individual. The pattern seen in the heat map and the proximity of the 2 data points shown in the MDS plot indicate that the reproducibility of the method is good.

We next analyzed the lacrimal biopsy specimens based on the pathologists’ scoring for fibrosis, granulomas, lymphoid nodules, or inflammation. In Figure 3A, the biopsy specimens have been grouped based on the gene expression cluster. Using χ² analysis, the clusters differ for each of these 4 pathology scores, with χ² = 32.2, P = .01 for fibrosis, χ² = 21.7, P = .001 for granulomas, χ² = 19.0, P = .002 for lymphoid nodules, and χ² = 49.2, P = .001 for inflammation. We then analyzed how the 4 pathology scores correlated with the histologic diagnosis. As shown in Figure 3B and as confirmed by χ² analysis, lymphoid nodules do not discriminate among the various diagnoses (χ² = 19.0, P = .75). The quantification of fibrosis tends to be a useful discriminator, although the differences did not reach statistical significance (χ² = 13.6, P = .09). On the other hand, either the detection of granulomas or the extent of inflammation had a nonrandom distribution among the diagnoses as determined by χ² analysis (granulomas χ² = 49.5, inflammation χ² = 58.2, P < .001 for both variables), although the inflammation pathology scores showed considerable overlap.

An important question to address is whether the results could be attributed to medication use. Data on prednisone use were known for 25 individuals in this study. For the 6 participants receiving prednisone, the dosages were moderately high and ranged from 20 to 60 mg/d (mean, 39.2 mg/d). With only 6 individuals receiving prednisone, statistical conclusions about the influence of the prednisone use should be tentative. However, the MDS plot shown in Figure 4 highlights those receiving prednisone therapy. Both of the participants with sarcoidosis who did not cluster with the other sarcoidosis biopsy specimens were taking prednisone at the time of the biopsy. The individual with GPA receiving prednisone appears to be shifted to the right as well. The use of corticosteroids is known to have major effects on gene expression²³; therefore, it is plausible to speculate that their use influenced some results herein.

Finally, potential uses of gene expression profiling are to predict prognosis or to identify therapy that has a high likelihood of success. Although the size of the database and the duration of follow-up data in this cohort study are not adequate to address these issues thoroughly, one clinical vignette might be instructive. An individual who was included in the study had a history of well-documented Graves disease more than a decade before the onset of orbital disease. Her orbital imaging showed bilateral but asymmetric inflammation that involved the inferior and superior rectus muscles as well as orbital fat, the lacrimal glands, and the cavernous sinus. The radiologist thought that the findings were atypical for thyroid orbitopathy due to the lacrimal and sinus involvement. The pathologist read the biopsy results as nonspecific inflammation. Her gene expression profile also supported this conclusion, with a pattern of expression that fell into cluster 1 (as shown in Figure 1) rather than with the TED biopsy specimens. The patient responded to rituximab therapy, an approach that was not effective for TED in a randomized controlled trial.²⁵

Our observations based on gene expression indicate clearly that nonspecific inflammation within the lacrimal gland is a heterogeneous collection of diseases. The data suggest that a minority of patients with nonspecific lacrimal gland inflammation might have a localized form of sarcoidosis even if granulomas are not present in the biopsy specimen examined. Other examples of NSOI resemble either GPA or TED in terms of the pattern of gene expression. More detailed clinical information and more extensive follow-up data would help determine if these individuals followed different clinical courses that correlated with the gene expression profile.

Our study found some degree of fibrosis, inflammation, and even lymphoid nodules among healthy controls, which corresponds well with what has been reported in minor salivary gland and lacrimal gland biopsy specimens.²⁶ Some degree of fibrosis and lymphoid nodularity is frequently detected in tissue samples from individuals who have no known abnormality regarding function of these glands. The clustering on the basis of gene expression subdivided the biopsy samples into groups that correlated with the pathology scores more consistently than categorizing the tissue on the basis of a diagnosis (ie, GPA, or sarcoidosis, or TED). Either grouping (ie, by gene expression or by diagnosis) reveals that the 4 pathology scores (fibrosis, granulomas, lymphoid nodules, and inflammation) do not reliably subdivide the tissue samples based on the overlap shown in Figure 3.

Our study has some limitations. The international centers might have varied in their clinical approach (eg, the acceptable duration of symptoms) before biopsy. They might also have differed as to pharmacologic treatment before biopsy. In addition, the length of time that the biopsy specimen was in fixative could have influenced the integrity of the RNA. Furthermore, our data on the clinical course of study participants were somewhat limited. Finally, some of our data suggest that oral corticosteroid use influences gene expression,²³ but we did not have adequate data to assess other medication use. We also emphasize that our observations on corticosteroid use are based on a small number of individuals receiving a high dosage of prednisone. In a prior study¹³ on orbital adipose tissue that focused on a different set of transcripts, our group could not show that corticosteroid use affected gene expression. Additional study of this issue is warranted.

Despite these limitations, this investigation is arguably the largest study to date on gene expression in lacrimal inflammation and the first, to our knowledge, to definitively show the heterogeneity of NSOI in the lacrimal gland. Prior studies on gene expression in the lacrimal gland²⁷ or in the lacrimal accessory gland²⁸ were limited to normal tissue samples. Our analysis supports the conclusion that some patients with nonspecific lacrimal gland inflammation might have a limited form of GPA, sarcoidosis, or even TED. Furthermore, our findings demonstrate how patterns of gene expression can provide data that complement the information gleaned by light microscopy.

Accepted for Publication: July 15, 2017.

Corresponding Author: James T. Rosenbaum, MD, Casey Eye Institute, Oregon Health & Science University, Mail Code L467AD, 3181 SW Sam Jackson Park Rd, Portland, OR 97239 (rosenbaj@ohsu.edu).

Published Online: September 21, 2017. doi:10.1001/jamaophthalmol.2017.3458

Author Contributions: Drs Rosenbaum and Planck had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Rosenbaum, Choi, Harrington, Grossniklaus, Maktabi, Dubovy, Planck.

Acquisition, analysis, or interpretation of data: Rosenbaum, Choi, Harrington, Wilson, Grossniklaus, Sibley, Salek, Ng, Dailey, Steele, Hayek, Craven, Edward, Al Hussain, White, Dolman, Czyz, Foster, Harris, Bee, Tse, Alabiad, Dubovy, Kazim, Selva, Yeatts, Korn, Kikkawa, Silkiss, Sivak-Callcott, Stauffer, Planck.

Drafting of the manuscript: Rosenbaum, Choi, Harrington, Al Hussain, Kikkawa, Planck.

Critical revision of the manuscript for important intellectual content: Rosenbaum, Choi, Harrington, Wilson, Grossniklaus, Sibley, Salek, Ng, Dailey, Steele, Hayek, Craven, Edward, Maktabi, White, Dolman, Czyz, Foster, Harris, Bee, Tse, Alabiad, Dubovy, Kazim, Selva, Yeatts, Korn, Silkiss, Sivak-Callcott, Stauffer, Planck.

Statistical analysis: Choi, Sibley, Salek, Dubovy, Korn.

Obtained funding: Rosenbaum, Choi, Planck.

Administrative, technical, or material support: Harrington, Wilson, Grossniklaus, Salek, Dailey, Hayek, Maktabi, Al Hussain, White, Czyz, Foster, Bee, Tse, Alabiad, Yeatts, Korn, Kikkawa, Silkiss, Sivak-Callcott, Stauffer, Planck.

Study supervision: Rosenbaum, Dubovy, Selva, Korn, Planck.

Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Dr Rosenbaum reported serving as a consultant for Genentech and reported being a coinvestigator on a study funded by Genentech to evaluate the use of rituximab for orbital inflammatory diseases. No other disclosures were reported.

Funding/Support: This study was supported in part by grants EY020249, EY010572, and RR024140 from the National Institutes of Health and by Research to Prevent Blindness, the William and Mary Bauman Foundation, the Mas Family Foundation, and the Stan and Madelle Rosenfeld Family Trust.

Role of the Funder/Sponsor: The funding organizations had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Meeting Presentation: Some of these data were presented at the 2017 Annual Meeting of the Association for Research in Vision and Ophthalmology; May 11, 2017; Baltimore, Maryland. The raw and normalized gene expression microarray data will be uploaded to the Gene Expression Omnibus (GEO) database.

Additional Contributions: Kristina Vartanian, BS, Integrated Genomics Laboratory, Oregon Health & Science University, provided technical support for the microarray work (without compensation). The RNA extraction and microarray assays were performed in the Oregon Health & Science University Gene Profiling Shared Resource.