Categorization of patients with systemic lupus erythematosus using disease activity, patient-reported outcomes, and transcriptomic signatures

Patients with systemic lupus erythematosus (SLE) display symptoms that are not always related to disease activity and may distort clinical trial results. Recently, a clinical categorization based on the presence of type 1 (inflammatory manifestations) and/or type 2 (widespread pain, fatigue, depression) symptoms has been proposed in SLE. Our aim was to develop a type 2 score derived from the Short-Form health survey (SF-36) to categorize SLE patients and to compare immunological and transcriptomic profiles between groups. Seventeen items from the SF-36 were selected to build a type 2 score for 50 SLE patients (100 visits; LUPUCE cohort), and the SLEDAI was used to define type 1 symptoms. Patients were categorized into four groups: minimal (no symptoms), type 1, type 2, and mixed (both type 1 and type 2 symptoms). Clinical, immunological, and transcriptomic profiles were compared between the groups. Type 2 scores ranged from 0 to 31, with a cutoff value of 14 (75th percentile). The sample categorization was minimal in 39%, type 1 in 37%, and type 2 in 9%, and mixed in 15%. Type 2 patients were older than minimal patients and had a longer disease duration than type 1 and mixed patients. Immunological data and modular interferon signatures did not differ between the groups. Patients with SLE can be categorized into four clinical groups using the SLEDAI score and our SF-36-derived type 2 score. This categorization is non-redundant with immunological or transcriptomic profiles and could prove useful to stratify patients in clinical trials. Key Points • A score derived from selected items of the SF-36 can be used to identify SLE patients with type 2 symptoms according to the Duke University categorization. • Using the SLEDAI and this type 2 score, SLE patients can be categorized into four clinical groups. • This categorization is not related to immunological activity or blood transcriptome profiles (and not to the interferon signature in particular). • This categorization could be useful in the daily care of patients as well as in clinical trials, for upstream patient stratification or for the interpretation of results. Key Points • A score derived from selected items of the SF-36 can be used to identify SLE patients with type 2 symptoms according to the Duke University categorization. • Using the SLEDAI and this type 2 score, SLE patients can be categorized into four clinical groups. • This categorization is not related to immunological activity or blood transcriptome profiles (and not to the interferon signature in particular). • This categorization could be useful in the daily care of patients as well as in clinical trials, for upstream patient stratification or for the interpretation of results.


Introduction
Systemic lupus erythematosus (SLE) is a chronic autoimmune disease, characterized by a wide range of symptoms. Some SLE symptoms are related to systemic and local inflammation resulting from disease activity, while others can affect quiescent patients. Although survival has been improved dramatically with medications [1], patients with SLE suffer from a persistently altered health-related quality of life (HRQoL) [2,3], mainly influenced by fatigue, reported by 67-90% of patients [4][5][6][7], chronic pain, and depression [8]. These symptoms correlate poorly with disease activity [7], do not respond to conventional immunosuppression, and are often difficult to manage. In addition, they can lead to the failure of therapeutic candidates to improve HRQoL in clinical trials [9]. Recently, the categorization of SLE symptoms between two non-exclusive types was proposed by Duke University [10] and tested in a cohort of 212 patients [11]. Type 1 symptoms include inflammatory manifestations related to disease activity, such as nephritis, arthritis, rash, serositis, and cytopenia; type 2 symptoms include widespread pain, fatigue, sleep disturbance, cognitive dysfunction, and depression. This allowed the categorization of patients at the time of clinical assessment into four distinct groups [10]: (i) "type 1," patients displaying only type 1 symptoms; (ii) "type 2," patients displaying only type 2 symptoms; (iii) "mixed," patients displaying both type 1 and type 2 symptoms; (iv) "minimal," patients with no symptoms. This classification was built upon the SLEDAI score (type 1 symptoms), which is widely available in SLE studies, and on American College of Rheumatology (ACR) fibromyalgia criteria [12] (type 2 symptoms) [10], which are less commonly used in SLE and are rarely available retrospectively.
Our aim was to determine whether selected items from the SF-36 questionnaire could allow the identification of type 2 symptoms and the retrospective categorization of patients from different SLE cohorts. We then aimed to evaluate whether these categories were associated with distinct immunological and transcriptomic profiles. In particular, we compared the interferon (IFN) modular scores of samples from these four groups through a modular framework of analysis described [13].

Patients and ethics
This study involved data from patients included in the LUPUCE study (NCT00920114) and the PSY-LUP study (NCT03913754). All patients gave their written informed consent before any study-related procedure. The study was conducted according to the Declaration of Helsinki.
LUPUCE was a transcriptomic study [13][14][15] comprising 62 consecutive SLE patients, of whom 50 had completed the self-administered SF-36 questionnaire on at least one visit and were included in the present study. Clinical, biological, and transcriptomic data were recorded at each visit. The study was approved in France by the Comité de Protection des Personnes Sud Méditerranée 1 (IDRCB 2009-A00257-50) and in the USA by the Institutional Review Boards (IRBs) of the Baylor Institute of Immunology Research (IRB 011-173) and the Benaroya Research Institute (IRB 12,085). PSY-LUP (NCT03913754) is a study evaluating the psycho-social consequences of SLE. PSY-LUP was independent of the LUPUCE cohort. In order to validate the type 2 score, we applied it to the PSY-LUP cohort. One-hundred SLE patients responded to several quality of life (QoL) questionnaires including the SF-36. The study was approved by the Comité de Protection des Personnes Ile de France I (IDRCB A02747-48).

Assessment of type 1 and type 2 symptoms
Patients were considered to have type 1 symptoms if they had a SLEDAI ≥ 6, a clinical SLEDAI ≥ 4, or active lupus nephritis documented by kidney biopsy [11].
The presence of type 2 symptoms was defined using a score built from the responses to specific items of the SF-36 questionnaire [16] (detailed in Table 1). Three physicians experienced in SLE (RA, LC, and NJC) manually selected 17 items/questions they considered the most relevant to express widespread pain, fatigue, and depression. According to their clinical experience and knowledge of the disease, the likelihood of the patients' responses item by item was assessed to select the relevant items. The selection process consisted of a consensus between the 3 physicians. Responses to these items were weighted and the global type 2 score ranged from 0 to 42. The 75th percentile of this score in the LUPUCE cohort was chosen as the threshold to consider that a patient displayed significant type 2 symptoms.

Immunological activity
The immunological activity was evaluated by the levels of the complement fractions C3 and/or C4 and of anti-dsDNA antibodies.

Blood modular transcriptional repertoire analyses
Whole-blood transcriptomic data collected in the LUPUCE study were investigated [13]. RNA was processed as described elsewhere using Illumina beadchips [13]. Data are deposited in the NCBI Gene Expression Omnibus (GEO, http:// www. ncbi. nlm. nih. gov/ geo, GEO Series accession number GSE49454). Analyses were performed using the second generation of a modular framework as described [13,17]. The level of regulation of each module was calculated as the percentage difference: % upregulated probes -% downregulated probes. A module was considered activated (upregulated) or inhibited (downregulated) if its value was, respectively, ≥ 20% or ≤ − 20%. A modular IFN score was calculated depending on the activation of three IFN modules (M1.2, M3.4, and M5.12). The IFN signature was absent, mild, moderate, or strong if 0, 1, 2, or 3 IFN modules were active (≥ 20%) respectively [13].

Statistical analysis
Quantitative variables are described as the median and interquartile range (IQR), or mean ± standard deviation (SD); categorical variables are described as numbers and percentages.
The four clinical groups were compared for patient characteristics, immunological activity, and level of regulation of IFN modules (modules M1.2, M3.4, and M5.12) using the Kruskal Wallis test and Tukey's test for pairwise comparisons. Categorical patient characteristics, frequency of IFN signature, and active modules were compared between the four

Type 2 score and patient distribution between clinical groups
All visits (n = 100) at which patients from the LUPUCE cohort answered a SF-36 questionnaire were considered to calculate the type 2 score (  Table 2). The type 2 score was calculated in an independent cohort of 100 SLE patients (PSY-LUP cohort). This showed a similar pattern of values to the LUPUCE cohort ( Fig. 1): the score ranged from 0 to 33, with a median of 8.3, and 23% of patients were above the threshold of 14. The distribution of patients from the LUPUCE cohort was compared with that from Duke University [11], in which type 2 symptoms were defined by ACR fibromyalgia criteria. A similar distribution was found in the different cohorts ( Table 2).
The characteristics of patients from the LUPUCE cohort at their first visit (n = 50 unique patients) and the characteristics of patient-by-visit by clinical groups are summarized in Table 3. The median time between each visit was 124 days [IQR: . Patients in the "type 2" group were older (52 vs. 31 years, p = 0.016) and more often menopausal (66.7% vs. 17.9%, p = 0.031) than patients in the "minimal" group, and they had a longer disease duration than patients in the "type 1" and "mixed" groups (10 vs. 5 years and 10 vs. 2 years, p = 0.011, respectively). Patients in the "type 2" group tended to have lower daily doses of corticosteroids than "type 1" or "mixed" patients.

Clinical groups and immunological activity
Immunological activity (elevated anti-dsDNA antibodies, anti-dsDNA level, low C3, and/or C4) did not differ between the four clinical groups (Table 3), although "type 1" patients tended to have more often a low C3 level.

Clinical groups and transcriptomic signatures
Modular IFN score, number of activated IFN modules, and level of regulation of each IFN module did not differ between the four clinical groups (Table 4). In the analyses of all modules, the values of two modules (M3.3, cell cycle signature, and M4.14, monocyte signature) were slightly higher in patients in the "mixed" group compared to other groups, but these modules were not significantly activated (Online Resources 1 and 2).

Discussion
This is the first study using items from the SF-36 to assess the ability of a type 2 score to categorize SLE patients according to the presence of symptoms of fatigue, widespread pain, and depression. Using the SLEDAI score to define type 1 symptoms, SLE patients from the LUPUCE cohort were categorized into the four groups defined by Duke University, with a similar patient distribution. The differences between these groups were then explored and apart from clinical differences, no significant immunological or transcriptomic profiles associated with the categories were found.
The initial definition of type 2 symptoms relied on the ACR fibromyalgia criteria [12]. Although these criteria are likely to reflect type 2 symptoms accurately, they are rarely available in clinical trials and in historical SLE cohorts. In contrast, SF-36 is widely used to assess HRQoL in SLE [3], as in many chronic diseases [18], and an item-level analysis has been shown to be useful to interpret HRQoL variations in SLE patients [19]. We chose to select the most relevant items from the SF-36 questionnaire to define type 2 symptoms and build a score that was easy to use in other cohorts. The values of the type 2 score observed in the LUPUCE cohort were very similar to those from an independent cohort of SLE patients, the PSY-LUP cohort. In addition, the proportions of patients categorized into each of the four clinical groups using the SLEDAI and the type 2 score were almost the same in the LUPUCE cohort and in the cohort Before the first description of type 1/type 2 symptoms by Duke University [10], some studies described SLE patients with active/inactive disease and high/low fatigue [7] and found quite similar proportions as with the type 1/type 2 categorization [19]. Interestingly, as in the cohort from Duke University, no significant difference was observed between the immunological profiles of the four clinical groups. This highlights the fact that categorization into these four clinical groups, instead of providing redundant information, is a complimentary tool for the care of SLE patients and for patient stratification in clinical trials. Patients with type 2 symptoms, regardless of clinical activity, are likely to report higher activity levels than that evaluated by physicians, which can result in discordant measures [10] challenging the interpretation of clinical trial results. This is the first study to investigate the association between transcriptional signatures, particularly IFN signature, and clinical type 1/type 2 SLE categorization. We could have expected a higher IFN signature in patients with active SLE (type 1 and mixed groups). However, there was no significant difference in IFN signatures between patients with active SLE and patients with inactive SLE. Alternatively, we could also have expected higher IFN signatures in patients reporting fatigue (type 2 group). Indeed, IFN signature was elevated in patients with fibromyalgia [20,21], and therapeutic IFN is associated with neuropsychiatric side effects such as fatigue and depression [22]. In contrast, we have previously shown a paradoxical association between HRQoL and IFN signatures, with patients with higher IFN scores displaying better HRQoL in some SF-36 domains (mental health and vitality) [14]. However here, although a majority of patients from the type 2 group had a moderate IFN score, no significant difference was observed with the other groups. This is consistent with the report of an absence of correlation between pro-inflammatory cytokines (including IFN) and HRQoL in SLE patients [23]. The same has been observed in patients with primary Sjögren syndrome in whom there was no correlation between IFN signature and HRQoL/fatigue [24]. Thus, the IFN signature, as an immunological marker, is not relevant to discriminate type 1 from type 2 profiles.
Our study has some limitations. First, we used items from the SF-36 questionnaire to assess fatigue and depression, and not more specific questionnaires dedicated to SLE patients (such as SLEQOL, LIT, L-QOL, LupusQOL, LupusPRO) [25]. However, the SF-36 has been used for a long time and more widely than specific scales and it is well validated in SLE [26] and widely available, even retrospectively, in SLE cohorts. Second, the limited number of patients could lead to a loss of power after the division of the cohort into the four clinical groups. New studies with larger populations are therefore needed to definitely validate our results. Cohorts such as the LUPUCE cohort, providing well-characterized clinical, immunological, transcriptomic, and HRQoL data, are very much needed. Third, using Physician Global Assessment (PGA) as an indicator to compare patients according to their  profile could have been interesting. Unfortunately, in the LUPUCE cohort, PGA data was not collected. A score derived from the SF-36 can be used to identify SLE patients with type 2 symptoms, and, with the use of the SLEDAI, to categorize patients into four clinical groups. This categorization, which is not related to immunological activity or blood transcriptomic profiles, could be useful in the daily care of patients, as well as in clinical trials, for the upstream stratification of patients and for the interpretation of results.