Systemic lupus erythematosus (SLE) is a chronic autoimmune disease that affects multiple organ systems with fluctuating disease activity and severity (1,2). The pathogenesis of SLE involves dysregulation of various aspects of the immune system, including elements of the adaptive (eg, T and B cells) and innate (eg, dendritic cells [DCs], neutrophils, and inflammatory cytokines) immune systems (3). Standard-of-care treatments for SLE, such as combinations of glucocorticoids, antimalarials, and immunosuppressants, can be effective in some patients, but their use is associated with adverse events (1,4).
Many novel-targeted agents have been evaluated for SLE, but belimumab and anifrolumab, which respectively target B-lymphocyte stimulator (BLyS) and type I interferon (IFN) receptor, are the only biological drugs approved by the US Food and Drug Administration for SLE to date (5,6). The limited success of new agents that have been investigated for SLE, as well as the relatively low proportion of patients meeting clinical trial endpoints, may reflect inherent heterogeneity in pathophysiology among patients with SLE (7). Consequently, there is significant need for precision medicine approaches that leverage novel biomarkers to select patients who are most likely to respond to targeted therapies based on their molecular profile.
BLyS and a proliferation-inducing ligand (APRIL) are vital cytokines for B cell homeostasis and function with key roles in the pathogenesis of diseases with autoreactive B cell involvement, such as SLE (8–10). Plasma BLyS protein levels correlate with disease activity in SLE, and elevated APRIL levels have been detected in the sera of patients with SLE (9,11–13). Thus, both cytokines are candidate therapeutic targets for SLE. As a key component of their overlapping functions, they share the ability to bind to the tumor necrosis factor receptor superfamily member and transmembrane activating factor and cyclophilin ligand interactor (TACI).
Atacicept, a human recombinant fusion protein of the extracellular domain of TACI and the fragment crystallizable portion of human IgG1, is an antagonist for all known conformations of BLyS and APRIL (14). Atacicept interferes with B cell maturation, differentiation, and function (15,16), leading to a diminution in B cell levels in patients with SLE (14,17). The ability to target both BLyS and APRIL differentiates atacicept from BLyS-targeted therapies (18). Because APRIL affects later stage B cell development and plasma cell antibody production, blocking APRIL may have additional benefits over blockade of BLyS alone for autoantibody-associated diseases, such as SLE. Atacicept had evidence of efficacy in patients with SLE in the phase II/III APRIL-SLE and phase IIb ADDRESS II trials, particularly for patients with high disease activity (HDA, defined as SLE Disease Activity Index 2000 [SLEDAI-2K] ≥10) at screening (17,19,20). However, a relatively high proportion of patients who received placebo achieved SLE Responder Index (SRI) responses, a problem seen in many SLE trials (5,21,22).
In these post hoc analyses of the atacicept studies in SLE, we aimed to address the biological heterogeneity of patients with SLE and determine whether clusters of patients with a greater treatment effect from atacicept could be identified. Five clusters were identified using immune cell deconvolution of gene expression data from APRIL-SLE. The gene expression profile, clinical characteristics, and response to atacicept of these patient clusters were characterized, and two subsets (superclusters) were defined and validated using an independent dataset from ADDRESS II.
MATERIALS AND METHODSAPRIL-SLE (NCT00624338) and ADDRESS II (NCT01972568) study designs were described previously (17,19). All patients enrolled in APRIL-SLE and ADDRESS II had a diagnosis of SLE satisfying at least 4 of the 11 American College of Rheumatology revised classification criteria (23) and a disease duration of 6 months or more. Because these trials had different designs and outcome measures, with primary endpoints being flare in APRIL-SLE and improvement in ADDRESS II, the expectation in the current study was that a biomarker-defined subgroup demonstrating fewer flares in the APRIL-SLE trial would demonstrate more improvement in the validation set.
APRIL-SLE was a 52-week, randomized, double-blind, placebo-controlled trial of atacicept in 461 patients with active SLE, defined as at least one British Isles Lupus Assessment Group (BILAG) A or B (17). To better evaluate the effect of atacicept on the development of new flares, patients received corticosteroids for 2 weeks with a dose reduction at the start of week 3. Patients achieving BILAG C or D scores in all systems at week 10 without any new A or B scores by week 12 while on 7.5 mg of daily prednisone for weeks 11 and 12 (n = 461) were included in the trial. Patients were randomized (1:1:1) to subcutaneous, twice-weekly placebo or atacicept 75 mg or 150 mg for the first 4 weeks of treatment and once weekly thereafter. The primary objective was to evaluate atacicept efficacy versus placebo in preventing new flares (BILAG A or B score due to items that were new or worse) in patients with SLE during the 52-week treatment period. Secondary objectives included the effect of atacicept on markers related to its mechanism of action and their correlation to disease activity and the identification of genetic variations associated with drug response.
The trial protocol for APRIL-SLE, and all substantial amendments, was approved by the relevant institutional review boards or independent ethics committees and by health authorities, according to country-specific laws. The APRIL-SLE trial was conducted in accordance with the protocol, the International Conference on Harmonisation (ICH) guideline for Good Clinical Practice, and applicable local regulations, as well as with the Declaration of Helsinki. All patients gave informed consent.
ADDRESS II was a 24-week, randomized, double-blind, placebo-controlled trial that included 306 patients with active, autoantibody-positive SLE (19). Patients were randomized (1:1:1) to receive once weekly subcutaneous injections of placebo, atacicept 75 mg, or atacicept 150 mg. The primary endpoint was SRI-4 response at week 24. Secondary endpoints included SRI-6 and BILAG-Based Combined Lupus Assessment (BICLA) responses at week 24. Responses were also analyzed for a subpopulation of patients with HDA at screening and a post hoc analysis on attainment of low disease activity (LDA, SLEDAI-2K ≤2) in the HDA subpopulation was performed (20).
The ADDRESS II study was performed in accordance with the Declaration of Helsinki, the ICH Note for Guidance on Good Clinical Practice (ICH Topic E6, 1996), and applicable regulatory requirements. All study sites received approval for the study from their local ethics board, and all patients gave informed consent.
See Supplementary Methods for information on the assessment of biomarkers (APRIL, BLyS, C3, C4, antinuclear antibodies and anti–double-stranded DNA [anti-dsDNA] antibodies) and interferon-gene signature (IFN-GS).
Immune cell deconvolution of gene expression data inGene expression data were derived from mRNA extracted from whole blood of patients with APRIL-SLE at baseline and analyzed by microarray (see Supplementary Methods for details).
A published cell deconvolution algorithm (24) as implemented in the CellMix R package (25) was applied to baseline gene expression data from APRIL-SLE to determine the relative proportions of 17 immune cell subset signatures. Immune cell subsets that had zero scores for all samples were filtered out, and DCs were removed for comparability with a similar study (26), resulting in the following nine immune cell subset signatures used in further analysis: neutrophils, monocytes, activated DCs, natural killer (NK) cells, activated NK cells, total B cells, plasma cells, T helper cells, and cytotoxic T cells. Based on baseline immune cell subset normalized enrichment scores, patients were grouped into clusters using a k-medoid clustering algorithm (27), in which the number of clusters, k, was chosen mainly based on the stability metric using ConsensusClusterPlus (28). Five clusters were identified (P1,2,3,4,5) and were described in terms of a dominant cell subset signature.
Flare rates and time to flare—BILAG A or B, as defined by the BILAG index (29)—were analyzed for each patient cluster, and two superclusters (P1,3 and P2,4,5) were defined based on responses. P1,3 included patients with high neutrophils, T helper cells and NK cells (P1), or high B cells and neutrophils (P3). P2,4,5 included patients with high plasma cells and activated NK cells (P2), high B cells and low neutrophils (P4), or high activated DCs, activated NK cells, and neutrophils (P5).
A simple classifier to predict supercluster membership based on the percentiles of cell subset signatures in the deconvolution scores was created so that superclusters P1,3 and P2,4,5 could be validated in another dataset. The percentile cutoffs were selected manually to match those that seemed to best define the clusters in the APRIL-SLE results. In the simple classifier, the criteria for inclusion in P2,4,5 were as follows: 1) plasma cell signature greater than the 93rd percentile, or 2) B cell signature greater than the 56th percentile and neutrophil signature less than the 31st percentile, or 3) activated NK cell signature greater than the 64th percentile and activated DC signature greater than the 68th percentile. Patients who did not fulfil these criteria were assigned to P1,3.
Cell deconvolution analysis forBased on the results from APRIL-SLE, a validation data set was generated from ADDRESS II by prospectively applying RNA sequencing to all available samples for which patients had given appropriate consent. In order to apply the classifying algorithm based on microarray data to the available RNA sequencing gene expression data, the probes in the APRIL-SLE microarray were mapped to Ensembl IDs, and the algorithm was modified to use an approach that combined transcripts into genes. Using the cellular signature modules that defined the patient clusters in APRIL-SLE, the simple classifier was applied to split ADDRESS II patients into analogous superclusters P1,3 and P2,4,5 using baseline RNA sequencing gene expression data.
Statistical analysesA subset of patients in the modified intent-to-treat (mITT) populations (defined as all randomized patients who received at least one dose of study medication) of APRIL-SLE and ADDRESS II was included in the immune cell deconvolution analysis. For APRIL-SLE, patients with gene expression and biomarker (C3, C4, dsDNA) data available at baseline who had been randomized at least 52 weeks before termination of the atacicept 150 mg arm, were included. For ADDRESS II, patients in the mITT population and HDA subpopulation (SLEDAI-2K score ≥10 at screening) with RNA sequencing data at baseline were included.
Demographic and baseline disease characteristics of different patient clusters were summarized descriptively and were not compared formally. Efficacy data were summarized based on data collected from each trial using descriptive statistics. For APRIL-SLE, incidence and time to BILAG A or B flare were compared between treatment groups using hazard ratios (HRs) and corresponding 95% confidence intervals (CIs), which were estimated using a Cox proportional hazards model. For ADDRESS II, the proportion of each group responding to atacicept was compared between the superclusters using SRI-4, SRI-6, BICLA, and LDA (SLEDAI-2K ≤ 2). Patients with missing clinical response data were imputed as nonresponders.
RESULTS Study populations and baseline disease characteristicsThis study included 105 of 461 (23%) patients from APRIL-SLE and 179 of 306 (59%) patients from ADDRESS II. The mean age of these patients was similar (APRIL-SLE, 39.0 years; ADDRESS II, 39.7 years), and the majority were female (93% and 94%) and White (71% and 70%) (Table 1). The mean (SD) duration of SLE disease was 6.0 (5.9) years in APRIL-SLE and 6.8 (7.2) years in ADDRESS II. Mean (SD) disease activity, as measured by Safety of Estrogens in Lupus Erythematosus National Assessment (SELENA)-SLEDAI in APRIL-SLE and SLEDAI-2K in ADDRESS II, was 11.4 (6.9) in APRIL-SLE and 9.7 (3.1) in ADDRESS II. The respective proportion of patients with BILAG A (severe) organ activity was 20.0% and 24.6%. The populations included were representative of the entire cohorts (Table 1).
Table 1 Patient demographics and clinical characteristics at screening for APRIL-SLE and ADDRESS II per superclusters P1,3 and P2,4,5
| APRIL-SLE | ADDRESS II | |||||||
| All n = 105a | P1,3 n = 55 | P2,4,5 n = 50 | Total N = 455a | All n = 179a | P1,3 n = 93 | P2,4,5 n = 86 | Total N = 306a | |
| Female, n (%) | 98 (93.3) | 53 (96.4) | 45 (90.0) | 424 (93.2) | 168 (93.9) | 90 (96.8) | 78 (90.7) | 280 (91.5) |
| Race, n (%) | ||||||||
| White | 75 (71.4) | 47 (85.5) | 28 (56.0) | 325 (71.4) | 125 (69.8) | 76 (81.7) | 49 (57.0) | 216 (70.6) |
| Black/African American | 3 (2.9) | 1 (1.8) | 2 (4.0) | 10 (2.2) | 10 (5.6) | 0 (0.0) | 10 (11.6) | 20 (6.5) |
| Asian | 19 (18.1) | 5 (9.1) | 14 (28.0) | 89 (19.6) | 27 (15.1) | 9 (9.7) | 18 (20.9) | 36 (11.8) |
| American Indian or Alaska Nativeb | NC | NC | NC | NC | 7 (3.9) | 4 (4.3) | 3 (3.5) | 11 (3.6) |
| Other | 8 (7.6) | 2 (3.6) | 6 (12.0) | 31 (6.8) | 10 (5.6) | 4 (4.3) | 6 (7.0) | 23 (7.5) |
| Hispanic or Latino, n (%) | 8 (7.6) | 4 (7.3) | 4 (8.0) | 75 (16.5) | 93 (52.0) | 50 (53.8) | 43 (50.0) | 153 (50.0) |
| Age, mean ± SD, y | 39.0 ± 11.7 | 41.7 ± 10.7 | 36.0 ± 12.0 | 39.0 ± 12.3 | 39.7 ± 12.4 | 44.4 ± 12.5 | 34.6 ± 10.1 | 38.9 ± 11.9 |
| Disease duration, mean ± SD, y | 6.03 ± 5.93 | 6.97 ± 6.50 | 5.00 ± 5.10 | 5.84 ± 5.49 | 6.76 ± 7.22 | 7.96 ± 7.85 | 5.45 ± 6.25 | 6.83 ± 7.14 |
| Disease activity | ||||||||
| SLEDAI, mean ± SDc | 11.4 ± 6.9 | 10.5 ± 6.2 | 12.5 ± 7.6 | 10.6 ± 5.9 | 9.7 ± 3.1 | 9.0 ± 2.8 | 10.5 ± 3.1 | 10.0 ± 3.1 |
| SLEDAI ≥10, n (%)d | 53 (50.5) | 24 (43.6) | 29 (58.0) | 229 (50.3) | 93 (52.0) | 35 (37.6) | 58 (67.4) | 157 (51.3) |
| BILAG A, n (%)d | 21 (20.0) | 9 (16.4) | 12 (24.0) | 89 (19.6) | 44 (24.6) | 22 (23.7) | 22 (25.6) | 68 (22.2) |
| BILAG 2B, n (%)d | 61 (58.1) | 30 (54.5) | 31 (62.0) | 252 (55.4) | 61 (34.1) | 36 (38.7) | 25 (29.1) | 121 (39.5) |
| BILAG B, n (%)d | 23 (21.9) | 16 (29.1) | 7 (14.0) | 114 (25.1) | 53 (29.6) | 23 (24.7) | 30 (34.9) | 92 (30.1) |
| Laboratory assessments, n (%) | ||||||||
| ANA titer ≥1:80 | 94 (89.5) | 47 (85.5) | 47 (94.0) | 411 (90.3) | 173 (96.6) | 91 (97.8) | 82 (95.3) | 293 (95.8) |
| Anti-dsDNA titer ≥15 IU/ml | 85 (81.0) | 41 (74.5) | 44 (88.0) | 328 (72.1) | 82 (45.8) | 25 (26.9) | 57 (66.3) | 147 (48.0) |
| ANA titer ≥1:80 and/or anti-dsDNA titer ≥15 IU/ml | 104 (99.0) | 54 (98.2) | 50 (100.0) | 447 (98.2) | 179 (100.0) | 93 (100.0) | 86 (100.0) | 305 (99.7) |
| C3 <0.9 g/l | 44 (41.9) | 19 (34.5) | 25 (50.0) | 180 (39.6) | 56 (31.3) | 12 (12.9) | 44 (51.2) | 101 (33.0) |
| C3 <0.9 g/l (day 1) | 37 (35.2) | 18 (32.7) | 19 (38.0) | 165 (36.3) | 52 (29.1) | 10 (10.8) | 42 (48.8) | 99 (32.6) |
| C4 <0.1 g/l | 28 (26.7) | 11 (20.0) | 17 (34.0) | 116 (25.5) | 34 (19.0) | 6 (6.5) | 28 (32.6) | 56 (18.3) |
| C4 <0.1 g/l (day 1) | 24 (22.9) | 12 (21.8) | 12 (24.0) | 91 (20.0) | 35 (19.6) | 6 (6.5) | 29 (33.7) | 56 (18.4) |
| Medication use, n (%) | ||||||||
| Prednisone ≥7.5 mg/day | 86 (81.9) | 47 (85.5) | 39 (78.0) | 355 (78.0) | 105 (58.7) | 53 (57.0) | 52 (60.5) | 181 (59.2) |
| Antimalarial agents | 61 (58.1) | 35 (63.6) | 26 (52.0) | 260 (57.1) | 137 (76.5) | 72 (77.4) | 65 (75.6) | 233 (76.1) |
| Immunosuppressive agents | 35 (33.3) | 19 (34.5) | 16 (32.0) | 128 (28.1) | 82 (45.8) | 44 (47.3) | 38 (4.2) | 156 (51.0) |
| Methotrexate | 16 (15.2) | 9 (16.4) | 7 (14.0) | 62 (13.6) | 21 (11.7) | 11 (11.8) | 10 (11.6) | 43 (14.1) |
| Azathioprine | 19 (18.1) | 10 (18.2) | 9 (18.0) | 70 (15.4) | 33 (18.4) | 18 (19.4) | 15 (17.4) | 61 (19.9) |
| Cyclosporine | 2 (1.9) | 1 (1.8) | 1 (2.0) | 2 (0.4) | NA | NA | NA | NA |
| Leflunomide | NA | NA | NA | NA | 1 (0.6) | 1 (1.1) | 0 (0.0) | 2 (0.7) |
| MMS/MPF | NA | NA | NA | NA | 28 (15.6) | 15 (16.1) | 13 (15.1) | 51 (16.7) |
Abbreviations: ANA, antinuclear antibodies; APRIL, a proliferation-inducing ligand; BILAG, British Isles Lupus Assessment Group; dsDNA, double-stranded DNA; MMS/MPF, mycophenolate mofetil, mycophenolate sodium, and mycophenolic acid; NA, not available; NC, not collected; P1-P5, patient cluster; SELENA, Safety of Estrogens in Lupus Erythematosus National Assessment; SLE, systemic lupus erythematosus; SLEDAI, Systematic Lupus Erythematosus Disease Activity Index; SLEDAI-2K, Systemic Lupus Erythematosus Disease Activity Index 2000.
a“All” refers to all patients in the gene deconvolution analysis. “Total” refers to all patients in the modified intent-to-treat populations of APRIL-SLE or ADDRESS II.
bAmerican Indian or Alaska Native was not collected as a race in APRIL-SLE.
cMeasured by SELENA-SLEDAI in APRIL-SLE and SLEDAI-2K in ADDRESS II.
dBILAG instrument was used in APRIL-SLE, and BILAG 2004 instrument was used in ADDRESS II.
Of the 158 patients from ADDRESS II with HDA at screening, 93 (59%) were included in this study. The mean (SD) duration of SLE disease of 6.8 (7.4) years was similar to the mITT population (Supplementary Table 1). Mean SLEDAI-2K for the HDA subpopulation was 11.9 (2.6), and 31.2% had at least one organ scored as BILAG A.
Identification and characterization of patient clusters inPatients in APRIL-SLE were clustered into five groups (P1,2,3,4,5) using the output of the deconvolution algorithm (Figure 1A). A total of 37 (35.2%) patients were included in cluster P1, 12 (11.4%) in P2, 18 (17.1%) in P3, 16 (15.2%) in P4, and 22 (21.0%) in P5. The distribution of cell type signature scores acrossthe five patient clusters suggested dominant cell types (Figure 1A and 1B): P1was distinguished from other clusters by high neutrophil, T helper cell and highNK cell signatures, P2 by expression patterns dominated by plasma cells andactivated NK cells, P3 by B cells and neutrophils, P4 by B cells but lowneutrophil signatures, and P5 by signatures indicating activated DCs, activatedNK cells and neutrophils.
Enlarge this image.Figure 1. Identification and characterization of patient clusters in APRIL-SLE by immune cell deconvolution. A, The heatmap shows relative immune cell profiles in the blood of 105 patients (columns) at baseline. The order of the patients in the heatmap is from clustering, and the panel below the heatmap shows which cluster each patient was assigned (P1: n = 37, P2: n = 12, P3: n = 18, P4: n = 16, P5: n = 22). Values are scaled within each cell type. B, Box plots show the distribution of the relative immune cell profiles per patient cluster. NK, natural killer; P1-P5, patient cluster.
Patient demographics and baseline disease characteristics were examined by cluster (Supplementary Table 2). Disease activity, measured by SELENA-SLEDAI, ranged from 9.9 (P3) to 13.8 (P5). Baseline BLyS and APRIL serum levels were assessed per patient cluster (Supplementary Table 3). The proportion of patients with high APRIL, greater than or equal to 2.2 ng/ml (30), was comparable across clusters (49%-64%). Between 44% and 92% of patients across clusters had a baseline BLyS of greater than or equal to 1.6 ng/ml (30); the proportions were highest in P2,5. The majority (84%) of patients in the analysis had high IFN-GS scores (Supplementary Table 3). All patients in clusters P2,5 had high IFN-GS.
Efficacy of atacicept inFlare rates and time to flare in the 52-week APRIL-SLE trial were assessed per treatment group and cluster. The proportion of placebo-treated patients with BILAG A/B flares was lower in clusters P1 (33%) and P3 (29%) than the other three clusters (P2, 100%; P4, 100%; P5, 83%). The proportion of patients in the atacicept 150 mg group with flares was also lower in P1 and P3 (9% and 13%) than in P2, P4, and P5 (33%, 20%, and 38%). Corresponding HRs with 95% CIs are shown in Table 2, and time to flare is in Figure 2. Atacicept was associated with fewer flares and longer time to flare relative to placebo in clusters P2,4,5 but not in clusters P1,3. These results led us to create the following two superclusters based on clinical response: P1,3 and P2,4,5.
Table 2 Risk of flare and time to BILAG A or B flare in the atacicept 75 mg and atacicept 150 mg groups compared with placebo at week 52 for clusters P1-P5 and superclusters P1,3 and P2,4,5 (APRIL-SLE)
| Number of patients | n (%) with a flare | Hazard ratio | 95% CI | P value | |
| P1 | |||||
| Placebo | 9 | 3 (33.3) | |||
| Atacicept 75 mg | 17 | 8 (47.1) | 1.37 | 0.36-5.18 | 0.64 |
| Atacicept 150 mg | 11 | 1 (9.1) | 0.23 | 0.02-2.21 | 0.20 |
| P2 | |||||
| Placebo | 2 | 2 (100.0) | |||
| Atacicept 75 mg | 7 | 3 (42.9) | 0.23 | 0.03-1.76 | 0.16 |
| Atacicept 150 mg | 3 | 1 (33.3) | 0.13 | 0.01-1.78 | 0.13 |
| P3 | |||||
| Placebo | 7 | 2 (28.6) | |||
| Atacicept 75 mg | 3 | 0 (0.0) | 2.44 × 10−9 | 0, NE | 1.00 |
| Atacicept 150 mg | 8 | 1 (12.5) | 0.41 | 0.04-4.55 | 0.47 |
| P4 | |||||
| Placebo | 6 | 6 (100.0) | |||
| Atacicept 75 mg | 5 | 4 (80.0) | 0.16 | 0.03-0.85 | 0.03 |
| Atacicept 150 mg | 5 | 1 (20.0) | 0.03 | 0.00-0.31 | 0.004 |
| P5 | |||||
| Placebo | 6 | 5 (83.3) | |||
| Atacicept 75 mg | 8 | 4 (50.0) | 0.39 | 0.10-1.49 | 0.17 |
| Atacicept 150 mg | 8 | 3 (37.5) | 0.36 | 0.08-1.53 | 0.17 |
| Supercluster P1,3 | |||||
| Placebo | 16 | 5 (31.3) | |||
| Atacicept 75 mg | 20 | 8 (40.0) | 1.27 | 0.41-3.87 | 0.68 |
| Atacicept 150 mg | 19 | 2 (10.5) | 0.29 | 0.06-1.52 | 0.14 |
| Supercluster P2,4,5 | |||||
| Placebo | 14 | 13 (92.9) | |||
| Atacicept 75 mg | 20 | 11 (55.0) | 0.30 | 0.13-0.69 | 0.005 |
| Atacicept 150 mg | 16 | 5 (31.3) | 0.15 | 0.05-0.44 | 0.001 |
Note: 95% CIs and P values were calculated using a Cox proportional hazards model.
Abbreviations: BILAG, British Isles Lupus Assessment Group; CI, confidence interval; NE, not estimated; P1-P5, patient cluster.
Enlarge this image.Figure 2. Flare rates and time to BILAG A or B flare by patient clusters according to treatment groups in APRIL-SLE. The lines represent the percentage of patients with a flare at each time point, and the shaded areas represent confidence intervals. The numbers to the right of the graphs are the number of patients in each treatment group. BILAG, British Isles Lupus Assessment Group; P1-P5, patient cluster.
The following two superclusters of gene expression patterns, manually defined based on clinical response, were used in follow-up analyses as a candidate biomarker: P1,3, which included 55 patients (52%), and P2,4,5, which included 50 patients (48%). Patient characteristics for superclusters P1,3 and P2,4,5 are provided in Table 1 and described in the Supplementary Results.
Flare rates and time to flare with atacicept versus placebo were calculated for the two superclusters (Table 2, Figure 2). In the P2,4,5 supercluster, flare rates were significantly reduced with both atacicept 75 mg (55%, HR: 0.30; 95% CI: 0.13-0.69; P = 0.005) and atacicept 150 mg (31.3%, HR: 0.15; 95% CI: 0.05-0.44; P = 0.001) compared with the placebo (92.9%). Flare rates for the P1,3 supercluster were not different from placebo (31.3%) with either atacicept 75 mg (40.0%, HR: 1.27; 95% CI: 0.41-3.87; P = 0.68) or 150 mg (10.5%, HR: 0.29; 95% CI: 0.06-1.52; P = 0.14). This suggested that the P2,4,5 supercluster might be an atacicept responsive group in APRIL-SLE.
Validation of the patient clustering algorithm usingThe simple classifier assigned 93 patients (52%) from ADDRESS II to the P1,3 supercluster and 86 (48%) to the P2,4,5 supercluster, demonstrating a similar distribution of patients in the superclusters as that obtained in APRIL-SLE (Table 1). For patients in ADDRESS II with baseline HDA, 35 (38%) were assigned to P1,3 and 58 (62%) to P2,4,5 (Supplementary Table 1).
Baseline disease characteristics in ADDRESS II patient superclusters P1,3 and P2,4,5 are shown in Table 1 and Supplementary Table 1 and are described in the Supplementary Results. More patients in P2,4,5 than P1,3 had high IFN-GS scores and elevated BLyS levels at baseline of ADDRESS II, as was seen in APRIL-SLE (Supplementary Tables 3 and 4).
Response with ataciceptThe percentage of patients with clinical response to atacicept versus placebo was calculated for each supercluster (Table 3). For the full biomarker population, lower proportions of placebo-treated patients in P2,4,5 than those in P1,3 had SRI-4 (39% and 53%), SRI-6 (21% and 44%), and BICLA (32% and 44%) responses at week 24. Accordingly, larger treatment effect sizes in terms of SRI-4, SRI-6, and BICLA were observed for patients on atacicept 75 mg and 150 mg versus placebo in the P2,4,5 group than the P1,3 group. For example, SRI-6 responses with atacicept 75 mg relative to placebo were +10.7% for the P2,4,5 group and –14.2% for the P1,3 group.
Table 3 SRI-4, SRI-6, BICLA, and LDA response at week 24 for the P1,3 and P2,4,5 superclusters in the total biomarker population and HDA subpopulation in ADDRESS II
| Total biomarker population,a % response (N) [difference vs. placebo] | HDA subpopulation,b % response (N) [difference vs. placebo] | |||||
| Placebo | Atacicept 75 mg | Atacicept 150 mg | Placebo | Atacicept 75 mg | Atacicept 150 mg | |
| SRI-4 | ||||||
| All (P1,3 and P2,4,5) | 46.7 (60) | 54.5 (55) [+7.8%]c | 48.4 (64) [+1.7%]c | 46.7 (30) | 54.8 (31) [+8.1%]c | 59.4 (32) [+12.7%]c |
| P1,3 | 53.1 (32) | 55.6 (27) [+2.5%]c | 52.9 (34) [–0.2%]c | 63.6 (11) | 54.5 (11) [–9.1%]c | 61.5 (13) [–2.1%]c |
| P2,4,5 | 39.3 (28) | 53.6 (28) [+14.3%]c | 43.3 (30) [+4.0%]c | 36.8 (19) | 55.0 (20) [+18.2%]c | 57.9 (19) [+21.1%]c |
| SRI-6 | ||||||
| All (P1,3 and P2,4,5) | 33.3 (60) | 30.9 (55) [–2.4%]c | 32.8 (64) [–0.5%]c | 30.0 (30) | 41.9 (31) [+11.9%]c | 53.1 (32) [+23.1%]c |
| P1,3 | 43.8 (32) | 29.6 (27) [–14.2%]c | 35.3 (34) [–8.5%]c | 54.5 (11) | 45.5 (11) [–9.0%]c | 61.5 (13) [+7.0%]c |
| P2,4,5 | 21.4 (28) | 32.1 (28) [+10.7%]c | 30.0 (30) [+8.6%]c | 15.8 (19) | 40.0 (20) [+24.2%]c | 47.4 (19) [+31.6%]c |
| BICLA | ||||||
| All (P1,3 and P2,4,5) | 38.3 (60) | 43.6 (55) [+5.3%]c | 42.2 (64) [+3.9%]c | 20.0 (30) | 48.4 (31) [+28.4%]c | 53.1 (32) [+33.1%]c |
| P1,3 | 43.8 (32) | 33.3 (27) [–10.5%]c | 38.2 (34) [–5.6%]c | 27.3 (11) | 27.3 (11) [0.0%]c | 46.2 (13) [+18.9%]c |
| P2,4,5 | 32.1 (28) | 53.6 (28) [+21.5%]c | 46.7 (30) [+14.6%]c | 15.8 (19) | 60.0 (20) [+44.2%]c | 57.9 (19) [+42.1%]c |
| LDA | ||||||
| All (P1,3 and P2,4,5) | 26.7 (60) | 25.5 (55) [–1.2%]c | 29.7 (64) [+3.0%]c | 10.0 (30) | 19.4 (31) [+9.4%]c | 31.3 (32) [+21.3%]c |
| P1,3 | 34·4 (32) | 37.0 (27) [+2.6%]c | 35.3 (34) [+0.9%]c | 18.2 (11) | 36.4 (11) [+18.2%]c | 30.8 (13) [+12.6%]c |
| P2,4,5 | 17.9 (28) | 14.3 (28) [–3.6%]c | 23.3 (30) [+5.4%]c | 5.3 (19) | 10.0 (20) [+4.7%]c | 31.6 (19) [+26.3%]c |
Note: N = the total number of patients in each group.
Abbreviations: BICLA, British Isles Lupus Assessment Group–Based Combined Lupus Assessment; HDA, high disease activity; LDA, low disease activity; P1-P5, patient cluster; SRI, Systemic Lupus Erythematosus Responder Index.
aThe total biomarker population included all randomized patients who received at least one dose of study medication and had baseline RNA sequencing data available.
bThe HDA subpopulation included those from the total biomarker population with SLEDAI-2K scores ≥10 at screening.
cThe difference in percentage response of the treated group versus the placebo group.
Similar results were observed in the HDA subpopulation (Table 3); patients in the P2,4,5 supercluster treated with atacicept showed greater treatment differences than those in the P1,3 group. For example, when evaluating the atacicept 150 mg dose versus placebo in the HDA subpopulation, SRI-4 response differences were +21.1% in P2,4,5 and –2.1% in P1,3. Treatment differences within the P2,4,5 supercluster were greater for the HDA subpopulation compared with the whole population as assessed by SRI-4, SRI-6, and BICLA.
For LDA, no consistent trends in placebo responders or treatment effect with atacicept were observed between superclusters P2,4,5 and P1,3 in either the full biomarker population or HDA subpopulation (Table 3).
DISCUSSIONDual blockade of APRIL and BLyS differentiates atacicept from other BLyS-targeted therapies. Atacicept affects multiple stages of B cell development and might have additional benefits for some patients with SLE (31). Although evidence of efficacy of atacicept was observed in two SLE trials (17,19,20), it was unclear whether patients with SLE subpopulations who would be most likely to respond could be defined. This exploratory post hoc analysis examined whether molecularly determined groups of patients with SLE varied in their response to atacicept. Five immunophenotypic patient clusters were identified that were further grouped into two superclusters based on efficacy trends observed in APRIL-SLE. The utility of these superclusters was then tested using data from ADDRESS II, an independent, nonoverlapping trial of atacicept in SLE.
The comparison of data from two trials is complicated by their different study designs. APRIL-SLE included patients with lower disease activity at the time of randomization to enable the assessment of flares, and ADDRESS II enrolled patients with currently active SLE and assessed efficacy by SRI-4 and other improvement measurements. In addition, gene expression in blood was assessed by microarray analysis in APRIL-SLE and RNA sequencing in ADDRESS II. In order to address these challenges, the immune cell deconvolution algorithm was modified and, based on the cellular modules that defined the patient clusters in APRIL-SLE, ADDRESS II patients were sorted into analogous superclusters P1,3 and P2,4,5 to examine treatment effects in these superclusters. Despite the different platforms used, the observed results were resonant between the studies, providing increased confidence in the conclusions.
The two superclusters of ADDRESS II patients were derived to mimic the molecularly defined superclusters from APRIL-SLE. The supercluster P2,4,5 included patients with high plasma cells and activated NK cells (P2), high B cells and low neutrophils (P4), or high activated DCs, activated NK cells, and neutrophils (P5). The supercluster P1,3 included patients who did not fulfill these criteria. Thus, the ADDRESS II superclusters were analogous to the two superclusters that were defined based on clinical response in a different trial.
Strengths of this study included the fact that the patients in the two trials were similar in terms of demographics, and the clinical disease activity (SLEDAI) was similar between the superclusters at baseline. Larger differences between placebo- and atacicept-treated patients were observed in the P2,4,5 supercluster than in the P1,3 supercluster in both studies across measures, including the finding that patients in P2,4,5 had fewer flares and greater proportions met the improvement criteria of SRI-4, SRI-6, and BICLA. Formal confirmation that patients with molecular signatures corresponding to the P2,4,5 supercluster are more likely to respond to atacicept would require a prospective study, but the current findings are among the first to demonstrate differences in treatment response based on molecular signatures in SLE, suggesting the prospect of biomarker-based treatment selection. A major benefit of analyzing immune cells using cell deconvolution of gene expression in addition to standard methods, such as flow cytometry, is that it is possible to retroactively evaluate immune cell subsets, such as activated DCs, activated NK cells, and neutrophils, that are highly relevant in lupus pathogenesis and were not evaluated by flow cytometry during the clinical trials.
The differential treatment effect between superclusters P1,3 and P2,4,5 was driven largely by the placebo response, which was much more striking than differences in the response to atacicept. In both studies, patients in the P2,4,5 supercluster had a much lower response to placebo (plus standard of care) than those in P1,3. Trends in the response to atacicept between patients in P2,4,5 and P1,3 were more complex; whereas patients in P2,4,5 had more flares than patients in P1 and P3 in APRIL-SLE, the differences in ADDRESS II were more nuanced, with patients in P2,4,5 sometimes having a similar or higher response to atacicept compared with P1,3. These results suggest that atacicept may provide substantial improvements over standard of care in P2,4,5 patients.
Other evidence adduced in this study suggests important differences between the superclusters. One of the prominent features of SLE is elevated expression of IFN-stimulated genes (32,33). The majority of patients in this analysis had high IFN-GS at baseline, but those with high IFN-GS were more likely to be in P2,4,5. Alt, the proportion of patients with high levels of anti-dsDNA was comparable between superclusters in APRIL-SLE, more patients in the P2,4,5 supercluster had high anti-dsDNA levels compared with the P1,3 supercluster in both the general population and HDA subpopulation of ADDRESS II. A higher proportion of patients in P2,4,5 than P1,3 had high BLyS levels at baseline in both the APRIL-SLE and ADDRESS II trials, as defined using a threshold previously established to be associated with response to atacicept (30). Conversely, the proportion of patients with high baseline APRIL levels was comparable between the superclusters in APRIL-SLE, and more patients in the P1,3 versus P2,4,5 had high APRIL levels in ADDRESS II. Although this observation is too preliminary to support any conclusions, it underscores the likely independence in the regulation of APRIL versus BLyS expression.
Our results suggest that patients in P2,4,5 may have higher levels of immune activation than those in P1,3, as evidenced by the greater abundance of pathological cell types, such as plasma cells, activated DCs and activated NK cells, and the fact that more P2,4,5 patients had high IFN-GS scores. This activated immune state may contribute to the lower response to standard of care in P2,4,5 versus P1,3 and suggests that P2,4,5 patients may benefit more from immune-modulating treatments. Further, patients in P2,4,5 may respond better to atacicept than those in P1,3 because of higher levels of activated DCs and plasma cells, which produce and respond to the atacicept targets APRIL and BLyS, respectively. Taken together, patient stratification was potentially based on a combination of 1) high immune activation; 2) presence of the drug target; and 3) presence of the cell type that responds to the drug target.
Immune cell deconvolution analysis of whole blood transcriptomic data has been used previously to identify a subset of patients with SLE more likely to respond to treatment with the CTLA3Ig construct abatacept, based on phase IIb data (26). That responsive patient cluster was characterized by high levels of plasma, activated DCs, activated NK cells and neutrophils, the highest baseline BILAG scores and anti-dsDNA levels, and shortest time to flare.
A limitation of this study was that data used to develop the initial algorithm were available from only 23% of the APRIL-SLE patients, resulting in small numbers in some of the individual clusters. Nevertheless, these patients were representative of the total population of the study. A further limitation was that gene expression was measured using different platforms in the two studies, but this challenge was addressed by developing a modified algorithm based on cell deconvolution data.
In conclusion, this exploratory study used immune cell deconvolution analysis of whole blood gene expression data from two trials to identify subsets of patients with SLE who had lower placebo responses and who were more likely to respond better to atacicept treatment than to placebo. This approach may have future clinical utility for improving treatment selection and optimizing dosing strategies for patients with SLE.
ACKNOWLEDGMENTSThe authors thank the patients involved in the APRIL-SLE and ADDRESS II trials and their families, as well as the study teams for their participation. Atacicept has been out licensed to Vera Therapeutics, South San Francisco, who reviewed the final version of this manuscript.
AUTHOR CONTRIBUTIONSAll authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. All authors 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 conception and designStudham, Samy, Haselmayer, Aydemir, Rolfe, DeMartino, Kao, Townsend.
Acquisition of dataStudham.
Analysis and interpretation of dataStudham, Vazquez-Mateo, Samy, Haselmayer, Aydemir, Rolfe, Merrill, Morand, DeMartino, Kao, Townsend.
ROLE OF THE STUDY SPONSOREMD Serono was involved in the study design, the collection, analysis, and interpretation of data, and the writing of the manuscript. The authors thank Peter Chang (Biostatistics, EMD Serono) for his input and advice on the analysis of the APRIL-SLE data. Medical writing support was provided by Samantha Lommano of Bioscript Group Ltd, Macclesfield, UK, which was funded by the healthcare business of Merck KGaA, Darmstadt, Germany.
© 2023. This work is published under http://creativecommons.org/licenses/by-nc-nd/4.0/ (the “License”). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.
Details
; Morand, Eric F 4
; DeMartino, Julie 1 ; Kao, Amy 1 ; Townsend, Robert 1 1 EMD Serono, Billerica, MA, United States
2 the healthcare business of Merck KGaA, Darmstadt, Germany
3 University of Oklahoma Health Sciences Center, Oklahoma City, OK, United States
4 Monash University School of Clinical Sciences, Clayton, Australia