Sarilumab and adalimumab differential effects on bone remodelling and cardiovascular risk biomarkers, and predictions of treatment outcomes

Efficacy endpoints included the following: proportion of patients achieving ≥ 20/50/70% improvement according to American College of Rheumatology criteria (ACR20/50/70), Clinical Disease Activity Index (CDAI) ≤ 2.8, CDAI ≤ 10, DAS28 using CRP (DAS28-CRP) or DAS28-ESR < 2.6 and DAS28-CRP or DAS28-ESR < 3.2.

PROs evaluated in the study were previously described for the overall intent-to-treat (ITT) population [17] and, evaluated as change from baseline at week 24, included Patient Global Assessment of disease activity visual analogue scale (VAS), Health Assessment Questionnaire-Disability Index (HAQ-DI), pain VAS, Functional Assessment of Chronic Illness Therapy (FACIT)-Fatigue, morning stiffness VAS, rheumatoid arthritis impact of disease (RAID) score and Medical Outcomes Study Short-Form (36-item) Health Survey (SF-36) physical (PCS) and mental (MCS) component summary scores, which include the physical functioning, role-physical, bodily pain, general health, vitality, social functioning, role-emotional and mental health domains.

Patients were selected for this biomarker analysis if they had been randomized and treated with sarilumab or adalimumab during the double-blind period and had provided written informed consent for future use of samples, with a serum sample drawn pre-dose (baseline) and evaluable (biomarker population). Serum samples were collected and stored frozen at baseline and post-treatment through week 24 from 307 patients in the ITT population (sarilumab, n = 153; adalimumab, n = 154).

Biomarkers were analysed retrospectively (except CRP) at one or two post-baseline timepoints through week 24 (Table S1). Timepoints selected for analysis were based on either previous data following sarilumab treatment [18, 19] or on literature suggesting either acute or latent effects of RA therapy on specific markers. The assay characteristics for most biomarkers have been described previously [18].

Baseline biomarker levels were compared between treatment groups using a Wilcoxon test. Spearman’s ranked correlations at baseline were computed in the overall biomarker population.

To evaluate pharmacodynamic changes in circulating biomarker concentrations between treatment groups at each timepoint, absolute and percentage changes from baseline were described. In addition, the percentage changes in biomarker concentrations were analysed using non-parametric methods because of non-normal distributions. For biomarkers measured once post-baseline, a rank-based analysis of covariance (ANCOVA) adjusted on baseline value was implemented. For biomarkers measured twice post-baseline, a mixed-effect model with repeated measures was performed on rank-transformed data (analysis of variance [ANOVA]-type method), with treatment, visit and treatment-by-visit interaction as fixed effects, baseline biomarker value transformed in rank, and baseline biomarker value transformed in rank-by-visit interaction as fixed covariates, assuming an unstructured covariance structure. The log-transformed RANKL/OPG ratio was analysed using a mixed model for repeated measures with response, visit and response-by-visit interaction as fixed effects, baseline biomarker value and baseline biomarker value-by-visit interaction as fixed covariates, and assuming an unstructured covariance structure separately by treatment group. p values were adjusted for false discovery rate (Benjamini–Hochberg 5% threshold). The number of patients with abnormal biomarker levels at baseline (according to the reference ranges provided by the testing laboratory) that normalized with treatment was compared between groups using a χ² test; nominal p values are reported.

For binary efficacy endpoints, predictive effects of baseline biomarker values on sarilumab efficacy vs. adalimumab were tested using a logistic regression with treatment group and region as fixed effects, baseline biomarker value as a continuous covariate and the baseline biomarker-by-treatment group interaction. For continuous PROs, a linear regression was used with the same effects as above, as well as the baseline PRO value as a covariate. Nominal values for the interaction are reported to assess the predictive value of the biomarkers. Similar analyses were performed after categorization of patients into high, medium and low biomarker levels at baseline using tertile values in the biomarker population. In addition, pairwise comparisons of responses between sarilumab and adalimumab were performed separately in patients with high, medium and low biomarker levels, and the Mantel–Haenszel estimates of odds ratios (ORs), stratified by region, and 95% confidence intervals (CIs) were derived and graphically represented using forest plots. For continuous PROs, a linear regression was performed separately in each biomarker tertile and differences in least squares mean (LSM) changes with 95% CI between both treatments were provided.

Differential combinations of circulating CXCL13 and sICAM-1 (low or high levels defined relative to baseline median levels) were assessed for prediction of response to sarilumab, using Mantel–Haenszel estimates of ORs derived for each combination.

All analyses were performed using SAS version 9.2 or higher (SAS Institute Inc., Cary, NC, USA).

Baseline demographics and disease characteristics of the biomarker population were generally similar to the overall ITT population (Table 1). Overall, efficacy and PROs were also generally similar between the ITT and biomarker populations (Table S2).

Baseline serum levels of most biomarkers were similar between treatment groups, except for Lp(a), which was higher in the adalimumab than the sarilumab groups (Lp [a]: median 235.5 vs. 179.0 mg/L, respectively; Wilcoxon test p value 0.039; Table S3).

Correlations between individual biomarkers at baseline were generally low or moderate (ρ < 0.5; Fig. 1). Correlation coefficients above 0.7 were observed for markers of inflammation (CRP and SAA; ρ = 0.81), bone formation (P1NP and OC; ρ = 0.82) and anaemia of chronic disease (ferritin and hepcidin; ρ = 0.74), as expected. Moderate correlations were observed between baseline CRP, SAA or MMP-3 with differential blood counts (leucocytes and neutrophils; ρ from 0.4 to 0.5) and, as expected, between iron and haemoglobin (ρ = 0.57; Figure S1).

To compare the effects of sarilumab and adalimumab treatment on biomarkers over time, the absolute (Table S4) and percentage changes from baseline in biomarker concentrations were analysed up to week 24. Greater reductions in biomarkers associated with the acute-phase response were observed at weeks 12 and 24 following treatment with sarilumab vs. adalimumab (adjusted p < 0.0001; Fig. 2a, b). Reductions in CRP were observed as early as week 4 with sarilumab vs. adalimumab and were sustained throughout the treatment period (Fig. 2a for median percentage change in CRP from baseline in biomarker population from week 4; Table S5 for percentage of patients with CRP ≤ 10 mg/L and ≤ 3 mg/L at weeks 12 and 24 [observed cases within the ITT population]).

At week 24, sarilumab treatment increased concentrations of P1NP, a marker of osteoblast activation, compared with adalimumab (adjusted p = 0.027; Fig. 2c). A numeric increase in OC, another marker of osteoblast activity, was also observed in sarilumab- vs. adalimumab-treated patients (Fig. 2d). Furthermore, reductions in total RANKL, a marker of bone remodelling, were observed as early as week 2 with sarilumab compared with adalimumab and persisted through week 24 (adjusted p < 0.0001; Fig. 2e); in addition, a numeric increase in total RANKL was observed after adalimumab treatment. A transient decrease in OPG, a decoy for RANKL, was observed after adalimumab treatment at week 2 but did not persist through week 24 (Fig. 2f). The log of the ratio of RANKL to OPG was also significantly decreased in patients treated with sarilumab vs. adalimumab (data not shown). Subgroup analyses revealed that this effect was significant through week 24 in patients who were not on steroids at baseline (nominal p = 0.0013). Additionally, greater reductions in MMP-3 were observed with sarilumab at week 24 (adjusted p = 0.020; Fig. 2g). Since corticosteroid use impacts bone remodelling, we compared the treatment effects in subgroups based on baseline corticosteroid use. At week 24 the differential effects of sarilumab on MMP-3 concentrations were significant (nominal p = 0.001) only in the subgroup of patients (sarilumab: n = 79, 51.6%; adalimumab: n = 88, 57.1%) who were not on concomitant steroids (data not shown). Reductions in total RANKL were significant in the sarilumab treatment group compared with adalimumab treatment irrespective of steroid use at baseline (nominal p < 0.01). Increases in P1NP, though numerically increased by sarilumab treatment relative to adalimumab, did not reach statistical significance in the steroid subgroups (median percentage change from baseline in P1NP [interquartile range]: with baseline steroid use, 24.6% [− 0.6 to 50.3%] with sarilumab and 7.5% [− 9.9 to 40.3%] with adalimumab [nominal p = 0.0757]; without baseline steroid use, 18.0% [− 3.5 to 47.2%] with sarilumab and 1.8% [− 13.9 to 31.9%] with adalimumab [nominal p = 0.1625]).

The effects of sarilumab and adalimumab on biomarkers associated with markers purported to reflect synovial lymphoid and myeloid cell infiltrates, CXCL13 and sICAM-1, respectively, were also examined. While greater reductions in these biomarkers were observed 2 weeks post-treatment with adalimumab vs. sarilumab, these effects did not persist through week 24 (Figure S2).

We also examined the effects of treatment on parameters associated with anaemia of chronic disease. Previously unpublished data showed that, in the overall safety population (sarilumab, n = 184; adalimumab, n = 185), sarilumab resulted in larger increases in haemoglobin levels vs. adalimumab [20] from baseline (mean 13.0 and 12.9 g/dL, respectively) at week 12 (LSM changes from baseline 0.53 vs. 0.12 g/dL, respectively; LSM difference 0.41 g/dL [95% CI 0.22–0.60; nominal p < 0.001]) and week 24 (LSM changes from baseline 0.59 vs. 0.08 g/dL, respectively; LSM difference 0.52 g/dL [95% CI 0.32–0.71; nominal p < 0.001]). Furthermore, in the overall ITT population, a numerically greater reduction from baseline in the proportion of patients with anaemia was observed with sarilumab vs. adalimumab from baseline: reductions were observed as early as week 2 and persisted through week 24 (Table S6). In this post hoc analysis, reductions in hepcidin and ferritin were observed at week 2 with both sarilumab and adalimumab. In contrast, increases in iron and TIBC were observed with sarilumab relative to adalimumab at week 2 post-treatment (Figure S3). Reductions in the lipid particle Lp(a) were observed with sarilumab vs. adalimumab at week 24 (adjusted p < 0.0001; Fig. 2h).

A subset of patients had abnormal baseline biomarker levels relative to reference ranges. In these patients, normalization of CRP and SAA was evident in a greater percentage treated with sarilumab than adalimumab at week 24 (nominal p < 0.0001). Normalization of total RANKL, OPG and Lp(a) occurred in a numerically greater percentage of patients treated with sarilumab vs. adalimumab at week 24 (Fig. 3).

To establish whether post-treatment changes in biomarker levels at week 24 were associated with clinical efficacy, changes were compared between sarilumab- and adalimumab-treated responders and non-responders. Median percentage changes at week 24 in total RANKL, OPG, P1NP, OC and Lp(a) did not differ greatly between responders and non-responders (data not shown). However, reductions in SAA from baseline at week 24 were greater in adalimumab ACR20 and DAS28-CRP < 3.2 responders than non-responders (− 33.3% vs. 0.0%, respectively; nominal p = 0.0038 and − 39.2% vs. 0.0%, respectively; nominal p = 0.0061, respectively). Greater reductions in MMP-3 were also observed in adalimumab ACR20 responders vs. non-responders (− 23.6% vs. 17.1%, respectively; nominal p < 0.0001). Associations between clinical efficacy and changes from baseline in SAA and MMP-3 were not observed in sarilumab-treated patients, and although both responders and non-responders had a ≥ 90% reduction in CRP, the p values for comparisons of responders vs. non-responders were > 0.05 across several parameters, including ACR20/50, DAS28-CRP < 3.2 and DAS28-CRP < 2.6 (data not shown).

The strongest correlations between baseline biomarkers and baseline disease activity were observed for SAA and CRP with DAS28-ESR (ρ = 0.26 and 0.31, respectively) and for CRP, SAA, MMP-3, hepcidin and CXCL13 with DAS28-CRP (ρ from 0.36 to 0.58). None of the biomarkers correlated with baseline PROs (all ρ < 0.3).

Baseline biomarker levels were analysed as continuous and categorical measures by tertiles (low, medium and high) because thresholds associated with clinical efficacy are not currently established, and treatment-by-biomarker interaction p values were calculated to assess the predictivity of the biomarker. Treatment-by-tertile biomarker interactions for efficacy endpoints at week 24 analysed by baseline biomarker in tertiles are shown in Fig. 4 and Table S7. Patients with the highest baseline concentrations of SAA who received sarilumab were more likely to achieve ACR20/50/70 or DAS28-CRP < 3.2 responses than with adalimumab compared with patients in the low tertile: ACR20 (OR [95% CI] 5.5 [2.1, 14.5]), ACR50 (5.4 [2.2, 13.2]), ACR70 (5.7 [1.8, 18.4]) and DAS28-CRP < 3.2 (6.1 [2.3, 15.7]) (Fig. 4 and Table S7). SAA was consistently predictive compared with high MMP-3 and CRP, which were only predictive of ACR20 and DAS28-CRP < 3.2 response (Table S7). Baseline levels of biomarkers associated with bone remodelling, synovial lymphoid and myeloid cell infiltrates and anaemia of inflammation were not predictive of efficacy at week 24, except for hepcidin and CXCL13, which were associated with ACR20 response.

The ability of baseline biomarker levels to predict PRO responses was also analysed by their respective tertiles and showed that sarilumab-treated patients with higher SAA, MMP-3 and hepcidin levels reported improved PRO responses including HAQ-DI (Fig. 5a) and pain VAS (Fig. 5b) scores compared with adalimumab-treated patients, as well as patient global VAS, morning stiffness VAS, SF-36 PCS and physical functioning domain and RAID. The p values for these interactions are included in Fig. 5 and Table S8 to demonstrate the differential efficacy predicted by high levels of these biomarkers compared with low levels. Baseline levels of markers associated with anaemia of chronic disease (hepcidin, ferritin and iron) were also associated with PRO improvements at week 24 (Table S8). Analysis of biomarkers as continuous measures also revealed interactions for SAA, MMP-3, CRP and P1NP with HAQ-DI at week 24 (interaction nominal p values < 0.01).

Baseline levels of CXCL13 and sICAM-1 were analysed to determine whether differential ratios of these biomarkers (high/high, high/low, low/high and low/low; using the median in the overall population as the cut-off) could predict clinical responses to sarilumab or adalimumab treatment at week 24. While CXCL13 high/sICAM-1 high and CXCL13 low/sICAM-1 low patients had greater ACR50 responses with sarilumab than adalimumab, the other combinations were not predictive (nominal p > 0.05; Figure S4). Additionally, CXCL13 high/sICAM-1 high patients had greater ACR20 responses with sarilumab than adalimumab (OR 3.8 [95% CI 1.5, 9.8]; nominal p = 0.004) but other combinations were not predictive (nominal p > 0.05).