An explorative study on deep profiling of peripheral leukocytes to identify predictors for responsiveness to anti-tumour necrosis factor alpha therapies in ankylosing spondylitis: natural killer cells in focus

Ankylosing spondylitis (AS) is a multifactorial chronic inflammatory rheumatic disease belonging to the group of rheumatic diseases known as spondyloarthritis (SpA) which primarily affects the axial skeleton [1]. AS has a prevalence of about 1.43 million in the European population [2] with an onset in adolescence [3] and a two-times higher occurrence in men than women [4]. The pathogenesis of AS is still obscure; it is assumed that AS is mainly caused by both genetic factors, which implies the expression of the major histocompatibility complex (MHC) I antigen human leukocyte antigen B27 (HLA-B27) [5], and also by environmental factors such as enterobacterial antigens [6]. To alleviate the axial symptoms of AS patients, non-steroidal anti-inflammatory drugs (NSAIDs) are delivered as first-line therapy. Anti-tumour necrosis factor (TNF) blocking therapy is applied only in patients with constantly high disease activity who are non-responsive to conventional NSAID treatment [7, 8]. At present, there are five anti-TNF-α agents approved for the treatment of AS: infliximab [9], a monoclonal chimeric antibody; etanercept, a soluble human TNF receptor (sTNFR)2 fusion protein [10]; adalimumab, a humanised monoclonal antibody [11]; golimumab, a fully human monoclonal antibody [12]; and certoluzimab, a Fab fragment of a humanised monoclonal antibody [13]. Most of these biologics are also successfully administered in rheumatoid arthritis (RA), psoriasis (Pso), juvenile inflammatory arthritis (JIA), and inflammatory bowel disease (IBD). The biological functions of TNF-α are mediated by binding to the membrane receptors TNFR1 (p55) or TNFR2 (p75). While TNFR1 is ubiquitously expressed in all lymphoid and myeloid immune cells and body cells [14], the expression of TNFR2 is mainly restricted to T cells [15] and natural killer (NK) cells [16]. In addition, TNFR2 can be found to be expressed in endothelial and mesenchymal cells, cells of the central nervous system (CNS), oligodendrocytes and cardiac myocytes [17], and a few other cell types [18]. According to their functions, TNFR1 is primarily associated with pro-apoptotic processes, while TNFR2 is responsible for processes ensuring survival of cells [19].

Although targeting of TNF-α is very effective in AS, around one-third of treated patients show only a poor response that can be partly attributed to the development of anti-drug antibodies (ADAb) resulting in reduced bioavailability [20]. The most likely precondition for swapping to another anti-TNF agent is partial or entire failure of effectiveness along with side effects [21]. With respect to these adverse reactions and the high costs of anti-TNF agents leading to high economic burden for the health care systems, it is desirable to stratify patients according to treatment predictors prior to biological therapy. Various demographic and clinical parameters such as high baseline disease activity, short disease duration, young age, male sex, and presence of HLA-B27 have been shown to correlate with adequate clinical short- and long-term response [22‐26]. In addition, modern imaging techniques, such as magnetic resonance imaging (MRI), are used to correlate bone and tissue destruction with treatment response [27]. However, these techniques are time consuming and expensive when used as a standard pre-treatment assessment. Another level of therapy response prediction was investigated when responsiveness to anti-TNF agents was related to the presence of different TNF-α genotypes. It was reported that patients with TNF-α–308 G/G, –857 C/C, or –1031 T/T genotypes showed a better response to anti-TNF agents than patients without these polymorphisms [28, 29]. Apart from these observations, there are no reliable predictive biomarkers for anti-TNF responsiveness in AS. Using a multi-parametric flow cytometric approach, we aimed to identify cell-based biomarkers in the peripheral blood of AS patients that are able to predict a successful therapeutic response to TNF inhibitors before starting therapy. As a result, we found that a low pre-treatment frequency of a CD8-expressing subpopulation of NK cells is associated with a lack of therapeutic response.

To our knowledge, this is the first study aiming to identify cellular biomarkers in peripheral blood to stratify AS patients upfront with regard to subsequent responsiveness to anti-TNF-α therapy. To date, there are no quantifiable laboratory parameters available that can be used for a personalised response prediction [34]. Although higher values of CRP, ESR, and serum amyloid A (SAA), together with the presence of HLA-B27, have been reported to be useful baseline predictors for a successful anti-TNF-α therapy response in AS, the robustness, sensitivity, and specificity fail if applied to individual patients [35, 36]. Even our data, such as the age of patients, disease duration, CRP levels, or blood sedimentation, did not allow any prediction of future responsiveness.

Other potential new biomarkers described, such as single nucleotide polymorphisms (SNPs) and activity of endoplasmic reticulum aminopeptidase (ERAP)-1 [37], serum levels of matrix metalloproteinase (MMP)-3 [38], or vascular endothelial growth factor (VEGF) [39], have not yet reached a standard in the clinical diagnostic routine of AS. Generally, it is challenging to correlate quantifiable parameters to the disease activity BASDAI score, which is calculated on the basis of a patient’s subjective assessment of well-being and therefore is only of limited value when used as an absolute variable.

In our explorative study presented here we used an integrated multi-parametric flow cytometry and a new unsupervised data clustering approach to identify possible cellular biomarkers that are qualitatively or quantitatively different in responders and non-responders. A similar approach has been successfully used to classify active AS patients from healthy donors according to specific phenotypical changes in blood [32]. A comprehensive overview about immunophenotypical changes described so far in different autoimmune diseases is given by Alegria et al. [40].

Principally, at first view, an increased frequency of the major CD56^dimCD16⁺ NK cell subset was indicative for a weak therapeutic anti-TNF response. This finding is in line with other reports showing that innate immunity and particularly NK cells may play a central role in the pathogenesis of various autoimmune diseases both in a protective and pathogenic manner. Their frequency and functionality were investigated in chronic inflammation, such as rheumatoid arthritis [41], multiple sclerosis [42], psoriasis [43], and systemic lupus erythematosus [44]. Despite some conflicting results, an overall decrease and a cytolytic impairment of circulating NK cells could be observed. In AS, the functionality of NK cells is described to be likewise impaired [45] but, in contrast to other autoimmune diseases, an increased NK cell number is reported [46, 47]. We could confirm this observation if comparing the non-responder group with normal donors (Fig. 3a); however, the NK cell frequency in the responder cohort was similar to the number of healthy individuals. The strong association between NK cells and the aetiology of AS is underlined by the capability of the disease-dependent elevated killer cell immunoglobulin-like receptors (KIRs) to recognise the HLA-B27 antigen, which is expressed in 80–90% of AS patients [48, 49]. Concomitant with a pathogenetic upregulated frequency of KIR3DL1⁺ NK cells in AS, interferon (IFN)γ production is correspondingly diminished [50]. Moreover, activated KIR3DL2⁺ NK cells are increased in SpA and may play a pathogenic role.

An in-depth analysis of the NK cell compartment revealed that the frequency of classical NK cells expressing the CD8 antigen showed a significant and positive correlation with anti-TNF responsiveness. It is known that about 40% of NK cells variably express CD8 in an α/α homodimeric form whereas the CD8-positive subset is described to exhibit enhanced cytotoxic features as compared with its CD8-negative counterpart [51, 52]. We could not detect significant age- or sex-related differences with respect to the frequency of CD8-positive NK cells either in the group of healthy controls or in that of AS patients. Thus, our data imply that NK cells expressing the CD8αα homodimer are directly or indirectly involved in the immunosuppressive effect exerted by anti-TNF-α blockers.

If CD8αα-expressing NK cells are directly involved it can be assumed that their increased cytotoxic activity and/or their diminished behaviour regarding cytotoxicity-induced apoptosis are responsible for the improved responsiveness of TNF blockers. Alternatively, it can be hypothesised that an engagement of CD8αα receptors on NK cells to molecules of the HLA-I family can cause an increased secretion of the pro-inflammatory cytokines TNF-α and IFNγ [53], which in turn promote both TNF receptor (TNFR) synthesis and its proteolysis to a soluble form [54]. By this mechanism, endogenously synthesised sTNFRs [55, 56] can co-operatively support the action of therapeutic TNF inhibitors [57, 58]. These non-signalling ‘decoy’ receptors are still competent for binding TNF and thus may function as a natural TNF antagonist comparable to ETN [59]. Since we have included adalimumab- and etanercept-treated patients in our study, it was interesting to know if differences in the prediction of responsiveness were detectable when either TNF-α was neutralised by ADA or scavenged by the sTNFR2 fusion protein (ETN). Although group size reduction was responsible for a less statistical power of prediction analysis, we could ascertain a better correlation with respect to ETN-treated patients compared with ADA-treated patients. Therefore, these findings are encouraging for validation in new independent cohorts of appropriate group sizes. If this result could be validated it would indicate diverse and more complex modes of action going beyond the mere neutralising effect of TNF-α blockers.

To elucidate possible differences in the TNFR expression in CD8-positive and CD8-negative NK cell subsets isolated from healthy individuals, we performed global gene expression analyses but could not detect any differential expression of TNFR1, TNFR2, or TNF-α (data not shown). Comparing the expression of TNFR1 and TNFR2 in healthy individuals, it was obvious that higher expression levels were detectable for TNFR2 (data not shown). Unfortunately, thus far we have had no opportunity for analysing cells isolated from AS patients to test if TNFRs or TNF-α were differentially expressed under chronic inflammatory conditions. Thus, according to our data, it can be postulated that CD8-positive NK cells are obviously capable of amplifying the neutralising effect of TNF-blockers, but it remains unclear why this amplification is preferentially observed in ETN- and not in ADA-treated patients.

Although our findings are promising, a further validation of CD8-positive NK cells as a potential biomarker for TNF responsiveness is necessary in an independent cohort of AS and other rheumatic and gastrointestinal diseases where anti-TNF blockers are successfully administered. Nevertheless, our study is a first proof of the concept that cellular response signatures can be identified in peripheral blood by an extensive immunophenotyping approach. Even though the exact mechanism of how CD8-positive NK cells and the therapeutic effect of TNF-α blockers are interrelated in AS is only poorly understood thus far, monitoring these cells by flow cytometry offers an interesting new diagnostic option with respect to the challenges of an individualised therapy concept, at least in the field of chronic inflammatory rheumatic diseases.