Failure to biological treatment is an important issue and challenge for today's rheumatology. The current literature focuses on identifying predictors of good response to bDMARDs, while there is a great need to find the causes of the resistance. In addition, only known and routinely assessed markers of disease activity are usually considered for research [29,30,31].

In our study, we went deeper into the pathogenesis of axSpA and decided to include factors that have not been considered so far in the context of biological treatment failure. The idea to explore factors related to the microbiome and increased intestinal permeability came from the constantly arising number of reports on the importance of the gut-joints axis [32,33,34,35]. Disruption of the intestinal barrier function has been shown to predict onset of arthritis and zonulin was the main agent associated with this process [34, 36].

In addition, zonulin was shown to be up-regulated in ileal samples of patients with AS. The authors demonstrated that zonulin is able to stimulate the expansion of macrophages with the M2 phenotype, which are involved in SpA gut inflammation and synovitis [13].

Interestingly, as in our previous study, higher levels of zonulin were associated with poor response to treatment in axSpA [22]. It appears that persistently increased intestinal permeability 'interferes' with therapy. This may be due to the continuous stimulation of the immune system by intestinal antigens, probably related to dysbiosis. It is possible that the condition of intestinal damage and disturbance of the microbiome itself is linked to NSAID use [37].

This is a very exciting result, especially in the context of the study showing that the use of the zonulin antagonist larazotide acetate can inhibit arthritis [36].

The lack of association of zonulin with gastrointestinal symptoms and IBD shows difficulty in selecting individuals with increased intestinal permeability based on the medical history and clinical symptoms. Similar results have already been reported in other studies [38, 39]. It is likely that the relationship of IBD and gastrointestinal with biological treatment failure did not depend only on a damaged intestinal barrier. In our study, these two factors were zonulin-independent predictors (OR = 7.56, 95% CI 1.06 − 54.06, p = 0.044; OR = 4.63, 95% CI 0.91 − 23.48 p = 0.064, respectively). It seems that healing the gastrointestinal tract, whatever the cause, may increase the possibility of therapeutic success in axSpA.

Our hypothesis related to the influence of the microbiome on response to treatment seems to be supported by two other factors: age and the history of frequent infections, which independently of zonulin were negative predictors (OR = 1.09, p = 0.047; 95% CI 1.00 − 1.18, OR = 8.63, 95% CI 1.43 − 52.21, p = 0.019, respectively). According to the results of another study, the composition of the gut microbiota at baseline may have a better predictive value for response to TNFi than indicators of disease activity including CRP.

A history of frequent infections increased the risk of treatment failure of bDMARDs by more than eightfold! This probably may stem from the disruption of the gut microflora caused by frequent antibiotic use.

Age, on the other hand, is linked to dysbiosis [40]. Previous reports have already shown a decrease in treatment effectiveness with age in axSpA, but this was usually associated with the presence of advanced degenerative changes in the spine or a higher degree of x-ray sacroiliitis [41,42,43]. In our study, x-ray sacroiliitis was not associated with response to treatment, whereas age was. Age was also not correlated with the degree of x-ray sacroiliitis.

Vallier et al. demonstrated that the composition of the gut microbiota at baseline in axSpA patients presented better predictive value for response to TNFi than indicators of disease activity including CRP [44].

Also, psoriasis treatment studies have noted differences in response to bDMARDs depending on the composition of the microbiome [45, 46].

Similarly, in another study, concomitant diseases were associated with worse treatment effects with TNFi in axSpA [47]. In our study concomitant diseases did not significantly increase this risk (OR = 2.03, 95% CI 0.22 − 18.77, p = 0.53).

As in our previous report, zonulin was detected in all patients, not only in Hp2 antigen carriers, and, as before, was highest in those with the Hp 1–1 phenotype (p = 0.91), demonstrating that the ELISA detected more than just the pre-Hp2 molecule [21, 22].

Zonulin was significantly correlated with CRP, although CRP alone was not associated with treatment response. This is a different result from most studies, which have shown the superiority of increased CRP in predicting good response to bDMARDs, especially TNFi [29, 30, 47, 48]. However, higher values of the inflammatory indices (WBC, CRP, ESR) reduced the risk of poor response, but not significantly. A substantially better predictive value demonstrated Hp level.

High levels of Hp were present in patients responding well to biological therapy. In fact, Hp concentrations above 400 mg/dl were the cut-off point for responders. The result was on the borderline of significance (p = 0.056), which may result from the small size of the group. It is possible that, like CRP, Hp level reflects inflammation and define those who may benefit from biological treatment.

Although there were significant differences in Hp levels between phenotypes none of the phenotypes proved to be a predictor of treatment failure to bDMARDs.

MRI sacroiliitis only occurred in patients who responded well to treatment, which is in line with other study results indicating MRI sacroiliitis as a predictor of good response to standard and biological therapy [22, 30, 49, 50]. In our analysis we had a lot of missing data, the result was not statistically significant. Also, all patients without radiographic changes in the sacroiliac joints responded well, reflecting the good efficacy of biological treatment at an early stage of the disease.

A strong part of this study is its observational character, which reflects ‘real-life’ clinical situations with consecutive sampling patients.

In addition, we analysed risk factors for a poor response to bDMARDs that had never been considered until now. We shed new light on certain aspects in the approach to treating patients with axSpA and identified factors that can be modified and increase the chance of therapeutic success.

Our study had also some limitations. Due to the COVID-19 pandemic, we were unable to collect more patients and the final sample size is smaller than intended. For this reason, we did not perform separate predictors analyses for different groups of bDMARDs. Although we were not able to do multivariate analysis due to small sample size, we performed a univariate analysis with zonulin referring to the factor of most interest.

In the assessment of axSpA disease activity, we did not use the Ankylosing Spondylitis Disease Activity Score (ASDAS), which is more appropriate for this purpose. Determination of inflammatory indices after 12 weeks of biological treatment was difficult due to pandemic.

It may seem that the inclusion of only hospitalised patients in the study raises the risk of a sample selection bias. However, in Poland, diagnosis of spondyloarthritis is most often made in the hospital setting due to limitations in the operation of outpatient clinics, whereas qualification for biological treatment can only take place in hospital. Instead, hospitalisation made it possible to quickly rule out possible other causes of the back pain.