We found that dysfunction of IL–18 negative feedback control is associated with disease severity and death in COVID–19 positive patients from symptom day 15 onwards. From symptom day 15 onwards, for every 37.7 pg/ml (Kd = 0.05nM) increase in highest per patient fIL–18, PFR (primary outcome) decreased by 100 mmHg (13.3 kPa) (p < 0.03). For each 50 pg/ml increase in highest fIL–18 per patient from symptom day 15 onwards, the adjusted OR for crude 60–day mortality was 1.41 (1.1–2.0] (p < 0.03). When the outcome was limited to those who died with hypoxaemic respiratory failure, the OR increased to 1.9 (1.3–3.1); p < 0.01. For each increase in highest fIL–18 by 50 pg/ml, the OR for requiring any organ support, defined as vasopressor therapy, artificial mechanical ventilation or renal replacement therapy, was 1.66 (1.1–2.9); p < 0.05. For each additional organ supported, fIL–18 increased by 50.7 pg/ml (p < 0.05) and when only patients who died with hypoxaemic respiratory failure were analysed, this increased to 63.67 pg/ml (p < 0.01).
This study adds to the literature in two significant ways. Firstly, this is the first report to characterise free IL–18 levels in COVID–19; previous studies characterised only the total IL–18 component [22, 23] which does not reflect the biologically active interleukin and may hide the true picture of inflammasome activation [13]. Secondly, this is one of the few studies to undertake interleukin analysis in the context of the disease course, through careful correlation to the first day of symptom onset.
The population under study were hospitalized adults who tested positive for SARS–COV–2 on PCR. The baseline features of the population relating to age and gender distribution were in keeping with other studies [24]. Ethnic differences between the mortality groups were not significantly different, but not representative of the region of the hospital, with minority groups being under–represented. It is possible that the “undisclosed” ethnicity category contained a larger proportion of ethnic minorities. There was no difference in median number of days from symptom onset, at the time of enrolment, between mortality groups, indicating that non–survivors do not present later to hospital than survivors. The mortality rate of patients not–ventilated (16%) and ventilated patients (37.5%) was higher than that described elsewhere (11.5% and 33%, respectively) [24], however, this may have been due to the higher median age of participants in our study. The distribution of comorbidities was as expected between the mortality outcomes, though of note, only the difference in chronic kidney disease reached the significance threshold; hypertension incidence did not. This appears to be due to a lower incidence of hypertension (58.5%) in our non–survivors compared to that described elsewhere (67.6%) [24]. During inpatient stay, non–survivors had significantly more blood samples taken per patient, as expected, with patients in intensive care being bled daily, and having a higher mortality rate than patients not admitted to intensive care. While daily dexamethasone was included in the standard of care by the time of enrolment, IL–6 blockade had not yet been included. Thus, the fIL–18 profile observed in this study is in the context of steroid–based immunosuppression.
Two inflammasome types, NLRP3 and NLRC4, have classically been the main focus of study. IL–18 production from the NLRP3 inflammasome, an intracellular sensor of anti–microbial signals found mainly in macrophages, and which activates in response to the pathogenically stimulated ASC protein scaffold, shows how IL–18 can be detrimental to clinical outcomes along two axes: level and persistence of elevation. In models of sepsis, for example, injection of a low or moderate dose of lipopolysaccharide (LPS) induces a moderate rise in IL–18 levels that enhances anti–bacterial host defenses, while injection with high doses results in sustained, high levels of IL–18, that impair host antibacterial defenses [25]. In models of lung infection, avian influenzae H5N1 and H7N9, which contain a PB1–F2 protein, persistently activate NLRP3, resulting in persistently elevated levels of IL–18, inducing IFN–gamma, and a subsequent cytokine storm [26, 27] in a manner reminiscent to severe acute respiratory syndrome (SARS) [28]. This profile of a sustained, elevated IL–18 level associated with poor clinical outcomes is in keeping with our findings that persistently elevated fIL–18 levels from symptom day 15 onwards are significantly associated with disease severity and 60–day mortality, in COVID–19.
While NLRC4 is also activated by pathogenic signals, specifically flagellin and components of the type III secretion system [29], it is also capable of activating caspase–1 and producing IL–18 independently of the ASC scaffold and thus, unlike NLRP3, is not dependent on pathogenic stimuli. Thus, NLRC4 mutations can result in overwhelming production of IL–18, as in systemic juvenile idiopathic arthritis (sJIA) [30] and adult onset still’s disease [31], in which NLRC4 mutations drive IL–18 production into the nanogram range, due to uninhibited production.
Hemophagocytic lymphohistiocytosis (HLH), a syndrome that bears key similarities to life–threatening COVID–19 [1–5], is characterised by phagocytosis in bone marrow and other tissues, of haemoglobin, white blood cells and platelets by histiocytes such as macrophages, under excess stimulation by IFN–gamma. The aetiology of excessive IL-18 production, and subsequent excess IFN–gamma stimulation, driving macrophage activation, differs by HLH type. For example, familial and secondary HLH, the latter also known as MAS, include macrophage activation due to lytic failure, by cytotoxic T–cells, of IFN–gamma producing antigen-presenting cells, due to intrinsic T–cell mutations [32]. MAS additionally is used to describe excessive release of fIL–18 due to uncontrolled inflammasome activation, due to NLRC4 mutations [33], again driving elevated IFN–gamma levels and macrophage activation. The third and final type, known as CpG–induced MAS, involves relentless antigenic stimulation of the NLRP3 inflammasome, through activation of toll–like receptor 9 (TLR9), which recognises DNA rich in unmethylated CpG–DNA motifs from bacterial or viral DNA, again driving elevated fIL–18 levels, and subsequent elevated IFN–gamma levels [34].
This third pathway of CpG–induced MAS through constitutive NLRP3 activation, may be the central pathomechanism of COVID–19. Early functional exhaustion of innate immunity, so crucial in early antigen–control, is seen in fatal COVID–19 [35]. Work by Waggoner et al has shown the essential role of Natural Killer (NK) cells in modulating CD4 + T Cells to prevent such functional exhaustion [36]. As NK cell function and number are impaired with age [37] and in those with metabolic syndrome conditions [38] we would expect to see greater functional exhaustion of lymphocytes in these groups, unchecked viral spread, and repeated inflammasome stimulation, driving CpG–induced MAS, with fatal outcomes.
Both Weiss et al [6] and Girard–Guyonvarc’h [34] et al have shown the essential role of IL–18bp in silencing IL–18 activity in MAS and CpG–induced MAS, respectively. Our findings demonstrate that elevated fIL–18 after symptom day 15 is driven by increased production of IL–18 without a commensurate increase in IL–18bp (Fig. 2). Why adequate levels of IL–18bp are not released from symptom day 15 onwards under its usual homeostatic mechanism is unclear. Neutralising auto–antibodies to IFN–alpha, seen in life–threatening COVID–19 [39] may explain the elevated levels of fIL–18; IFN–alpha both diminishes IL–18 production from macrophages and is an important inducer of IL–18bp [40]. Interestingly, the PB1–F2 protein in H5N1 and H7N9, cited earlier, which drives a cytokine storm through excessive IL–18 production from mass activation of NLRP3 inflammasome [27], also inhibits IFN–alpha production [41]. Persistently elevated fIL–18 in severe COVID–19 may underlie the finding that elevated IFN–gamma after day 10 of symptoms is independently associated with death [42]; IFN–gamma potently activates macrophages [43] and macrophage–mediated destruction of lung architecture via infiltration of extra pulmonary tissue is a hallmark of fatal COVID–19 [44].
Weaknesses of this study include, firstly, the inherent limitations of being a single–centre, prospective observational study. Recognition of this early in study design was attempted to be mitigated through selection of a site which serves a racially diverse population. A second weakness relates to the use of PF ratios derived from Sa02 when Pa02 values were not available, albeit through a validated mathematical model. Right–shift in the haemoglobin dissociation curve in critically ill patients may have resulted in under–estimation of the PFR in the critically ill cohort, potentially under–estimating the slope of the association between highest fIL–18 and PFR from symptom day 15 onwards. This is unlikely to have played a significant role in introducing bias however, since PFR was calculated from measured Pa02 values directly, in all critically ill patients. A third weakness relates to the incompleteness of our data on ferritin measurement. Hyperferritinaemia forms a part of the H-score used to diagnose MAS, and is particularly elevated with NLRC4 mutation-driven MAS. Though the ferritin levels seen in this study are certainly elevated (Table 2), the averaged values per time-bin do not reach the required threshold for contributing to the H-score (> 2000 ng/ml). This may be due to incomplete data collection, dexamethasone-mediated suppression, or simply because life-threatening COVID-19 may not conform to all the diagnostic features of MAS as currently formulated, despite its similar clinical features and the association of elevated fIL–18 with disease severity and mortality.
Areas of further research include, firstly, validating these results in a separate cohort. Secondly, comparing fIL–18 levels in patients with COVID–19 against other conditions, which may help clarify conflicting results [45, 46], though of note, these cited studies did not analyse the free IL–18 portion. Additionally, concurrent analysis of IL–6, though not within the scope of our research question, could be an avenue of further research. Since IL–18 stimulates IL–6 release [47], it is unlikely IL–6 blockade would attenuate the fIL–18 profile; current understanding is that the IL–1β/IL–6/CRP and IL–18/ferritin inflammatory axes are separate [48], supported by the lack of association between fIL-18 and CRP in our study. Finally, though we focused our regression analyses on the period from symptom day 15 onwards on the basis of our research question, our longitudinal data indicates that non-survivors have higher IL–18bp early in the disease course (days 5–9). High baseline IL-18bp, as seen in metabolic syndrome conditions [49], may prevent a sufficient rise in fIL-18 necessary to facilitate a strong Th1 response for early antigen-control, resulting in antigen escape from symptom day 15 onwards, and persistent inflammasome-mediated IL-18 release. This is an area requiring further research.
This study demonstrates the potential utility of fIL-18 as a biomarker of disease, from symptom day 15 onwards in patients with COVID-19. While causation cannot be established in this observational study, our findings provide hypothesis–generating evidence for modulation of IL–18 from symptom day 15 onwards in patients with COVID–19. Our finding that for every 37.7 pg/ml (Kd = 0.05 nM) increase in highest fIL–18, PFR declines by 100 mmHg after symptom day 15 (Fig. 3), provides both a time frame and a theoretical approach for fIL–18 blockade based on the degree of hypoxaemic respiratory failure. Potential drug candidates for such an intervention include Tadekinig Alfa (AB2Bio), a recombinant human interleukin–18 binding protein, having shown efficacy in conditions with elevated fIL–18, such as MAS [50] and sJIA [30], and caspase–1 inhibitor, Belnacasan (Roivant), currently in Phase 2 trials for COVID–19 [51], having shown efficacy in reducing pulmonary inflammation in animal models [52].