A rationale for using steroids in the treatment of severe cases of H5N1 avian influenza

Cytokine families and functions

Cytokines are signalling proteins involved in the response to pathogens and subsequent inflammatory processes. Of the six classes, the interferons (IFNs), interleukins (ILs) and tumour necrosis factors are the most important. A clinical and experimentally useful classification system divides cytokines into those that promote the proliferation and function of T-helper type 1 cells and those that do the same for T-helper type 2 cells. Examples of Th-1 cytokines include IL-1 and IFN-γ, and examples of Th-2 cytokines include IL-4, IL-10, IL-13 and TGF-β.

The IFN family comprises several types: type I, of which IFN-α and IFN-β are the most well known, have antiviral activity and are associated with the regulation of natural immunity, while type II IFNs, such as IFNγ, are associated mainly with cellular immunity (Sladkova & Kostolansky, 2006). More recently, a type III family has been described that is evolutionarily distinct from type I IFNs but appears to have a similar function to them, and whose members are structurally related to IL-10 cytokines (Onoguchi et al., 2007).

Dozens of IL entities have been described, of which the most important are IL-1 (α and β), IL-6, IL-10, IL-11, IL-12 and IL-18. When viral infection occurs, the action of these entities can be generally described as pro-inflammatory (with the exception of IL-10), and sometimes overlapping in effects, which can be local or systemic. Conversely, depending upon the phase of the infection (early or late), some of these cytokines, such as IL-18, can act in an anti-inflammatory capacity through feedback to prevent excessive cytokine production (Liu et al., 2004; Sareneva et al., 1998). In some instances, however, this feedback breaks down, and dysfunctional regulation occurs. The result is an imbalance that is thought to lead to chronic hypercytokinaemia in such conditions as rheumatoid arthritis, Crohn's disease and other autoimmune syndromes, or acute hypercytokinaemia and life-threatening events when potent viruses are involved (Hussell et al., 2001; Palucka et al., 2005; To et al., 2001). A demonstration of the power of acute hypercytokinaemia was recently illustrated in a clinical phase 1 trial of the anti-CD28 mAb TGN1412 in six young healthy males (Suntharalingam et al., 2006). Induction of TNF-α, IFN-γ, IL1-β, IL-2, IL-10 and other ILs occurred within 1–2 h, and resulted in pulmonary infiltrates, lung injury, renal failure and disseminated intravascular coagulation 12–16 h after infusion that required immediate and extensive medical treatment.

Effect of influenza viruses on cytokines

Certain strains of influenza viruses can also elicit acute hypercytokinaemia. For example, using mice infected with recombinant influenza viruses containing the haemagglutinin and neuraminidase genes from pandemic 1918 influenza strains, Kash et al. (2004) demonstrated that dozens of genes associated with cytokines, apoptosis, lymphocytes, activated T cells, macrophages and oxidative damage were upregulated with lethal consequences.

The mechanism by which influenza viruses can effect hypercytokinaemia is poorly understood. Cells sense the presence of viral material through a variety of pattern recognition receptors, such as the Toll-like receptors (TLR). Both TLR7 and TLR8 are involved in the recognition of single-stranded RNA, and TLR3 in the recognition of double-stranded RNA (Garcia-Sastre, 2006; Matikainen et al., 2006). Induction of IFN-α/β follows recognition, although influenza A viruses are able to prevent this via the antagonistic action of the NS1 protein (Garcia-Sastre, 2006), which also inhibits adaptive immunity by attenuating human dendritic cell maturation (Fernandez-Sesma et al., 2006).

In the case of H5N1, both the induction of TNF-α and IFN-β/γ have been observed in alveolar and bronchial epithelial cells (Yuen & Wong, 2005). Induction of IL-6, IP-10 (a macrophage attractant) and RANTES, a member of the IL-8 family and a selective attractant for memory T lymphocytes and monocytes, has been additionally witnessed in cell culture experiments (Chan et al., 2005) and patients infected with H5N1 (de Jong et al., 2006). Moreover, de Jong et al. (2006) reported that the highest levels of IL-6, IL-10, MIG and MCP-1 (the latter two cytokines are also monocyte and macrophage attractants) were found in those patients who subsequently died, highlighting the concentration–response relationship.

The production of these early response cytokines then causes the infiltration of macrophages, and attraction of monocytes, eosinophils, basophils and T cells. Transcription factor NF-κB also seems to play a central role in development of the cytokine cascade, as well as upregulation of the mitogen-activated protein kinases, which are strongly associated with the production of TNF-α at the transcriptional and posttranscriptional levels (Lee et al., 2005).

Autopsy findings confirm TNF-α involvement in hypercytokinaemia induced by H5N1 avian influenza strains, with its presence in the lungs (Uiprasertkul et al., 2005). In addition, in influenza-associated (non-H5N1) encephalopathy, serum levels of both IL-8 and TNF-α were found to be significantly higher than for controls (influenza, but no encephalopathy). However, serum cytochrome c levels were nearly 90-fold higher, which seems indicative of apoptosis in several organs, including the cerebrum and liver (Nunoi et al., 2005).

While these studies show that H5N1 can increase pro-inflammatory activity, the virus can lower the cell's ability to produce antiviral responses. For example, it appears that TNF-α can depress generation of plasmacytoid dendritic cells (a source of IFNαβ), as well as suppressing IFNα secretion by immature cells of this type (Palucka et al., 2005). It should be pointed out in this context that while IFNs are generally regarded as pro-inflammatory, they exert a powerful antiviral affect (Sladkova & Kostolansky, 2006). Therefore, suppressing IFN gene expression would be counter-productive to clearing viral infections. In this regard, it should be noted that corticosteroids can augment the action of exogenously given IFNs because they help suppress the NFκB-mediated downregulation of IFN gene expression (Loutfy et al., 2003; Wei et al., 2006). Interestingly, low-dose steroids were also found to counter the influenza-like side effects of exogenously administered IFN (IFNβ-1b) in patients suffering from relapsing–remitting multiple sclerosis (Rio et al., 1998). In addition, dexamethasone was found to selectively suppress the pro-inflammatory gene expression induced by IFNγ, including cyclo-oxygenase-2 expression in human bronchial epithelial cells (Pawliczak et al., 2005).

The H5N1 virus can also affect both non-specific and specific immunity. Thus, Hsieh and Chang (2006) found that perforin expression appears to be suppressed, and the cytotoxicity of CD8+T cells is reduced in the presence of H5-bearing cells. Furthermore, this led to lymphoproliferation, the overproduction of IFNγ and macrophage activation.

Hypothalamic–pituitary–adrenal (HPA) axis failure

Under normal circumstances, corticotropin-releasing hormone is produced in the pituitary gland in response to a variety of factors. In turn, corticotropin-releasing hormone stimulates ACTH production in the anterior pituitary, and this hormone stimulates cortisol secretion in the adrenal glands. Increased cortisol levels can then modify the original response. In critically ill patients, cortisol levels can be extremely high and mirror the severity of the disease, but often the response to the stressor is inadequate, leading to what is termed HPA axis failure or adrenal insufficiency. What constitutes adrenal insufficiency is still a matter for debate. One definition suggests that 30–60 min following a 250 μg dose of ACTH, blood cortisol levels should rise at least 9 μg dl⁻¹. If they do not, this would constitute adrenal insufficiency (Annane et al., 2000). However, some investigators have criticized the supraphysiological-dose ACTH test and believe that the amount of ACTH administered should be near physiological (1–2 μg), or that a random plasma cortisol sample of <25 μg dl⁻¹ under situations of extreme, sepsis, illness or infection is a more realistic marker of adrenal insufficiency (Marik & Zaloga, 2002).

The HPA axis and cytokines

In cases of infection by viruses or bacteria, corticosteroids are produced in response to pro-inflammatory cytokines, such as IL-1 and IL-6, and downregulate these entities, ensuring a controlled feedback loop (Padgett et al., 2000). An extensive review of early work (mostly cell culture studies) concerning the effect of cytokines, such as IL-1 (α and β), IL-6 and TNFα, showed stimulation of the HPA axis in the majority of cases (Besedovsky & del Ray, 1996), but some investigations have demonstrated that inhibition of the axis can also occur (Bateman et al., 1989; Gaillard et al., 1990; Jaattela et al., 1990, 1991; Mastorakos et al., 1993).

Chronic inflammation, which is the result of high, long-term cytokine levels, is one example in which suppression of the HPA axis is likely. For example, rheumatoid arthritis patients, who demonstrate secondary adrenal insufficiency, also show IFNγ/IL-10 ratio peaks during the early morning hours when cortisol levels are at their lowest (Cutolo et al., 2006). In other words immunomodulation of the IFNγ does not occur, because of HPA axis failure.

In acute hypercytokinaemia resulting from H5N1 influenza, as evidenced by the raised levels of IFN-γ, IL-6, IL-8, IL-10 and other factors in blood, which are correlated with viral load (de Jong et al., 2006), what happens to the HPA axis? Initially we might suspect that the axis would be strongly activated. However, since this influenza strain causes such a high case fatality rate, we might presume that either the suppression induced by the elevated cortisol levels is insufficient to depress the high cytokine levels, or the HPA axis collapses to produce adrenal insufficiency. In order to understand the possibilities, insight can be gained by studying the sepsis/shock literature, in which similar parallels take place.

Steroid treatment in sepsis and shock

For many decades, steroids have been empirically used to treat patients with shock or sepsis to reduce mortality, with unclear results. By 2002, however, it was becoming apparent to some researchers that low, rather than high doses of steroids, might be the correct way to treat patients with demonstrated adrenal insufficiency. Not all investigators agreed. In summing up the state of research in 2002, Bernard (2002) said, ‘For now, the data indicate that we should not administer steroids to patients who are in shock’. However, a systematic review and meta-analysis conducted in 2004 showed that for post-1995 reports separated into trials that used (a) short courses of high-dose steroids (6 trials, n=511), and (b) long courses of low-dose steroids (5 trials, n=465), there was a difference (Annane et al., 2004); mortality remained the same for group (a) but was significantly reduced for group (b). In conclusion, Annane et al. (2004) recommended giving patients 200–300 mg hydrocortisone daily for 5–11 days when adrenal insufficiency was demonstrated (typical percentage of patients ∼50 %, condition for adrenal insufficiency was a rise of ≤9 μg dl⁻¹ after ACTH administration or a random cortisol level of ≤15 μg dl⁻¹).

Should a daily 250 mg dosage of hydrocortisone be considered physiological? Studies conducted by Samuels (2000) suggested that in primary adrenal insufficiency, 19 mg hydrocortisone per day is a sufficient replacement dosage. Taking into account the increase in cortisol of 200–300 % often observed following a systemic response to stress (Marik & Zaloga, 2002), 250 mg hydrocortisone in the sepsis-shock model is clearly supraphysiological, and not a replacement level, even if it is regarded as a low dose.

Mechanisms of HPA axis failure

Wheatland (2004) proposed that SARS and influenza A cause adrenal insufficiency, because portions of viral amino acid sequences mimic human ACTH sequences (amino acid residues 25–39) less conserved between species. This may result in the host producing specific antibodies against these viral epitopes, also resulting in lower ACTH levels. A search of influenza A H5N1 HA gene GenBank sequences showed similar amino acid identities as detected by Wheatland (2004) with various influenza A non-H5N1 isolates and the 1997 Hong Kong H5N1 isolate (see Table 1⇓).

Wheatland (2004) cites additional evidence for his hypothesis from data reported by Jefferies et al. (1998), who demonstrated that influenza-infected patients (H3N2) had normal cortisol levels but lowered levels of ACTH 1–3 days after onset of illness (cortisol levels in infected patients 13.7±1.4 μg dl⁻¹; cortisol levels in controls 10.8±1.0 μg dl⁻¹, P=not significant; ACTH levels in infected patients 13.5±2.1 pg ml⁻¹; ACTH levels in controls 23±3.2 pg ml⁻¹, P=0.02). However, ACTH levels rebounded to normal approximately a week later. Wheatland (2004) argues that the data are the result of auto-antibodies generated during the influenza infection that compete with assay antibodies in the immunoassay, causing erroneously low ACTH readings. Moreover, he suggests that experiments in which influenza-infected mice were treated with antilymphocyte serum, which resulted in 50 % less mortality (Suzuki et al., 1974), are indicative that the T lymphocytes induce antibodies to ACTH. This interpretation is questionable, since it can take up to 10 days to generate an immune response, although it does not invalidate the hypothesis itself. This leaves us with the puzzling data of Jefferies et al. (1998), which shows a transitory depression of ACTH but no substantial effect on cortisol. Normally, stimulation by the influenza infection of both ACTH and cortisol levels would be expected. However, since this is not the case (assuming there were no methodological problems with the measurement of the ACTH), it has to be assumed that the virus has induced suppression of ACTH but not cortisol. As no other investigators have reported ACTH or cortisol data for influenza-infected patients, this unexpected finding must be accounted for when developing a rationale for steroid use.

An interesting report that sheds further light on the subject is the case of a 32-month child who was treated for persistent reactive airway disease with long-term systemic and inhaled steroids. One day after influenza vaccination, the child presented with acute adrenal insufficiency, confirmed by plasma cortisol levels before and after ACTH administration (0.6 μg dl⁻¹ and 8.4 μg dl⁻¹, respectively) (Kennedy et al., 2002). Following hydrocortisone replacement therapy and tapering fluticasone treatment, the child completely recovered. In this immunocompromised child, we can speculate that influenza antigens further reduced an already low ACTH level and cortisol levels. However, it is clear from the timescale that this must have involved a cytokine response, rather than a humoral (adaptive auto-antibody generation) response of the kind suggested by Wheatland (2004). In fact, from the time of influenza infection, it could take up to 10 days for the kind of auto-antibodies that Wheatland (2004) suggests to form, and given the timeline of H5N1 cases presented in the literature (Chotpitayasunondh et al., 2005; Hien et al., 2004; Kandun et al., 2006), it is certain that in many cases of H5N1, any adrenal insufficiency would have to result from hypercytokinaemia. This case of the child also points out the hazard of using prolonged high-dose corticosteroids, which can cause direct HPA axis failure.

In reviewing acute respiratory distress syndrome cases that fail to resolve, Meduri & Yates (2004) suggest an alternative mechanism based on excess activation of NF-κB compared to the anti-inflammatory transcription factor glucocorticoid (GC) receptor alpha. This imbalance causes an inability to downregulate the transcription of inflammatory cytokines despite elevated cortisol levels, which the authors term ‘systemic inflammation-induced acquired GC resistance’. However, the imbalance can be alleviated via prolonged GC administration (Meduri & Yates, 2004). Other explanations for this condition, which also appears to be the peripheral GC resistance encountered in the sepsis/shock literature (Ali et al., 1991; Chadda & Annane, 2002; Meduri & Chrousos, 1998; Molijn et al., 1995), include dysfunctional transcription of the GR gene and dysregulation of the mineralocorticoid receptor enzyme, 11-β-HSD (Pretorius et al., 2006). Activator protein AP1, a regulator of cell proliferation, inflammation and immune response similar to NF-κB, also appears to be repressible by corticosteroids via GC receptors (Reily et al., 2006).

While the GC resistance mechanism appears to provide a better rationale for the use of supraphysiological low-dose steroid levels, there are caveats. For example, repeated cycles of restraint stress – an exogenous stressor used in mice and rat models – can produce high enough corticosterone levels to compromise the overall immune response to an influenza infection, perhaps by oversuppressing cytokine production (Padgett et al., 2000). Furthermore, differential suppression of cytokines can also occur, as was witnessed in a murine influenza A/PR8 infection following restraint stress that suppressed IL-1α, but not IL-6 (Konstantinos & Sheridan, 2001). This is interesting, because both IL-1 and IL-6 can stimulate adrenal GC production.

If stress-mediated responses are present in the early stages of influenza-infected patients, and the doses of corticosteroids are employed are too high, the result might be oversuppression of the immune response and a further cytokine imbalance, essentially a forward feedback that leads to fatality. Such scenarios might explain apparent steroid treatment failure in some severe cases of SARS (Li et al., 2004; details cited in Stockman et al., 2006; Poutanen et al., 2003) and avian influenza (Hien et al., 2004). Conversely, in the absence of an a priori stress-mediated response and no exogenous steroids, a virulent influenza strain can prevent corticosteroid production with the result of uncontrolled hypercytokinaemia and poor prognosis.

Influenza and steroid treatment

Few animal studies have been reported regarding steroid use in influenza. Ottolini et al. (2003) used daily dosages of 1, 4 and 16 mg triamcinolone kg⁻¹ to treat rats infected with a strain of H3N2 (A/Wuhan/359/95). The two highest dosages were effective in reducing pulmonary lesions and suppressing IFNγ mRNA, which led to shutdown of macrophage activation. However, the lowest dose was ineffective. While all the doses are supraphysiological – 1 mg kg⁻¹ translates to 70 mg per day for a 70 kg patient, or approximately 350 mg hydrocortisone daily – this experiment appears to demonstrate that there is a relatively high threshold steroid concentration necessary for successful treatment. This is a key observation. However, what we cannot extrapolate from these experiments is the threshold steroid dosage for humans infected with H5N1 strains even though the cotton rat model is a good influenza model compared to mice and ferrets (Ottolini et al., 2005).

Reports of the use of steroids in human avian influenza treatment are limited, and the case numbers few (Table 2⇓). Confounding factors also render analysis extremely difficult. Although steroid treatment has not appeared to alter mortality rates, there are several factors that must be borne in mind. First, the steroids were often given in high doses, some 7–8 times higher relative to the recommendation of Annane et al. (2004). For example, Hien et al. (2004) treated four H5N1-infected patients with 5 mg methylprednisolone kg⁻¹ daily. Another three patients were treated with 6 mg methylprednisolone kg⁻¹ daily. Second, to be effective in an immunomodulation capacity, steroid treatment must be initiated relatively early and for 10 days minimum. This was illustrated in SARS patients in whom it was found that 7–10 days of treatment was needed to alter cytokine levels (IL-6, IL-8 and IL-10; Ng et al., 2004). Third, steroid monotherapy cannot clear the virus from patients but must be used in conjunction with an antiviral, such as oseltamivir, in a timely fashion. This did not occur in many patients (Hien et al., 2004; Kandun et al., 2006). Thus it has to be concluded that in published reports of patients treated with steroids in H5N1 cases, steroids were administered in a far-from-optimal manner.

SARS and steroid treatment

With many more cases of SARS than H5N1, of whom a substantial proportion were treated with steroids, one might think that more substantial conclusions would have been reached regarding steroid treatment. Unfortunately, this is not the situation (Gomersall et al., 2004). Recently, Stockman et al. (2006) reviewed the literature to date on the subject and decided that because of confounding factors, and the fact that no large randomized clinical trials were conducted, it was not possible to reach any conclusions. The study of Zhao et al. (2003) is illustrative. These investigators describe patients being randomly allocated to one of three groups (a later group of 60 patients was created). Group A received ribavirin and cefoperazone/sulbactam, group B received fluoroquinolone, azithromycin, IFNα and restricted steroids (80–160 mg methylprednisolone per day) 14 days after treatment was initiated, group C received the same antibiotics as group B but only some had IFN, and the same steroid regimen was only given for 2–3 days in the most severe cases, group D received levofloxacin and azithromycin with 75 % being given IFN, and an unknown number methylprednisolone for 5–14 days at a dose of 160–1000 mg per day (individual patients received varying dosages depending upon condition). Although group D fared the best, it is impossible to draw from the study the contribution of the steroids.