Discussion
In addition to re-establishing the hypo- and hyper-inflammatory COVID-19 ARDS phenotypes, our results segregate the hypoinflammatory group into low and high viral load subphenotypes. We have also transitioned to a more accurate clustering by sampling the lung compartment as opposed to the plasma. The potential advantage of our approach is obtaining hypothesis-driven clusters by linking the semi-quantitative viral loads and immune activation status. This may permit the treating physician to utilize a different balance between immune modulators and antiviral therapies as patients in the hypoinflammatory high viral load group may exhibit pathophysiology secondary to direct viral cytopathic effects rather than immunopathology. Inhibiting the immune responses further in these patients may ultimately prove detrimental.
Few studies have comprehensively described the clinical phenotypes in COVID-19 ARDS. Sinha et al confirmed the traditional hypo- and hyperinflammatory phenotypes in a cohort of 39 patients by using clinical variables and plasma IL-6 and TNFR1 concentrations (4). Cluster 2, the hyperinflammatory phenotype, was observed in 10-21% of the patients and was characterized by higher organ failure and mortality. In a subsequent study, the authors overlapped the ARDS phenotypes defined by systemic inflammation and multiorgan failure to severe COVID-19 and reassessed the presence of the 2 phenotypes by using latent class analysis (5). 14% and 19% of patients from the 2 clusters were misclassified pointing towards noteworthy overlap between the clusters. The hyperinflammatory phenotype had higher proinflammatory markers, lactate and lower bicarbonate. The response to corticosteroids (nonrandomized, various formulations) was more prominent in the hyperinflammatory group. Interestingly, the nasopharyngeal SARS-CoV-2 RT-PCR Ct was associated with mortality only in the hypoinflammatory phenotype. Our study defines the high viral load subgroup part of the hypoinflammatory phenotype which may be a contributor to detrimental outcomes.
Bos et al employed group - based trajectory models to define COVID-19 ARDS phenotypes by looking at longitudinal data (12). Subphenotype 2 (33% of the patients) was characterized by higher minute ventilation - a marker of increased dead-space-, more venous thrombotic events, acute kidney injury and higher mortality. Cytokine concentrations were not available, but IL-6 has been linked with endothelial dysfunction and thrombosis in previous studies.
The main debate in severe COVID-19 remains: what is the driver of ARDS and bad outcomes? An exaggerated immune response or uncontrolled viral replication similar to other respiratory viral infections such as RSV?(13) It is known that coronaviruses have the capability to inhibit and delay the type I IFN immune response. In turn, this will promote continued viral replication and possibly immunopathology (14). Several autopsy reports have contributed to this debate: studies confirmed the presence of SARS-CoV-2 in lung specimens while others failed to find significant direct viral cytopathic effects (6–8).
The immune inflammatory response has been sought as a viable explanation and potential treatment target in severe COVID-19. Yet, numerous studies failed to detect a “cytokine” storm fingerprint in the plasma of patients with severe COVID-19 (3, 15). Usually, the plasma cytokine concentrations were similar or even 10 times lower in COVID-19 patients than in septic or ARDS patients (3). In a comprehensive study, Wauters et al analyzed the interplay between the innate and adaptative immune responses using single cell transcriptomics on BALs and found abundant and dysregulated T cells along with hyperinflammatory monocytes in critical COVID-19 (16). Viral- RNA tracking demonstrated infected lung epithelial cells along with neutrophils and macrophages involved in viral clearance.
Several other authors have argued instead for stimulating the immune response in critically ill patients with COVID-19 and not suppressing it. Remy et al compared the T cell subsets and quantified the T cell IFN-γ and monocyte TNF-α production in patients with severe COVID-19, sepsis, critically ill non-septic and healthy volunteers (2). Patients with severe COVID-19 were characterized by impaired immune effector cell function which was partially restored ex vivo by administering IL-7. Analyzing these discrepancies and also the improved mortality with several immune modulators prompted our second hypothesis that using the bronchoalveolar lavage will give us a more accurate image of the lung-centric immunity.
In our previous study, we described the inflammatory immune dysregulation occurring in the lower airways where 28/35 cytokines had elevated levels. Among those only IP-10 BAL levels correlated with plasma levels (11). Therefore, most peripheral blood cytokine concentrations serve as inaccurate/ weak markers in quantifying the intensity of the immune response during severe COVID-19. This may contribute to the discrepancy between low blood cytokine concentrations but positive clinical response to corticosteroids and other immune modulators. The RECOVERY trial was the first randomized controlled trial to show a significant mortality reduction in patients treated with dexamethasone (17). A randomized Bayesian trial showing a trend towards improved organ support- free days was stopped early after the results of the RECOVERY trial were released (18). Given these results and the fact that the hyperinflammatory phenotype characterizes only a minority of patients (4), it results that dexamethasone improved mortality in patients with hypoinflammatory phenotype as well. This opens the hypothesis of treatment heterogeneity within the phenotypes, hypothesis that our study tried to ellucidate.
The second most studied class of immune modulators in COVID-19 are the IL-6 blockers. Many studies missed to detect a significant IL-6 concentration in plasma (5, 15, 19–21). Multiple non-randomized studies of IL-6 blockade suggested a benefit in reducing intubation rates and even mortality (22, 23) yet the initial randomized controlled trials failed to prove a benefit in moderate/ severe COVID-19 pneumonia (24, 25). Meta-analyses found that IL-6 blockers decreased the need for mechanical ventilation and mortality in both critically ill and non-critically ill patients albeit with a small effect size with mortality risk ratio of 0.89, (95% CI 0.82 to 0.96) (26, 27).
After dexamethasone became the standard of care in patients diagnosed with COVID-19 requiring oxygen, several studies tried to quantify the benefit of adding a second immune modulator agent to corticosteroids. Despite the debate on systemic inflammation, several other immune modulators acting on distinct pathways have shown additional benefits in hospitalized patients with COVID-19 treated with steroids, mainly in reducing rates of intubation. These include tocilizumab, an antagonist of the IL-6 receptor (28), baricitinib, a JAK 2 inhibitor (29), lenzilumab, a Granulocyte-macrophage colony-stimulating factor (GM-CSF) blocking antibody (30) and anakinra, an IL-1 blocker (31).
Similar to the small signal to harm in the hypoinflammatory phenotype (5), the small effect sizes and partial lack of replication point towards lack of power or heterogeneity of treatment effects and mixing responders and non-responders as possible causes (32). This may be explained by differences in semi-quantitative SARS-CoV-2 RT-PCR. So far, the SARS-CoV-2 RT-PCR tests have been developed as qualitative tests and the relationship between Ct and viral load at very low and very high viral loads is not linear. Ct thresholds have been linked to severity of illness, transmissibility and differentiating replicating virus from persistent shedding after a resolved infection (33, 34). We used the same test in all of our patients and also a very standardized bronchoscopy procedure and sampling technique. The results for theS and Orf gene targets were very similar in our sample with less than 0.5 variation in Ct. While the qualitative outputs of the nasopharyngeal swabs and BAL correlate in terms of positive and negative, no published reports establish the correspondence between Ct values in the upper vs lower airways (35, 36). As such, we elected to use SARS-CoV-2 RT-PCR Ct in the BAL as a surrogate marker for the amount of virus.
The most important limitation of our study is the small sample size. It is possible that differences between groups may have been missed. The similar distribution and signal described in other published related studies lends external validity to our results. As it has been noted, identifying groups based on theory should take precedence since clustering/ phenotyping techniques will always find groups in the data (37). A recent study by Sinha et al questions the use of machine learning without biomarker data in grouping patients to explain heterogeneity of treatment effects (38). When analyzing 3 RCT in ARDS by using 9 clustering methodologies, the authors found a wide variation to identify clusters which was only exacerbated in the absence of biomarkers. Our study pushes phenotyping one step further by using a biologically plausible hypothesis (i.e. the infection matters in explaining heterogeneity of treatment effects) and providing a measurable marker in semi-quantitative SARS-CoV-2 in BAL. Our results expand on previous studies by showing the role of SARS-CoV-2 RT-PCR in phenotyping patients with COVID-19 ARDS and providing evidence that the lower airways are the best compartment to sample for accurately assessing the intensity of the immune response. Since we do not have BAL or plasma samples prior to dexamethasone to verify the “native” clusters, it is possible that more patients could have been grouped under the “hyperinflammatory” phenotype earlier in the disease process. We provide a homogenous cohort of mechanically ventilated patients with COVID-19 ARDS but this may also be labelled as a convenience sample since these were patients undergoing bronchoscopy based on clinical reasons. Yet, only 1 patient in cluster 1 had cryptococcal pneumonia and 1 in cluster 2 had Staphylococcus aureus pneumonia, no patients in cluster 3 exhibited positive BAL cultures. It is unlikely that the use of remdesivir altered our results as previous reports did not find a difference in nasopharygeal Ct values when patients were treated with remdesivir (39). We counteracted the different timeline of sampling by standardizing all variables per week.
Conclusion: Our hypothesis-driven study adds the infection dimension to ARDS phenotypes. We identified a unique group of hypoinflammatory with high viral load patients that is biologically plausible and where additional immune modulators may have a detrimental role. Future studies are needed to establish a threshold for SARS-CoV-2 RT-PCR and to assess the immune cascade before and after administration of additional immune-modulating therapies.