Mapping the disease course of COVID-19 survivors at ICU
To unravel the sequential recovery of the immune system throughout the course of a severe COVID-19 infection, we have analysed a cohort of 40 patients that survived a stay at the intensive care unit (ICU) at the University Hospitals Leuven, Belgium. These patients were enrolled in the COntAGIouS trial (NCT04327570) through which longitudinal whole blood samples were collected during their ICU stay (see methods and Table 1 for patient demographics and characteristics; a set of healthy controls was included for reference).
For each patient, a detailed timeline of their ICU stay was generated displaying the most clinically relevant events including onset of symptoms, hospital admission, ICU admission, start/stop of therapeutic interventions and finally discharge from ICU and hospital. If applicable, discharge from a specialized revalidation centre where the patient continued his/her recovery after their hospital stay, is also included. Occasionally, blood was drawn at additional time points (such as during bronchoscopy), which is indicated on the timeline. Once individual time points were plotted (Figure 1A), a large range in the duration of ICU stay (from 2 to 72 days) was found, indicating a highly variable recovery rate across patients. From the 40 enrolled patients, only 9 were women, corresponding to previous observations of males experiencing increased severity and hospitalization rates once infected with COVID-19 [28–30](Figure 1B). The age range and overall time spent at ICU of the admitted patients was however comparable for men and women (Figure 1B and Table 1). Within this critically ill patient cohort, we did not find a correlation between BMI and the overall duration of ICU stay (r=-0.15, p=0.35), despite >40% of the patients having a BMI above 30 (Table 1, Figure 1B) [31].
This timeline was further extended with the level of respiratory support that was required at concomitant sampling time points (Figure 1C, D). Here, respiratory support levels were classified from 0-5 where level 0 indicates no support, while level 4 and 5 indicate the need for mechanical ventilation (with level 5 indicating patients that were ventilated in prone position and/or requiring inhalation NO therapy or extracorporeal membrane oxygenation (ECMO)) (Table 1). Subsequent correlation analysis showed that the maximal level of respiratory support correlated well to the overall duration of ICU stay (r = 0.67, p = 7e-17, Figure 1C).
Identifying peripheral blood profiles using longitudinal sampling
Next, we performed deep immune profiling on collected blood samples using high dimensional cytometry by time-of-flight (CyTOF) analysis. From this data, 33 surface markers were used to map changes in composition and phenotype of white blood cells during ICU stay. All data were mapped using unsupervised gating methods and manually annotated to identify the immune cell subtypes to reconstruct potential dynamic changes. However, given the variation in duration of hospitalization at the ICU in this patient cohort, mapping immune profiles on an absolute time scale did not produce coherent results. Alternatively, we ranked samples based on the neutrophil-to-lymphocyte ratio (NLR) and defined 4 equal groups (Figure 2A-C) across all collected samples, regardless of when a sample was taken. As such, we defined 4 “NLR” stages (R1-4) ranging from severe lymphopenia and neutrophilia (profile R4, which showed the highest NLR values) to immune profiles resembling those of healthy controls, defined by the lowest NLR values (R1) (Figure 2A, B and Suppl. Figure 1). In addition, a high NLR (R4) also correlated to increased levels of the acute phase reactants C-reactive protein (CRP) and ferritin (Figure 2C, D), markers that are commonly monitored in ICU patients and define a critical state.
Integrating longitudinal immune-profiling with clinical features
Next, we assessed how the identified NLR groups correlated to a variety of clinical parameters. As indicated above, the level of respiratory support was measured at each sampling time point and correlation analysis of this feature with the 4 NLR profiles revealed that patients exhibiting R4 required increased respiratory support at that moment compared to lower level NLRs (Figure 2E). Further correlation analysis also showed that the overall clinical status of each patient upon admission to ICU, as determined by calculating the Sequential Organ Failure Assessment Score (SOFA) [32, 33], was significantly correlated to the NLR groups (Figure 2F; r=-0.23, p=0.03)[33]. We did not find any correlations of age or BMI with the NLR groups (Suppl. Figure 3A, B).
Using these groups, we reconstructed a detailed sequence of events for each individual patient and investigated how these evolved along their ICU stay (Figure 1E, 2G, Suppl. Figure 4). For 36 out of 40 patients, we observed a recovery (overall downwards shift of the NLR score) or stabilisation of the immune system (and NLR score) along their stay at ICU (Suppl. Figure 4), with an enrichment of R1 towards ICU discharge (Figure 2G). Strikingly, the R values at ICU discharge still ranged between R1-4 suggesting that the physical condition of these patients was still heterogeneous at that moment. In line with this observation, we found that patients exhibiting a higher NLR value upon discharge from ICU required significantly longer revalidation (either at the regular hospital ward and/or a specialized revalidation center) following their stay at ICU (Figure 2H, r=-0.4, p=0.02). Additionally, Charlson comorbidity indices [34] were low for all patients (Table 1), indicating that the majority of patients admitted to ICU were in good general health before their infection with SARS-CoV2. We also did not observe correlations of the Charlson comorbidity indices to either the NLRs or the required level of respiratory support (Suppl. Figure 3C, D).
Finally, we tracked alterations in patient’s immune profiles who were treated with diverse immunomodulatory regimens. Commonly, the glucocorticoid methylprednisolone (MP) was given as a monotherapy (n=19). In some cases, MP was prescribed in a combination with either anti-IL-1 (anakinra; n=3) or anti-IL-6 (tocilizumab; n=5). One patient received anti-IL-6 exclusively, while the remaining 12 patients in our cohort relied on supportive care alone without immunomodulatory treatments (Table 1; Figure 1F). Unsurprisingly, our data analysis confirmed that patients who underwent steroid treatment, had a greater tendency towards a longer ICU stay. Correlation analysis of the NLR profiles with MP treatment, revealed that patients receiving MP exhibited a higher NLR upon ICU discharge, as opposed to patients that did not receive MP (Suppl. Figure 3E). Finally, while the cohort of patients that received anti-IL-1 and anti-IL-6 were small, the reconstitutional trajectory of their immune systems followed a similar pattern to those patients receiving standard-of-care treatment without immunomodulatory drugs and/or MP.
Reconstructing the cellular recovery of critically ill COVID-19 patients
Considering that the NLR levels correlated well with the clinical condition of the patients (see above), we next investigated the recovery of more specific immune cell populations to eventually define a sequence of events. To do so, we used a dual approach: first, we performed curve fitting along the trajectory of the four NLR profiles, in which we determined the inflection points of each identified immune cell population to estimate the moment recovery would begin (Figure 3A, B, Suppl. Table 2). The data from this approach were subsequently combined with the statistical comparison of the 4 NLR groups (Figure 2A, suppl. Table 3) to define the eventual order by which specific cell populations were recovering. We found that, on average across this cohort of patients, classical monocytes were the first immune cell population to recover followed by naive CD8+ T cells, naive and Th2-polarized CD4+ T cells. As indicated by our analysis, effector and memory T cell populations were only restored at later stages, including the anti-viral Th1-polarized CD4+ T cells [35]. From the other professional antigen presenting cells, myeloid dendritic cells (mDC) were recovering immediately after the early T cell response, while plasmacytoid DCs (pDC) would only recover towards R1, similar to overall B and NK cell populations. This suggests that monocytes, as implied by other studies, could become important targets in both understanding COVID-19 disease progression, as well as, improving recovery at the early phase of ICU admission [36].
Functional recovery of monocytes
Antigen-presentation by monocytes, pDCs and/or mDCs plays a vital role in the initiation of an efficient adaptive immune response to viral infections [37–41]. As previously shown, declined levels of HLA-DR expression and hence a reduced antigen presentation seems to be an early hallmark of a trajectory towards a severe COVID-19 disease course compared to a mild/moderate one [10, 12]. Considering that monocytes are among the first immune cell types to recover, we further investigated the functionality of these cells across the 4 NLR profiles during recovery. As such, we observed a regain in both the numbers and the antigen presenting capabilities of the monocytes towards discharge (figure 3C). The first subset to re-establish were the classical HLA-DR+ monocytes (Cluster (Cl)3, Cl33 and Cl12; figure 3C), while it was only in later stages of disease progression that the non-classical HLA-DR+ monocytes were restored (see Cl11, Figure 3C; see also statistics in Figure 2A).
The early phase of severe COVID19 is characterized by a functional shift towards a more immunosuppressive spectrum of monocytes, as seen by a downregulation of HLA-DR and an enrichment of CD163+ monocytes in R4 (Figure 3) [10, 12]. However, longitudinal follow-up shows that this shift is reversible, as seen by the recovery of HLA-DR expression and a relative reduction of the amount of immunosuppressive CD163+ monocytes ( see Cl53 and Cl32; Figure 3C), leading to a regain of the antigen-presenting phenotype of these cells.
Longitudinal cytokines profiling
Next to immunophenotyping, we also performed multiplexed analysis of soluble serum proteins, which included 32 pro-inflammatory cyto- and chemokines in serum samples at concomitant sampling times. The levels of these analytes were subsequently correlated to the NLR profiles. In line with previous observations, the initial R4 stage of patients arriving at ICU was characterized by an increase in IFN-ƴ, TNF-α, IL-2, IL-6, IL-7, IL-10, IL-15, IP-10, MCP-1, MIP-1α, MIP-1β levels and a reduction in TARC, MDC and IL1-α, suggestive of a pro-inflammatory cytokine signature, often referred to as a “cytokine storm” (Figure 4A). This signature was steadily reversed as patients attained a normal R1 stage. Similarly, CRP and ferritin were also significantly higher in R4 and gradually decreased as patients reached R1 suggestive of a return to baseline following an acute phase induction of the immune system (Figure 2D,E). Clustering analysis to uncover patterns in cytokine profiles across the various NLR groups further confirmed that pro-inflammatory cytokine levels collectively reduced along the duration of an ICU stay (Figure 4B). On the other hand, MDC and TARC, two constitutive chemokines designated as CCL17 and CCL22 that are regulated at a post-translational level, increased to normal levels [42].
Even though the levels of cytokines within the R1 profile remained comparable between COVID-19 patients and healthy individuals, we observed altered expression levels of TNF-α, IL-18, IL1-α, MCP-1 and MIP-1α, between the two study groups(Figure 4A; healthy controls are indicated as hollow black circles in the figure). It remains to be studied how long it would take for levels to normalise, however, in spite of their aberrant profiles, patients were still able to leave the ICU for their recovery.
Integrative mapping of the immune response during recovery
Considering the important interplay between immune cells, cytokines and chemokines during severe COVID-19, we also performed an integrative, similarity matrix-based correlative statistical modelling analyses (Figure 5), to uncover associative patterns along the duration of a patient’s recovery at ICU. As recently determined, critically ill COVID-19 patients are characterized by a strong dysregulation of the immune reaction against the SARS-CoV2 virus, whereby normally highly concerted interplays of cyto/chemokines and specific immune cell populations are now disentangled [8]. In line with above observations, our similarity matrix analyses also revealed an intense correlation between various immune cells and cyto/chemokines (a putative indication of ongoing immunological interactions) in the R4 critically-ill COVID-19 patients (Figure 5A), such that these correlations progressively “normalized” or decreased in terms of number of intense clusters when traversing from R4 to R1 (Figure 5A-D), thereby indicating that apart from above quantitative shifts, there was also a qualitative shift in possible immune-interactions when going from critically-ill to recovering COVID-19 patients.
Furthermore, the immunological characteristics of these clusters within the four blood profiles were of particular interest considering the importance of both neutrophils and lymphocytes in marking the health status of COVID-19 patients. Interestingly, neutrophils exhibited proficient correlation with mainly adaptive immunity-relevant cytokines (e.g., IFN-ƴ, IP-10/CXCL10, IL-2, IL-13, IL-15), pro-neutrophilic IL-8 and CD163+ intermediate-monocytes in R4 (Figure 5A). However, in R3, apart from the above-mentioned cytokines, neutrophils gained correlation with TNF-α, IL-10, MCP-1/CCL2, ƴδT cells, IL-17 and effector/naive CD4 T cells, a sign of slightly better (albeit pro-inflammatory) cross-talk with adaptive immune cells and some degree of immunoregulation (IL-10/IL-17) (Figure 5B). However, the neutrophil-centred clusters considerably “contracted” in R2/R1 (Figure 5C-D); ultimately resulting in a more neutrophil-specific immune-cluster consisting of the pro-neutrophilic chemokine (IL-8) and inflammasome footprint (IL-1β/IL-1Rα), a potent sign of anti-pathogenic neutrophil activity [43].
On the level of lymphocytes, the most striking phenotypes were observed in the R4 profile (Figure 5A) whereas patterns in R3/R2 were largely transitional and ultimately culminated into “contracted” lymphocytic correlative-compartment in R1 (Figure 5B-D). More specifically, in R4 patients (Figure 5A), the typically infection-resolving lymphocytic compartment (e.g. Th1/Th2 cells, effector/memory CD4/CD8 T cells) had relatively less correlations with various effector-function cytokines including IFN-ƴ, TNF-α, IL-2, IL-15, thereby indicating a certain degree of immunological dysregulation. Most of these lymphocytes also had considerable negative correlation with neutrophils. Interestingly TNF-α, Th17 and Tregs formed a cluster together which might be a sign of pro-inflammatory signalling since Tregs and Th17 have been shown to reciprocally stimulate each other via TNF-signalling pathway [44]. We believe that in the current context, this crosstalk might play a disease-potentiating role in COVID-19 severity. These discords were largely ameliorated from R3 to R1 (e.g., Th17 gaining correlation with Th1/Th2/Tfh and activated-CD8 T cells in R1) thereby indicating better lymphocytic activity/regulation to be beneficial for recovery of COVID-19 patients.
In conclusion, above qualitative analyses revealed that neutrophil/lymphocyte-associated inflammation undergoes considerable changes in terms of co-associative immune-components between R4 to R1 such that a relatively contracted neutrophil cluster with a chemotactic/inflammasome-footprint and a contracted lymphocyte cluster defines favourable recovery for COVID-19 patients.