COVID-19 is an acute illness caused by the SARS-CoV-2 virus. Clinically and immunologically COVID-19 is heterogeneous1. Around 10% of SARS-CoV-2 infections develop a more severe illness, with one or more organ dysfunction requiring intensive care (ICU) admission and organ support (referred to as critical COVID-19 in this manuscript). Whilst it is postulated that the time from onset of SARS-CoV-2 infection to developing critical COVID-19 varies between patients, studies to-date have not tested the hypothesis as to how the early to late immune responses change in critical COVID-19. We tested this by studying the longitudinal changes in immune responses in multiple immunological domains amongst patients with critical COVID-19, where such assessments are limited. We reasoned that early assessments are ICU admission assessments, but immunologically patients may be at different stages in the evolution of their responses, which will be highlighted with longitudinal assessments. Patients with critical COVID-19 patients have numerous clinical risk factors such as older age, male sex, comorbidities including immune comorbidities2, which are confounders when studying immune responses, which we address with our control population. Further, we report an integrated analysis of longitudinal multidomain immunological data to report dominant factors explaining immunological heterogeneity, using multi-omic factor analysis (MOFA)3.
Our cohort study consisted of adults with critical COVID-19 without any immune comorbidities as defined in the acute physiology and chronic health evaluation score (APACHE II score). Blood samples were collected from critical COVID-19 patients and patients undergoing elective uncomplicated cardiac surgery, pre-pandemic to use as demographically similar controls for comparisons (referred to as controls in this manuscript), after securing ethics approvals and informed consent (ref:19/SC/0187)4. Our cohort consisted of 28 patients with critical COVID-19, and 30 controls (Table-1). The longitudinal multidomain immunological assessments in critical COVID-19 patients were at four clinically relevant time points (admission, day-3, day-5 and ICU discharge day). At each time point we measured, plasma cytokines (n=28), immune cell subsets (n=30) identified using mass cytometry and pan-leukocyte transcriptome (Figure-s1).
Table 1: Baseline characteristics of acute COVID-19 cohort and healthy controls
Characteristic
|
Acute COVID-19
(n=28)
|
Control (n=30)
|
Age (years), median (IQR)
|
59.5 (8.25)
|
70 (10.75)
|
Female, n (%)
|
7 (25%)
|
8 (26.7%)
|
Weight (kgs), median (IQR)
|
82.5 (18.5)
|
81 (19.75)
|
Pre-existing conditions
Hypertension
COPD
Diabetes
Renal disease
CVA
Congestive cardiac disease
|
16 (57.1%)
6 (21.4%)
10 (35.7%)
1 (3.6%)
1 (3.6%)
0 (0%)
|
17 (56.7%)
6 (20%)
7 (23.3%)
3 (10%)
3 (10%)
1 (3.3%)
|
APACHE II score, median (IQR)
|
14.5 (4)
|
N/A
|
SOFA score, median (IQR)
|
4 (2)
|
N/A
|
CRP (mg/L), median (IQR)
|
171.5 (97.25)
|
N/A
|
Albumin (g/L), median (IQR)
|
35.5 (4)
|
N/A
|
Bilirubin (umol/L), median (IQR)
|
10 (5.75)
|
N/A
|
Creatinine (umol/L), median (IQR)
|
72.5 (24.5)
|
N/A
|
Outcomes
ICU LOS (hrs), median (IQR)
ICU mortality, n (%)
Hospital LOS (days), median (IQR)
Hospital mortality, n (%)
|
462.5 (653.25)
4 (14.3%)
25 (33.75)
5 (17.9%)
|
N/A
N/A
N/A
N/A
|
Longitudinal changes in cytokines, and histones
Compared with controls (Figure-1a & Figure-s2), patients with critical COVID-19 had higher levels of CXC and CC family of chemokines (CXCL10, CXCL11, CCL4), B-cell survival factors (BAFF, APRIL), interleukins (IL-6, IL-9, IL-10, IL-13, IL-17A) and lower levels of chemokines CXCL5 and CCL17, on admission day (Figure-1b), and these differences largely persisted at ICU discharge, albeit at lower magnitude (Figure-1c), with the addition of interferon gamma (IFN-γ) elevation. Comparing ICU discharge with ICU admission timepoint, CXCL10 was lower at discharge (Figure-1d). Correlations between cytokines were positive, and significant, at all-time points (Figure-1e), consistent with our previous report5. On admission day there were two notable negative correlations. IL-6 was negatively correlated with CCL3, CXCL5, Il17A, IL13, and IL9; and CCL4 was negatively correlated with BAFF, IL17A, IL13, IL9, IL2, and IL17F. Similarly, at ICU discharge, CCL3 was negatively correlated with IFNG, IL.17A, IL-9, IL-2, IL-5, IL-13, IL-6. The radar plot highlights abnormal cytokine levels even at ICU discharge, when critical COVID-19 patients were considered ‘clinically’ better (Figure-1f). Compared with controls, patients with critical COVID-19 had higher histones (H3.1 and H3R8) at ICU admission day, which increased further over time, before normalising by ICU discharge (Figure-1g, 1h), implying ongoing neutrophil activation and degranulation6.
Longitudinal changes in immune cell subsets
Unsupervised clustering and normalised expression of markers mapped with standardised human immunophenotyping for the Human Immunology Project7 identified 5 innate and 25 lymphocyte subsets (Figure-2a-h). Whilst longitudinal changes differed in these subsets, there were no discernible patterns across subsets. Both myeloid and plasmacytoid DCs are lower at all timepoints compared with controls (Figure-2i). Classical monocytes and CD16+ NK cells remain unchanged across all timepoints (Figure-s3). CD16- NK cells are higher across all timepoints compared with controls (Figure-2i). Amongst the eight B cell subset comparisons, patients with critical COVID-19 had higher proportions of plasmablasts, lower proportions of class switched memory and IgM+ memory, lower proportions of transitional B cells, higher proportions of activated naïve B cells, on admission day measurements compared with controls (Figure-2j & Figure-s3). Changes over time differed by different B cell subsets, with plasmablasts decreasing, memory B cell subsets increasing, and naïve B cells proportions changes were minimal (Figure-2j & Figure-s3). The longitudinal changes in BAFF and APRIL provides biological plausibility to the observed B cell subset changes8. Amongst the nine CD4 subsets, compared with controls, patients with critical COVID-19 had similar proportions of effector memory, central memory, Treg memory, and naïve cells (Figure-2k and Figure-s3), that showed minimal changes over time. We observed higher proportions of exhausted naïve cells (Figure-2k) across all ICU timepoints. Memory Tregs decreased over the course of ICU stay, from admission day samples to discharge (Figure-2k). Amongst the eight CD8 subsets, compared with controls, patients with critical COVID-19 had higher proportions of exhausted activated effector memory cells, exhausted effector memory cells and naïve cells, and lower proportions of central memory and effector memory cells on admission day measurements (Figure-2l & Figure-s3). The exhausted activated effector memory cells and the naïve cells increased over time in patients with COVID-19 (Figure-2l & Figure-s3).
At all ICU time points in patients with critical COVID-19 compared with controls, the expression of the activation marker HLA-DR was lower, and the exhaustion markers PD1/PDL1 were higher in all immune cell subsets, and often within the same subset (Figure-2m-p and Figure-s3), which has a well described association with higher IL-6 levels9, 10 and high IL-6 levels were observed our data. Amongst the B cell subsets, at all ICU timepoints compared with controls, patients with COVID-19 had lower CD45RA, a marker of pre-B and B cells prior to differentiation towards plasma cells11, CCR6, a marker of maturation and activation12, CXCR5, responsible for migration of B cells in secondary lymphoid organs13, and CD25, a marker of maturation and memory and cells with a better proliferative and antigen-presenting capacity14 (Figure-2n & Figure-s3). CCR7, a marker of two functionally distinct memory subsets15, was higher in admission day, day 3 and day 5 samples compared with control and discharge day samples (Figure-s3). Amongst the T cell subsets, in both CD4+ and CD8+ cells, CD127, a marker of memory and effector populations16controls (Figure-2o,po and and Figure-s3).
Longitudinal changes in pan-leukocyte transcriptome
We had hypothesised that longitudinal sampling will highlight time course in immune responses, but also cluster by similarities across time-points. Principal component analyses of pan-leukocyte transcriptome shows that controls group separately to patients with critical COVID-19, regardless of sequencing batch (Figure-3b). The top 1000 most variable genes highlight 3 clusters (Table-s1 & Figure-3c) that separate out on PCA (Figure-3b), with 87% (26/30) of patients in cluster 1. All time-points in all patients with critical COVID-19 were represented either in cluster 2 or cluster 3. Cluster 2 consists of mostly ICU admission day and day 3 samples from patients with critical COVID-19. Cluster 3 consists of mostly ICU day 5 and discharge samples from patients with critical COVID-19. Comparing critical COVID-19 samples with controls, on admission day there were 2472 DEG, with 1537 upregulated and 935 downregulated genes (Figure-3d). For longitudinal changes, when comparing critical COVID-19 admission day samples with other time points, by day-3 there were 132 DEG, with 75 upregulated and 57 downregulated genes (Figure-3e), by day-5 there were 441 DEG, with 273 upregulated and 168 downregulated genes (Figure-3f) and the largest differences were seen at ICU discharge, with 1034 DEG, with 547 upregulated and 487 downregulated genes compared to ICU admission (Figure-3g). From comparisons made between admission day COVID-19 samples with controls (Figure-3d), the DEG mapped to pathways involved in immunoglobulin production, adaptive immune responses, and mitotic spindle assembly checkpoint signalling (Figure-3h). Most genes in these pathways were upregulated, and the biological inference being pathways associated with initiation of an immune response following infection. Enriched pathways differ between admission vs. control, and discharge vs. control comparisons (Figure-3i & Figure-s4). The number of patients in each cluster based on their clinical time point is highlighted in Figure-3j and Table-s1.
Immunological features of transcriptomic clusters
Immunologically, the transcriptomic clusters within critical COVID-19, highlight early and later immune responses, with early represented by cluster 2. Compared with cluster 1, there was elevated plasma inflammatory cytokines in both cluster 2 and cluster 3 (Figure-4a,b). There were no significant differences in the cytokine profile between cluster 2 and 3 (Figure-4b), explained by the non-synonymous relationship between gene and protein levels. Compared with cluster 1, in cluster 2 and 3, the proportion of T cells and innate cells (CD3-CD19-) were lower while B cells were higher (Figure-4c). The proportion of gamma delta T cells (γδ T cells) was lower in cluster 3 compared with cluster 1 and 2. Most T cells were alpha beta T cells (αβ T cells), and their proportion was lower in cluster 2 compared with cluster 1 (Figure-4d). The proportion of CD4+ T cells was lower in cluster 2 compared with cluster 1, while in CD8+ T cells, cluster 3 was lower compared with cluster 1 (Figure-4d). In the innate cell populations, the proportion of dendritic cells was lower in cluster 2 and 3 compared with cluster 1 and the proportion of NK cells was lower in cluster 3 compared with cluster 1 (Figure-4e). Compared with cluster 1, cluster 2 had 2616 DEG (1162 upregulated, 1454 downregulated) associated with enrichment in early innate, and adaptive immune response pathways amongst others (Figure-4f). Comparing with cluster 1, cluster 3 had 1152 DEG (908 upregulated, 244 downregulated) associated with enrichment in reactive oxygen and inflammation pathways, suggesting these patients remain in an altered immune state despite their clinical improvement (Figure-4g). Compared with cluster 2, cluster 3 had 1113 DEG (956 upregulated, 157 downregulated), associated with enrichment in metabolism and apoptosis pathways (Figure-4h). Alluvial plot highlights a transition from cluster 2 to cluster 3 over clinical time (Figure-4i), with no patients transitioning from cluster 3 back to cluster 2. Of the 5 deceased critical COVID-19 patients, 2 were assigned to cluster 3 throughout, while 2 transitioned from cluster 2 to cluster 3, suggesting this cluster is associated with a worse outcome.
Multi-omic factor analysis (MOFA) distinguishes clusters based on interferon signalling genes
MOFA identified 15 factors responsible for 78% of variation, 70% of which was within RNA (cytokines 1%, cells proportions 7%). Within RNA, factor 1 accounted for around 18% of variance (Figure-5b), had the largest association with cluster designation (Figure-5c) and clearly separated cluster by factor value (Figure-5d). Gene set enrichment analysis indicated that genes with positive weights were enriched for factor 1, these included genes involved in interferon and immune system signalling (Figure-5e). Investigation of gene expression associated with factor 1 showed cluster 2 had the highest RNA expression of SIGLEC1, OAS3 and IFIT with cluster 3 having the lowest expression (Figure-5f). In agreement with previous studies, our results show that impaired type 1 interferon (IFNα/β) responses correlate with worse immune state in COVID-1917.