Downregulation of MALAT1 is a hallmark of tissue and peripheral proliferative T cells in COVID-19

Abstract T cells play key protective but also pathogenic roles in COVID-19. We studied the expression of long non-coding RNAs (lncRNAs) in COVID-19 T-cell transcriptomes by integrating previously published single-cell RNA sequencing datasets. The long intergenic non-coding RNA MALAT1 was the most highly transcribed lncRNA in T cells, with Th1 cells demonstrating the lowest and CD8+ resident memory cells the highest MALAT1 expression, amongst CD4+ and CD8+ T-cells populations, respectively. We then identified gene signatures that covaried with MALAT1 in single T cells. A significantly higher number of transcripts correlated negatively with MALAT1 than those that correlated. Enriched functional annotations of the MALAT1- anti-correlating gene signature included processes associated with T-cell activation such as cell division, oxidative phosphorylation, and response to cytokine. The MALAT1 anti-correlating gene signature shared by both CD4+ and CD8+ T-cells marked dividing T cells in both the lung and blood of COVID-19 patients. Focussing on the tissue, we used an independent patient cohort of post-mortem COVID-19 lung samples and demonstrated that MALAT1 suppression was indeed a marker of MKI67+ proliferating CD8+ T cells. Our results reveal MALAT1 suppression and its associated gene signature are a hallmark of human proliferating T cells.

Long non-coding RNAs (lncRNAs) are regulatory non-coding RNAs, longer than 200nt. In most cases, lncRNAs show low-medium expression with poor conservation across species often acting as scaffolds for recruitment, sequesters for chromatin-modifiers, or RNA binding proteins to specific genomic sites [17,18]. LncRNAs may be cis-or trans-acting wherein the former influences transcription by affecting the loci near their transcription site (enhancer-like) while the latter transcripts leave the transcription site to affect gene expression (mRNA-like) via transcriptional or post-transcriptional mechanisms [19].
LncRNAs play essential roles in adaptive immunity, particularly in lymphocyte activation, signaling and effector functions [20]. For example, lncRNAs such as lncHSC-2 commit HSCs to lymphoid specification as B or T cells [21]. T-cell development is regulated by Notch1 signaling [22] whose expression is in turn regulated by the lncRNA NALT1 [23]. As T cells mature, their activation is triggered by T-cell receptors (TCRs) upon MHC-mediated antigen presentation that is further modulated by co-stimulatory or co-inhibitory ligands. This activation leads to a switch to glycolysis [24] which is in turn is influenced by the lncRNA PVT1 [25,26]. Indeed, sets of lncRNAs specifically regulate lineage-specific gene expression in activated T cells [27] such as Th1 [28], Th2 [29], Th17, [30] and Treg [31] programs.
Profiling lncRNA expression in immune cells during the response to infection can provide insights into key transcriptional and post-transcriptional mechanisms operating in health and disease. Of note, even though the transcriptomes of tissue and peripheral T cells during responses to infection, and more specifically SARS-CoV-2 have been extensively studied [32], the study of T-cell lncRNA profiles has been limited [33,34].
We explored T-cell lncRNA profiles from three publicly available datasets from individuals with COVID- 19 identifying several lncRNAs that are detectable in lung T cells during infection. We particularly focused on MALAT1, a long intergenic non-coding RNA (lincRNA, a sub-class of lncRNAs) remarkably conserved in vertebrates [35]. LincRNAs such as MALAT1 do not overlap with proteincoding genes and can have various regulatory effects on gene expression [36]. Localized in nuclear speckles [37, MALAT1 is known to work in a variety of ways, such as through binding splicing factors [37], controlling the function of proteins involved in transcription [38], miRNA sequestration [39], and associating with proteins [40]. MALAT1 has been associated with positively regulating cell cycle progression in cancer tissues [41] a loss of which impairs cell proliferation [23] MALAT1 has been shown to regulate T-cell function, predominantly in animal models of infection or immunopathology [42][43][44][45]. In a previous study, in CD4+ T cells, we reported that MALAT1 downregulation is a hallmark of naïve CD4+ T-cell activation and that MALAT1−/− CD4+ T cells express lower levels of IL-10, an anti-inflammatory cytokine resulting in enhanced inflammation or immunity in experimental models of leishmaniasis and malaria [46].
Here, we examined COVID-19 single-cell RNA sequencing (scRNA seq) datasets from bronchoalveolar lavage (BAL) [47,48], explant/post-mortem lung cells [49], and peripheral blood [50] and discovered that MALAT1 was negatively correlated with cell cycle progression and proliferation in CD4+ and CD8+ T cells of severe COVID-19 patients. Performing RNAscope on COVID-19 post-mortem lung tissue from individuals who died of COVID-19, we confirmed that MKI67expressing CD8+ T cells had lower levels of MALAT1 mRNA in situ. Overall, our findings reveal that MALAT1 expression in T cells from COVID-19 patients is linked to a specific gene signature and that low MALAT1 expression is a hallmark of proliferative T cells.

MALAT1 is differentially expressed in CD4+ and CD8+ subpopulations
We integrated T-cell BAL scRNAseq datasets [47,48] to look at highly expressed lncRNAs in T cells from healthy volunteers and individuals with COVID-19 (Methods; Fig. 1A). We found MALAT1 to be the highest expressed lncRNA with similar distribution in both datasets which is ubiquitously found across all T cells (Fig. 1A). We then normalized and integrated the two datasets (see Methods) and clustered them at a low resolution to infer coarse-grained T-cell heterogeneity ( Fig. 1B; Supplementary Fig. S1). Cells visualized on UMAP showed both the datasets to be similarly spread across UMAP space indicating similar composition (Fig. 1C left). We then used cell type metadata [47,48] to obtain a finer-grained T-cell phenotyping (Fig. 1C middle). We found that there were marked differences between T cells based on disease severity (Fig. 1C right).
Importantly, we found that MALAT1 is differentially expressed within unbiased clusters, especially cluster 2 (Fig. 1D left) and within imputed T-cell subpopulations. We found that Th1 cells (CD4_TH1, inflammation-associated TNF/IFNγ expressing effector cells) demonstrate lower MALAT1 levels with respect to naïve CD4+ T cells (CD4_N, immature cells with no exposure to cognate antigen), confirming previous findings in mouse Th cells [46]. CD4_Treg (regulatory T cells) showed the highest MALAT1 levels. We also observed differences in MALAT1 expression within CD8+ T cells, with CD8_RM subset showing the highest MALAT1 expression compared to all other subsets. The difference in MALAT1 expression between CD8_RM (tissue-resident memory CD8) and CD8_EM/CD8_EMRA (memory cells/recently activated memory cells in periphery) may mark how a memory T cell is poised toward tissue homing [51]. While exhausted CD8+ T cells (CD8_EX, activated cells with exhausted effector function) had a lower median value of MALAT1 than naïve CD8+ cells (CD8_N), this was not significant. However, compared to CD8_EM, CD8_EX had lower MALAT1 levels. It is notable that the lower quartile of CD8_EX cells was the lowest among all CD8 subsets (Fig. 1D right). In our data integration (see Methods), we retained cell cycle genes, as MALAT1 has been previously linked to the cell cycle [23,41]. In doing so, and as suggested [47,48], we found cluster 2 ( Fig. 1B) to be a mix of CD4_TH1 and CD8_EX T cells (Fig. 1C, middle panel). Interestingly, MALAT1 expression was reduced in T cells from BAL from severe patients in both the datasets (Fig. 1E), although we note that this may be biased due to the low proportion of cells from non-severe patients (Fig. 1F). Interestingly, among the top 10 highly expressed lncRNAs ( Fig. 1A; Supplementary Fig. S2) only MALAT1 seemed to be down-regulated in severe cases with respect to both healthy, mild/moderate cells ( Fig. 1E versus Supplementary Fig. S2).

MALAT1(anti-)correlated gene lists identify CD8 + T EX CD4+ T TH1 Cells
To understand the effect of variability in MALAT1 expression (Fig. 1D) across coarse-and fine-grained T-cell heterogeneity we looked at how MALAT1 gene expression correlated against all other genes across all T cells, or only CD4+ T cells or CD8+ T cells, respectively. Keeping a significance score of P = 0.05 and the positive correlation value > 0.1 or negative correlation value < −0.1 as a cut-off, we found that ~80% of the genes that significantly co-vary with MALAT1 are those anti-correlated to its expression (all T cells, Fig. 2A). This percentage is ~88% for CD4+ T cells and ~65% for CD8+ cells when correlations were calculated separately for CD4+ and CD8+ cells ( Fig. 2A).
Upon analyzing the intersection of these gene lists we found that out of the genes that correlated positively with MALAT1 for CD4+ and CD8+ cells, there were 79 genes that were common between the two T-cell types while there were over four times as many uniquely MALAT1 correlated genes in    Fig. 2B). While CD4+ and CD8+ cells shared a high number of genes that were anticorrelated to MALAT1, the number of genes that were unique in anti-correlation lists were four times as many in CD4+ T cells than CD8+ T cells (Fig. 2B).
We next identified whether the top 25 MALAT1-correlated and top 25 MALAT1-anti-correlated genes (based on correlation value) in both CD4+ T cells (Fig. 2C) and CD8+ T cells (Fig. 2D) were differentially expressed in clusters identified previously (Fig. 1D  In a similar manner in CD8+ T cells, cluster 2 is characterized by genes that are MALAT1 anti-correlated (bottom 25 genes, Fig. 2D). When grouped by T-cell subpopulations, CD8+ T EX -cells appeared to be enriched in the MALAT1 anti-correlated signature (Fig. 2D). When grouped by cluster identities, MALAT1 anti-correlated genes were expressed in cluster 2 of CD8+ T cells (Fig. 2D) as with CD4+ T cells (Fig.  2C). Interestingly, CD8+ T EX -population appears heterogeneous in terms of expression of MALAT1 anti-correlating and correlating genes (Fig. 2D) which may explain why MALAT1 expression is not significantly different between CD8 T N and CD8 T EX cells (Fig. 1D right). In addition, MALAT1 correlated signature is enriched in the resident memory subset (CD8+ T RM ) and effector memory (CD8+ T EM ) sub-populations (Fig. 2D).
Importantly, MALAT1 is anti-correlated with MKI67, a commonly used marker of T-cell proliferation which is really a graded marker of the same and also marks T cells that may have recently divided [52], in both CD8+ and CD4+ and its expression is increased in cluster 2 ( Fig. 2C and D) potentially indicating the proliferative nature of cells in this cluster. Overall, these findings identified a core gene signature that anti-correlates with MALAT1 expression in T cells and indicated that these genes were highly expressed in proliferative CD4_TH1 and CD8_EX cells.
MALAT1 anti-correlated genes include a core proliferation and cell-specific signature in T cells Next, we used STRING-DB to perform network analysis for the top 100 genes (corresponding to approximately the top 25th percentile of all correlation values) that anti-correlate with MALAT1 in both CD4+ and CD8+ T cells as networks (Fig. 3A). Upon clustering these using k-means (k = 3), the resulting clusters showed FDR corrected enrichment for "Cell Division", "Oxidative phosphorylation," and "Response to Cytokine" (Fig. 3A).
Next, we investigated the gene lists using gene set enrichment analysis [53] using the hallmark gene sets to look at signatures within our gene lists. MALAT1 anti-correlated genes were significantly enriched for cell-cycle targets of E2F transcription factors, genes regulated by MYC, progression through cell division (G2M) for CD4+ and CD8+ T-cells suggesting the MALAT1 anti-correlated signature might play a role in proliferation and cell cycle progression (CD4 and CD8, Fig. 3B). Further, these gene sets showed enrichment for hypoxia, oxidative phosphorylation, and glycolysis for both CD4+ and CD8+ T cell whereas genes for DNA repair were only enriched in MALAT1 anti-correlated gene list for CD4+ T cells (CD4, Fig. 3B).
Genes upregulated in response to IFN-γ and IFN-α signaling were hallmarks uniquely associated with CD4+ T cells (unique to CD4, Fig. 3B). Genes involved in complement were associated with both CD4+ and CD8+ T cells while genes associated with xenobiotic metabolism were associated with CD4+ T cells (unique to CD4, Fig. 3B). Further, MALAT1 anti-correlating genes uniquely in CD8+ T cells were enriched for genes involved in the p53 pathway and those regulated in response to TNF via NF-κB (unique to CD8, Fig. 3B).
We looked at the top 20 genes uniquely anti-correlated to MALAT1 in CD4+ T cells for STRINGDB interactions and found that genes related to response to TNF/IL-1 such as PSMA5, NFKBIA, and Ubiquitin cross-reactive protein (ISG15) (Fig. 3C). On the other hand, in CD8+ T cells, MALAT1 uniquely anti-correlates with genes associated with membrane targeting of proteins along with genes involved in CD8+ T-cell exhaustion like GNLY, GZMB, and HAVCR2 (Fig.3D).

MALAT1 and MKI67 anti-correlate in COVID-19 post-mortem lung tissue
The above cell-type gene signature and pathway analyses indicated a potential link between MALAT1 expression and T-cell proliferation. To further test this, we checked if the MALAT1 anti-correlated gene list signature common to CD4+ and CD8+ T cells (Fig. 3A) identified in BAL samples was sufficient to mark proliferating T cells in lung tissue. For this purpose, we analyzed a COVID-19 explant/post-mortem lung scRNA seq dataset [49]. We pre-filtered barcodes labeled as "T cells" from the dataset and used the top 100 common genes that are anti-correlated with MALAT1 (Fig. 3A), of which 88 genes were found in Bharat et al., to calculate the 'area under recovery curve' or AUC [54] for each cell to calculate enriched gene set activity per cell (histogram, Fig. 4A) to identify gene list enrichment. Thresholding the AUC score (at AUC >= 0.39) based on the bimodality in AUC distribution (histogram, Fig. 4A), the cells were highlighted on a UMAP plot (Fig. 4A). The high AUC score highlights proliferating T cells as indicated by their corresponding MKI67 expression in UMAP space and lower MALAT1 levels is associated with cells with high MKI67 levels (Fig. 4B).
Interestingly, the proliferative T cells in post-mortem lung tissue appeared diverse in terms of their position in UMAP space. To investigate this further, we examined a subset of these cells (AUC scores >= 0.39) and visualized canonical T-cell markers in two-dimensional UMAP space (Fig. 4C). We further re-clustered these cells (Fig. 4C, bottom right) to understand whether the heterogeneity in proliferative T cells with a high AUC score (Fig. 4A) translates in terms of differential gene expression (Fig. 4D).
We then tested whether MALAT1 expression levels were consistently lower in proliferative cells and whether this was dependent on their cell cycle state. We calculated a score based on genes involved in cell cycle progression [55] including those involved in the S, G2/M, and G1 phases. In Fig. 4E, we show how these proliferative T-cell clusters (Fig. 4C, bottom right) comprise cells in S, G2/M, and G1 phases. We then calculated Spearman's correlation between the imputed cell phase score and MALAT1 and found that MALAT1 levels are anti-correlated with the imputed S phase score and strongly positively correlated with the G2M score (Fig. 4F).
To test if our findings were limited to tissue T cells, we examined a COVID-19 PBMC dataset [50], using the abovedefined top 100 genes that anti-correlate with MALAT1 we identified a small proportion of cells within this dataset that expressed these genes differentially (99/100 genes were found in the dataset, Fig. 4G). In fact, this signature also picks out PBMCs from influenza patients, suggesting that this is a hallmark feature of T-cells responding to infection. As in the case of lung T-cells, these AUC > 0.42 cells (Fig. 4G and H) are found to be neighborly in UMAP space and express MKI67 (Fig. 4H).
To test the above findings in situ, we examined post-mortem lung sections from the UK Coronavirus Immunology Consortium (UK-CIC) (patient_meta_data, Supplementary Table S1 and Milross et al., in prep). We analyzed lung autopsy sections (n = 6) and representative sections stained with DAPI are shown in Fig. 5A. We concentrated on CD8+ T cells due to their roles in COVID-19 pathology [16] and the fact that MALAT1 expression co-varied with both proliferation and exhaustion markers in these cells (Fig. 3D). We determined CD8 expression and MKI67 (as a marker for non-senescent cells that may be in any of G1, S, G2, and M phases) by immunofluorescence along with MALAT1 by RNAScope. We found, qualitatively, that MALAT1 was seldom co-expressed with MKI67 unless MKI67 levels were high (Fig. 5B-D). Interestingly, this suggested some correlation between MALAT1 and MKI67 when the latter was more highly expressed. This may be related to the MALAT1 expression correlation we observed with the G2M phase T-cell score (Fig. 4F). We next performed quantitative analysis using QuPath (Fig. 5E). For all tested samples we observed distinct MKI67-hi/MALAT1-lo populations, with the majority of highest MKI67 expressing CD8+ T cells (mean nuclear intensity > 2000) showing low MALAT1 levels. We also found double-positive CD8+ cells that co-express MALAT1 and MKI67 (Fig. 5F). These double-positive cells may be explained by the particular phase of the cell, as it has been shown that MKI67 is not a binary marker for proliferation but a graded marker for proliferation/senescence [56]. In general, however, we found that when MALAT1 expression is high then MKI67 expression is low and vice versa (Fig.  5F) across all tested samples.

Discussion
MALAT1 is one of the most abundant non-ribosomal RNA transcripts in mammalian transcriptomes. Despite an increasing understanding of how MALAT1 upregulation contributes to cancer development and progression [57], less is known about its physiological functions in non-transformed cells. Recent work from our and other laboratories has indicated that MALAT1 plays a role in T-cell function and that, in preclinical models, antigenic activation of naïve T-cells results in suppression of MALAT1 expression [42][43][44][45][46]. Here, we used published annotated transcriptomic datasets in COVID-19 to specifically look at T-cell phenotypes ranging from naïve to effector memory and exhausted and found that MALAT1 is negatively correlated with a core gene signature in T cells, which in turn is linked to cellular proliferation. Using post-mortem lung autopsy samples, we experimentally validated this association and showed that MKI67+ proliferative CD8+ cells are characterized by low MALAT1 expression.
T-cell proliferation can be spontaneous or homeostatic [58] and the conditions that regulate the same vary between CD4+ and CD8+ T cells [59]. CD8+ T-cell proliferation is essential with rapid proliferation in response to interaction with a foreign peptide but also during homeostasis if T-cell numbers fall below a threshold [60]. The former, however, progresses through to a CD8 effector memory phenotype [61]. In fact, it has been demonstrated in CD8+ T cells, that a T-central memory phenotype is marked by a higher number of prior divisions than the effector memory T-cell pool [62][63][64]. The replicative history of T cells is closely connected to its functional repertoire [64]. Interestingly, CD8+ T exhausted cells in COVID-19 are connected via the CD8 T N , CD8+ T EM lineage (using pseudo time analysis) and have higher levels of proliferation markers [48]. Indeed, using Wauters Mol, et al., 2021 dataset, we note that CD8 + T EX have a corresponding lower MALAT1 level with increased expression of MALAT1 anticorrelating genes (Fig. 2C).
MALAT1 has been long associated with enhanced proliferation in cancer [35,65] and the lack of the gene is shown in human diploid lung fibroblasts to have a reduction in their proliferation with an arrest at the G1/S phase with an increase in genes involved in the p53 pathway [41]. Interestingly, in T cells we observe a physiological downregulation of MALAT1 that anticorrelates with the S-phase score of cells (Fig. 4F), suggesting MALAT1 suppression may be a consequence of T-cell proliferation. Interestingly, overall MALAT1 levels anti-correlate to HALLMARK_P53_PATHWAY (Fig. 3B) and is unique to CD8 + T cells. While MALAT1 in this work has been shown to anti-correlate with a cell's S-phase score (Fig.  4F), it has been shown that many lincRNAs peak during the S phase in human epithelial cells leading to transcriptional regulation during cell cycle progression [66]. In that study, it was found that MALAT1 peaks close to the beginning of G2/M [66]. In this respect, we found MALAT1 levels to correlate with G2M score in T cells (Fig. 4F), which indicates the similarity of T cells to epithelial cells in terms of MALAT1 expression during the cell cycle.
We find that more genes anti-correlate with MALAT1 than those that correlate ( Fig. 2A). Whether this is due to the direct effects of MALAT1 through its roles in gene regulation [67] will need to be further tested. However, it suggests that physiological regulation of MALAT1 levels may alter gene expression of T cells that are known for their plasticity [68]. Further still, as cell proliferation is central to T-cell activation [69,70], it will be interesting to investigate how a lack of MALAT1 during proliferation may shape T-cell function upon subsequent activation and differentiation. We have previously reported that a lack of MALAT1 results in lower levels of MAF and IL10 in mice and as a consequence, greater host resistance to infection or increased immunopathology [46]. Others have reported impaired CD8+ T-cell function upon MALAT1 loss [45]. Interestingly, MALAT1 mediates its function through interactions with proteins and potentially RNA, interactions which based on the results presented here would be expected to be altered in MALAT1-lo proliferating T cells. Genes that may be associated with shaping T-cell function post proliferation may indeed lie amid the MALAT1 anti-correlated signature that we find in CD4+ and CD8+ T cells, especially those involved in cytokine response and oxidative phosphorylation (Fig. 3A). As an example we find HAVCR2 (TIM-3) which is a marker for T-cell exhaustion [71] anti-correlates with MALAT1 (unique to CD8+ T cells, Fig. 3D). How these genes may vary between T-cell subsets such as CD8+ T-central memory where lowly divided cells are capable of mounting a better effector response upon re-infection [64] and exhausted CD8+ T-cell population with increased cell cycle markers like MKI67 [3,48] requires further investigation.
Taken together our results reveal that suppression of MALAT1 expression is a feature of proliferating activated T cells. This means that MALAT1-associated functions are likely to be suppressed in proliferating T cells, but not necessarily that MALAT1 suppression drives the proliferation. There is a long list of reports supporting that MALAT1 promotes cell proliferation at least within the context of cancer cells [72]. Based on this, we speculate that one possibility is that MALAT1 downregulation following T-cell activation can be a potential mechanism to limit uncontrolled T-cell proliferation. This however will need to be experimentally confirmed in future studies. Mechanistically, MALAT1 might affect T-cell activation, proliferation, or differentiation through its role in post-transcriptional regulation, for example through direct interaction with several RNA-binding proteins [73], many of which are involved in T-cell proliferation and differentiation [74,75]. The MALAT1-linked gene signatures identified here provide an initial insight into the potential functional consequences of MALAT1 suppression in human T cells, forming the foundation for further mechanistic studies on the function of this highly expressed lincRNA in T cells within and beyond viral infection.

Datasets
Single-cell RNA seq data from healthy and COVID-19 patients from gene expression omnibus accession number GSE145926 which is referred to throughout the paper as Liao et al. [47] and T-cell barcodes (using metadata from the original publication) were subset and used further for analysis. Dataset Wauters et al. [48] was obtained from https:// lambrechtslab.sites.vib.be/en/data-access. Specifically, the file T_NKT_cells.counts.rds was downloaded to use as counts matrix. Single-cell data for barcodes with 'COVID19' as metadata were included in the downstream analysis from the Wauters et al. dataset.
Finally, accession number GSE158127 [49] was used to analyze post-mortem T cells from the lungs and for validation. Further, GSE149689 [50] was used to look at MALAT1 signatures in PBMCs and in flu.
Cell cycle genes were not regressed prior to dimensionality reduction and downstream analysis in any of the datasets to show proliferating T cells as a separate cluster owing to their distinct cell cycle-related gene expression.

In silico T-cell quality check and phenotype identification
Single T-cell transcriptomes from Liao et al. and Wauters et al. were loaded as Seurat (v4.0.5) objects and the latter Seurat object's metadata describing T-cell phenotypes were used to impute T-cell phenotypes in Liao et al. using the functions FindTransferAnchors() and Transferdata(). Next, single transcriptomes with greater than 5% mitochondrial genes were discarded from downstream analysis. Next, counts from both Seurat objects were regressed using percentage of mitochondrial genes, ribosomal genes, total RNA count, and number of unique features using method "glmGamPoi" which is available as an R package with the same name (https:// bioconductor.org/packages/release/bioc/html/glmGamPoi. html). Finally, the anchors between the two Seurat Objects were found (functions SelectIntegrationFeatures() and FindIntegrationAnchors()) to then integrate (IntegrateData()) them into a single integrated Seurat object.

Dimensionality reduction
Principal components analysis was performed on the integrated Seurat object (3000 variable features). Top 30 PCA components were used to cluster the data by a K-nearest neighbor clustering using FindClusters() with a resolution parameter of 0.8. UMAP was performed on the PCA space and single cells were represented on UMAP axes and colored by their cluster membership.

Correlation analysis
The correlation of all genes with MALAT1 was calculated using cor.test() from the stats package in R (4.1.1) implemented with Spearman's ranked correlation method. The level of significance associated with a correlation was set at 0.05. Correlation values between −0.1 and 0.1 (both included) were excluded.

Network and gene set enrichment analysis
Network analysis of gene lists was performed on String-DBGene set enrichment analysis was performed on STRING (https://string-db.org/). Gene set enrichment analysis (GSEA, http://www.gsea-msigdb.org/gsea/msigdb/annotate.jsp) using the option to 'Investigate Gene Sets' to search for significant (P-value corrected) overlaps with Hallmark gene sets, GO biological process, cellular component, and molecular function.

Area under curve
Gene list enrichment in cells was calculated using the R package AUCell, originally published as a part of SCENIC [54]. Expression matrices as obtained from the Seurat object were provided to the function AUCell_buildRankings() to build cell rankings which were then used to calculate an 'area-under-recovery-curve' for the provided gene list. AUC score thresholds were selected based on visual inspection and are indicated in the relevant figure.

RNAScope and immunofluorescence
Post-mortem autopsy sections from UK-CIC first wave cohort (CIC003-9) were obtained on glass slides and stained for CD8, MKI67, and DAPI. MALAT1 was probed on the same section using RNAScope (Bio-techne) FISH assay as per the manufacturer's instructions.

QuPath
All images were acquired on a Zeiss AxioScan.Z1 slide scanner. Exposure times and threshold settings for all three channels were used for each of the images. Images in the CZI format were loaded on QuPath-0.3.2 [76]. Whole images were analyzed for co-expression of MKI67, CD8, and MALAT1 at single-cell resolution, and count data were analyzed. CD8 + cells were detected using the module 'positive cell detection' using DAPI as a counterstain to draw nuclei and cell boundaries. Cellular CD8 intensity was then used to detect positive cell types. Data was exported and then further investigated in R. Cells with a circularity score of less than 0.75 were excluded and expression positivity for MKI67 and MALAT1 was determined by selecting only those cells that had a maximum pixel intensity greater than the minimum detected intensity.

Supplementary data
Supplementary data is available at Clinical and Experimental Immunology online.