Study cohort

In December 2020, the Norwegian Institute of Public Health established a population-based cohort directed towards people aged 65–80 yrs (The Senior Cohort). The aim was to measure the health consequences of the SARS-CoV-2 pandemic in this age group16. Initially, 1373 adults in the age group of 65–80 years and living in Oslo, Norway were randomly selected from the Norwegian Population Register and were invited to donate blood in the first round, of whom 488 provided consent. Blood samples were obtained from 412 participants, and PBMCs, serum, plasma, and DNA were isolated. Sampling was performed pre-vaccination (Dec 8, 2020, to May 4, 2021), and after the second vaccine dose (June 13 to Oct 27, 2021). All participants provided electronic informed consent. Participant data and national registry data were linked with the Norwegian unique personal identification number to obtain COVID-19 vaccination dates from the Norwegian Immunization Registry (SYSVAK); and dates of PCR-confirmed SARS-CoV-2 infections from the Norwegian Surveillance System for Communicable Diseases (MSIS). At the time of sampling, PCR testing and reporting to MSIS was mandatory. The post-dose 2 samples from the 14 participants included in this study were collected at least 14 days post-vaccination. None of the 14 participants had a record of SARS-CoV-2 infection at the time of sampling pre- and post-vaccination and were therefore considered uninfected. Subsequent analysis of anti-nucleocapsid (N) antibodies confirmed this observation, (i.e., anti-N antibodies were not detectable). Circulating antibodies against anti-RBD of SARS-CoV-2 spike protein were quantified as described previously16. Spike-specific T cell responses were quantified according to a previous report16. The Regional Committees for Medical and Health Research Ethics Southeast (229359) approved the study.

Human peripheral blood mononuclear cells and cell stimulation

Thawed cryopreserved peripheral blood mononuclear cells (PBMC) (1–3 × 106/well) were rested for 1 h at 37 °C in complete medium (RPMI 1640 Medium, 10% fetal bovine serum, 1 mM sodium pyruvate, 1% MEM non-essential amino acids, 12 μg/mL gensumycin (all ThermoFisher Scientific, Waltham, MA, USA), and 50 nM 1-thioglycerol (Sigma-Aldrich, Merck, Darmstadt, Germany), and then stimulated for 22 h with either SARS-CoV-2 spike peptides (Wuhan-Hu-1 strain; immunogenic (130-126-700) and complete (130-127-953), 0.3 nM/ml, Miltenyi Biotech, Germany), Cytostim (Miltenyi Biotech, Germany) or medium only (unstimulated). Brefeldin A (BD Golgi Plug, BD Biosciences, San Jose, CA, USA) was added after 2 h.

Mass cytometry

Stimulated and non-stimulated samples were washed in PBS (Standard BioTools) prior to staining with Cell-ID™ Cisplatin-194Pt (Standard Biotools, 1:2000) for 10 min on ice. Cells were washed in Cell Staining Buffer (CSB (Standard BioTools)), Fc blocked (FcX, Biolegend) for 10 min at room temperature (RT) and each sample barcoded by staining with a unique mix of CD45 antibodies with three metals (Standard BioTools, Supplementary Table 1) for 30 min (RT). The individual barcoded samples were consolidated, counted (Muse® Guava®), and distributed into tubes of no more than 3 × 106 cells/tube. The combined samples were stained with the surface antibody cocktail (Supplementary Table 1) in CSB for 30 min (RT) before being fixed using FIX I buffer (Standard BioTools) and permeabilized in methanol overnight (−20 °C). The next day the cells were washed in PBS and Perm-S buffer (both Standard BioTools) and stained with the intracellular antibody cocktail (Supplementary Table 1) in Perm-S for 30 min (RT). After staining the samples were washed in CSB and fixed using freshly made 1.6% formaldehyde in PBS (Thermo Scientific, Standard BioTools) for 10 min (20 °C). The samples were centrifuged and immediately resuspended in intercalation solution (0.5X Cell-ID Intercalator in Fix and Perm buffer, Standard BioTools) for 20 min (RT). The samples were washed twice in CSB, re-consolidated, and frozen in 10% DMSO in Fetal Calf Serum and stored at −80 °C. On the day of acquisition, the samples were thawed and washed twice in Maxpar® water (Standard BioTools) and filtered through a 35 μm cell strainer. The barcoded samples were acquired on a Helios mass cytometer (Standard BioTools).

Data preprocessing

Raw mass cytometry data was bead normalized and concatenated using the CyTOF software (Standard Biotools). Initial preprocessing of normalized files involved filtering of debris through sequential gating of various Gaussian parameters, performed with the CyTOFClean package in R (https://github.com/JimboMahoney/cytofclean). Single cells were identified by gating on a bivariate plot of Ir-191 and Ir-193, followed by dead cell removal of Cisplatin-194 positive cells. Gating of single cells and exclusion of dead cells was done with a custom-written R script. Live single cells were then debarcoded using the premessa package in R (https://github.com/ParkerICI/premessa).

Unsupervised clustering and annotation

Live single cells were initially clustered by FlowSOM with a 14 × 14 SOM grid and 30 meta-clusters, using all markers excluding the cytokines. The resulting clusters were visualized on UMAP-embedded space. For a coarse-grained analysis, the expression of lineage markers was visually inspected and the 30 meta-clusters were manually merged to obtain B cells (CD19+), CD4 T cells (CD3+CD4+), CD8 T cells (CD3+CD8+), γδ T cells (CD3+TCRγδ+), NK cells (CD56+), NK T cells (CD3+CD56+), and Myeloid cells (CD3-CD19-CD56-CD11b+). We also detected a very small fraction of cells that demonstrated inconclusive marker profiles (Undefined).

In the second step, further unsupervised clustering of each of the major cell types was performed by a 10 × 10 SOM grid and up to 80 meta-clusters starting with 2 metaclusters, using all markers including the cytokines. Thereafter, 20 metaclusters in each of B cells, CD4+ T cells, and γδ T cells were manually merged into 7, 17, and 9 annotated clusters, respectively, based on antigen expression visualized in marker histograms. Cell counts in each cluster per sample were then exported for further downstream analysis. FlowSOM clustering and UMAP visualization was done with the CATALYST R package. For annotating the clusters, we chose a combination of the natural separation of the cells by visual inspection of the UMAP embeddings and monitoring canonical marker expression in marker histograms. The scheme for annotation of the major clusters and their subtypes is listed in Supplementary Table 2.

Deeper unsupervised clustering was performed for the B cells, according to the following steps:

i.

Significant differences between LR and HR were estimated at different metaclustering levels (k = 10, 20, 30, 40, 50, 60, 70, 80, and all 100 FlowSOM clusters).

ii.

Then, unique clusters at each splitting step were selected by a marker similarity match. For example, if cluster no. 1 at k = 10 was phenotypically similar to cluster no. 5 at k = 20, then these two clusters were considered to be the same cluster. This way, we were able to identify clusters that are phenotypically unique at each value of k.

iii.

The clusters significantly differing between the groups were FDR-adjusted.

iv.

Steps i–iii were repeated three more times with different FlowSOM seeds.

v.

All significant clusters across all FlowSOM runs were plotted on a UMAP plot, colored by the number of FlowSOM run. Only the clusters that appeared together in 3 out of 4 FlowSOM runs on the UMAP plot were finally reported to be significantly different (Supplementary Fig. 7).

vi.

Significant clusters were annotated by examining marker expression histograms.

Statistical analysis

Comparison of cluster frequencies between vaccine low and high responders within a given experimental condition was done by negative binomial regression of cell counts data using the total cells per sample as an offset, according to the following:

$$}_} \sim \left(\log \left(}_}\right)\right)+$$

Significant differences between low and high responders were reported by adjusting for multiple testing by a false discovery rate (FDR) cutoff of <0.1.

For testing if vaccine-induced change in the frequency of a cell cluster was significantly different between low and high responders, we modeled the cell counts data of the cluster from post vaccination samples by regressing against vaccine response category, while offsetting for the frequency of the same cluster from the pre vaccination samples, according to the following:

$$_} \sim \left(\log \left(_}\right)\right)+\left(\log \left(_}\right)\right)+$$

To test for consistency across clustering owing to the high number of clusters tested for B cells, we performed FlowSOM clustering with four different seeds and tested for differences by negative binomial regression. Only clusters that were significantly different at least three out of the four FlowSOM runs were considered biologically different and thus reported. Negative binomial regression was done with the Mass package in R (v4.3.2).