Uniform manifold approximation and projection (UMAP) analysis of the 27 healthy adults, 20 older adults and 39 oncohaematologic patients at 2 different time points – before and after vaccination – (with a loss of 10 samples due to insufficient quality) identified four major continents and three smaller islands ( Figure 1A ). The relative expression of each marker from the UMAP analysis is shown in Figure 1B . The main continent on the left represents cytotoxic T-cells together with Tγδ cells, while the main continent on the right comprises helper T-cells. The smaller island on the bottom is mainly composed of B-cells and the two islands in the middle represent monocytes. Natural killer cells (NK) cells are in the top island together with innate lymphoid cells (ILCs), while the small top left island comprises mixed myeloid antigen-presenting cells.

To refine our analysis, we used the FlowSOM algorithm to find similar cell clusters and integrate them into metaclusters in an unsupervised manner ( Figure 1C ). We identified a total of 48 clusters according to the expression of the surface markers as shown in the heatmap ( Figure 1D ). Table 1 shows an in-depth characterisation of the phenotype of all clusters, which allowed the identification of 46, because clusters 38 and 39 could not be clearly identified. In the case of cluster 38, although it is close to monocytes in the UMAP plot, there is no CD14 or CD16 expression. Hence, cluster 38 could be conventional dendritic cells (cDCs) as they are negative for almost every marker except HLA-DR – although they are also CD11c^–, so their true nature remains elusive. Cluster 39 comprises intermediate (CD14⁺CD16⁺) and non-classical (CD14⁺CD16^–) monocytes together with other HLA-DR⁺CD11c⁺ that could resemble cDCs.

Of note, the same cell population can be divided into further subsets as shown in Table 1 based on the expression of several surface markers. For example, TEMRA Tγδ (CD45RA ⁺ CCR7 ^– ) cells can be divided into 7 different populations based on the expression of the surface markers CD56, CD127, CD27 and CD8 together with NKG2D, FAS and CCR6, among others. Finally, we overlaid all of clusters onto the UMAP plot ( Figure 1E ) to determine not just how the relate one to each other, but also to display their pseudoevolution.

To validate these findings, we used the hierarchical or classical gating strategy to identify different immune cell subsets, as shown in Supplementary Figure S1A , with T-cells shown in Supplementary Figure S1B . Given that some of the identified clusters can be found within the same subset ( Table 1 ; Supplementary Figure S1D ), Supplementary Figure S1C displays the required gating strategy to identify these subsets while Supplementary Figure S1D shows the identification of the 7 TEMRA Tγδ cell subsets.

Having described the global leucocyte subset composition ( Figure 1 ), we next examined the differences between the cohorts at baseline ( Figure 2A ). The UMAP plot revealed a deficit of classical monocytes in older adults as well as in lenalidomide- and ibrutinib-treated oncohaematologic patients (clusters 40 and 41 in Figure 1E ). As expected, rituximab-treated patients displayed a lack of B-cells (clusters 34–36 in Figure 1E ), except for plasmablasts (cluster 37 in Figure 1E ), of which there was a deficit in healthy adults, older adults and lenalidomide- and rituximab-treated oncohaematologic patients.

Differences according to surface marker expression within the cohorts are shown in the heatmap ( Figure 2B ). Older patients are closer to lenalidomide- and rituximab-treated oncohaematologic patients, while untreated and ibrutinib-treated oncohaematologic patients are closer to healthy adults. To validate and quantify these differences, we performed a paired comparison of each cohort relative to the healthy controls. We found that a third of the total identified clusters were differentially expressed among older adults and healthy controls: older adults displayed a significant expansion of CD4⁺ and CD8⁺ effector T-cells and a deficit of monocytes and immature T-cells ( Figure 2C ). To further confirm these results, we employed a classical gating strategy that revealed older adults have an expansion of mature NK cells together with several subsets of effector CD4⁺ and CD8⁺ T-cells, and a deficit of immature T-cells (mainly CD4^–CD8^– T-cells and naïve CD8⁺ T-cells), CCR7^–CD45RA^– Tγδ and plasmablasts ( Supplementary Figures S2A, B ). Moreover, and given that all clusters identified in Figure 1 and Table 1 do not always resemble a whole population as they can be further divided into subsets, we validated the specific clusters identified in Figure 2C by following the gating strategy displayed in Supplementary Figures S1C, D . Interestingly, even though early effector CD4⁺ T-cells did not differ between healthy and older adults, subset 1 of this population (cluster 2 in Table 1 ) was expanded in older adults, whereas its second subset (cluster 3 in Table 1 ) was reduced in older adults ( Supplementary Figure S2C ). Mature NK cell expansion in older adults ( Supplementary Figure S2A ) was specifically due to an expansion of subset 1 of mature NK (cluster 44 in Table 1 ), while the second subset of this population (cluster 45 in Table 1 ) remained unchanged. Therefore, this approach confirms the relevance of validating differences between as well as within different subsets.

We performed analyses for oncohaematologic patients based on their treatment status. Untreated oncohaematologic patients had 14 differentially expressed clusters compared with healthy adults ( Figure 2D ). Further analysis of this cohort revealed that these patients had a deficit of non-classical monocytes, TEMRA Tγδ cells, CD45RA^–CD39⁺ regulatory T-cells (T_regs) and early-like effector CD4⁺ T-cells, together with an expansion of terminal effector CD4⁺ T-cells and CD4^–CD8^– T-cells ( Supplementary Figure S2D ). Moreover, when we performed this analysis based on the specific clusters ( Supplementary Figure S2E ), we found that subset 2 of early-like effector CD4⁺ T-cells (cluster 3 in Table 1 ) was reduced, while cluster 2 of the terminal effector CD4⁺ T-cells (cluster 7 in Table 1 ) was expanded. Further analysis also revealed a deficit of subset 1 of non-memory B-cells (cluster 35 in Table 1 ) in this population together with subset 2 of CD4^– ILCs (cluster 47 in Table 1 ). Finally, analysis of TEMRA Tγδ cells revealed that subset 2 of this population (cluster 24 in Table 1 ) was decreased while subset 5 (cluster 27 in Table 1 ) was expanded.

The lenalidomide-treated oncohaematologic patients displayed specific differences in half of the analysed clusters compared with the healthy controls ( Figure 2E ). Using the classical gating strategy, we found a loss in several CD4⁺ and CD8⁺ T-cell populations, together with non-classical memory B-cells, memory B-cells and CD45RA^–CCR7^– Tγδ cells. On the contrary, these patients had an expansion of CD39⁺CD45RA^– T_regs, TEMRA Tγδ cells, and CD4^– ILCs ( Supplementary Figures S2F, G ). Further analyses revealed the specific clusters responsible for these differences ( Supplementary Figure S2H ).

For ibrutinib-treated oncohaematologic patients, 44% of the clusters were differentially expressed compared with healthy adults ( Figure 2F ). Following hierarchical gating identification of populations, it was clear that these patients exhibited a deficit of memory B-cells, non-classical monocytes, naïve CD8⁺ T-cells, CD45RA^-CD39⁺ T_regs, CD45RA^–CCR7^– Tγδ cells and total ILCs as well as CD4^– ILCs ( Supplementary Figure S2I ). On the other hand, they displayed an expansion of terminal effector CD4⁺ and CD8⁺ T-cells, early-like effector CD8⁺ T-cells, CD4^-CD8^- T-cells and TEMRA Tγδ cells. The specific clusters responsible for these differences are shown in Supplementary Figure S2J .

Moreover, we examined differences between rituximab-treated oncohaematologic patients and healthy adults at baseline. The volcano plot shows a 29% difference in the total identified clusters including, as expected, total B-cell depletion ( Figure 2G ; Supplementary Figure S2K ). In addition, these patients had a deficit of CD8⁺ naïve T-cells and an expansion of CD8⁺ early-like and terminal effector T-cells, together with CD45RA^–CD39⁺ T_regs. Further analysis revealed that although the CD4⁺ T-cell populations were not altered in this cohort, several of their specific clusters did show changes ( Supplementary Figure S2M ).

Having described the cohorts at baseline, we next studied the acquired humoral and cellular immunity 3 months after full vaccination with messenger RNA (mRNA) vaccines. We assessed SARS-CoV-2 neutralising spike protein (S) antibodies (anti-S IgG and IgA) and nucleocapsid protein (N) antibodies (anti-N IgG) in plasma before and 3 months after full vaccination (two doses of the vaccine) ( Figure 3A ). A small fraction of healthy adults (14%), older adults (5%) and oncohaematologic patients treated with lenalidomide (11.1%) and ibrutinib (7%) had anti-N IgG antibodies before vaccination, suggesting an unnoticed previous asymptomatic infection. Of note, vaccination in healthy adults and older adults triggered anti-S IgG antibodies, confirming successful vaccination; however, the percentage was lower in oncohaematologic patients and virtually absent in rituximab-treated patients ( Figure 3A ). In addition, given that vaccination was intramuscular, IgA antibody production was mostly induced in healthy adults; it was not triggered in older adults or oncohaematologic patients. It is well known that vaccination efficiency in elderly individuals is often reduced due to immunosenescence. For example, the capacity of influenza vaccines to induce immune protection is age-related, with efficacy ranging from 70% to 90% in young people but decreasing to 30–50% in people over 65 years (14). Finally, we assessed induced cellular memory and found a strong response in all cohorts (including rituximab-treated patients), although it was smaller in lenalidomide-treated patients and absent in ibrutinib-treated patients ( Figure 3B ).

Having observed the differences before vaccination in all cohorts, we next analysed the differences before and after full vaccination ( Supplementary Figure S3A ). Healthy adults showed an expansion of the proportion of subset 3 of central memory CD4⁺ T-cells (cluster 10 in Table 1 ) and CD45RA^–CD39⁺ T_regs ( Supplementary Figure S3B ). However, we could not validate this change with classical gating approaches.

In the case of older adults, vaccination induced changes in 15% of the clusters ( Supplementary Figure S3C ): it expanded the proportion of circulating CD4^–CD8^– T-cells and early-like effector CD4⁺ T-cells and decreased classical and non-classical monocytes ( Supplementary Figure S3D ). Further analysis revealed that both subsets of classical monocytes decreased after vaccination while both CD4^–CD8^– T-cell subsets, subset 2 of early effector CD4⁺ T-cells (cluster 3 in Table 1 ) and subset 2 of early-like effector CD4⁺ T-cells (cluster 5 in Table 1 ) were expanded ( Supplementary Figure S3E ).

For untreated oncohaematologic patients, vaccination decreased the levels of circulating memory B-cells and plasmablasts ( Supplementary Figure S3F ), although we could not confirm these changes with hierarchical gating approaches. On the contrary, vaccination induced changes in 4 clusters in the lenalidomide-treated patients ( Supplementary Figure S3G ), revealing a decrease in CD4^–CD8^– T-cells and CD45RA^–CD39⁺ T_regs after vaccination ( Supplementary Figure S3H ). The decrease in the CD4^–CD8^– T-cells was due to a reduction in the second subset of this population (cluster 22 in Table 1 ) while these patients also expanded subset 5 of TEMRA Tγδ cells (cluster 27 in Table 1 ). For the ibrutinib-treated patients, vaccination decreased the levels of circulating monocytes ( Supplementary Figure S3J ), although we could not confirm this change with classical gating approaches. Finally, vaccination induced changes in 4 clusters of the rituximab-treated patients ( Supplementary Figure S3K ), confirming that these patients had a trend for a decreased level of subset 2 of the non-memory B-cells (cluster 36 in Table 1 ) following vaccination ( Supplementary Figure S3L ).

Having assessed the vaccine-induced immunity after vaccination, we performed a clinical follow-up of all individuals during an 18-month period. We found that 30.4% of the healthy adults displayed subsequent SARS-CoV-2 infection as defined by a positive PCR ( Table 2 ). This percentage, however, was much lower in the older adults (15%) as they were protected in a nursing home environment. Finally, 41.12% of the oncohaematologic patients had an infection (1 untreated patient, 6 ibrutinib-treated patients, 1 lenalidomide-treated patient and 1 rituximab-treated patient). Of note, given the low number of infected patients, we considered them as a single cohort in the subsequent analyses (irrespective of their treatment).

The acquired humoral and cellular memory following immunisation did not predict subsequent infection for any of the analysed cohorts ( Table 3 ). Nevertheless, when we evaluated the cellular immunome post-vaccination of all individuals in the context of subsequent infection, we found differences in the UMAP analysis ( Figure 4A ), specifically in 5 clusters ( Figure 4B ). Infected individuals had lower levels of CD4⁺CD8⁺ T-cells and a trend towards higher levels of TEMRA Tγδ cells and terminal effector CD8⁺ T-cells ( Supplementary Figure S4A ), due to an expansion in the second case of subset 1 (cluster 15 in Table 1 ) of this population ( Supplementary Figure S4B ). For a deeper insight into these observations, we performed additional analysis of healthy adults and oncohaematologic patients. However, given the low number of post-vaccinated infected older adults (n = 1), we excluded this cohort from the analysis.

The UMAP analysis revealed visual differences between the infected and non-infected healthy adults and oncohaematologic patients ( Figure 4C ). Infected healthy adults had higher levels of 3 different TEMRA Tγδ cell subsets ( Figure 4D ), although we could not confirm these changes with hierarchical gating approaches. Infected oncohaematologic patients had differences in 15% of the total clusters ( Figure 4E ). Similarly to healthy adults, they showed an expansion of TEMRA Tγδ cells ( Supplementary Figure S4C ), specifically subsets 1, 2 and 7 (clusters 23, 24 and 29 in Table 1 ) of this population ( Supplementary Figure S4D ).

Although vaccine-induce humoral and cellular immunity does predict subsequent infection, the circulating levels of the different TEMRA Tγδ cell subsets seem to anticipate it. Having said that, we cannot discard the possibility that these individuals were more exposed to the virus and thus were infected more often. Therefore, for a deeper insight into these mechanisms, we addressed whether these immune differences could be also observed before vaccination.

Indeed, we found differences in the immunome composition between subsequently infected and not-infected individuals, even before vaccination ( Figure 5A ) since 21% of the total clusters were differentially expressed between them ( Figure 5B ). Subsequently infected individuals had higher levels of circulating terminal effector CD8⁺ T-cells and plasmablasts, coupled with a trend towards higher levels of TEMRA Tγδ cells, and lower levels of early NK and total ILCs before vaccination ( Supplementary Figure S5A ). Further analysis confirmed that subset 3 of TEMRA Tγδ cells (cluster 25 in Table 1 ) was expanded in infected individuals even before vaccination ( Supplementary Figure S5B ).

After noting these differences, we performed additional analysis within the 3 cohorts ( Figure 5C ). Although healthy adults only had a difference in 1 cluster based on subsequent infection ( Figure 5D ), hierarchical gating revealed a trend towards higher levels of circulating plasmablasts and non-classical monocytes ( Supplementary Figure S5C ), with a deficit of subset 1 of CD4^–CD8^– T-cells (cluster 21 in Table 1 ) and a trend towards lower levels of subset 4 of TEMRA Tγδ cells (cluster 26 in Table 1 ) in subsequently infected individuals ( Supplementary Figure S5D ). For older adults, only 2 clusters were differentially expressed before vaccination between subsequently infected and non-infected individuals ( Figure 5E ). For this cohort, there was a trend towards lower levels of subset 5 of TEMRA Tγδ cells (cluster 27 in Table 1 ) in subsequently infected individuals ( Supplementary Figure S5E ). Finally, it is clear that immunocompromised patients are more likely to get infected due to an overall reduced immune response. Indeed, 17% of the clusters were differentially expressed in the oncohaematologic patients ( Figure 5F ): these patients had higher levels of TEMRA Tγδ cells and a trend towards lower levels of early NK cells ( Supplementary Figure S5F ). Moreover, there was a specific expansion before vaccination of 1 of the TEMRA Tγδ cell subsets (cluster 23 in Table 1 ) in subsequently infected patients ( Supplementary Figure S5G ).

We found that 3 of the 13 infected oncohaematologic patients had vaccine failure as defined by SARS-CoV-2-induced hospitalisation ( Table 2 ). Thus, considering that the immunome composition before vaccination could predict subsequent infection, we addressed whether the same could be true within the oncohaematologic cohort to predict vaccine failure. The UMAP immunome analysis of this cohort divided into mild disease (no hospitalisation) and severe disease (required hospitalisation) displayed evident differences between them ( Figure 6A ). A total of 8 clusters were differentially expressed in these patients even before vaccination ( Figure 6B ). Patients with a severe outcome displayed an expansion of T_regs, non-classical monocytes and CD8 ⁺ effector T-cells. However, we could not confirm these findings with classical gating approaches, likely due to the low number of patients with a severe outcome (data not shown). Nevertheless, patients with a severe outcome displayed a trend towards a deficit of classical monocytes ( Figure 6C ) due to a deficit of subset 1 (cluster 40 in Table 1 ) ( Figure 6D ), confirming the relevance of this cell type to control subsequent infection.

To evaluate whether mRNA vaccines are equally effective in immune-compromised patients, two cohorts of these patients were recruited, including older adults and patients with lymphoid cancer (oncohaematologic patients); these cohorts were subsequently compared with a control cohort. In all cases, individuals with a previous PCR-confirmed diagnosis of COVID-19 were excluded from the study. Older adults (2 males, 18 females) were recruited from the Orpea residential nursing home (Valladolid, Spain). The mean age of the older adults was 88.1 years (all over 70 years old). All of them had been vaccinated with BNT162b2 (Pfizer-BioNTech). Thirty-nine oncohaematologic patients treated at the Department of Hematology (Hospital Clínico Universitario de Valladolid, Valladolid, Spain) were also recruited. Seven of them had chronic lymphocytic leukaemia (CLL) or follicular lymphoma without active treatment, 10 had NHL treated with rituximab, 14 had CLL treated with ibrutinib and 8 had myeloma treated with lenalidomide. With the exception of 4 patients vaccinated with BNT162b2 (Pfizer-BioNTech), the remaining were vaccinated with mRNA-1273 (Moderna). This cohort showed homogeneity in terms of sex and had an average age of 63.5 years. Finally, a total of 27 healthy controls with no known inflammatory, autoimmune or malignant diseases were recruited from the occupational risk prevention service (Hospital Clínico Universitario de Valladolid, Valladolid, Spain). The controls were age- and sex-matched to the other cohorts. All of them were vaccinated with BNT162b2 (Pfizer-BioNTech). An 18-month clinical follow-up was performed in all individuals to further address (following vaccination) subsequent PCR-confirmed SARS-Cov-2 infection. Additional information about patient demographics and subsequent infection can be found in Table 2 . Ethics approval was obtained from the local ethics committee from Valladolid Este (PI 21-2098).

Blood samples from all individuals were obtained before the first vaccine dose (between January and April 2021) and 3 months following full vaccination (i.e., 3 months after the second vaccine dose). Hence, the SARS-CoV-2 B.1.1.7 strain was predominant in Spain at the time of the pre-vaccine samples while the B.1.617.2 strain was predominant by the time that the post-vaccination sample was obtained.

In all cases, blood was collected in LH Lithium Heparin separator tubes. Subsequently, peripheral blood mononuclear cells (PBMCs) were isolated using Cytiva Ficoll-Paque™ PLUS (Cytiva 17-1440-03). Blood was slowly poured into a centrifuge tube with Ficoll-Paque™ without mixing (3 ml of Ficoll for 5 ml of blood) and centrifuged at 800 g for 30 min at 4°C (Fisherbrand™ GT2) with acceleration set to maximum and deceleration to minimum. PBMCs were collected from the interface between the Ficoll-Paque™ and plasma layers. PBMCs were centrifuged again in RPMI at 400 g for 5 min at 4°C to wash them. The resulting pellet was suspended in freezing medium (90% foetal bovine serum [FBS] + 10% dimethyl sulphoxide [DMSO]) to cryopreserve the cells in liquid nitrogen until use. Plasma samples were also obtained and immediately preserved at -80°C until use.

The determination of specific anti-S IgG and IgA antibodies was performed by electrochemiluminescence immunoassay (Elecsys Anti-SARS-CoV-2 S, Roche Diagnostics, Mannheim, Germany). The results are expressed in binding antibody units (BAU). In addition, the presence of anti-N IgG antibodies was evaluated with an enzyme-linked immunosorbent assay (ELISA) (COVID-19 ELISA IgG, Vircell Microbiologists, Santa Fe, Granada, Spain).

The magnitude and kinetics of the cellular response to SARS-CoV-2 was tested with an ex vivo IFN-γ ELISpot assay. PBMCs (both before vaccination and 3 months following full vaccination) were thawed in sterile tubes with 10 ml of RPMI 1640 without L-glutamine (Gibco, Paisley PA49RF, Scotland, United Kingdom) and centrifuged at 400 g for 5 min at 4°C. After removal of the supernatant, 1 ml of AIM-V serum-free medium with L-glutamine, 50 μg/ml streptomycin sulphate and 10 μg/ml gentamycin sulphate (Gibco) was added to count the cells in a BLAUBRAND^® Neubauer counting chamber in the presence of trypan blue. Cells were cultured in duplicate (100,000 viable cells in 200 μl of AIM-V medium) in 96 U-bottom culture plates for a total period 48 h in the presence of 2 μg/ml of a pool of SARS-CoV-2 spicule S1 domain peptides (Mabtech). As a positive control, total PBMCs were stimulated with a polyclonal stimulus of anti-CD3 and anti-CD28 (Mabtech), at concentrations of 0.2 and 0.02 μg/ml, respectively; unstimulated cells served as a negative control. Following culture, secreted interferon gamma (IFN-γ) was detected by adding 1 μg/ml anti-IFN-γ monoclonal antibody (7-B6-1-ALP, Mabtech) and incubating for 2 h in the dark. The plates were developed using BCIP/NBT-plus, according to the manufacturer’s instructions. The results were obtained as spot-forming units (SFU).

After thawing, and in parallel to determining cellular memory, 2 × 10⁶ PBMCs were stained with monoclonal antibodies ( Supplementary Table S1 ) to be subsequently characterised by spectral cytometry (CyTek Aurora 5-laser) following the OMIP-069 protocol and analysis panel, with slight variations (31).

Briefly, before staining the PBMCs, the Live/Dead Fixable Blue Dead Cell Stain Kit (Molecular Probes, Thermo Fisher Scientific) was added to exclude dead cells from the analysis. Brilliant Stain Buffer and True-Stain Monocyte Blocker were also added prior to staining with the antibodies to obtain optimal fluorescence of the desired cells. The PBMCs were washed with fluorescence-activated cell sorting (FACS) buffer (500 ml phosphate-buffered saline [PBS] + 10 ml filtered FBS + 0.1 g NaN₃ + 2.5 ml sterile ethylenediaminetetraacetic acid [EDTA]) and incubated in the dark at room temperature during the staining process. Finally, the cells were fixed with 0.8% paraformaldehyde in FACS buffer in the dark for 10 min, washed with FACS buffer and stored at 4°C. Cells were acquired within 48 h in a 5-laser spectral cytometer (Aurora, Cytek).

The OMIQ Data Science platform (Omiq, Inc. 2022) was used following transformation of the data; the scale, parameters and cofactors were set as suggested by the platform. The data-cleaning FlowAI algorithm was applied to remove outlier events in spectral cytometry data files due to abnormal flow behaviour resulting from clogs and other common technical problems. Subsequently, a manual discard was performed to eliminate cell debris and doublets and to select viable leucocytes (CD45⁺ cells), which were used for subsequent analysis. After cleaning, a total of 10 samples did not fulfil the required quality criteria, so they were excluded from the analysis (3 pre-vaccine healthy adults, 2 pre-vaccine older adults, 3 post-vaccine ibrutinib-treated patients and 2 pre-vaccine rituximab-treated patients).

Due to the large amount of data obtained with this panel, it is not advisable to examine the results exclusively through traditional manual identification due to their subjectivity. Therefore, an unsupervised approach applying the UMAP algorithm was used for the exploratory analysis. Briefly, this algorithm uses a non-linear method based on graphs to represent information in multiple dimensions, and then reconstructs the results into a two-dimensional map, preserving the multidimensional structure. In this way, the algorithm finds similarities between cells in all dimensions. These dimensions are, in this case, the intensity of the markers that they express. The algorithm returns a two-dimensional map where the proximity of cells reflects their distances in multidimensional space, such that cells with similar patterns of expression are located very close to each other. This distance/similarity relationship is respected within and between each group or islet. A prior subsampling or random selection of events was performed until a total of 4 × 10⁶ events was reached to ensure that each cohort was equally represented.

Subsequently, the FlowSOM algorithm was used to find similar cell clusters and to separate them into groups in an unsupervised manner. This algorithm analyses the expression of all the selected markers in each of the cells of each sample and then groups them into metaclusters according to their expression level. In this way, it not only allows for the visualisation of cells in typical biological groupings, but also for the detection of new or unexpected clusters. However, this algorithm only displayed metaclusters that would represent the large subsets of the immune system present in the sample. The visual representation of the two algorithms allows one to further to subdivide these metaclusters into clusters that provide a more accurate representation of all the phenotypic and functional subsets of the human immunome. A clustered heatmap was created using the clusters obtained in the previous point. This heatmap graphically represents the level of expression of each phenotypic marker into each cluster. Dendrograms grouped clusters and phenotypic markers associated by similarity (distance). This approach permits one to identify the immune cell subsets represented by each cluster based on the expression levels of their markers. In this way, if a specific cluster is associated with a condition under study, its phenotype could be elucidated to identify it using classical supervised approaches where it would otherwise have gone undetected. Finally, the refined results of the FlowSOM algorithm were mapped on the UMAP plot to observe their distribution.

For the computational cytometric data, volcano plots were constructed with the edgeR algorithm to compare cluster differences. Once the clusters showing significant differences were identified, the data were validated with classical hierarchical analysis. Using a modified gating strategy of the OMIP-69 panel, the percentages within the total viable leucocyte fraction (CD45⁺) of those clusters that stood out in the previous analysis were obtained, and then GraphPad Prism 9 (GraphPad Software, San Diego, CA, USA) was used for statistical analysis. Quantitative variables are expressed as mean and standard deviation (because they followed a normal distribution). One-Way analysis of variance (ANOVA), Fisher’s exact test and t-test comparisons were applied as detailed in each figure legend. In all cases, p < 0.05 was considered statistically significant.