All patients in this study were enrolled on the NCRI phase III RIAltO trial (Randomised Investigation of Alternative Ofatumumab-based regimens for less fit patients with CLL; NCT01678430; EudraCT 2011-000919-22; Research Ethics Committee (REC) ref. 11/NW/0548). RIAltO is an open-labelled, multicentre randomised controlled trial (RCT) investigating the second generation anti-CD20 monoclonal antibody ofatumumab combined with either chlorambucil or bendamustine in patients with CLL for whom FCR was considered unsuitable (13). 521 patients were enrolled between December 2011 and April 2018 including 145 between September 2014 and March 2016 who received idelalisib (first-in-class phosphoinositide 3-kinase delta (PI3Kδ) inhibitor) or placebo in addition to chemoimmunotherapy. Follow-up data was available to April 2021. The CONSORT diagram in Supplementary Figure S1 shows how samples were selected. The discovery cohort comprised patients who were randomised to receive bendamustine or chlorambucil, with or without idelalisib, and for whom suitable samples were available (n=79). The validation cohort comprised randomly selected patients who received bendamustine and ofatumumab only and for whom suitable samples were available (n=59). The study was approved by the UK CLL Biobank (REC ref. 19/NW/0573) and performed in accordance with the Declaration of Helsinki and relevant ethical guidelines for research in humans, with written consent obtained from all patients.

Grade 3-5 infections (which included febrile neutropenia) and SPMs were defined in accordance with the Common Terminology Criteria for Adverse Events (CTCAE) v4.0. Grade 3-5 infections were reported from the point of informed consent until 6 months after the last dose of anti-CLL treatment and censored at that point in the analysis. In contrast, grade 3-5 SPM were reported until the end of the study. Short-term therapy effectiveness was measured as end of treatment (EoT) measurable residual disease (MRD) in the bone marrow quantified by flow cytometry and using cut-offs corresponding to <1 CLL cell per 10² leukocytes (MRD2), <10^-3 CLL cells (MRD3) and <10^-4 CLL cells (MRD4) (14). Long-term therapy effectiveness was measured as time to progression (TTP) using iwCLL criteria (15).

Blood samples from CLL patients were collected in EDTA and transported within 24 hours to the UK CLL Biobank (REC ref 14/NW/1014), where peripheral blood mononuclear cells (PBMC) were isolated by centrifugation over Lymphoprep and cryopreserved in 10% DMSO at -80°C. Healthy control samples, obtained from the Liverpool Blood Disease Biobank (REC ref 16/NW/0810), were also included to enable batch alignment. Prior to analysis, samples were thawed at 37°C, diluted incrementally in ice-cold cell culture media consisting of RPMI-1640 and 10% FBS and washed before counting.

PBMC were washed with purification buffer consisting of phosphate-buffered saline (PBS; pH of 7.2) supplemented with 0.5% bovine serum albumin and 2mM EDTA, resuspended with a Fc receptor blocker FcX (BioLegend, UK) at a ratio of 1:4, and incubated for 10 mins at 4°C. Anti-CD19 microbeads (Miltenyi Biotech, UK) were added and samples incubated for 15 mins at 4°C before being washed in PBS and passed through a negative selection column using the QuadroMACS™ Separator system (Miltenyi Biotech, UK).

Cell suspensions were washed and resuspended in cell culture medium (RPMI-1640 supplemented with Pen/Strep, L-glut, 10% FBS) at a density of 1 x 10⁶/ml and aliquoted into a 24-well plate. CD3/CD28 Dynabeads^® (Gibco, UK) were added at a ratio of 1:1 and the samples incubated for 48 hours at 37°C in humidified CO₂ (0.2%). GolgiPlug™ (BD biosciences, UK) was then added to each well at a concentration of 1μl/ml and incubated for 4-6 hours at 37°C prior to analysis.

Individual PBMC samples were incubated for 30 minutes at 4°C with a CD45 antibody conjugated with different combinations of ⁸⁹Yb, ¹⁰⁶Cd, ¹¹⁰Cd, ¹¹¹Cd, ¹¹³Cd, ¹¹⁴Cd, ¹¹⁵In, and ¹¹⁶Cd (HI30, Biolegend). Barcoded samples were then washed (550xg/4°C/5 minutes) twice with Maxpar Cell Staining Buffer (CSB; Standard Biotools™, US) and pooled together to create a single, multiplexed sample.

Pooled samples were incubated for 45 minutes at 4°C with a cocktail of antibodies reactive to surface epitopes ( Supplementary Tables S1 , S2 ) diluted to their respective optimal concentrations, washed with CSB followed by Maxpar phosphate buffered saline (PBS; Standard Biotools™, US), resuspended in a PBS-cisplatin solution (prepared by diluting Cell-ID™ cisplatin-195Pt (Standard Biotools™, US) in PBS at a ratio of 1:1000), and incubated for 5 minutes at room temperature (RT). Cisplatin was then quenched by adding RPMI supplemented by 20% FBS at 5 times the total PBS-cisplatin volume used. Cells were then fixed by incubating overnight at 4°C with 1.6% formaldehyde (ThermoFisher Scientific, UK), permeabilised using the eBioscience™ Foxp3/Transcription Factor Staining Buffer Set (ThermoFisher Scientific, UK) as per manufacturer’s instructions, and incubated for 30 minutes at RT with a cocktail of antibodies reactive to internal epitopes ( Supplementary Tables S1 , S2 ). After washing, cells were incubated for a further 60 minutes at RT in cell intercalation solution consisting of 125nM Maxpar^® Intercalator-Ir and Maxpar^® Fix and Perm Buffer (Standard Biotools™, US) at a 1:2000 ratio. Cells were then washed with Maxpar^® Cell Acquisition Solution (Standard Biotools™, US) before adding EQ calibration beads a 1:10 ratio. Finally, samples were loaded into the Helios mass cytometer (Standard Biotools™, US) and cell events recorded at a rate of 200-400 cells/second. A summary of the experimental steps involved for CyTOF analysis is provided in Supplementary Figure S2 .

Raw data was first normalised against EQ beads using the CyTOF Software v8.0 and analysed using the online Cytobank software (Beckman-Coulter Life Sciences, US). Doublets were removed by the gating of Intercalator-Ir and Gaussian parameters while viable cells were identified by gating 195^Pt negative events ( Supplementary Figure S3A ). Samples were then debarcoded in the R environment using FlowCore and CATALYST package (16). Subsequently, samples were batch-aligned by applying the CytoNorm package (17), using designated anchor samples comprising biological replicates of a healthy donor that were included in every batch. Unsupervised analysis involving UMAP and FlowSOM was performed using the CATALYST package while supervised analysis using Boolean gating was performed using the Cytobank software. Heatmaps were generated using the ComplexHeatmap package.

Data visualisation and statistical analyses were performed using R software. Mann-Whitney U test was used to compare differences in T cell population frequencies. Uni- and multivariable analyses of FlowSOM cluster frequencies were performed using Cox proportional hazard regression (coxph; using survival R package) in censored outcomes and logistic regression in uncensored outcomes. The Kaplan-Meier method and log-rank test were used when comparing cumulative risk of infections or SPMs, as well as overall survival (OS) between patient groups. A p-value of 0.05 was used throughout to determine statistical significance.

Datasets analysed during the current study are available from the corresponding author on reasonable request.

We first sought to establish the extent of variation in T-cell subpopulations among CLL patients requiring initial therapy. To do so, cryopreserved PBMCs obtained from 79 patients enrolled in the RIAltO trial ( Table 1 ) were subjected to deep immunophenotyping using mass cytometry and a panel of 37 antibodies ( Supplementary Table S1 ). CD4⁺ and CD8⁺ T-cells were individually organised into 30 clusters respectively using the FlowSOM clustering algorithm (18). Those clusters expressing similar markers were then merged, and minimally abundant clusters (3 or more events in fewer than 10% of samples) excluded. A detailed description of our clustering approach is provided in Supplementary Figure S4 . Among CD4⁺ T-cells, 17 distinct clusters were identified and categorised as regulatory T-cells (Treg; R1-4) based on CD25, FoxP3 and CD127 expression, and T helper cells (Th; H1-13) for the remaining clusters. One (H10) cluster contained naïve cells (CD45RA⁺CCR7⁺), four (H8, H9, H13, R3) co-expressed CD45RO and CD45RA, while the rest comprised effector memory (EM) cells (CD45RO⁺CCR7^-). CD8⁺ T-cells were divided into 22 clusters, categorised as NK/T-cells (NKT1-4) based on CD8 and CD56 co-expression, and cytotoxic T-cells (CTLs; T1-18) for the remaining clusters. Most clusters comprised EM cells, with the exception of one (T6) comprising naïve cells, six (NKT1-NKT3, T7, T12, T14) comprising terminal effector (TE) cells (CD45RA⁺CCR7^-), and four (T15-T18) co-expressing CD45RO and CD45RA ( Figures 1A–D ). To understand the extent to which T-cell subpopulations varied among individual patients, the relative size of each cluster within its parental CD4⁺ or CD8⁺ population was calculated for each sample and the spread of data visualised. Significant variation was found in the size of all clusters ( Figures 1E, F ), thereby justifying correlation with clinical events.

We first explored the association between pre-treatment T-cell subpopulations and the subsequent occurrence of severe infections. Univariable coxph analysis, incorporating all relevant co-variates, identified an association between the risk of subsequent infections and pre-treatment haemoglobin levels, as well as the pre-treatment frequencies of the C1 and H1 CD4⁺ clusters. However, only the H1 cluster was considered further as it consistently retained significance in multivariable analyses, particularly after adjusting for clinically significant covariates such as age, performance status and treatment allocation ( Supplementary Table S3 , Figure 2A ). In line with this, stratification of patients based on the median size of the H1 cluster showed that those with a higher frequency of H1 cells had a significantly lower risk of grade ≥3 infections (HR 0.082 [95% CI: 0.01 – 0.643], P = 0.002; Figure 2B ). Given the heterogeneity of the analysed CLL cohort, which included patients with different baseline characteristics receiving different treatment regimens, we performed a sensitivity analysis confirming that the pre-treatment H1 cluster size, using the median cut-off point, independently predicted infection risks despite adjusting for baseline demographics, CLL disease characteristics, treatment allocation and subsequent disease response ( Supplementary Table S4 ).

When examining the marker expression profile of H1, we found that it consisted of non-senescent (CD27⁺CD28⁺) effector memory (EM; CD45RO⁺CCR7^-) Th cells co-expressing activation markers ICOS and HLA-DR, exhaustion markers PD-1, TIGIT and TOX, and the T-box transcription factor T-bet. To define this subpopulation prospectively, we employed a Boolean gating strategy that reflected the cluster’s main non-redundant features (ICOS⁺HLA-DR⁺PD1⁺TIGIT⁺Tbet⁺ Th; Figure 2C ). A significant correlation was observed between the size of the H1 cluster and that of the prospectively defined subpopulation (R² = 0.6, P < 0.0001; Figure 2D ), thereby validating the gating strategy. In keeping with our findings for H1, there was a trend for a lower risk of grade ≥3 infection in patients with a greater proportion (25^th centile or higher; ≥ 0.072%) of ICOS⁺HLA-DR⁺PD1⁺TIGIT⁺Tbet⁺ Th cells (HR 0.358 [95% CI: 0.109 – 1.186], P = 0.076; Figure 2E ). These findings were further validated in a separate cohort of 59 patients ( Table 1 , validation cohort) using the same methodology, gating strategy and cut-off value, confirming an association between ICOS⁺HLA-DR⁺PD1⁺TIGIT⁺Tbet⁺ Th cells and a lower risk of grade ≥3 infections (HR 0.245 [95% CI: 0.061 – 0.981], P = 0.031; Figure 2F ).

We next explored the association between pre-treatment T-cell subpopulations and the subsequent occurrence of grade ≥3 SPMs using the same analytical approach. Univariable analysis showed that pre-treatment frequencies of clusters R2, T10, and T13 correlated with the risk of SPMs, though only T10 consistently retained significance in the multivariable analyses ( Supplementary Table S5 , Figure 3A ). Stratifying patients based on the median cluster size showed that those with higher T10 levels had a significantly lower risk of grade ≥3 SPMs (HR 0.108 [95% CI: 0.013 – 0.881], P = 0.011; Figure 3B ). Sensitivity analysis confirmed that the pre-treatment T10 cluster size, using the median cut-off point, independently predicted risk of SPMs despite adjustments for baseline demographics, CLL disease characteristics, treatment allocation and subsequent disease response ( Supplementary Table S6 ).

Examination of the marker expression profile of T10 revealed that it comprised EM (CD45RO⁺CCR7^-) CTLs co-expressing exhaustion markers PD-1, TIGIT and TOX, cytotoxic marker Granzyme B, and T-box transcription factors T-bet and Eomes. The loss of CD28 also suggests a shift towards a senescent phenotype, however further confirmation with CD57 and KLRG-1 is required. When a Boolean gating strategy reflecting its main non-redundant features (CD27⁺CD28^-PD1⁺Tbet⁺Eomes⁺ CD8⁺; Figure 3C ) was applied, a significant correlation was observed with the size of the T10 cluster (R² = 0.82, P < 0.0001; Figure 3D ). In line with our findings for T10, patients with more (50^th centile or higher; ≥ 5.8%) CD27⁺CD28^-PD1⁺Tbet⁺Eomes⁺ CD8⁺ cells exhibited a significantly lower risk of developing grade ≥3 SPM (HR 0.089 [95% CI: 0.011 – 0.728], P = 0.005; Figure 3E ). However, validation of these results was not possible due to the limited number of grade ≥3 SPMs in the validation cohort, where only two patients were affected.

The relationship between pre-treatment T-cell populations and overall survival was subsequently examined. In the univariable analysis, IGHV mutational status, haemoglobin levels and pre-treatment cluster sizes of T12 and T13 showed borderline to significant correlations with OS; however, only IGHV and T12 consistently retained statistical significance in the multivariable analyses ( Figure 4A ; Supplementary Table S7 ). In keeping with this, stratification of patients into two groups based on the median cluster size indicated a trend towards longer survival in the group with higher T12 levels (HR: 0.440 [95% CI: 0.184 – 1.051], P = 0.058; Figure 4B ). This trend persisted despite adjustments for baseline demographics, CLL disease characteristics, treatment allocation and subsequent disease response in the sensitivity analysis ( Supplementary Table S8 ).

The T12 cluster consisted of TE (CD45RA⁺CCR7^-) TOX^loGrymB⁺Tbet⁺Eomes⁺ CTLs exhibiting features (CD27⁺CD28^-) suggestive of cell senescence. To prospectively define this T-cell subpopulation, a Boolean gating strategy was applied (CD27⁺CD28^-GrymB⁺Tbet⁺Eomes⁺ CD8⁺ TE; Figure 4C ), where a significant correlation between the size of this prospectively defined subpopulation and that of T12 was observed (R² = 0.61, P = <0.0001; Figure 4D ). Consistent with our previous findings, separation of patients based on the 25^th centile value (1.85%) showed that survival was significantly longer in the group with more CD27⁺CD28^-GrymB⁺Tbet⁺Eomes⁺ CD8⁺ TE T-cells (HR: 0.404 [95% CI: 0.173 – 0.947], P = 0.031; Figure 4E ). A similar correlation was also observed when the same gating strategy/cut-off value was applied to pre-treatment samples from the validation cohort (HR: 0.352 [95% CI: 0.123 – 1.005], P = 0.042; Figure 4F ), confirming the association between this subpopulation and OS.

We next sought to elucidate any correlations between T-cell subpopulations and the effectiveness of the CLL treatment administered. Each cluster was related to EoT MRD status as a measure of initial cytoreduction, and time to progression as a measure of sustained effectiveness ( Supplementary Table S9 ). Three clusters were identified that separately correlated with the attainment of MRD2 (H6), MRD3 (NKT3) and MRD4 (T1), respectively. However, since none of these clusters correlated with the attainment of MRD at more than one level, they were not considered clinically significant. Similarly, no correlation was seen between any cluster and TTP ( Supplementary Table S10 ).

To gain insight into the functional characteristics of the three clinically significant T-cell subpopulations identified in this study, PBMCs from a randomly selected subset of 30 CLL patients in the discovery cohort were stimulated using anti-CD3/CD28 microbeads and analysed for the expression of seven different cytokines using a modified CyTOF antibody panel ( Supplementary Table S2 ). Apart from TGF-β in ICOS⁺HLA-DR⁺PD1⁺TIGIT⁺Tbet⁺CD4⁺ Th cells and IFN-γ in CD27⁺CD28^-GrymB⁺Tbet⁺Eomes⁺ TE CD8⁺ T cells, the expression of pro- and anti-inflammatory cytokines was significantly higher in all three clinically significant T-cell subpopulations compared to their parental CD4⁺ or CD8⁺ populations ( Figure 5 ).

Given the role of polyfunctional T-cells in eradicating pathogen and tumour (19–22), we next sought to establish the proportion of cells in the clinically significant T-cell subpopulations that expressed multiple cytokines. Pooled analysis of all CLL samples by UMAP showed that all three subpopulations were enriched with cells co-expressing two or more cytokines ( Figure 6 ; Supplementary Figure S5 ). Furthermore, analysis of cytokine expression in individual samples showed that a significantly higher proportion of cells within the clinically significant Th subpopulation expressed three or more cytokines compared to the parent CD4⁺ population, and that cells co-expressing two or more cytokines were over-represented within the two clinically significant CTL subpopulations compared to the parent CD8⁺ population. Moreover, the CTL subpopulation associated with a lower risk of SPM showed enrichment of cells co-expressing four or more cytokines ( Figure 7 ). When focussing solely on pro-inflammatory cytokines (IFN-γ, TNF-α, and IL-2), clinically significant T-cell subpopulations exhibited a lower proportion of cells lacking any pro-inflammatory cytokine expression, along with a trend towards a higher proportion of cells co-expressing two or more pro-inflammatory cytokines, compared with their parental CD4 or CD8 populations ( Supplementary Figure S6 ). Together, these findings indicate that all three clinically significant T-cell subpopulations are enriched with cells that are primed and, upon stimulation, exhibit poly-functional properties.