Frozen peripheral blood mononuclear cells (PBMC) from 32 patients were obtained from a prospective pilot cohort (cf. Supplementary Material , patient characteristics in Table). Patients at least 18 years old, diagnosed with CML in chronic phase (CML-CP), who had been treated with TKI (tyrosin kinase inhibitor) and being at least 2 years with DMR, were included by the Poitiers CHU Hematology and Cellular Therapy Service, except for 10 patients from the OPTIM DASATINIB clinical trial (NCT01916785) (17). Among 32 patients, 8 relapsed, and 24 did not relapse after 12 months of TKI cessation. The term “relapsed patients” refers to patients having experienced MR (molecular recurrence), defined by the loss of MMR (major molecular response) during the 12 months following TKI cessation (18). The term “non-relapsed” patients refers to those having maintained their MMR during the 12 months following TKI cessation, the objective being to focus on early relapses. Venous blood was collected before treatment discontinuation (D0), and 3, 6 and 12 months after treatment discontinuation. PBMCs were isolated by density gradient centrifugation (Lymphocyte Separation Medium, Eurobio Scientific, CMSMSL01-01), resuspended in 90% fetal calf serum with 10% DMSO, and cryopreserved in liquid nitrogen until use. Frozen PBMCs from healthy control donors were obtained from the Blood Transfusion service [Etablissement Français du Sang (EFS Lyon)]. The study protocol was approved by the ethics committee at Poitiers University Hospital (Ethics Committee Ouest III, Poitiers, France, ministerial declaration n° DC 2008-565) and the study was conducted in compliance with Good Clinical practice guidelines. All participants gave written informed consent before inclusion in the trial.

PBMCs were cultured in RPMI 1640 medium supplemented with antibiotics (penicillin and streptomycin) and 10% heat-inactivated fetal bovine serum. For IL-12/IL-18 stimulation, PBMCs from CML patients and healthy control donors were seeded at 1.10⁶ cells/mL into 24-well plates with IL-12 and IL-18 (both at 20 ng/mL, R&D systems) for 48 hours. GolgiSTOP (BD Biosciences) was added for the least 4 hours of culture.

Phenotypic analyses by flow cytometry were performed either ex vivo or after culture in the presence of IL-12 + IL-18. Expression of the different markers was assessed by multicolor flow cytometry from staining panels with appropriate combinations of the following antibodies: anti-TCR-αβ BV510 (clone: IP26, BioLegend), anti-CD8 PE-Cy7 (clone: RPA-T8, BD Pharmingen), anti-Eomes eFluor^® 660 (clone:WD1928, eBiosciences), anti-Eomes PerCP-eFluor^® 710 (clone:WD1928, eBiosciences), anti-CD45RA PE (clone: HI100, BioLegend), anti-CD45RA BV510 (clone: HI100, BioLegend), anti-CCR7 PerCP-Cy5.5 (clone: G043H7), anti-TCR-γδ PE-Cy7 (clone: B1), anti-TCR Vα24-Jα18 PE (clone: 6B11, BioLegend), anti-CD3 PerCP-Cy5.5 (clone: UCHT1, BioLegend), anti-CD56 BV510 (clone: HCD56, BioLegend), were used to identify our populations of interest. Anti-PD-1 APC (clone: EH12.2H7, BioLegend), anti-Ki67 BV421 (clone: Ki-67, BioLegend), anti-Perforin FITC (clone: δG9, BD Biosciences), anti-IFNγ Alexa Fluor^® 488 (clone: 4S.B3, BioLegend), anti-panKIR2D PE (clone: NKVFS1, Miltenyi Biotech), anti-KIR3DL1/KIR3DL2 (CD158e/k) PE (clone: 5.133, Miltenyi Biotech), and anti-NKG2A (CD159a) PE (clone: REA110, Miltenyi Biotech), were used as read-out data. KIR/NKG2A referred to staining with a mix of anti-KIR2D, anti-KIR3DL1/KIR3DL2, and anti-NKG2A antibodies. For nuclear and intracytoplasmic staining, cells were fixed and permeabilized with an anti-human FoxP3 staining kit (eBiosciences) before staining with intracellular antibodies. Dead cells were excluded using the Zombie NIR™ Fixable Viability Kit (BioLegend).

Data were acquired on a FACSVerse™ cytometer with FACSuite™ software (BD Biosciences), and analyzed with the FlowJo software version 10 (BD Biosciences). The gating strategy used to define populations of interest is given in Supplementary Figure S1 . Samples that failed to meet quality control criteria were excluded from analysis.

Statistical analysis was performed using GraphPad Prism software version 9.5 (GraphPad, Dotmatics). Statistical significance of differences in frequencies and mean values was analyzed by the Mann-Whitney non-parametric test. The Spearman correlation test was used to assess the association between expressions of PD-1 and perforin. Results were considered to be statistically significant when p < 0.05. P-values between 0.05 and 0.15 are shown on the histograms. P-values below 0.05 are indicated on figure legends for healthy control donor and patient comparisons. Kaplan–Meier curves were plotted to assess the probability of molecular recurrence in patients with low versus high levels of the cell subsets studied. Immunological biomarkers were categorized in high and low levels building receiver operating characteristics (ROC) curves and calculating Youden index to choose the best cut-off value. A Log-Rank test was used to compare the probability of molecular recurrence between patients with high and low levels of each biomarker.

Relative proportion of NK (CD3(-)CD56(+)) cells did not significantly differ between non-relapsed and relapsed patients ( Figure 1A ). CD8 T (TCR-αβ(+)CD8(+)) cells are functionally divided into four subpopulations, according to cell surface expression of CCR7 and CD45RA: naive (CCR7(+), CD45RA(+)); central memory (CM, CCR7(+)CD45RA(+); effector memory (EM, CCR7(-)CD45RA(-)); and terminally differentiated effector memory (EMRA, CCR7(-)CD45RA(+)) (19). There was no significant difference between the two groups of patients in the percentages of total CD8 T-cells or the four CD8 T-cell subpopulations ( Figure 1B ).

As for NK cells and the four CD8 T-cell subpopulations, percentages of non-conventional effector T-cell populations, namely innate CD8 T (TCR-αβ (+)CD8(+))Eomes(+)KIR/NKG2A(+)) cells ( Figure 1C ), iNKT (CD3(+)TCR-Vα24-Jα18(+)) cells ( Figure 1D ), and γδ-T (CD3(+)TCR-γδ(+)) cells ( Figure 1E ), did not significantly differ between non-relapsed and relapsed patients.

We next studied the functional status of circulating immune effector cells by analyzing ex vivo perforin expression ( Figure 2 ), a key cytotoxic player known to be mainly expressed by NK cells and effector T-cell populations.

In both healthy control donors and non-relapsed patients, perforin was shown to be expressed by a large majority of NK cells (nearly 80% of positive cells), a proportion that dramatically decreased (only 30-35% of positive cells) in relapsed patients ( Figure 2A ). When considering T-cell compartments ( Figures 2B–D ), in healthy control donors, the propensity to express perforin was an expected common feature of effector T-cells, i.e. EMRA, EM, and innate CD8 T-cell subpopulations ( Figures 2B, C ), iNKT cells ( Figure 2D ), and γδ-T cells ( Figure 2E ), with a proportion of perforin-expressing cells ranging from 7 to 40%, compared with less than 2% for the other T-cell subsets (naive and CM CD8 T-cells). In non-relapsed patients, among all effector T-cell subsets, only the innate CD8 T-cell pool contained a significantly higher proportion of perforin-positive cells, compared with relapsed patients. Remarkably, the proportion of perforin-positive innate CD8 T-cells was even higher in non-relapsed patients than in healthy control donors (p-value: 0.0015). Another effector activity of innate CD8 T-cells is their ability to produce IFN-γ in response to innate stimulation by IL-12/IL-18. However, this NK-like activity of innate CD8 T-cells did not appear to significantly differ between non-relapsed and relapsed patients, with IFN-γ levels similar to those found in healthy control donors ( Supplementary Figure S2 ).

Finally, the higher perforin expression frequency evidenced at the moment of TKI cessation in NK cell and innate CD8 T-cell subsets from non-relapsed patients appeared relatively robust. Indeed, for both cell subsets, when patients were dichotomized to low- and high- perforin expressing-cell-frequency groups according to ROC curves and the Youden index, the difference in molecular relapse-free survival at 6 months between the two groups was significant (87.5% vs 28.5% (p=0.001) and 94.1% vs 40.0% (p=0.003) for NK cell and innate CD8 T-cell frequencies, respectively) ( Figure 3 ).

Furthermore, over time (3, 6, and 12 months) after TKI cessation, the frequency of cells expressing perforin remained significantly higher among the NK cell subset (p-values: 0.0264, 0.0048, 0.0645, respectively), and tended to remain higher among the innate T-cell subset (p-values: 0.2523, 0.1409, 0.1353, respectively), compared with relapsed patients at the moment of TKI cessation ( Supplementary Figure S3 ).

In our cohort, the majority of non-relapsed patients were treated by dasatinib, which is known for targeting both NK cells and effector T-cells (20–24). As shown in Supplementary Figure S4 , frequencies of perforin-expressing innate CD8 T-cells and NK cells did not differ with regard to the type of TKI received by the non-relapsed patients, thereby ruling out a selective effect on perforin expression.

All in all, these data lead us to consider the increased proportion of perforin-expressing NK cells and innate CD8 T-cells as functional signatures of TFR.

We next assessed the expression of PD-1, an activation marker also routinely used to evaluate T-cell exhaustion during chronic pathological situations, including cancers (25) ( Figure 4 ).

In relapsed patients, when compared to healthy control donors, PD-1 appeared increased, but not significantly, in terms of cell-surface expression level on several effector T-cell subsets, namely EMRA CD8 T-cells, iNKT cells and γδ T-cells. In clear contrast, as compared to the same healthy control group, non-relapsed patients showed a dramatic reduction in PD-1 expression level on EMRA CD8 T-cells (p-value: 0.017), EM CD8 T-cells (p-value: 0.004), and iNKT cells (p-value: 0.0005), providing evidence that down-regulation of PD-1 is systemic in effector T-cells from this group of patients. As a result, cell-surface MFI level of PD-1-expressing cells among EMRA CD8 T-cells (p-value: 0.015), EM CD8 T-cells (p-value: 0.006), and iNKT cells (p-value: 0.0025) significantly decreased in non-relapsed patients compared to their relapsed counterparts. Moreover, for total CD8 T-cells, and iNKT cells, when patients were dichotomized to low- and high- PD-1 MFI groups according to ROC curves and the Youden index, the difference in molecular relapse-free survival at 6 months between the two groups was significant (100.0% vs 36.3% (p=0.0006), and 94.1% vs 20.0% (p=0.0002) for CD8 T-cell, and iNKT-cell PD-1 MFI, respectively) ( Figures 5A, B ). Furthermore, this phenomenon was maintained over time (3-to-12 months) after discontinuation of TKI therapy (see legend of Supplementary Figure S5 ).

In innate CD8 T-cells, a decrease in PD-1 surface-expression level was also observed in non-relapsed patients compared to their relapsed counterparts, albeit less marked than for other effector CD8 T-cell subpopulations. This conclusion applied both at the moment of TKI discontinuation (p-value: 0.026) and over the following months (p-values for 6 and 12 months: 0.0813 and 0.0392, respectively) ( Supplementary Figure S5 ). However, when patients were dichotomized to low- and high- PD-1 MFI groups, on innate CD8 T-cells, according to ROC curves and the Youden index, the difference in molecular relapse-free survival at 6 months between the two groups was significant (100% vs 53,3% (p=0.0146) ( Figure 5C ).

Lastly, as for perforin, type of TKI, which was used as a treatment prior to its discontinuation, was found to have no impact on PD-1 expression level in immune effector cells from non-relapsed patients ( Supplementary Figure S6 ).

To test whether the higher proportion of perforin-expressing CD8 T-cells was associated with a lower exhaustion-like status in non-relapsed patients, we analyzed correlations between percentages of perforin-expressing cells and PD-1 cell-surface MFI levels among the three CD8 T-cell effector subpopulations at the time of TKI discontinuation ( Figure 6 ). We found a significant negative correlation between perforin- and PD-1 expression by innate CD8 T-cells ( Figure 6A ), a phenomenon that was not found when considering the total CD8 T-cell pool or EMRA and EM CD8 T-cell effector subpopulations ( Figure 6B ).