CD158k and PD-1 expressions define heterogeneous subtypes of Sezary syndrome

Sezary syndrome (SS) is a rare leukemic form of cutaneous T-cell lymphoma (CTCL), defined by erythroderma, pruritus, lymphadenopathy, and clonal T cells in the skin, lymph nodes, and peripheral blood. SS accounts for <5% of CTCL, occurs in aged adults, and has a male predominance.

SS, mycosis fungoides (MF), and erythrodermic inflammatory diseases share overlapping clinical and biological features, often making diagnosis difficult. Because prognosis differs between diseases, defining and harmonizing the diagnosis process have been a real challenge for many years.¹ 

Blood tumor burden evaluation mainly depends on the detection of Sezary cells by flow cytometry (FC) but results are not specific enough to be unequivocal.^2,3 

Currently, flow cytometry is used to quantify CD4⁺ lymphocyte subpopulations that lack T-cell marker expression, such as CD7 and CD26 (International Society for Cutaneous Lymphomas, ISCL).^3,4  Thus, because of the low specificity of these univariate criteria, the diagnostic thresholds are high (either 40% of CD4⁺CD7⁻ or 30% of CD4⁺CD26⁻ or respective absolute count >1000/µL).

Several studies have focused on the discovery of SS hallmarks. T central memory cell markers have been investigated because of assumed cell of origin of SS (CCR7, CD27, CD45RO, CD45RA, CD62, CD95, and CD103).^5,6  Finally, percentage of peripheral T central memory cells in patients with SS is variable, and SS cells exhibit a phenotypic plasticity for these markers. Skin homing receptor studies showed similar results (CCR4, CCR6, CCR10, CXCR3, and CLA).^5,6  Others have identified an expansion of a CD60⁺CD26⁻CD49d⁻ T-cell clone in Patients with SS, but they used only 10 healthy donors as controls.⁷  A study performed in 8 Patients with SS screened 240 markers and concluded heterogeneity in surface marker expression.⁸  The high difficulty in defining a single marker that could characterize SS may be the consequence of different pathological subtypes.

Nevertheless, a few markers seemed promising, like KIR3DL2/CD158k and PD-1/CD279.^2,9 

KIR3DL2/CD158k is a member of the killer cell immunoglobulin-like receptor (KIR), identified by transcriptomic analyses. It was described as a reliable SS⁺ marker, but some studies suggested a lack of sensitivity and the need for further studies in normal CD4⁺ T cells.^2,3,9,10  Similarly, PD-1/CD279 is nearly systematically expressed in skin biopsies of patients with SS in contrast to MF cases.¹¹  A study on peripheral lymphocytes confirmed these data but only on a few patients.¹² 

In this study, leaving aside the classical biphenotypic analysis, we proposed a new approach using unsupervised multivariate FC analyses. We chose to investigate the expression of classical CD3, CD4, CD7, and CD26 and the new association of the 2 markers CD158k and PD-1/CD279. We performed lymphocyte computational phenotypic analyses during diagnosis and follow-up of patients with SS to better define Sezary cells and improve the sensitivity of the diagnostic and follow-up FC method.¹³ 

Peripheral blood cells from 231 patients admitted to the dermatology department of Toulouse University Hospital between 2015 and 2020 were assessed: 18 patients with SS, 39 patients with MF, 18 patients with other CTCL (without circulating lymphoma T cells), and 156 patients without CTCL (mainly erythrodermic inflammatory diseases or B-cell lymphomas). Patients were classified according to their complete medical records and final clinical diagnosis.

A full list of SS patient characteristics, including treatments, is shown in Table 1. The study was conducted in accordance with the Declaration of Helsinki.

Fresh samples were analyzed at diagnosis and/or at different time points during the follow-up. Ten-color labeling was performed with the following antibodies: anti–CD3-APCR700, anti–CD26-FITC, anti–CD30-APC, anti–CD279-BV421 (BDBiosciences), anti–CD158ek-PE (Miltenyi Biotec), anti–CD4-PC7, anti–CD7-PerCP-Cy5.5, anti–CD28-PE-CF594, anti–CD45-BV510 (BioLegend), and anti–CD8-AA750 (Beckman Coulter). Anti–CD158ek-PE was chosen because of its commercial availability. There is no evidence of KIR3DL1 (CD158e) expression on either Sezary cells or normal T CD4⁺.¹⁴  Anti–T-cell receptor (TCR)-Vβ antibodies (Beckman Coulter) were used when clonal circulating cells were suspected. Red blood cells were lysed with BD FACS Lysing Solution (BDBiosciences). Acquisition of a minimum of 50 000 T cells was performed using Navios Flow Cytometer (Beckman Coulter). Counting beads were used for absolute values assessment (FlowCount Fluorospheres; Beckman Coulter). Data were analyzed using Kaluza software (Beckman Coulter).

Computational analysis of lymphocytes was performed using FlowSOM method in R plug-in on Kaluza as described by Lacombe et al.¹⁵  Briefly, for the antibody combination, all listmode (lmd) files were processed using the FlowSOM module (Bioconductor version 3.3.2 with flowSOM and flowCore packages) integrated to the analysis software Kaluza (Beckman Coulter). In a first step, compensations were checked, and the 12 parameters of each lmd file were normalized. Normalized files of the 18 SS were then merged and processed to obtain a frozen reference minimal spanning tree (MST) of 100 nodes. The lmd file of merged SS was then processed together with the normalized 35 merged controls (non-CTCL) and, one by one, each normalized SS lmd file for each patient, in the FlowSOM module. As in other diseases, FlowSOM method allows a clear and visual representation of how all markers are behaving on all cells.¹⁶ 

Sixteen skin biopsies were blindly reviewed, without considering cytometry data. Several parameters were evaluated: distribution and density of the lymphoid infiltrate, size of tumor cells, epidermal associated changes, cellular environment (eosinophils, plasma cells, reactive CD19⁺ B cells, and reactive CD8⁺ T cells), and PD-1 expression and intensity on tumor cells. A semiquantitative estimation of the number of B cells compared with the whole infiltrate (0: no B cells; 1: <5%; 2: >5%) and of CD8⁺ T cells compared with CD3⁺ cells (0: no CD8⁺ cells; 1: <10%; 2: >10%) was done.

Comparisons were performed using a Mann-Whitney test or Kruskal-Wallis test for continuous variables and Fisher’s exact test, χ² test, or 1-way analysis of variance (ANOVA) or 2-way ANOVA for categorical variables with GraphPad Prism. Statistical test results are graphically expressed: *P ≤ .05, **P ≤ .01, ***P ≤ .001, ****P ≤ .0001.

Peripheral blood cells from 231 patients (18 SS, 39 MF, 18 other types of CTCL [Table 2],¹⁷  and 156 non-CTCL) were analyzed for CD4/CD8 ratio and CD4⁺ T-cell expression of CD26 and CD7. In agreement with the literature and ISCL criteria, CD4/CD8 ratio, the percentages and absolute values of CD4⁺CD26⁻ T cells and CD4⁺CD7⁻ T cells were significantly increased in the 18 patients with SS.^3,4,18,19  The CD4/CD8 ratio was 7 times higher in patients with SS (mean = 27.2; interquartile range [IQR] = 24.4) compared with the non-CTCL group (mean = 3.7, IQR = 2.7; Figure 1A). The percentage of CD4⁺CD26⁻ lymphocytes was 5 times higher in patients with SS (mean = 60.9; IQR = 48.8) compared with the non-CTCL group (mean = 12.7; IQR = 6.0; Figure 1B). The percentage of CD4⁺CD7⁻ lymphocytes was 5 times higher in patients with SS (mean = 27.7; IQR = 70.35) compared with the non-CTCL group (mean = 6.0, IQR = 5.0; Figure 1C).

Then, we determined thresholds with the Fayyad and Irani method for each criterion (SS vs no SS).²⁰  Despite the small size of the SS cohort, we found similar thresholds as defined in the literature except for the percentage of lymphocytes of CD4⁺CD7⁻ (Table 3).^3,4  For this parameter, we obtained a threshold of 66% with the Fayyad and Irani method, whereas the ISCL threshold is set at 40%.

As expected, our data showed excellent sensitivity (>90%) and specificity (>90%) for the threshold of 27.5% CD4⁺CD26⁻ cells (30% according to the ISCL criteria).²¹  There was an excellent specificity for the threshold of 66% CD4⁺CD7⁻ cells and the threshold of 12.7 for the CD4/CD8 ratio but an obvious lack of sensitivity (Table 3). Negative predictive values were >94% for all criteria, but positive predictive values were low, certainly because of the size of the SS cohort. In this cohort, the percentage of CD4⁺CD26⁻ cells was the best performing marker among the classical ISCL criteria (area under the receiver operating characteristic [ROC] curve [AUC] = 0.9723; Figure 2A). At the opposite extreme, the percentage of CD4⁺CD7⁻ was the least powerful marker among the classical ISCL criteria to diagnose SS (AUC = 0.6799; Figure 2A).

These results validated the cohort and allowed us to explore new markers and a new analysis method.

Concomitantly, peripheral blood cells were analyzed for the expression of CD158k and PD-1/CD279 on Sezary-like CD4⁺CD26⁻ cells in order to improve the sensitivity and specificity of the already effective CD4⁺CD26⁻ criterion. Classical supervised dot-plot analysis was performed. As expected, we found that the percentage and absolute values of CD4⁺CD26⁻CD158k⁺ cells are significantly higher in patients with SS compared with other categories¹⁰  (Figure 1D,H). We observed the same pattern regarding PD-1 expression even if there were more overlapping phenotypes between the different categories of patients¹²  (Figure 1E,I).

With the Fayyad and Irani method, we defined 4 threshold values with the best SS diagnostic performances²⁰  (Table 3).

The threshold for absolute CD4⁺CD26⁻CD158k⁺ cells values of 107/µL showed a sensitivity of 77.8% and a specificity of 96.4%, making it the most specific univariate marker with acceptable sensitivity (AUC = 0.9050; Table 3; Figure 2A).

The threshold for absolute of CD4⁺CD26⁻PD-1⁺ cell values of 82/µL showed a sensitivity of 100% and a specificity of 84%, making it the most sensitive marker in our study but with a persistent lack of specificity (AUC = 0.9537; Figure 2A).

These data led us to combine different markers in a multivariate model to improve the diagnostic performance of FC. For this purpose, we used the FlowSOM algorithm.

Gated lymphocyte phenotypic data of 18 patients with SS and 35 non-CTCL patients were included in the model.

The FlowSOM algorithm allowed for cells to be clustered according to their phenotype.

First, the best seed of nodes, allowing the best cell separation was generated on merged SS files (Figure 2B).

To evaluate the specificity of FlowSOM nodes, we compared their frequency to those from 35 non-CTCL patients. Using a volcano plot and multiple Student t tests, we identified 15 nodes of interest among 100 nodes, including 5 nodes significantly (P < .01) overrepresented in patients with SS (Figure 2C; supplemental Figure 1).

Using the ROC curve, we studied diagnostic performances of each node overrepresented in patients with SS. The 5 selected nodes, even cumulated, do not show better performances than the classical ISCL criteria (Figure 2A).

To characterize those nodes overrepresented in patients with SS, we studied the median fluorescence intensities of CD4, CD26, CD7, CD158k, and PD-1 (supplemental Figure 2). We observed that there is no unique profile; each node showed a variable expression of CD4, CD26, CD7, CD158k, and PD-1.

Then, considering heterogeneity and the performance diagnostic of individual nodes of interest, we decided to classify the nodes into different subsets according to their common expression profile.

The expression of the markers was highlighted on frozen FlowSOM MST graphs to delineate different subsets of lymphocytes (Figure 3A-F, supplemental Figure 3). Nine subsets of nodes were defined according to their expression profile (Figure 3G).

The frozen FlowSOM MST and subsequently defined subsets of nodes were also applied to lymphocytes from merged non-CTCL patient lmd files and compared with those from patients with SS (Figure 3H). Five overrepresented subsets of nodes were identified in patients with SS (2-way ANOVA, α = 0.05; Sidak multiple comparison test) (Figure 4A).

As expected, the subsets were CD4⁺ T cells repressing the expression of CD26. Specifically, 2 subsets of 19 nodes strongly expressed CD158k (z score mean fluorescence intensity [MFI] mean ≥1) (subsets 1 and 2), and 1 subset of 13 nodes highly expressed only PD-1 (z score MFI mean ≥1) (subset 4) (Figure 3; supplemental Figure 3; Table 5). CD7 expression was heterogeneous (Figure 3; supplemental Figure 3; Tables 3 and 4).

In order to study the phenotypic profiles of SS, we focused on the 51 nodes of the 5 identified subsets. We classified the 18 SS according to the frequency of nodes in each subset: 4 patients overexpressed subset 1; 5 patients overexpressed subset 2; 1 patient overexpressed subset 3; 4 patients overexpressed subset 4; and 4 patients overexpressed subset 5, demonstrating high heterogeneity (Figure 4B).

We defined 3 phenotypic classes of SS: CD158k⁺ SS (majority of subsets 1 and 2), CD158k⁻PD-1⁺ SS (majority of subset 4), CD158k, and PD-1 double-negative SS (majority of subsets 3 and 5) (Figure 4B).

It is worth noting that clonal proliferation of T-cell lymphoma was phenotypically evaluated by measuring the expression of the TCR Vβ repertoire.²²  Among the 18 patients, we identified a TCR Vβ clone in 56% (10/18) and a hole in the TCR repertoire in 33% (6/18). We confirmed the correlation between the Vβ clone or the repertoire hole proportion and the predominant subset proportion in each patient, although the predominant subset does not represent all the Vβ clone (Figure 5; supplemental Figure 4).

To complete the characterization of the different phenotypic classes of SS, diagnostic skin biopsies of 16 patients were blindly reviewed (Figure 6).

The immunohistochemical expression of PD-1 by SS cells correlated with the phenotypic classes (Figure 6A). The study of the cellular environment showed that the CD158k⁺ profile had significantly (Fisher exact test, P < .05) less reactive CD8⁺ T cells in the skin, and the CD158k⁻PD-1⁺ profile appeared to have more reactive CD19⁺ cells (Fisher exact test; P = .06). To summarize, we identified 3 SS classes according to their cellular environment in correlation with the 3 phenotypic profiles (Figure 6C-D).

Interestingly, the study of circulating cells showed that the CD158k⁻PD-1⁺ SS class had significantly more circulating B cells (supplemental Figure 4). There was no significant difference considering circulating CD8⁺ T cells, but they were threefold increased in the CD158k⁻PD-1⁺ class, without reaching statistical significance (P = .09), certainly because of the low number of patients (supplemental Figure 5). The results were therefore similar between skin and blood.

Because we identified 3 classes of SS with heterogeneous immune profiles, we wondered if they could correlate with treatment response.

Biological follow-up of patients with SS was monitored in blood by measuring the level of CD4⁺ SS cells (ie, CD26⁻ or CD7⁻).³  To evaluate whether CD158k or PD-1 could be used as an SS cell readout, we followed their expressions during treatment. Whereas lymphocyte count fluctuated, the phenotype of SS cells was stable in all SS classes (supplemental Figure 6). Therefore, the residual circulating disease could be efficiently evaluated by absolute count of SS cells that could be identified by their expressions of CD158k and PD-1 at diagnosis⁴  and that were consistently and robustly expressed during treatment.

We evaluated the treatment response as the absolute value of SS CD4⁺CD26⁻ cells. In the SS cohort, at the 1-year follow-up evaluation, the absolute count of CD4⁺CD26⁻ cells in the CD158k⁻PD1⁻ class decreased 1.6 times more than that of the CD158k⁺ class and significantly 2.5 times more than that of the CD158k⁻PD-1⁺ class (Figure 7). The absolute count of CD4⁺CD26⁻ cells in the CD158k⁺ class decreased 1.5 times more than that of the CD158k⁻PD-1⁻ class without reaching statistical significance (Figure 7).

The first step was to validate our cohort. This step was important because of the SS cohort size, which could bias statistical analyses. Indeed, it was a small cohort of fresh samples, but comparable to what has already been described in the literature (eg, 16 blood and skin-paired samples of patients with SS).^5,8,23-25  In contrast, our non-CTCL cohort is substantial and allowed us to define reliable diagnostic thresholds. We therefore began by studying the classical ISCL criteria. As described, measurement of CD4⁺CD26⁻ cells percentage was the most reliable marker but could not be used as a stand-alone diagnostic criterion because 5.5% of non-CTCL patients (9 on 156) had a value above the 30% ISCL threshold. It should be noted that CD7 loss is not sensitive enough in our cohort. We defined a threshold of 66% for the CD4⁺CD7⁻ percentage compared with 40% in the literature. This difference could be explained by the fact that CD7 loss is frequently associated with cell senescence and that the percentage of CD4⁺CD7⁻ cells increases with age.^26,27  Therefore, the percentage of CD4⁺CD7⁻ cells may vary in non-CTCL patients depending on the age of the patients. Furthermore, expansion of CD4⁺CD7⁻ cells may also be observed in reactive dermatoses.²⁸  In addition, CD7 loss is also described in other T-cell neoplasms, which we have included to calculate the thresholds in our study.²⁹  Finally, as described in the literature, CD7 expression depends on the stage of the disease.³⁰  Considering those data, CD4⁺CD7⁻ percentage, although considered a classical feature of CTCL, should be used with caution in SS diagnosis.

Once our cohort validated, the first aim of the study was to evaluate CD158k and PD-1/CD279 coexpression for the diagnosis of SS. We showed a significant increase of CD4⁺CD26⁻CD158k⁺ cells and CD4⁺CD26⁻PD-1⁺ cells in patients with SS compared with other patients. We defined thresholds for percentages and absolute values with good diagnostic performances. CD158k is highly discriminating with only 3.7% of patients without CTCL (6 on 156) expressing SS-like CD158k⁺ cells above the threshold of 100/µL. These data confirm previous reports.^10,21  It can be noted that we found a slightly lower threshold in our cohort. This may be due to a different gating strategy or a difference between the patients included in the control group.²¹  These results also confirm that the anti-CD158ek (commercially available) can be incorporated in FC panels for the diagnosis of SS. PD-1 is extremely sensitive but lacks specificity probably because it is an activation and exhaustion marker also expressed by normal lymphocytes during aging.^31,32 

The second and major aim of our study was to evaluate the interest of using unsupervised multivariate analysis in SS diagnosis. Multivariate analysis allowed for the simultaneous study of markers of interest, such as CD158k and PD-1 on SS cells, with has never been described to our knowledge. This method allowed for the identification of SS cells, distributed heterogeneously in 5 subsets of nodes. It was an easy way to evaluate blood involvement by SS cells, as their phenotype was unchanged during follow-up. Moreover, if the phenotype realized at diagnosis is available, this type of representation allows a better sensitivity and could be used to assess minimal residual disease, as in other disease.¹⁶  With this method, we could also easily imagine adding other markers to increase analytic performances and refine diagnosis.

In an interesting way, using multivariate analysis, we identified 3 classes of SS according to their phenotypic profile, which had never been described to our knowledge: CD158k⁺, CD158k⁻PD-1⁺, and CD158k⁻PD-1⁻ SS.

In light of this discovery, we tried to characterize these 3 classes.

For this purpose, we compared our data with pathological data from skin biopsies. We observed that there was no significant difference between patients with SS regarding cutaneous lymphoma infiltration, but there were notable differences regarding the cellular environment (B cells and CD8⁺ T cells). The 3 phenotypic classes did not have the same cellular environment, suggesting a different immune response to tumor cells or to infections. Interestingly, according to the literature, bacterial sepsis, related to cellular immunity impairment, is the main cause of death in a patient with SS.³³ 

Some studies were interested in adaptive immunity in SS involving circulating CD8⁺ T cells. One recent study showed a chronic activation of CD8⁺ T cells, with exhaustion marker expression and attenuated cytotoxic profile in patients with SS.³⁴  Another found that CD8⁺ T cells downregulate the growth of tumor cells.^35,36  Moreover, high density of CD8⁺ T-cell infiltrate is associated with a more favorable prognosis.³⁷  This could explain the better treatment response of CD158k⁻PD-1⁻ patients with SS in which numerous CD8⁺ T cells were observed in skin biopsies. In particular, photosensitive lesions are characterized by a CD8⁺ T-cell–dominant infiltrate, so a phenotypic profile with more CD8⁺ T cells in the skin could be a better responder to phototherapy.³⁸ 

In contrast, there is no literature reporting the involvement of B cells in immune response in SS. We know that T follicular helper cells play a role in stimulating B-cell response.³⁹  CD158k⁻PD-1⁺ SS cells show a T follicular helper–like profile with a strong expression of PD-1; interestingly, we found in the skin biopsies of those patients a high CD19⁺ infiltrate.³⁹  The impact of B-cell expansion on the outcome of patients with SS remains to be determined.

Regarding PD-1 expression, its immunosuppressive nature makes it an interesting therapeutic target in SS.¹²  The existence of 3 SS classes with variable intensity of expression could eventually predict the outcome and/or response to different therapies, including anti-PD-1.^12,40 

Obviously, further studies on larger cohorts are necessary to better define these different profiles, but there appear to be several subtypes of SS with different biological behavior especially with respect to the tumor microenvironment.

The need for surrogate biomarkers such as tumor environment has already been discussed to help distinguish patients for whom ECP monotherapy is sufficient from those who may benefit from ECP in combination with other agents.⁴¹  To further investigate this question, it would be interesting to compare responses to treatment between each SS class. Such a study could help to define biological markers of susceptibility/resistance to treatment, including ECP or immunotherapy (eg, anti-KIR3DL2).⁴²  Then, phenotypic profiles could help to choose better targeted therapy for a patient.

Contribution: I.V. analyzed the data and wrote the paper; C.D.-B. collected and integrated the data; S. Boulinguez provided clinical data and followed patients; J.-B.R. and J.-P.V. helped to write the paper; R.B., S. Boudot, A.C., S.C., M.D., S.E., N.L., and M.-L.N.-T. performed the cytometry analysis; E.T. performed pathological analysis; L.L. performed pathological analysis and helped to write the paper; and F.V. designed research, analyzed the data, and wrote the paper.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Inès Vergnolle, Laboratoire d’Hématologie, Institut Universitaire du Cancer de Toulouse Oncopole, 1 Ave Irène Joliot-Curie, 31059 Toulouse Cedex 9, France; e-mail: vergnolle.ines@iuct-oncopole.fr; and François Vergez, Laboratoire d’Hématologie, Institut Universitaire du Cancer de Toulouse Oncopole, 1 Ave Irène Joliot-Curie, 31059 Toulouse Cedex 9, France; e-mail: vergez.francois@iuct-oncopole.fr.