PD-1 expression identifies proliferating malignant CLL B cells and is a potential biomarker of response to BTK inhibitor therapy

Significance

Chronic lymphocytic leukemia (CLL) is an incurable B cell malignancy where biomarkers that predict response and resistance to treatments are lacking. We report that nearly all activated and proliferating CLL cells express PD-1 in vivo. B cell receptor and toll-like receptor-9 signaling readily increased PD-1 expression on CLL cells ex vivo, an effect that was blocked by Bruton’s tyrosine kinase inhibitors (BTKi). Indeed, the percentage of circulating PD-1⁺ CLL cells correlated with BTKi treatment response and progression, suggesting that PD-1 expression could be a useful biomarker to predict response and resistance to BTKi therapy. Additionally, eliminating PD-1⁺ CLL cells with depleting anti-PD-1 antibodies should be explored as a potential therapeutic strategy.

Abstract

Chronic lymphocytic leukemia (CLL) remains incurable despite treatment advances, and a major challenge is that biomarkers that predict response and resistance to current therapies are lacking. We report that activated and proliferating malignant CLL B cells in circulation express PD-1, a protein normally expressed in T cells. PD-1 expression is absent in circulating B cells from healthy controls and nonmalignant B cells from patients with CLL. Circulating PD-1⁺ CLL cells are found in all treatment naïve patients, regardless of immunoglobulin heavy-chain variable region gene mutation status or cytogenetic abnormalities. PD-1⁺ CLL cells are transcriptionally distinct compared to PD-1⁻ CLL cells and upregulate genes associated with cell activation, proliferation, and B cell receptor (BCR) and toll-like receptor (TLR) signaling. Indeed, ex vivo stimulation of the BCR and TLR9 readily increased PD-1 expression in CLL cells from treatment-naïve patients within 24 h, an effect that was blocked by Bruton’s tyrosine kinase inhibitors (BTKi). More importantly, patients initiating BTKi therapy experienced profound reductions in circulating PD-1⁺ CLL cell numbers within 1 mo, which is in line with reduction in Ki-67⁺ CLL cells. Elevated percentages of circulating PD-1⁺ CLL cells also preceded a clinical diagnosis of disease progression in patients receiving BTKi. Thus, our findings indicate that PD-1 expression is a potential biomarker to identify proliferating CLL cells in vivo and will be useful to predict response and resistance to BTKi. In addition, eliminating PD-1⁺ CLL cells with depleting anti-PD-1 monospecific or bispecific antibodies should be explored as a potential therapeutic strategy.

Chronic lymphocytic leukemia (CLL) is the most common leukemia diagnosed in the western hemisphere, characterized by the accumulation of mature B cells expressing CD19 and CD5, and low levels of CD20 in the peripheral blood (PB), bone marrow, and lymphoid organs (1). The treatment of CLL has changed dramatically with the introduction of covalent Bruton’s tyrosine kinase inhibitors (BTKi) such as ibrutinib and acalabrutinib, and Bcl-2 antagonists like venetoclax, resulting in significant improvements in clinical outcomes (2). Despite these advances, however, few biomarkers of response or early progression currently exist for these therapies. Furthermore, CLL remains incurable, and patients who progress after treatment with these agents have limited treatment options and poor outcomes, highlighting the need for new and better treatments (3–5).

CLL cells receive stimulation from the microenvironment and proliferate within lymphoid tissues while circulating cells are largely thought to be in a resting state (6, 7). Many signaling pathways are involved in promoting CLL proliferation and survival, including the activation of the B cell receptor (BCR) and toll-like receptors (TLR) like TLR9 (8, 9). BCR signaling activates a series of kinases that eventually lead to the activation of the NF-κB pathway (2). BTK plays a central role in this signaling cascade, and its inhibition results in CLL cell death (2). Microenvironmental interactions between CLL cells and T cells are also essential for tumor growth (9–11), though many of these interactions are incompletely understood.

Patients with CLL often have an abnormal T compartment, with an inverted CD4:CD8 ratio, lower numbers of naïve cells, higher proportion of regulatory T cells, and higher frequencies of T cells with an exhausted phenotype, including the overexpression of PD-1 and other immune checkpoint molecules (12–14). PD-1 is normally expressed on T cells and functions to reduce T cell effector function. In this study, we detected the expression of PD-1 in a significant fraction of circulating malignant CLL cells, determined the factors that promote its expression in these cells, and revealed its biological and clinical relevance in relation to BTKi therapy.

Results

PD-1 Is Expressed in a Subset of Circulating CLL Cells.

Peripheral blood (PB) samples from 73 treatment-naïve CLL patients were collected for analysis (Fig. 1A). Median age for this cohort was 65 y, 48% were female, and 78% White (SI Appendix, Table S1). Individual cytogenetic abnormalities, IGHV gene somatic hypermutation (SHM) status, and karyotype are listed in SI Appendix, Tables S1 and S2. Flow cytometry analysis of the PB showed an elevated percentage of B cells in the blood, most of them expressing CD5 and low levels of CD20, consistent with their diagnosis of CLL (SI Appendix, Fig. S1). Interestingly, we observed that 10 to 85% of circulating CLL cells expressed PD-1 while PD-1 expression was observed in <10% of non-CLL B cells (CD20^hiCD5^-) from the same patients and was rarely observed in B cells from age-matched individuals without CLL (Fig. 1 B and C). Indeed, the percentage of circulating B cells expressing PD-1 in healthy individuals was low regardless of the B cell subset analyzed (SI Appendix, Fig. S2). PD-1⁺ CLL cells were detected in all patients regardless of immunoglobulin heavy-chain variable region gene (IGHV) somatic hypermutation status or cytogenetic abnormalities. Patients with del13q, the most common chromosomal abnormality in CLL, were found to have a higher percentage of circulating PD-1⁺ CLL cells, but no correlation with other known chromosomal aberrations (del11q, del17p, and trisomy 12) was observed (SI Appendix, Table S3). Patients with mutated IGHV appear to have higher circulating percent PD-1⁺ CLL cells, but this finding was not significant in our sensitivity analysis (SI Appendix, Table S4).

Fig. 1.

PD-1⁺ CLL cells from treatment-naïve patients show phenotypic markers consistent with recent proliferation. (A) Study design. (B) Representative flow cytometry plot of PD-1 expression in CLL cells (CD5⁺CD20^low, cyan) and non-CLL B cells (CD5^-CD20^hi, magenta). Numbers = % of parent cells. (C) Percent of B cells expressing PD-1 in treatment-naïve CLL patients or healthy controls. (D) Representative FACS plots showing expression of PD-1 and Ki-67 in CLL cells. Numbers = % of CLL cells. (E) Percent of circulating Ki-67⁺ CLL cells with or without PD-1 expression. (F) Geometric mean fluorescence intensity (MFI) of PD-1 in PD-1⁺ CLL cells based on Ki-67 expression as measured by flow cytometry. Median MFI fold change between PD-1⁺Ki-67⁺ and PD-1⁺Ki-67⁻ CLL cells was 2.25. (G) Percent of cells expressing Ki-67 among PD-1⁺ and PD-1⁻ CLL cells. (H) Correlation between percent PD-1⁺ and percent Ki-67⁺ CLL cells. (I) Histogram plots showing expression of other relevant surface markers in PD-1⁺ (red) and PD-1⁻ (blue) CLL cells. CD5⁻ B cells (gray) shown as control. (J) Quantification of MFI fold change of surface markers shown in I between PD-1⁺ and PD-1⁻ CLL cells. Error bars = median ± IQR. For all other graphs, error bars = mean ± SEM. N = 73. *** = P < 0.001, **** = P < 0.0001 by the Kruskal–Wallis test and adjusting for multiple comparisons using a Dunn’s test (C) or paired t test (D–G).

Interestingly, nearly all CLL cells expressing Ki-67 also expressed PD-1, indicating that PD-1 is expressed in proliferating CLL cells (Fig. 1 D and E). Indeed, PD-1 was also more abundant in the surface of Ki-67⁺ CLL cells, with geometric MFI values increasing by a median of 2.25-fold (Fig. 1F). Ki-67 expression was found in up to 45% of PD-1⁺ CLL cells, but in <3% of cells lacking PD-1 (PD-1⁻, Fig. 1G). Interestingly, the percentage of circulating PD-1⁺ CLL cells correlated with the percent of CLL cells expressing Ki-67, indicating that patients with higher percentages of PD-1⁺ CLL cells have a disease with higher proliferation rates (Fig. 1H).

On further phenotypic characterization, we observed that circulating PD-1⁺ CLL cells had higher expression of CD5 and CD27 compared to PD-1^- cells (Fig. 1 I and J). Conversely, expression of the chemokine receptor CXCR4 was lower in PD-1⁺ cells. IgM and IgD expression did not differ significantly. This phenotype suggests that PD-1 expression is observed in cells within the proliferative fraction (6).

PD-1⁺ CLL Cells Upregulate the Cellular Activation and Proliferation Program.

To further characterize PD-1⁺ CLL cells, circulating cells with a PD-1 MFI above the 85th percentile (PD-1^hi cells) from 10 treatment-naïve patients were sorted by flow cytometry, and their transcriptional profile was compared with sorted PD-1⁻ cells (Fig. 2A). We detected 6,580 differentially expressed genes (DEGs), of which 3,412 were overexpressed in PD-1^hi cells using a false discovery rate of 0.05 (Fig. 2 B and C and Dataset S1). Indeed, unsupervised principal component analysis of differentially expressed genes revealed that PD-1^hi cells are transcriptionally distinct, clustering closer to each other than to their PD-1⁻ counterparts regardless of IGHV gene SHM status or cytogenetics (Fig. 2D). Consistent with our phenotypic analysis, PD-1^hi cells had higher gene transcripts for MKI67, PDCD1, CD5, and CD27, and lower transcripts for CXCR4. Other genes involved in proliferation and survival (MYCN, CDK4, CDKN1A, CDKN2A, and BIRC5), chemotaxis (CCL3, CCL4, and CCL5), and cytokine signaling (CD126/IL6R) were also differentially expressed. Interestingly, lower transcripts were observed for genes associated with transforming growth factor-beta (TGF-β) signaling, including SMAD3, SMAD4, TGFBR1, and TGFBR2. This is consistent with previous findings that TGF-β signaling is downregulated upon disease progression (15). Concordance of protein expression of a subset of DEGs was confirmed by flow cytometry (Fig. 3).

Fig. 2.

PD-1⁺ CLL cells are transcriptionally distinct from PD-1⁻ cells and show expression of pathways associated with B cell activation. (A) CLL cells expressing PD-1 MFI above the 85th percentile (PD-1^hi) and PD-1⁻ cells were sorted for RNA-Seq analysis. Sample plots before and after sorting are shown. (B) Heatmap of DEGs (fold change ≥ 2.8, FDR <0.05; paired Student’s t test) 3,412 and 3,168 genes upregulated and downregulated in PD-1^hi, respectively. Each column represents individual samples (n = 10), and each row represents normalized gene expression, with blue and red denoting low and high relative expression, respectively. * = genes with confirmed protein expression as shown in Fig. 1 and Fig. 3. (C) Volcano plot showing log2 fold change versus -log(P-value) for each gene. (D) Unsupervised principal component analysis (PCA) of gene expression in PD-1^hi and PD-1⁻ CLL cells from 10 treatment-naïve CLL patients. (E) Heatmap of select key gene set modules found to be differentially enriched between PD-1^hi and PD-1⁻ CLL cells. Columns represent individual samples, and rows represent normalized sample-level enrichment score. Relevant pathways include BCR signaling, TLR signaling, PD-1 signaling, cell activation, cell cycle, and oxidative phosphorylation. Relative expression of module genes is depicted in a gradient, with blue and red representing a lower or higher expression, respectively.

Fig. 3.

Transcriptional changes between PD-1⁺ and PD-1⁻ CLL cells correlate with protein expression. (A) Histogram plots showing protein expression of a subset of DEGs found on RNA-Seq between PD-1⁺ (red) and PD-1⁻ (blue) CLL cells, as assessed by mean fluorescence intensity using flow cytometry. (B) Quantification of MFI fold change of surface markers shown in A between PD-1⁺ and PD-1⁻ CLL cells. Error bars = median ± IQR.

To better understand the biological relevance of PD-1, we used Gene Set Enrichment Analysis (GSEA) to investigate gene expression differences between PD-1^hi and PD-1⁻ cells. A total of 3,789 gene sets were found to be enriched in PD-1^hi cells, which were condensed into 178 gene set modules using EnrichmentMap (16). These modules included pathways associated with B cell activation, proliferation, migration, signaling, apoptosis, and metabolism (Fig. 2E). Importantly, BCR, TLR, and PD-1 signaling, cell activation, and cell cycle pathways were all upregulated in PD-1^hi cells. We also observed upregulation of oxidative phosphorylation (OXPHOS) genes, which are regulated by BCR signaling and associated with a proliferative drive (17–19). Thus, our analyses showed that PD-1^hi cells carry a phenotypic and transcriptional signature suggestive of recent activation and proliferation.

BTK-Mediated Signaling Modulates PD-1 Expression in CLL Cells.

The factors that lead to the expression of PD-1 in CLL cells are unknown, but we hypothesized that PD-1 expression in CLL cells might be induced through BCR and TLR signaling given that PD-1 expression in T cells is induced upon T cell receptor stimulation (20) and our transcriptomic analysis showed upregulation of BCR and TLR pathway-related genes. To test this, we stimulated CLL cells ex vivo either through the BCR using a polyclonal anti-IgM antibody or through TLR9 using the agonist CpG. We then quantified the percentage of CLL cells expressing PD-1 and their PD-1 MFI by flow cytometry at different timepoints (Fig. 4A). No significant change in PD-1 expression was observed after 8 h of stimulation with either anti-IgM or CpG (Fig. 4 B–F). However, the percentage of PD-1⁺ CLL cells and their PD-1 MFI significantly increased 24 h after stimulation with either agent. Interestingly, stimulation with CpG appeared to exert a stronger effect on PD-1 expression compared to anti-IgM, with 80 to 100% of cells becoming PD-1⁺ within 24 h.

Fig. 4.

PD-1 expression in CLL cells from treatment-naïve patients is induced by BCR and TLR9 stimulation. (A) Study schema. (B) Representative flow cytometry histogram showing the expression of PD-1 after stimulation with anti-IgM or CpG at various timepoints in 2 patient samples. Percentage of PD-1⁺ CLL cells after stimulation with anti-IgM (C) or CpG (D) over time. (E and F) PD-1 MFI in CLL cells after stimulation with anti-IgM (E) or CpG (F) over time. N = 7. ** = P < 0.01, *** = P < 0.001, **** = P < 0.0001 by one-way ANOVA using Sidak to correct for multiple comparisons.

To further study the effect of stimulation on PD-1 expression, we sorted PD-1^hi and PD-1⁻ CLL cells by flow cytometry, stimulated them ex vivo with anti-IgM or CpG for 24 h in the presence or absence of ibrutinib, and then quantified the percentage of PD-1⁺ cells (Fig. 5A). Treatment of flow-sorted PD-1⁻ CLL cells with either anti-IgM or CpG for 24 h resulted in a significant increase in the percentage of PD-1⁺ cells, with CpG again inducing higher percentages of PD-1⁺ CLL cells than anti-IgM (Fig. 5 B–D). Notably, inhibition of BTK-dependent signaling by ibrutinib resulted in near-complete abrogation of the effect of anti-IgM and a marked decrease of the effect of CpG on PD-1 expression. These data indicate that PD-1 expression in CLL cells is induced by B cell activation through a BTK-dependent pathway.

Fig. 5.

BCR- and TLR9-dependent PD-1 expression in CLL cells is abrogated by BTK inhibition. (A) Schematic of the experimental approach. CLL cells from treatment-naïve patients were sorted by flow cytometry based on PD-1 expression, treated with the TLR9 agonist CpG or anti-IgM for 24 h in the presence or absence of the BTKi ibrutinib, and then assessed for PD-1 expression. (B) Representative flow cytometry plots of sorted PD-1^– cells after stimulation with anti-IgM or CpG (vs no stimulation control) for 24 h in the presence or absence of ibrutinib. (C and D) Percent of sorted PD-1⁻ CLL cells that subsequently expressed PD-1 following stimulation with anti-IgM (C) or CpG (D) in the presence or absence of ibrutinib as described in A. Percentage of PD-1⁺ cells increased after stimulation. Ibrutinib blocked the effect of CpG and anti-IgM on PD-1 expression. (E and F) Percent of sorted PD-1⁺ CLL cells that subsequently expressed PD-1 following treatment with anti-IgM (E) or CpG (F) in the presence or absence of ibrutinib as described in A. N = 10. * = P < 0.05, ** = P < 0.01, **** = P < 0.0001 by mixed-effects analysis using Holm–Sidak to correct for multiple comparisons.

Stimulation of previously sorted PD-1⁺ cells with anti-IgM and CpG was also performed (SI Appendix, Fig. S3). We observed no significant changes in PD-1 expression after anti-IgM stimulation in the presence or absence of ibrutinib (Fig. 5E). Higher percentages of PD-1⁺ cells were observed after treatment with CpG, but this was not observed when cells were treated with ibrutinib (Fig. 5F and SI Appendix, Fig. S3). Interestingly, sorted PD-1⁺ cells largely retained PD-1 expression even after treatment with ibrutinib for 24 h, suggesting that BTK-mediated signaling is required for PD-1 expression but lack of signaling through BTK does not result in its immediate downregulation once expressed.

PD-1 Is Not Expressed in Circulating CLL Cells from Patients Responding to BTKi Treatment and Reemerge at the Time of Progression.

BTKis including ibrutinib and acalabrutinib are a mainstay of therapy in CLL. Thus, to study the effects of these treatments on CLL cells in vivo, we collected PB from 40 patients receiving and responding to either ibrutinib or acalabrutinib for at least 1 mo (responders) and measured the percentage of PD-1⁺ CLL cells (Fig. 6A and SI Appendix, Tables S1 and S5). Unlike in treatment-naïve CLL patients, PD-1 expression was detected in <10% of the circulating CLL cells in 38 of the 40 BTKi-treated patients. One patient with >10% circulating PD-1⁺ CLL cells was treated with a 50% dose reduction due to intolerance and the other patient had a history of nonadherence to treatment (Fig. 6 B and C). In contrast, all 22 patients treated with the Bcl-2 inhibitor venetoclax for at least 1 mo and had detectable disease in the PB had >10% circulating PD-1⁺ CLL cells (Fig. 6 A−C and SI Appendix, Tables S1 and S6). We also collected PB from 18 patients receiving BTKi therapy who were subsequently diagnosed with disease progression within 3 mo of sample acquisition (progressors) and measured the percentage of circulating PD-1⁺ CLL cells (Fig. 6A and SI Appendix, Tables S1 and S7). PD-1 expression was observed in 13 to 90% of circulating CLL cells from these patients (Fig. 6 B and C).

Fig. 6.

Circulating PD-1⁺ CLL cells inform treatment response in patients receiving BTKi therapy. (A) Study design for patients receiving and responding to BTKi (ibrutinib or acalabrutinib, N = 40) for at least 1 mo, within 3 mo from progression from BTKi therapy (N = 18), or responding to venetoclax (N = 22). (B) Percent PB PD-1⁺ CLL cells in patients responding to or progressing from BTKi, or responding to venetoclax. PD-1 expression was detected in <10% of circulating CLL cells in most responders while all progressors showed >10% circulating PD-1⁺ CLL cells. Purple = patient with well-documented treatment nonadherence. Blue = patient treated with 50% dose reduction of ibrutinib due to side effects. Error bars = mean ± SEM. **** = P < 0.0001 by the unpaired t test with Welch’s correction. (C) Representative flow plots showing PD-1 expression in PB CLL cells of patients responding to or progressing from BTKi or responding to venetoclax. (D) Study design in patients followed longitudinally before and after initiation of BTKi therapy (N = 21). (E) Flow cytometry plots showing Ki-67 and PD-1 expression in CLL cells over time in two patients who initiated ibrutinib therapy. In patient 26 (orange), circulating PD-1⁺ CLL cells were detected within 2 wk after treatment was discontinued due to side effects. (F) Percent circulating PD-1⁺ CLL cells in patients before and after initiation of BTKi therapy. N = 21. Orange = patient who discontinued BTKi therapy after development of side effects.

To determine the temporal effect of BTKi therapy on the percentage of circulating PD-1⁺ CLL cells in vivo, we measured circulating PD-1⁺ CLL cells in 21 patients prior to and in serial time points after initiating BTKi therapy (Fig. 6 D and E). The percentage of circulating PD-1⁺ CLL cells decreased within 1 mo of initiating BTKi therapy and remained low through the first year (Fig. 6F and SI Appendix, Fig. S4). No significant changes in the expression of CD5, CXCR4, or CD27 were observed with treatment initiation (SI Appendix, Figs. S4 and S5). Of note, CD5 and CXCR4 expression were no longer good markers to identify proliferating subsets in patients receiving BTKi therapy (SI Appendix, Fig. S4). Interestingly, one of the study participants discontinued BTKi therapy due to intolerance, and PD-1 expression was seen in over 41% of the circulating CLL cells within 2 wk of treatment discontinuation (Fig. 6 E and F, orange).

The remarkable decrease in circulating PD-1⁺ CLL cells observed in patients receiving BTKi prompted us to question whether this phenotype can be overcome with additional stimulation. To test this, we measured PD-1 expression in CLL cells from 7 responders after ex vivo stimulation using anti-IgM or CpG. Treatment with anti-IgM resulted in no significant increase in the percentage of PD-1⁺ cells while CpG resulted in <5% of cells expressing low levels of PD-1 (SI Appendix, Fig. S6). No increases were observed after CpG stimulation when cells were also incubated with ibrutinib. Overall, our results indicate that BTKi therapy suppresses PD-1 expression in vivo.

Circulating PD-1⁺ Cells at Progression Shared the Transcriptional Signature of PD-1⁺ Cells of Treatment-Naïve Patients.

To further characterize the PD-1⁺ cells at the time of progression, we performed RNASeq on circulating PD-1^hi and PD-1^- CLL cells from 5 progressors and compared their gene expression profile with CLL cells from 5 responders, which only had PD-1^- cells in circulation (Fig. 7A). There were 1961 and 680 DEGs between cells from responders and PD-1^hi and PD-1⁻ cells from progressors, respectively. We identified 1,425 DEGs when comparing the 3 groups (Fig. 7B). Cells from responders largely shared the signature PD-1⁻ cells from progressors while PD-1^hi cells clustered together (Fig. 7C).

Fig. 7.

PD-1^hi CLL cells reemerge at the time of progression on BTKi therapy and regain transcriptional signature of PD-1^hi cells. (A) CLL cells from patients with CLL responding to and progressing from BTKi therapy were sorted by flow cytometry based on PD-1 expression, and their transcriptional profile was obtained. (B) Heatmap showing DEGs between PD-1^hi and PD-1⁻ CLL cells from progressors and CLL cells from responders. Each column represents individual samples, and rows represent normalized gene expression. Relative expression is depicted using a color gradient, with blue and red denoting lower and high expression, respectively. Column annotation tracks (Top) represent the source of each sample. (C) Unsupervised principal component analysis of CLL cells from patients responding to BTKi therapy (green) and PD-1^hi and PD-1⁻ cells from patients with progressive disease while on BTKi therapy (red and blue, respectively). (D and E) The number of DEGs (D) and differentially overexpressed pathways (E) for PD-1^hi and PD-1⁻ cells from treatment-naïve and patients progressing on BTKi therapy are shown in these Venn diagrams. Minimal DEG overlap was observed between PD1^hi and PD-1⁻ cells in treatment-naïve and progressors. Only 2 pathways overlapped between treatment-naïve PD-1^hi and PD-1⁻ cells from progressors. (F) Differentially enriched gene set modules between treatment-naïve PD-1^hi and PD-1⁻ CLL cells were also differentially enriched in PD-1^hi and PD-1⁻ cells from progressors. Circulating CLL cells from BTKi responders are transcriptionally more similar to PD-1⁻cells. Each column represents individual samples, with column annotation tracks showing the sorted cell source and phenotype.

When comparing DEGs between PD-1^hi and PD-1⁻ cells from treatment-naïve patients and progressors, we observed that only three genes were shared between treatment-naïve PD-1⁻ and PD-1^hi cells from progressors (ZBTB7A, PICK1, and LMF2) and 2 genes were shared between treatment-naïve PD-1^hi and PD-1⁻ cells from progressors (MTMR1 and NAA50, Fig. 7D). Comparisons between these populations at the gene set level showed only two overlapping pathways between treatment-naïve PD-1^hi and progressor PD-1⁻, none of which are associated with CLL or B cell biology (“Rickman Head and Neck Cancer E” and “GSE22601 Immature CD4 Single Positive vs Double Positive Thymocyte Up”, Fig. 7E). GSEA revealed that the transcriptional signature of treatment-naïve PD-1^hi cells reemerged in PD-1^hi cells at the time of progression, with overexpression of genes associated with cell activation, cell cycle, oxidative phosphorylation, cytokine response, and PD-1, BCR, and TLR signaling (Fig. 7F). These findings suggest that the transcriptional activity associated with PD-1 expression is retained in this subset of cells at the time of progression.

Discussion

In this study, we showed that a subset of circulating CLL cells in all treatment-naïve patients expressed PD-1, a protein that is not typically expressed in B cells. Importantly, we showed that PD-1 is expressed in the fraction of circulating CLL cells that are proliferating and demonstrated that PD-1 expression is induced in a BTK-dependent manner via BCR and TLR signaling. Our data also demonstrate that circulating PD-1⁺ CLL cells are observed when BTK-mediated signaling is not completely inhibited, implying that PD-1 could be a marker of CLL cell activation and that the presence of circulating PD-1⁺ CLL cells in patients receiving BTKi therapy may be an early biomarker of BTKi resistance.

Signaling through both the BCR and TLR pathways has been shown to play key roles in the activation of CLL cells (9, 21). While stimulation through either BCR or TLR9 increased the percentage of PD-1⁺ CLL cells and transcriptomic signatures suggest that both pathways were activated in PD-1^hi cells, treatment with ibrutinib completely blocked the effect of BCR activation and only partially blocked the effect of TLR9 signaling. This partial inhibition of TLR9 signaling by ibrutinib supports a cross-talk between the BCR and TLR signaling pathways, as previously observed in aggressive lymphomas, autoreactive B cells, and in CLL (21–24). However, given that patients on BTKi rarely achieve undetectable minimal residual disease, the robust induction of PD-1 expression after TLR9 stimulation raises the possibility that TLR9 signaling may provide enough stimulation to maintain tumor viability in the presence of BTKi. Therefore, combining BTKi with inhibitors of TLR9 signaling may be a strategy that warrants further investigation as it may synergize and improve the clinical efficacy of BTKi alone in the treatment of CLL.

Several aspects of the biology of PD-1 in CLL cells are still unknown. B cells expressing PD-1 have been observed in chronically inflamed human tonsils (25). Given that lymph nodes are the primary site where CLL cells receive BCR signaling and proliferate (2), it is likely that PD-1⁺ CLL cells are also present in lymph nodes. This would be consistent with PD-1 expression being observed in nearly all circulating Ki-67⁺ CLL cells and in cells with higher expression of CD5 and lower expression of CXCR4, which suggests recent lymph node egress (6). Furthermore, the transcriptional signature observed in PD-1^hi cells highly resembled the signature of lymph node CLL cells reported by Herishanu, et. al. (9). However, the degree of cell activation and proliferation cannot be reliably assessed using CD5 and CXCR4 expression once patients started treatment with BTK inhibitors. PD-1 expression, on the other hand, is a more reliable marker of response to BTKi therapy and more clearly correlates with the presence of proliferating cells.

By analyzing circulating CLL cells with high PD-1 expression, we detected the subset of highly active cells that are more likely found in lymph nodes, providing an opportunity to further study the biology of proliferating cells without resorting to invasive lymph node biopsies. Detailed analysis of PD-1^hi cells uncovered several differentially expressed proteins that could inform future therapeutic strategies. Among these proteins is LILRB4, a surface protein commonly expressed by antigen-presenting cells and contributes to T cell dysfunction in acute myeloid leukemia (26). In CLL, LILRB4 may also help regulate BCR signaling (27). The predilection of LILRB4 expression toward PD-1⁺ CLL cells suggests that LILRB4 may be an attractive therapeutic candidate to target earlier, proliferating cells for elimination.

PD-1 plays a central role in regulating T cell function, but its function in CLL cells is unknown. PD-1 may have an important biological function because its expression is retained when CLL transforms into aggressive lymphoma (i.e. Richter transformation) (28). Indeed, interactions between CLL cells expressing PD-1 and surrounding cells expressing PD-L1/L2 may help generate a microenvironment conducive for CLL growth, as suggested by the effects of PD-L1-blocking antibodies in a mouse model of CLL (29). Interestingly, while PD-1 could itself represent a therapeutic target in CLL, PD-1 blocking antibodies showed promising activity in Richter’s transformation but not in CLL (28, 30–33), suggesting that PD-1 blockade alone is not sufficient in CLL and other strategies may need to be pursued. More studies are needed to understand the role of PD-1 signaling in the biology of CLL and Richter transformation, its possible contribution to immune dysregulation, and to elucidate the best strategy to exploit the atypical expression of PD-1 and its signaling pathway as a therapeutic target.

CLL remains incurable despite our current therapeutic advances. In fact, patients who progress through BTKi and BCL-2 inhibitors or develop Richter transformation have very limited treatment options, and their outcomes are very poor. Additionally, CLL patients have a very dysfunctional immune system that results in lower efficacy of cellular therapies and vaccines, contributing to the morbidity and mortality in these patients (34–36). These limitations highlight the need to better understand the factors that promote CLL proliferation, treatment resistance, and immunosuppression. Our data show a simple way to identify a subset of circulating CLL cells that can offer insight into the biology of CLL in lymph nodes while serving as a potential biomarker to monitor the efficacy of BTKi therapy and detect the development of resistance to BTK inhibition.

Materials and Methods

Patient Samples and Collection of Human Peripheral Blood Mononuclear Cells (PBMCs).

Samples were collected from patients who received care at the Winship Cancer Institute of Emory University under a protocol approved by the IRB of Emory University. After written informed consent was obtained, blood samples from patients were collected either in sodium citrate CPT cell preparation tubes (BD) or in standard EDTA tubes. Up to 1 mL of blood was removed for whole blood staining, and the remaining sample was immediately processed to separate the PBMCs and the plasma. PBMCs were washed three times with Dulbecco’s phosphate buffer saline without calcium or magnesium (DPBS) plus 2% fetal bovine serum (FBS) and 2 mM EDTA (MACS buffer) after lysis of red blood cells using ACK lysing buffer (Lonza), suspended in 90% FBS plus 10% DMSO, frozen in freezing chambers at −80˚C, and then transferred to liquid nitrogen for long term storage. Plasma was stored in aliquots at −80˚C. Control samples from healthy individuals were obtained from individuals who provided informed consent and were enrolled in several Emory University Institutional Review Board–approved protocols studying immune responses.

Flow Cytometry.

FACS analysis was performed using either fresh or cryopreserved mononuclear cells. Cells were incubated with appropriate surface antibody mix for 15 min then 1x FACS/Lyse (BD) was added to each sample. After a 10-min incubation, the sample was centrifuged, and pelleted cells were washed once with MACS buffer, permeabilized with the FOXP3 buffer kit (eBiosciences) per the manufacturer’s protocol to allow for staining of intracellular markers. Following a 15-min incubation with antibodies against intracellular markers, cells were washed once with 1x FOXP3 wash buffer and once with MACS buffer prior to sample analysis on a BD LSR II or a Cytek Aurora cytometer. Antibodies used in this study are listed in SI Appendix, Table S8. Acquired data were analyzed using FlowJo v10 (BD).

RNA Sequencing Analysis of Flow-Sorted CLL Cells.

Previously frozen PBMC samples were thawed and stained with Live/Dead Fixable Near-IR Dead Cell Stain (Invitrogen) followed by antibodies directed against CD3, CD19, CD20, CD5, and PD-1. Cells were sorted into RLT buffer at the Emory University School of Medicine Flow Cytometry core using a BD FACSAria II. RNA isolation, library preparation, and sequencing were performed by the Emory University National Primate Center Nonhuman Primate Genomics Core at Emory University. Patient effects from sorted samples were batch corrected using ComBat (37). Raw FASTQ files were processed with the Sekaly lab pipeline on the Emory AWS platform: After sequencing, reads are processed to remove Illumina adapters and low-quality 3′-end bases using the Trimmomatic software (38), and then mapped to the reference human genome version GRCh38 using the RNA-seq optimized software STAR (39). RSeQC was then used to assess strand-specificity of reads for all transcripts (40). Transcript abundance was then estimated from unique mapped reads into raw counts using HTSeq (41). R package DESeq2 (version 1.40.2) was then used to normalize read counts among samples and to identify differentially expressed genes between biological samples (42). For the pretreatment samples, PD-1^hi/− sorted subsets were compared in a paired manner, using patients in the design. For on-treatment samples, the three conditions (Ibrutinib treated, PD1^high and PD1⁻) were compared in an unpaired manner, as well as a using the full 3-way model using the Log Rank Test. Unless otherwise specified, a Wald test was used to evaluate the statistical relevance of the observed variations given its reproducibility between biological replicates, and a Benjamini–Hochberg correction for large number of measurements was applied to obtain adjusted P-values. Genes of interest were selected based on statistical significance (adjusted P-value <0.05), and Bayesian shrinkage estimation, with the apeglm method (43), was applied to the fold change to estimate effect size more accurately. Hierarchical clustering with the ward D2 method linkage was performed using Euclidean distance and displayed using the ComplexHeatmap R package.

Preranked GSEA was performed for each contrast and/or correlation against gene sets extracted from the MSigDB (BROAD Institute) (44), and CHEA (45, 46) databases. The shrunken fold change was used as a ranking metric. Gene sets found to be significantly enriched (adjusted P-value < 0.05) associated with the breadth of the response were considered as differentially activated pathways. EnrichmentMap was then used to reduce pathway redundancy of enriched pathways by generating gene set pathway modules of overlapping gene sets on the basis of shared genes using a Jaccard distance cutoff of < 0.25 (47). Manual curation was used for naming the modules, on the basis of member gene set names and biological role of core genes.

GSVA (Gene Set Variation Analysis) R package (48) was then used to compute a sample-level gene set enrichment z-score for significant EnrichmentMap modules to be significantly enriched using GSEA, on the basis of the expression level of the core genes. Sample-level z-scores were then used for correlation with other OMICs.

Ex Vivo Stimulation of CLL Cells.

Previously frozen PBMC samples were thawed, washed once with MACS buffer, and resuspended in B cell medium (RPMI plus 10% FBS plus 1% pen/strep plus 1% L-Glutamine plus 55 µM 2-mercaptoethanol) plus stimulating agent, transferred to round-bottom 96-well plates, and incubated at 37 °C until time of analysis. For experiments requiring cell sorting, previously frozen PBMC samples were thawed and stained with Live/Dead Fixable Near-IR Dead Cell Stain (Invitrogen) followed by antibodies directed against CD3, CD19, CD20, CD5, and PD-1. Cells were subsequently washed once with MACS buffer and sorted into FBS at the Emory University School of Medicine Flow Cytometry core using a BD FACSAria II as shown in Fig. 3A. Sorted cells were washed once with MACS buffer, resuspended in B cell medium (RPMI plus 10% FBS plus 1% pen/strep plus 1% L-Glutamine plus 55 µM 2-mercaptoethanol) plus stimulating agent and/or inhibitor, transferred to round-bottom 96-well plates, and incubated at 37 °C until time of analysis. The following concentrations of stimulating agents and/or inhibitors were used: 5 µg/mL CPG ODN 2006 (Invivogen), 20 µg/mL anti-IgM (Jackson ImmunoResearch), 5 µM ibrutinib (Selleckchem). At the time of analysis, cells were pelleted, stained with antibodies directed against CD3, CD19, CD20, CD5, and PD-1, and analyzed on a BD LSR II or a Cytek Aurora instrument. Acquired data were analyzed using FlowJo v10 (BD).

Statistical Analysis.

Statistical analysis was conducted using GraphPad Prism 10.0 (RRID:SCR_002798) or R. Significance level was set at P < 0.05, two-tailed, for all analyses. Descriptive statistics were performed to tabulate patients’ demographic and clinical characteristics. Frequency and percentage, mean and SE of the mean or median with interquartile range (IQR) were reported based on the data structure of variable. Statistical differences were assessed by group using Kruskal–Wallis, Brown–Forsythe, and Welch ANOVA test, repeated measures one-way ANOVA, mixed effect analysis, or Student’s t test, when appropriate. Multiple comparison is accounted for by utilizing Dunnett T3, Bonferroni, or Dunn’s test based on the selected statistical test and validity of assumptions. A least squares regression was used to determine the relationship between PD-1 and Ki-67 expression. Welch’s t test was used to compare the mean percentage of PD-1 + CLL cells in circulation based on cytogenetic and IGHV mutation status. Given that patients with unknown IGHV mutation status were present, a sensitivity analysis assuming that they had mutated (m) or unmutated (u) IGHV was also performed. The plots of the residuals (Q–Q plots) from each variable were used to examine to determine violations of assumptions and selection of appropriate statistical methods e.g., parametric or nonparametric statistical methods. Analysis of transcriptomic data was performed as outlined above.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Data, Materials, and Software Availability

All code used for the analysis can be found at https://github.com/achang06/PD-1-CLL.git (49). Anonymized RNA Sequencing data have been deposited in GEO (Accession: GSE283045) (50).