Development, Function, and Clinical Significance of Plasmacytoid Dendritic Cells in Chronic Myeloid Leukemia

BCR-ABL1 is a fusion oncogene, which arises from the t(9;22)(q34;q11) chromosomal translocation in a hematopoietic stem cell (1, 2). Chronic myeloid leukemia (CML) seems to be sensitive to immune therapy, such as with IFNα, which stimulates CML-specific T-cell responses (3, 4). Mounting of a specific T-cell response requires, first, recognition through the T-cell receptor of peptide–MHC class I complexes on the surface of antigen-presenting dendritic cells (DC). However, there is conflicting evidence regarding the capability of BCR-ABL–expressing leukemic CML stem cell–derived DCs (5–9) to trigger CML-specific T-cell immunity.

Little is known about the origin, regulation, and function of a specialized type of DCs, plasmacytoid DCs (pDC) in CML. Normal pDCs are the main source of a key Th1 immune-stimulatory cytokine, IFN (10–12). pDCs regulate innate and adaptive immune responses (12, 13) and contribute to immune activation (13), but also tolerance (14, 15). Untreated patients with CML were reported to have reduced pDC counts (16–18), which might involve BCR-ABL–mediated suppression of interferon regulatory factor 8 (IRF8, ICSBP) gene expression (19–23), a supposed pDC-fate–defining transcription factor (24, 25). Although BCR-ABL–induced Irf8 suppression was linked to impaired pDC development based on a murine model of CML (20), recent genetic in vivo evidence demonstrated that pDC development is Irf8-independent (26), essentially requiring only E2-2 (Tcf-4) and Zeb-2 (27, 28). However, Irf8 is clearly of key importance for the function of pDCs—the most prominent being IFN production (24–26). In a translational study of the multicenter tyrosine kinase inhibitor (TKI) discontinuation trial, EUROSKI, we recently observed unusual high counts of mature (CD86⁺) pDC in patients with CML in deep molecular remission (MR). Their abundance was associated with CD8⁺ T-cell exhaustion and predictive of a significantly lower chance of sustaining treatment-free remission (TFR) after TKI stop (29).

Here, we characterized the origin, genetic, and biological characteristics of pDCs in untreated CML (CML-pDCs) and corroborated specific aspects of CML-pDC development and function using a CML mouse model. Furthermore, we prospectively investigated qualitative and quantitative pDC changes under nilotinib versus nilotinib plus IFN therapy within a large multicenter CML trial. We find that CML-pDCs emerge from low-level BCR-ABL⁺ stem cells, produce inflammatory cytokines, and retain important functional properties, such as IFN secretion, maturation, and elicitation of CML-specific immunity. Data support for the first time that CML-pDCs are a source of inflammation in CML and likely involved in the regulation of treatment response and control CML-specific immunity.

WT C57BL/6J mice, male at 6 to 10 weeks of age and female mice at 12 to 14 weeks of age, were used. Animal experiments were performed according to German law and approved by the regional board “Regierungspräsidium Giessen” (animal proposal # V54-19c20-15(1)MR20-36 Nr.07/2010 and # V54-19c2015h01MR Nr47/2014). 32D cell line was purchased from the “Deutsche Sammlung von Mikroorganismen und Zellkulturen GmbH” (DSMZ), and Phönix eco cell line was obtained from Allele Biotechnology. Cells were tested periodically and confirmed as Mycoplasma free by the PCR-based method (myco: 5′-gggacgaaacaggattagataccct-3′ and 5′-tgcaccatctgtcactccgttaacctc-3′), and authentication was not conducted, unless by morphology check due to microscope. Both cell lines were maintained according to the supplier's recommendations and not cultured more than 2 months.

The retroviral vectors pMIGp210 (encoding the BCR-ABLp210 and gfp), referred to as BCR-ABL, and pMIG (control, encoding only gfp) were a kind gift from Dr. W. Pear (The University of Pennsylvania, Philadelphia, PA; ref. 30). Using this vector as basis, the [BCR-ABL]-2A-[IRF8]-2A-[GFP] plasmid (BARF) was constructed, coexpressing BCR-ABL and Irf8 and gfp. Retroviral particles were generated with Phoenix eco cell line as described previously (31).

Bone marrow of wild-type mice was harvested, enriched for pDCs (pDC isolation kit II, Miltenyi), and spinoculated with either BCR-ABL– or BARF-containing retroviral particles. During spinoculation, pDCs were cultured in RPMI Medium 1640 (Gibco) and supplemented with 10% FCS and 1% penicillin/streptomycin, 20 mmol/L HEPES, 1 mmol/L sodiumpyruvat, 50 μmol/L mercaptoethanol (all Gibco), and 100 ng/mL FLT3-ligand (FLT3-L; Roche). GFP-positive pDCs were sorted and then i.v. injected into the tail vein of wild-type recipient mice (1,000–3,000 cells). pDC injections were repeated weekly in each mouse, for 3 times.

CD8⁺ T cells were enriched at a purity of up to 80% from spleens of pDC-immunized animals, and 2.5 × 10⁶ of these were adoptively transferred i.p. the day after CML stem cell transplantation of independent recipient animals. To control for engraftment, blood samples from mice tail vain were analyzed weekly for proportions of GFP⁺ cells.

Western blotting of 32D cells and primary murine cells was performed as previously described. The following primary antibodies were used: anti-IRF8 goat, clone C-19 and anti-c-Abl mouse, clone 24-11 (both Santa Cruz Biotechnology) and anti-Actin mouse, clone AC-15 (Sigma-Aldrich). The secondary antibodies were horseradish peroxidase–conjugated polyclonal rabbit anti-goat and polyclonal goat anti-mouse antibodies (both Dako).

Peripheral blood samples of patients with primary diagnosis of CML or during follow-up were received through the immunologic substudy of the German multicenter CML-V study. In accordance to the Declaration of Helsinki, all patients gave written informed consent to participate in this substudy of CML-V or EUROSKI. We confirm that both studies were approved by the local Ethic Committees [NCT01657604 (CML-V) and NCT01596114 (EUROSKI)]. Primary blood mononuclear cells (PBMC) were isolated by Biocoll density gradient separation (density 1.077 g/mL, Biochrom).

Flow cytometry was performed on the LSRII (BD Biosciences), and data were analyzed using the FlowJo software (FlowJo LLC). For fluorescence-activated cell sorts, cell populations were enriched prior to sorting using magnetic bead enrichment (Miltenyi). Sorting was performed using the MoFlo XDP (Beckman Coulter). For analysis of maturation status, the PBMCs were cultured for at least 18 hours with and without 2 μmol/L CpG 2006 in RPMI Medium 1640 (Gibco) and supplemented with 10% FCS and 1% penicillin/streptomycin, prior to antibody staining. Two four-color stainings were performed for each, untreated and treated cells, which included FcR blocking, BDCA-2 (PE), CD123 (FITC), and either HLA-DR (APC) + CD86 (TRI-COLOR) or CD80 (APC) + CD40 (TRI-COLOR). Due to very low cell numbers, isotype controls for APC- and TRI-COLOR stainings were performed on CpG-treated cells and used for both untreated and treated cultures. A list of used antibodies and enrichment kits is attached (Supplementary Tables S1 and S2). For details on flow cytometry (sample preparation, gating, and abundance of populations), see Supplementary Table S3.

For intracellular staining of primary human pDCs, enriched cells (Supplementary Table S2) were cultured in 100 μL RPMI Medium 1640 (Gibco) and supplemented with 10% FCS and 1% penicillin/streptomycin, in 96-well U-bottom plates for 5 hours with and without 5 μmol/L CpG 2216 (TIB MOLBIOL). After 2-hour incubation, 1 μL Golgi-Plug (BD Biosciences) was added into each well. Then the cells were stained for surface markers and afterward fixed with 100 μL Medium A for 15 minutes at room temperature (dark). This was followed by permeabilization with 100 μL Medium B (both Thermo Fisher Scientific) and simultaneous staining with the intracellular IFNα antibody, FITC-conjugated (clone 225.C, Chromaprobe) for 20 minutes at 4°C. The cells were resuspended in 2% formaldehyde (Fischer) and analyzed on the LSRII.

Bone marrow of wild-type mice was harvested by flushing femura and tibiae. Mononuclear cells were obtained after density-gradient centrifugation (density 1.086 g/mL, Pancoll, PAN-Biotech). Stem cells for transplantation were enriched from bone marrow mononuclear cells using immunomagnetic column separation (lineage depletion) of lineage-positive cells (Lineage Cell Depletion Kit, Miltenyi) and then enrichment of Sca-1⁺ cells (easy sep mouse SCA1-positive selection kit, stem cell). Isolated stem cells had a purity of 85% to 95%, were precultured for 18 hours in IMDM (Gibco, Thermo Fisher Scientific), and supplemented with 20% FCS, 1% penicillin/streptomycin (both PAN Biotech), 50 μmol/L Mercaptoethanol (Gibco), 10 ng/mL recombinant murine IL3, 50 ng/mL recombinant IL6, and 50 ng/mL recombinant murine SCF (all Immunotools). Cells were transduced with either pMIG, BCR-ABL-, or BARF-retroviral particles by spinoculation (in the presence of 5 μg/mL polybrene) for 2 consecutive days. GFP positivity was assessed by flow cytometry. Note that 3 × 10⁴ GFP⁺ cells were injected intravenously via tail vein into sublethally (7 Gy) irradiated female recipient mice. Animal experiments were performed according to German law and approved by the regional board “Regierungspräsidium Giessen” (animal proposal # V54-19c20-15(1)MR20-36 Nr.07/2010 and # V54-19c2015h01MR Nr47/2014). For details regarding animal experimentation (sample size calculation, data exclusion, replication, randomization, and blinding), see Supplementary Table S4.

BCR-ABL FISH was performed as previously described (32).

Murine RNA was isolated and reverse transcribed using RNeasy Micro Kit and Omniscript RT Kit (both Qiagen) according to the manufacturer's protocol. RNA of human samples was isolated and transcribed using TriFast (peqlab) and SuperScript III reverse transcriptase (Thermo Fisher Scientific). Real-time quantitative PCR data for IRF8 gene expression were performed using the ΔΔC_t method, with GAPDH as the housekeeping gene. Samples were analyzed in triplicates using QuantiTect SYBR Green Mix (Qiagen). The following primer pairs were used: mouse gapdh: 5′-catggccttccgtgttccta-3′ and 5′-cctgcttcaccaccttcttgat-3′; mouse irf8: 5′-tgccggcaagcaggattaca-3′ and 5′-ccacgtggctggttcagctt-3′; human gapdh: 5′-ctcctccacctttgacgctg-3′ and 5′-accaccctgttgctgtagcc-3′; human irf8: 5′-gtcccaactggacatttccg-3′ and 5′-cattcacgcagccagcag-3′. Quantitation of BCR-ABL levels in patients was performed following procedures and definitions of molecular response as previously described. Results were reported analogous to the European LeukemiaNet criteria as a normalized transcript number of the international scale, IS (33). Absolute BCR-ABL transcript numbers were assessed relative to GUS expression.

For competitive BCR-ABL and gfp PCR, we used genomic DNA of either spleen or chloroma tissue. For amplification, we used DreamTaq DNA Polymerase (Thermo Fisher Scientific), and the following 2 primer pairs: ba-1 5′-ttcagaagcttctccctggcatccgt-3′ and a-1 5′-ggtaccaggagtgtttctccagactg-3′, generating a 487 bp fragment, and gfp-for 5′-cgtaaacggccacaagttca-3′ and gfp-rev 5′-tcttgtagttgccgtcgtcc-3′, generating a 260 bp product.

Cytological smears were stained by the panoptic method of Pappenheim and analyzed using the Olympus BH-2 microscope (Olympus), oil objective, ×100. Images were done with the DHS picture database software (Dietermann & Heuser Solution GmbH).

Next-generation sequencing of ASXL1 from genomic DNA of CD34⁺ enriched cells was performed as described previously (34). For conventional sequencing, gDNA from sorted pDCs and CD34⁺/CD38⁺ cells was preamplified with GoTaq DNA Polymerase (Promega) using the following primers: human ASXL1 mut2: 5′-cacttacaaaagaccagagcca-3′ and 5′-ctggatggagggagtcaaaa-3′. The resulting 237 bp DNA product was cleaned up by the NucleoSpin Gel and PCR Clean-up Kit (Machery and Nagel) and then sequenced with the human ASXL1 mut2 reverse primer by Seqlab. In a patient harboring the BCR-ABL F359I mutation, cDNA of FACS-sorted pDCs and PBMCs was used in PCR with Phusion Green High Fidelity DNA-Polymerase (Thermo Fisher Scientific). The primers of ABL-A, 5′-acagcattccgctgaccatcaataag-3′ and 5′-atggtccagaggatcgctctct-3′, amplified a 1,718 bp fragment. Using this fragment, a nested PCR was performed with the primers pair: human BCR-ABL F359I 5′-catgacctacgggaacctcc-3′ and 5′-ggccaaaatcagctaccttc-3′. A 200 bp PCR product was sequenced with the BCR-ABL F359I forward primer.

Sequences were analyzed for mutations using the BLAST Align software (www.ncbi.nlm.nih.gov/BLAST/) and are shown as electropherogram.

Total RNAs were isolated from sorted pDCs of healthy donors and the first diagnostic patients with CML with RNeasy Micro Kit (Qiagen), according to the manufacturer's protocol. RNA libraries were generated with the NEBNext Ultra RNA Library Prep Kit for Illumina (New England Biolabs Inc.), according to the manufacturer's protocol. The libraries were sequenced on an Illumina HiSeq 2000, requested Read Length HiSeq 2 × 75 bp with paired end.

Data were analyzed with GraphPad Prism 6 software (GraphPad Software, Inc.). Statistical significance of data was analyzed by the unpaired t test or ANOVA, after testing for normal distribution, unless indicated otherwise. Significance of survival difference was analyzed by the Mantel–Cox test. P value is shown as: ns, nonsignificant, P > 0.05; *, P < 0.05; **, P < 0.01; ***, P < 0.001; and ****, P < 0.0001.

Flow cytometry data were acquired with BD FACSDIVA software (Version 6.1.3; BD Biosciences) and analyzed by FlowJo software (either on windows version 7.6 or Mac version 9.5) from FlowJo LLC.

The Summit software (version 4.1; Cytomation) was used for acquiring cell sorts.

For Pappenheim image acquisition and processing, the DHS picture database software was used (Dietermann & Heuser Solution GmbH).

Statistical analysis was performed with GraphPad Prism 6 software (GraphPad Software, Inc.).

Patients with untreated CML were previously reported to have strongly reduced pDC counts (16–18). However, here we show that patients with treatment-naïve CML only have lower pDC proportions in the peripheral blood (% pDC in white blood cells; Fig. 1A and B), whereas their absolute pDC numbers (pDCs per μL) are not reduced (Fig. 1C). Moreover, we found that CML-pDCs are BCR-ABL–positive (BCR-ABL⁺) by FISH (median: 81%; range: 57%–100%, n = 8; Fig. 1D). They also express normal IRF8 levels when compared with normal donor pDCs (Fig. 1E). Thus, human data do not show a link between BCR-ABL, IRF8 expression, and pDC counts in CML.

To study the origin of CML-pDCs and a previously proposed dependence of CML-pDC development on Irf8, we asked whether a block of CML-pDC transdifferentiation from leukemic stem cells is caused by BCR-ABL–mediated Irf8 suppression in vivo. The murine transduction/transplantation (tt)-CML model was employed (35). In the tt-CML mouse model, leukemogenesis depends on the engraftment of BCR-ABL–transformed Lin⁻Sca1⁺c-Kit⁺ cells (LSK; refs. 36, 37). BCR-ABL⁺ hematopoiesis can be tracked using GFP expression as a marker. Irf8 was strongly reduced by BCR-ABL⁺ in CML stem cells (CML-LSK), common lymphoid (c-Kit⁺Sca-1⁺IL7-receptor alpha⁺GFP⁺; CLP), and myeloid progenitors (c-Kit⁺Sca-1⁻CD16/32⁻CD34⁺GFP⁺; CMP; Supplementary Fig. S1A), but, in contrast to patients with CML (Fig. 1C), CML mice showed a severe pDC loss (Supplementary Fig. S1B). Moreover, restoring IRF8 expression in BCR-ABL⁺ CML stem cells in vivo using a BCR-ABL/Irf8 coexpression construct (BARF; Fig. 2A) did not rescue CML-pDC maturation, suggesting that the CML-pDC maturation block is causally linked to BCR-ABL overexpression, but not IRF8 suppression (Fig. 2B). This was further supported by the finding that BCR-ABL caused an expansion of CML-LSK, T cells (CD3⁺), B cells (B220⁺), and myeloid cells (Gr1⁺), but selectively blocked maturation of pDCs irrespective of restored Irf8 levels (Fig. 2C). Of note, the few GFP⁺ pDCs, which transdifferentiated from GFP⁺ stem cells, consistently expressed much lower BCR-ABL levels (GFP) than the remaining leukemic bulk in the same animals (Fig. 2D). Likewise, human CML-pDCs expressed in median 3.5-fold lower BCR-ABL mRNA (range: 1.2–6.8) than the leukemic bulk (Fig. 2E). This supports that high BCR-ABL expression thresholds, not IRF8 loss, suppress CML-pDC maturation. We concluded that the reason of maintained CML-pDC maturation in human CML as opposed to CML mice is clonal BCR-ABL expression heterogeneity (31, 38, 39). CML-pDCs likely originate from stem cells with lower-level BCR-ABL expression, which is rare in tt-CML mice. Indeed, a heterogeneous evolutional trajectory of CML bulk cells and CML-pDCs was supported genetically in 2 patients with CML. One patient carried an ABL-resistance mutation, detectable only in the sorted CML bulk, but not in the sorted pDC fraction (Fig. 2F). The other patient had an ASXL1 mutation that was detectable in sorted CML-CD34⁺ cells, but not in CML-pDCs (Fig. 2G).

In order to obtain insights into the activation- and functional status of CML-pDCs, we compared the gene expression profiles of CML-pDCs and normal donor pDCs using RNA sequencing (Fig. 3A). Gene expression profiles of CML-pDCs (n = 6) differed significantly from normal pDCs (n = 3; Fig. 3B; Supplementary Table S5). With a median sequencing depth of 127.2 × 10⁶ mapped reads (CML-pDC samples) and 154.1 × 10⁶ mapped reads in normal pDC samples, 3,109 genes were found to be significantly regulated (log₂FC ≥ 1) in CML-pDCs (Fig. 3B; Supplementary Data File S1). The majority of these differentially regulated genes (63.5%) were upregulated (Fig. 3C). Significantly more of these genes (20.5%) were strongly upregulated in CML-pDCs, whereas only a minority (5.8%) were strongly downregulated. Gene set enrichment analysis (40) showed enrichment in CML-pDCs of nine classes of genes involved in antigen capturing, processing, and presentation by MHC class I and II, as well as IFN, IL, and Toll-like receptor (TLR) signaling (Fig. 3D). The leading-edge subsets of genes in these nine classes were commonly overlapping between 2, 3, 4, or more classes, consistent with their critical importance (Fig. 3D; Supplementary Data File S2).

IFN secretion is a key function of activated normal pDCs and Irf8-dependent (26, 27). mRNA sequencing shows that CML-pDCs constitutively express higher basal levels of IFNs than normal donor pDCs in an IRF8-dependent manner (Fig. 4A). IFN secretion is also promptly induced upon in vitro stimulation with the TLR9 ligand (oligonucleotide CpG 2216; Fig. 4B). In contrast, healthy donor-derived pDCs do not express or secrete IFN unless stimulated with CpG (Fig. 4A and B). In addition, also other cytokines, which are implemented in the elicitation of T-cell responses (e.g. ILs, CXCL9, 10, 11, CCL 22) or CML stem cell survival (TNFα, TGFβ; refs. 37, 39, 41), are overexpressed in CML-pDCs. TGFβ is strongly expressed also in normal pDCs (Fig. 4C). Intriguingly, although normal pDCs mature and then upregulate MHC class II and T-cell costimulatory molecules (CD80, CD86, and CD40) in response to CpG-induced TLR signaling (Fig. 4D, top), CML-pDCs are frequently spontaneously mature/activated (CD86⁺, CD40⁺) and thus resemble a recently described subpopulation of normal BDCA-2⁺CD123⁺ pDCs, so-called Axl⁺Siglec-6⁺ DCs (AS-DCs) or pre-DCs (Figs. 3A and 4D, bottom; Supplementary Fig. S2; refs. 42, 43) and display activated TNF, JAK, and NF-κB signaling (Supplementary Fig. S3), suggesting a constitutive autocrine stimulation of CML-pDCs by secretion of TNF, IFNs, and ILs. Only a minority of patients with de novo CML (exemplary shown pat. # 009-501) do not show spontaneous maturation and CD86, CD80, or CD40 expression (Supplementary Fig. S2). However, also immature CML-pDCs (pat. #009-501) retain their IFN secretion capacity and mature upon CpG treatment (Supplementary Fig. S2).

In vivo treatment with the BCR-ABL–specific TKI nilotinib reduced CML-pDC counts within months of therapy (Fig. 5A). This suggests an antiproliferative effect of TKI on CML-pDC precursors. A more potent reduction in pDC counts occurred under cotreatment with nilotinib and IFN within the CML-V trial (Fig. 5A). This combined therapy also reduced the number of mature (CD86⁺) pDC more efficacious than nilotinib alone. This could be clinically important, because CD86⁺pDCs were previously linked to a lower TFR probability (29). Of note, there is a rapid decline of the CD86⁺pDC counts after 6 to 12 months under nilotinib and more pronounced with nilotinib plus IFN. CD86⁺pDC counts were comparably low in patients with long-term imatinib treatment (median, 7.5 years) within the EUROSKI TKI discontinuation trial (Fig. 5A and B). More than 10 CML-pDCs/μL at diagnosis—corresponding to the median number of pDC detected in MR (Fig. 5A)—were associated with a reduced probability to achieve a MR4 after 12 months of therapy with nilotinib (Fig. 5C). Interestingly, if patients with CML-pDC counts > 10/μL were randomized to receive nilotinib plus IFN, an inferior molecular response was no longer seen (Fig. 5C). PDCs from patients in MR were BCR-ABL–negative by FISH (n = 8/8), which enabled to analyze approximately only 100 pDCs per patient (Fig. 5D). However, when using absolute quantification and a standardized, international scale BCR-ABL^IS PCR (sensitivity 5 × 10⁻⁵), a median of 7,700 BCR-ABL copies (range, 470–17,000) were detected in every patient. Thus, numbers of residual CML-pDCs remain detectable by PCR in all 18 analyzed patients—even in very deep MR of CML (Fig. 5E). This suggests that a significant proportion of BCR-ABL⁺ CML-pDCs is TKI-insensitive.

To experimentally assess the functional competence and biological relevance of persisting BCR-ABL⁺ pDCs in vivo, wild-type C57BL/6J mice were vaccinated with BCR-ABL⁺ pDCs or different types of control pDCs. Vaccination occurred intravenously weekly for 3 successive weeks using 1 to 3 × 10³ in vitro–generated control pDCs (pDC^wt), BCR-ABL⁺ pDCs (pDC^BCR-ABL), BARF⁺ pDCs (pDC^BARF), or gfp⁺ Mig-control pDCs (pDC^Mig), respectively (Fig. 6A).

CD8⁺ cytotoxic T cells from these vaccinated mice (Tc^BCR-ABL, Tc^BARF, Tc^wt, and Tc^Mig) were sorted and adoptively transferred into other recipients: tt-CML-mice (Fig. 6A). These mice succumbed from GFP⁺ leukemia, unless they had received Tc^BCR-ABL or Tc^BARF, which potently suppressed the outgrowth GFP⁺ leukemia (Fig. 6B–D). Of note, pDC^Mig, which do not express BCR-ABL, but only gfp, also primed a leukemia-protective T-cell response. Presumably, GFP is the target of this T-cell response, because leukemic relapse after Tc^Mig transfer was gfp-negative and BCR-ABL–positive, suggesting an immune escape through loss of gfp expression. In contrast, mice relapsing despite Tc^BCR-ABL or Tc^BARF lymphocyte infusions were never gfp-negative (Supplementary Fig. S4). This would be in line with an immune escape, supporting that BCR-ABL⁺ pDCs could be functional regulators of CML immunity. However, it cannot be ruled out that GFP-derived neoantigens or BCR-ABL–dependent leukemia-associated antigens are also released from GFP⁺ dying cells and then cross-presented by the recipient's own professional DCs such as cDCs, but also pDCs.

BCR-ABL–induced IRF8 suppression has previously been causally linked to pDC loss, pDC dysfunction, and a lack of immune surveillance in CML (16–18, 20, 44). Our results do not support this concept. First, while acknowledging that BCR-ABL overexpression suppresses Irf8 in primitive murine hematopoiesis and abrogates pDC development from BCR-ABL⁺ LSCs in mice (Fig. 2B and C; Supplementary Fig. S1; refs. 20, 45), there is no causal link between IRF8 loss and pDC loss in vivo. Secondly, there is no pDC loss in human CML, although CML-pDCs are of BCR-ABL–positive origin (Figs. 1C and 5A).

Data rather support that modeling human CML-pDC development in tt-CML mice is flawed by BCR-ABL overexpression, which is a prerequisite for LSC transformation in mice and potently abrogates pDC transdifferentiation from LSC (Fig. 2D). In contrast, CML-stem cell evolution is characterized by clonal heterogeneity (38, 39, 46), including BCR-ABL expression variability (31). Data imply, therefore, that pDC development in human CML is maintained from low-level BCR-ABL–expressing precursors, which are not progenies of the CML LSC clone producing the bulk of leukemia cells (Fig. 2F and G). Here, we showed that in spite of their BCR-ABL⁺ origin, CML-pDCs retain important functional pDC-properties, namely, IFN secretion and maturation capacity. At the same time, CML-pDCs aberrantly express multiple inflammatory cytokines and chemokines. Autocrine constitutive activation of JAK-STAT signaling in CML-pDCs could be a consequence (Supplementary Fig. S3), possibly leading to their peculiar mature phenotype in steady state (Fig. 4D; Supplementary Fig. S2). Very recently, a subpopulation of BDCA-2⁺ CD123⁺Axl⁺Siglec-6⁺ pDCs (AS-DC) also referred to as pre-DCs have been identified in normal individuals. These cells are very potent CD4⁺ and CD8⁺ T-cell stimulators. This raises the intriguing possibility that BCR-ABL leads to an expansion of BDCA-2⁺CD123⁺ CML-pDCs, which are AS-DC-/pre–DC-like and hence, unlike BDCA-2⁺CD123⁺ conventional pDCs, T-cell stimulatory (42, 43).

Via paracrine effects, CML-pDC–derived TNF and tumor-derived growth factor (TGF) might also promote CML stem cell survival (47, 48), limit TKI response, and contribute to CML-stem cell persistence (39). Supporting this, we noticed that higher CML-pDC counts at diagnosis were associated with an inferior depth of nilotinib response. However, when patients with high CML-pDCs at diagnosis were treated with a nilotinib/IFN combination within the randomized German CML-V study, total pDCs including CD86⁺pDCs were strongly reduced, and an association between high CML-pDC counts at diagnosis and inferior depth of response was no longer seen. Based on our recent results showing that CD86⁺pDCs are associated with lower probability of TFR after TKI cessation (29), this implies that an IFN plus nilotinib-based induction therapy leads to a higher rate of TFR, which is the primary endpoint of the CML-V study (NCT01657604). To study this, it will be tested whether TFR risk is associated with the number of CD86⁺pDCs and exhausted T cells measured at diagnosis and prior to TKI stop.

Chronic inflammation has detrimental effects during tumorigenesis. It promotes tumor evolution and is immune suppressive. We showed here that low-level BCR-ABL–expressing CML stem cells might trigger chronic inflammation by transdifferentiation into CML-pDCs as a source for many inflammatory chemokines and cytokines. Indeed, there is precedence that persistent high counts of activated CD86⁺pDCs during chronic infection are immune suppressive. Mechanistically, pDCs can induce immune-suppressive regulatory T cells (15, 49) and attenuate natural killer– and T-cell functions (50).

In contrast, in the early (preleukemic) evolution of CML, low frequencies of BCR-ABL⁺ pDCs might be tumor-suppressive. When low numbers of BCR-ABL⁺ pDC were injected into wild-type mice, they acted as vaccines and induced antileukemic T cells, which indeed protected from leukemia in vivo, when adoptively transferred into CML mice. This suggests that CML-CD86⁺pDCs—like their normal AS-pDC counterparts—are capable of priming T-cell responses to endogenously expressed antigens (12, 51, 52). Because a significant proportion of CML-pDCs are TKI-insensitive and persist during TKI-induced MR (Fig. 7), we propose that they elicit anti-CML CD8⁺ T-cell responses in MR via exploiting a restored and less exhausted remission T-cell repertoire (53, 54). It will now be important to validate that BDCA-2⁺CD123⁺CD86⁺ CML-pDCs correspond to the CD86⁺AS-DC/pre-DC population (42, 43). Beyond the scope of CML, it is tempting to speculate that other oncogenic driver mutations, which lead to an increased stem cell fitness and stem cell expansion such as in benign clonal hematopoiesis (55) or myelodysplastic syndrome, cause the emergence of a higher CD86⁺AS-DCs frequency. This hypothesis can be tested. Myeloid driver mutation–expressing stem cells, via maturation into CD86⁺pDCs, might supposedly engage extrinsic tumor suppression before cell-intrinsic tumor-suppressive signaling cascades such as ARF/TP53 can sense oncogenic stress and become activated. Oncogenic stress-sensing tumor-suppressive pathways usually require high-level oncogenic signaling output (56–58), suggesting that CML-pDC–induced extrinsic tumor suppression can be effective long before intrinsic tumor suppression is engaged.

Together, qualitative (pDC-activation and BCR-ABL-status) and quantitative (high versus low pDC counts) factors govern the outcome of pDC biology in CML, which can be tumor-suppressive or -promoting. Further studies are warranted to eventually exploit oncogene-expressing pDCs as leukemia vaccines.

S. Saussele reports receiving commercial research grant from Novartis and BMS, and has received honoraria from the speakers' bureau of Novartis, BMS, Incyte, Pfizer, and Roche. A. Hochhaus reports receiving commercial research grant from Novartis, BMS, Incyte, and Pfizer. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S. Inselmann, Y. Wang, M. Huber, A. Hochhaus, A. Burchert

Development of methodology: S. Inselmann, Y. Wang, S. Liebler, T. Ernst, C. Brendel, D. Pavlinic, A. Burchert

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Inselmann, S. Saussele, L. Fritz, C. Schütz, M. Huber, T. Ernst, S. Botschek, C. Brendel, D. Pavlinic, V. Benes, A. Hochhaus, A. Burchert

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Inselmann, Y. Wang, S. Saussele, L. Fritz, S. Botschek, R.A. Calogero, E.T. Liu, A. Neubauer, A. Hochhaus, A. Burchert

Writing, review, and/or revision of the manuscript: S. Inselmann, S. Saussele, E.T. Liu, A. Neubauer, A. Hochhaus, A. Burchert

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Wang, D. Cai, C. Brendel, V. Benes, A. Neubauer, A. Burchert

Study supervision: A. Hochhaus, A. Burchert

We thank C. Haferlach (MLL Munich) and D. Haase (University Hospital Göttingen) for performing BCR-ABL FISH. We thank C. Fabisch for coordinating the CML-V study. We thank Prof. Dr. Zink and Prof. Dr. Engenhart-Cabillic for help in performing animal radiation. We thank S. Hühn for excellent help with patient sample preparation, cell sorting, and mice work. We thank S. Sopper (University Innsbruck) and D. Wolf (University Bonn) for helpful discussions.

This work was supported by the “Deutsche Forschungsgemeinschaft “(DFG), Klinische Forschergruppe 210 “Genetics of Drug resistance in Cancer”, and the Deutsche José Carreras Leukämiestiftung (AR12/12) and the “Anneliese Pohl Stiftung.” The CML Study V has been supported by Novartis Oncology and MSD Sharp & Dohme GmbH. Scientific subprojects related to this study have also been supported by EUTOS.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.