Abstract
Acute myeloid leukemia (AML) originates from hematopoietic stem and progenitor cells that acquire somatic mutations, leading to disease and clonogenic evolution. AML is characterized by accumulation of immature myeloid cells in the bone marrow and phenotypic cellular heterogeneity reflective of normal hematopoietic differentiation. Here, we show that JAM-C expression defines a subset of leukemic cells endowed with leukemia-initiating cell activity (LIC). Stratification of de novo AML patients at diagnosis based on JAM-C–expressing cells frequencies in the blood served as an independent prognostic marker for disease outcome. Using publicly available leukemic stem cell (LSC) gene expression profiles and gene expression data generated from JAM-C–expressing leukemic cells, we defined a single cell core gene expression signature correlated to JAM-C expression that reveals LSC heterogeneity. Finally, we demonstrated that JAM-C controls Src family kinase (SFK) activation in LSC and that LIC with exacerbated SFK activation was uniquely found within the JAM-C–expressing LSC compartment. Cancer Res; 77(23); 6627–40. ©2017 AACR.
Introduction
Acute myeloid leukemia (AML) emerges from a hematopoietic stem or progenitor cell (HSPC) that has acquired genetic and/or epigenetic aberrations, resulting in uncontrolled expansion of cells blocked at different stages of myeloid differentiation (1). AML is hierarchically organized with leukemia-initiating cells (LIC) at the apex of the hierarchy that generate non-self-renewing blasts and differentiated progeny cells (2, 3). Although intensive chemotherapy is initially effective in most cases of AML, drug-resistant LICs can repopulate the disease, leading to subsequent relapse and dismal prognosis.
Experimentally, LICs are functionally defined by their ability to initiate, generate, and maintain leukemia in xenograft models. Two landmark studies have identified LICs within a cellular compartment harboring the CD34+CD38low phenotype (2, 3), suggesting that LICs are present within a cell subset endowed with stem cell properties, the leukemic stem cell (LSC) compartment. However, other studies have suggested that LIC-enriched populations in AML are highly heterogeneous and are not necessarily confined to the CD34+CD38low fraction (4–6). Such difficulties in finding a consensus for phenotypic definition of LICs likely reflect the fact that they are not defined based on their differentiation state but on their engraftment properties. The latter do not only rely on acquired somatic mutations, but also depend on the ability of LSCs to interact with bone marrow (BM) microenvironment of immunodeficient mice. Thus, although stem cell gene expression signatures have been correlated with poor disease outcome and increased leukemia-initiating activity (7–9), isolation of LSCs highly enriched in LICs has remained challenging. Many efforts have been made to identify cell surface markers that discriminate between normal and leukemic stem cells such as CD123 (10), CD96 (11), CD44 (12), CLL-1 (13), CD47 (14) CD93 (15), TIM-3 (16), or CD98 (17). Unfortunately, differences between LSC and HSC surface antigen expression are not absolute and may vary according to AML subtypes or during the course of the disease. Thus, it is imperative that we identify new surface markers that could be used by flow cytometry to unravel leukemic cellular heterogeneity with respect to leukemia-initiating activity.
Adhesion molecules specifically expressed by HSPC subsets and involved in human HSCs engraftment in NOD/SCID/IL2Rγc-deficient mice (NSG) represent attractive markers to identify LICs. We have recently demonstrated that the junctional adhesion molecule-C (JAM-C) represents a functional marker for engraftment of human HSCs in NSG mice (18). JAM-C is an adhesion molecule expressed by HSC that interacts with JAM-B expressed by BM stromal cells. Previous studies have revealed that JAM-C/JAM-B interaction contributes to the maintenance of HSC homeostasis (19, 20). Here, we tested whether JAM-C could be expressed by leukemic subsets in AML and may play a role in leukemic maintenance.
Materials and Methods
Human samples
This study was performed after approval by our institutional review board: COS. Patient's samples were obtained after informed consent in accordance with the Declaration of Helsinki and stored at IPC/CRCM Tumor Bank (14-001; 11-007) and in HIMIP collection.
Cell culture
KG1 cells were cultured in IMDM GlutaMAX medium supplemented with 10% FBS and 1% Penicillin Streptomycin.
Clonogenic assays
MS-5 mouse BM stromal cells were plated in 96-well format (2 × 104 per well in IMEM containing 10% FBS) and kept at 37°C/5% CO2. One day later, 15, 10, 5, and 1 CD45dimCD34+CD38lowCD123+CD41−JAM-CPos or JAM-CNeg cells sorted from three AML patient samples were seeded respectively in 12 wells in 100 μL of coculture media (α-Eagle MEM, 12.5% FBS, 12.5% horse serum, 200 mmol/L glutamine, 1 mmol/L monothioglycerol, 1 μmol/L hydrocortisone, and 20 ng/mL human recombinant IL-3). Six days later, the wells were rinsed and fresh media was added. The cocultures were then maintained with one half media change every week and assessed for CAFC after 5 weeks. A CAFC was defined as at least six tightly packed cells beneath the MS-5 stromal monolayer. For colony forming unit (CFU) assays, 1 × 103 KG1/JAM-CPos or KG1/JAM-CNeg were seeded in methylcellulose medium H4230 (Stem Cell) with or without 1 μmol/L of peptides (Covalab), SU6656, or PP2 (Calbiochem) following the manufacturer's instructions. Quadruplicate cultures were incubated at 37°C/5% CO2 and colonies (>20 cells) were scored after 14 days.
Mice and xenograft experiments
NSG mice (005557) were obtained from the Jackson Laboratory and bred in a pathogen-free environment. NSG-Jamb−/− immune-deficient mice were generated by crossing male NSG mice with Jamb−/− mice backcrossed for more than six generations on BALB/cByJ background. F1 male animals carrying the Il2rgtm1Wjl and Jamb+/− mutation were backcrossed with NSG female mice to produce N1 generation, which was characterized by microsatellites and Prkdc locus sequencing. Accelerated backcross using microsatellites was then performed for six generations before intercross and production of NOD.Cg-Prkdcscid Il2rγtm1Wjl Jam2Tm1.1Maal/SzJ used in this study. All experiments were performed in compliance with the laws and protocols approved by animal ethics committees (Agreement No. 02294.01). To assess AML patients sample and KG1 variant cells engraftment, adult mice (6–10 weeks old) were sublethally irradiated with 2 Gy of total body irradiation 24 hours before injection of 1 × 106 cells, or with 350, 200, 100, 50, or 10 sorted cells mixed with 1 × 106 NSG BM cells, or with 2 × 105 KG1/JAM-CPos or KG1/JAM-CNeg, or 1,3 × 105 CD34+ cells. Weight loss >20%, ruffled coat, hunched back, weakness, reduced motility were monitored daily and considered as endpoints.
Flow cytometry
Antibodies and method used for flow cytometry are described in the Supplementary Methods.
Single-cell qPCR and data analysis
For single-cell analysis, cells were directly sorted using the autoclone module on ARIA SORP sorter into C1 chip (Fluidigm). Single-cell loading, cellular lysis, and preamplification steps using Taqman primers were performed using C1 auto-prep system and further processing on Biomark was performed according to manufacturer's instruction (Fluidigm). Results were further analyzed using R software version 3.2.2 and standard packages. Cells lacking detectable signals for housekeeping genes (ACTB and GAPDH) were excluded from analysis and relative expression values were defined as 40−Ct. Data have been generated after three independent cell sorting and captures of frozen vials from the same patient sample. Absent values were replaced by a Gaussian distribution centered on zero with a standard deviation identical to the corresponding dataset. Normalization between datasets was done with the GeNorm algorithm and ACTB was identified as the less variable housekeeping gene. The volcano plot was obtained by plotting the negative log10 P values (calculated from t test for each gene) against the fold-change (ratio) between CD34+ and CD34− cells. Genes significantly upregulated in CD34+ compartment were used to perform a principal component analysis (PCA). Two principal components (PC1 and PC2, representing the largest two variances) were plotted using custom function from ggbiplot R package and normal confidence interval of 80% was drawn as an ellipse identifying 2 different clusters. Violin plots were generated for clusters 1 and 2 using custom function based on geom_violin of the ggplot2 R package.
Microarray and bioinformatics analysis
For gene expression studies, KG1/JAM-CPos and KG1/JAM-CNeg cells were directly sorted in RLT lysis buffer before extraction of total RNA using RNeasy Mini Kit according to manufacturer instructions (Qiagen). mRNA quality was evaluated on Agilent 2100 (Pico Chip) and three samples for each KG1 variant were sent to the GenomEast Platform for hybridization using human Gene 2.0 ST arrays (Affymetrix). Microarray data were normalized with robust multiarray average (RMA) algorithm and processed with ComBat to remove batch effects. Heat-map of microarray was generated using Gene-E software and shows genes differentially expressed between KG1 variants (at least 1.3-fold) and tested for significance with a 5% FDR with Significance Analysis of Microarray. GO enrichment analysis was performed for KG1/JAM-CPos and KG1/JAM-CNeg by hypergeometric testing associated with the custom multiple testing correction procedure g:SCS implemented in gProfiler-R version 5.3 under R 3.1.4. P values threshold was fixed at 5% and reference annotation set size was set to the annotated domain.
Adhesion assay
Maxisorp flat-bottom 96-well plates (Thermo Fisher Scientific) were coated overnight at 4°C with 10 μg/mL of recombinant soluble JAM-B (R&D Systems). Wells were washed twice in PBS and saturated with HBSS 5% BSA for 1 hour at 37°C. 17648 ± 4289 KG1/JAM-CPos cells or 16549 ± 2799 KG1/JAM-CNeg cells were resuspended in 200 μL of HBSS 2% BSA in presence or not of 0.5 μg/mL blocking anti-JAM-C antibody (clone no. 208206, R&D Systems) or isotype control (clone no. 11711; R&D Systems). Cells were allowed to adhere for 2 hours at 37°C before washing three times with PBS. Adherent cells were removed by adding 30 μL of trypsin and resuspended in a final volume of 250 μL of PBS. Absolute numbers of cells were quantified using the HTS coupled to the FACS LSR Fortessa (BD Biosciences).
Immunoblotting analysis
KG1/JAM-CNeg, KG1/JAM-CPos, KG1/JAM-CPos shCTR, KG1/JAM-CPos shJAM-C #1 and #2 cells were starved 2 hours in 0.5% FBS IMDM GlutaMAX medium with or without 1 μmol/L of a permeant peptide, Dasatinib (Sequoia Research Products), SU6656 (Calbiochem), PP2 (Calbiochem), or Sorafenib (Sequoia Research Products) and lysed in HNGT buffer (Hepes pH 7, 50 mmol/L, NaCl 150 mmol/L, glycerol 10%, Triton 1%, MgCl2 1.2 mmol/L, EGTA 1 mmol/L, 1× protease inhibitors and sodium orthovanadate 100 μmol/L). Phosphotyrosine (4G10), ERK2 (#C-14; Santa Cruz Biotechnology), phospho-Src Family (Tyr416; Cell Signaling Technology), phospho-Akt (Cell Signaling Technology), phospho-STAT1 (Cell Signaling Technology), phospho-STAT3 (Cell Signaling Technology), and phospho-STAT5 (Cell Signaling Technology), Akt (Santa Cruz Biotechnology), STAT1 (Santa Cruz Biotechnology), STAT3 (Santa Cruz Biotechnology), and STAT5 (Santa Cruz Biotechnology) antibodies were used for immunoblotting.
Survival analysis
Frequencies f of JAM-C expressing CD45dimCD34+CD38lowCD123+CD41− cells were binned into a histogram, with a step size of 0.25. Histogram values were next fitted by least square curve fit, using the lsqcurvefit Matlab function, with the sum of two normalized Gaussian distributions, centered on frequencies f1 or f2, with standard deviation of σ1 or σ2, according to the following formula: y = α/σ_1√2π exp(−(f − f_1)^2/2σ_1^2) + (1 − α)/σ_2 √2π exp(−(f − f)^2/2σ_2^2). Initial values were set to f1 = mean(f), f2 = 2 mean(f), σ1 = std(f), σ2 = 2 std(f), and α = 0.7. The threshold between the two Gaussian components was determined as the value corresponding to the crossing of the two Gaussian functions.
Overall survival (OS) was defined as the duration from the date of diagnosis to the date of death/last follow-up or allograft stem cell transplantation in first complete remission. Leukemia-free survival (LFS) was defined as the duration from the date of complete remission to the date of relapse or death/last follow-up or allograft stem cell transplantation in first complete remission. Survival curves (OS, LFS) were estimated by the Kaplan–Meier method and compared using the log-rank test. Multivariate Cox and Fine&Gray regression models were used to assess the predictive value of high % of JAM-C expressing LSC in the population (age at diagnosis, WBC at diagnosis, and cytogenetic prognosis).
Statistical analysis
Statistics were calculated using Prism 6 software (GraphPad Software, Inc.). Data were presented as the mean ± SEM or median ± interquartile range. The significance of the differences between groups was determined via paired or unpaired t tests, Mann–Whitney, or Wilcoxon signed-rank tests.
Results
JAM-C expression predicts poor disease outcome in de novo AML
JAM-C expression by leukemic subsets in CD34+ AML blood samples at diagnosis was tested by flow cytometry using anti-JAM-C, CD45, CD33, CD34, CD38, and CD123. JAM-C expressing cells were identified in all CD45dimCD33+ leukemic fractions irrespective of CD34 or CD38 expression (Supplementary Fig. S1A). However, the frequency of JAM-C expressing cells was slightly higher in the CD34+CD38low compartment as compared with other subsets. More precisely, JAM-C–expressing cells were significantly enriched in the CD34+CD38lowCD123+ compartment as compared with CD34+CD38lowCD123− fraction (Supplementary Fig. S1B and S1C). In agreement with previous reports (21, 22), the CD34+CD38lowCD123+ compartment did not contain CD90+CD45RA−CD34+CD38low normal residual HSPCs (Supplementary Fig. S1D and S1E). Thus, we can exclude that JAM-C expressing HSPCs are identified with our gating strategy. This prompted us to test JAM-C expression in a large cohort of 71 de novo AML patient blood samples. The main characteristics of patients are reported in Supplementary Table S1. In all tested samples, JAM-C expression was found in the CD45dimCD34+CD38lowCD123+ fraction as illustrated for two representative patients (Fig. 1A and B). This was not due to JAM-C expression by platelets and their sticking to leukemic cells since JAM-C and CD41 expression were mutually exclusive. For three patients, JAM-C expression was tested in peripheral blood and BM paired samples and no difference was found in frequencies of JAM-C–expressing cells (Supplementary Fig. S2A and S2B). Frequencies of JAM-C–expressing cells ranged from 0.02% to 26.56% (median, 0.47%) of the CD45dimCD34+CD38lowCD123+ compartment and presented a bimodal distribution with a calculated threshold value of 0.4214% obtained by curve fitting (Fig. 1C). This statistically defined value was used to stratify patients into “JAM-C low” and “JAM-C high” frequencies. The median percentages of CD45dimCD34+CD38lowCD123+JAM-CPos cells were respectively 0.21% (range, 0.02–0.36) and 1.08% (range, 0.42–26.56) for “JAM-C low” and “JAM-C high” groups. No differences were found between these two groups in terms of age, sex, leukocyte count, NPM1, and FLT3 mutational status, type of stem cell transplantation received (autologous or allogeneic), or FAB classification (Supplementary Tables S1 and S2). Analysis of JAM-C fluorescence intensity on LSCs and normal HSPCs revealed that JAM-C protein expression was higher in leukemic cells than in HSPCs isolated from healthy donors (Supplementary Fig. S2C). High frequency of CD45dimCD34+CD38lowCD123+JAM-CPos cells was associated with lower OS and reduced LFS (Fig. 1D and E). This was confirmed by reduced cumulative relapse incidence (Supplementary Fig. S2D). Multivariate analysis with age at diagnosis, cytogenetic prognosis, WBC count, or other covariates indicated that high frequencies of CD45dimCD34+CD38lowCD123+JAM-CPos cells remained an independent prognostic factor for OS and LFS (Supplementary Table S3). In addition, analysis of paired patient samples at diagnosis and first relapse revealed that frequencies of CD45dimCD34+CD38lowCD123+JAM-CPos cells at relapse were systematically increased whereas frequencies of CD45dimCD34+CD38low cells were not significantly changed (Fig. 1F–H).
Frequencies of JAM-C–expressing CD45dimCD34+CD38lowCD123+ AML cells correlate with poor prognosis. A and B, Representative flow cytometry analysis of JAM-C expression by CD45dimCD34+CD38lowCD123+ cells of de novo AML samples from patient #1 (A) and #2 (B) at diagnosis. Lymphocytes are used as negative controls and are superimposed on CD34/CD38 and CD34/CD123 profiles. Isotype control is used as negative control for JAM-C staining. C, Distribution of CD45dimCD34+CD38lowCD123+CD41−JAM-CPos cell frequencies in 71 AML patient samples at diagnosis. Inset, curve fitting to distribution of CD45dimCD34+CD38lowCD123+CD41−JAM-CPos cell frequencies reveals two groups of samples separated by a threshold value of 0.4214%. D and E, Comparison of overall (D) and leukemia-free (E) survival of AML patients according to CD45dimCD34+CD38lowCD123+CD41−JAM-CPos cell frequencies. Dashed lines represent patients with percentage of JAM-C expressing CD45dimCD34+CD38lowCD123+CD41− below 0.4214% (low), whereas solid lines represent patients with percentage above 0.4214% (high). F, Representative flow cytometry analysis of AML patient samples at diagnosis and relapse. G and H, Summary of CD45dimCD34+CD38lowCD123+CD41−JAM-CPos (G) and CD45dimCD34+CD38low (H) cell frequencies quantified in 10 paired AML patient samples at diagnosis and relapse. **, P = 0.0020 and ns, nonsignificant, respectively (Wilcoxon signed-rank test).
Frequencies of JAM-C–expressing CD45dimCD34+CD38lowCD123+ AML cells correlate with poor prognosis. A and B, Representative flow cytometry analysis of JAM-C expression by CD45dimCD34+CD38lowCD123+ cells of de novo AML samples from patient #1 (A) and #2 (B) at diagnosis. Lymphocytes are used as negative controls and are superimposed on CD34/CD38 and CD34/CD123 profiles. Isotype control is used as negative control for JAM-C staining. C, Distribution of CD45dimCD34+CD38lowCD123+CD41−JAM-CPos cell frequencies in 71 AML patient samples at diagnosis. Inset, curve fitting to distribution of CD45dimCD34+CD38lowCD123+CD41−JAM-CPos cell frequencies reveals two groups of samples separated by a threshold value of 0.4214%. D and E, Comparison of overall (D) and leukemia-free (E) survival of AML patients according to CD45dimCD34+CD38lowCD123+CD41−JAM-CPos cell frequencies. Dashed lines represent patients with percentage of JAM-C expressing CD45dimCD34+CD38lowCD123+CD41− below 0.4214% (low), whereas solid lines represent patients with percentage above 0.4214% (high). F, Representative flow cytometry analysis of AML patient samples at diagnosis and relapse. G and H, Summary of CD45dimCD34+CD38lowCD123+CD41−JAM-CPos (G) and CD45dimCD34+CD38low (H) cell frequencies quantified in 10 paired AML patient samples at diagnosis and relapse. **, P = 0.0020 and ns, nonsignificant, respectively (Wilcoxon signed-rank test).
JAM-C expression reveals KG1 cell line heterogeneity
Because of the paucity of CD45dimCD34+CD38lowCD123+JAM-CPos cells in AML patient samples, we searched for cell lines that would present similar phenotype to CD34+ AML patient samples. We tested 10 cell lines and found that KASUMI1, KG1, and TF1 cells contained a fraction with the expected CD45dimCD34+CD38lowCD123+ phenotype (Fig. 2A and Supplementary Fig. S3A). Because TF1 and KASUMI1 cell lines are difficult to engraft due to their respective dependence on growth factors (23) and poor expansion properties (24), we focused on KG1 cell line. JAM-CPos and JAM-CNeg CD45dimCD34+CD38lowCD123+ fractions were isolated from KG1 cells by flow cytometry. These two LSC-like variant KG1 cells are referred to as KG1/JAM-CPos and KG1/JAM-CNeg cells hereafter. In agreement with the LSC model, JAM-C expression was progressively lost in KG1/JAM-CPos cells cultured over time (Fig. 2B), whereas KG1/JAM-CNeg cells never produced KG1/JAM-CPos cells. Therefore, all experiments were performed with freshly isolated KG1/JAM-CPos and KG1/JAM-CNeg variant cells. Global gene expression analysis revealed 138 differentially expressed genes (log2 >1.3) between KG1/JAM-CPos and KG1/JAM-CNeg cells (Fig. 2C and Supplementary Fig. S3B). Gene ontology (GO) enrichment analysis of the differentially expressed genes showed that pathways related to protein interactions, kinase inhibitor activity or cation transport were significantly enriched in KG1/JAM-CPos cells. Conversely, KG1/JAM-CNeg cells expressed genes related to myeloid cell differentiation such as MHC class II protein complex (Fig. 2D). Homogeneity of the two variant cells was then tested by single-cell PCR analysis using probe sets selected from microarray analysis. As expected, unsupervised hierarchical clustering identified two major clusters corresponding to differential expression of JAM3, encoding JAM-C. Upregulation of HLA gene expression in single cells lacking JAM3 expression and increased expression of CHRNA5, HPGD, and BACE2 in cells expressing JAM3 were confirmed (Fig. 2E and Supplementary Fig. S3C). This shows that JAM-C identifies distinct subset of cells within the CD45dimCD34+CD38lowCD123+ compartment of KG1 cells.
Differentially expressed genes between KG1/JAM-CNeg and KG1/JAM-CPos cells. A, Flow cytometry gating strategy used to isolate KG1/JAM-CNeg and KG1/JAM-CPos variant cells. Isotype controls are used as negative controls. B, Flow cytometry profiles showing JAM-C expression on KG1/JAM-CPos cells over the course of 9 weeks in culture after sorting. C, Scatter plots of mean RMA values obtained for 138 genes differentially expressed between KG1/JAM-CNeg and KG1/JAM-CPos cells. Mean RMA values are calculated from triplicate experiments. D, GO analysis of differentially expressed genes between KG1/JAM-CNeg (white bars) and KG1/JAM-CPos cells (black bars). E, Heat-map representation of single cells gene expression unsupervised clustering in admixed KG1/JAM-CNeg and KG1/JAM-CPos cells. Each column represents a single cell and each row corresponds to the indicated gene. Color code represents (40-Ct) values showing high expression in red and low expression in blue.
Differentially expressed genes between KG1/JAM-CNeg and KG1/JAM-CPos cells. A, Flow cytometry gating strategy used to isolate KG1/JAM-CNeg and KG1/JAM-CPos variant cells. Isotype controls are used as negative controls. B, Flow cytometry profiles showing JAM-C expression on KG1/JAM-CPos cells over the course of 9 weeks in culture after sorting. C, Scatter plots of mean RMA values obtained for 138 genes differentially expressed between KG1/JAM-CNeg and KG1/JAM-CPos cells. Mean RMA values are calculated from triplicate experiments. D, GO analysis of differentially expressed genes between KG1/JAM-CNeg (white bars) and KG1/JAM-CPos cells (black bars). E, Heat-map representation of single cells gene expression unsupervised clustering in admixed KG1/JAM-CNeg and KG1/JAM-CPos cells. Each column represents a single cell and each row corresponds to the indicated gene. Color code represents (40-Ct) values showing high expression in red and low expression in blue.
JAM-C expression reveals LSC heterogeneity in AML patient sample
To test if such a result holds true for the rare leukemic cells expressing JAM-C in patient samples, single-cell gene expression study was performed. To this end, CD45dimCD34−, CD45dimCD34+CD38lowCD123+CD41−JAM-CPos, or CD45dimCD34+CD38lowCD123+CD41−JAM-CNeg cells were isolated by flow cytometry and captured in individual microfluidic chambers of C1TM Fluidigm IFC plates. Single-cell gene expression analysis was then performed using TaqMan probe sets selected from differentially expressed genes between KG1/JAM-CPos and KG1/JAM-CNeg cells and from publicly available data sets of LSC gene expression signatures (Supplementary Table S4). In agreement with the known LSC nature of CD45dimCD34+CD38lowCD123+CD41− cells (10, 21), most of the selected genes were differentially expressed between CD34+ and CD34− cells (Fig. 3A). To further analyze LSC heterogeneity with respect to JAM3 expression, single-cell gene expression analysis was restricted to CD45dimCD34+CD38lowCD123+CD41− cells and to probe sets significantly enriched in CD34+ cells (red dots in Fig. 3A). PCA of gene expression revealed two distinct overlapping clusters (Fig. 3B). Although the overlap contained an equivalent number of JAM3 expressing and non-expressing cells, cells belonging to cluster 1 only were significantly enriched for JAM3-expressing cells, whereas cells present in cluster 2 lacked JAM3 expression (Fig. 3B, insert). We found that cells in cluster 1 expressed genes related to adhesion such as ALCAM, F11R, ESAM, CD47, CD302, ITGA4, ITGA6, CTNNAL1 and CLN3 in contrast to cells belonging to cluster 2 (Fig. 3C and Supplementary Table S5). Finally, in addition to JAM3, most of the cells in cluster 1 also expressed PTK7, PMP22, HPGD, and CHRNA5 as compared with cells in cluster 2, suggesting that JAM-C expression identifies distinct LSCs subsets.
Single cell gene expression reveals heterogeneity of CD45dimCD34+CD38lowCD123+ cells in AML patient. A, Volcano plot showing single-cell gene expression differences between CD34+ and CD34− cells from patient #4. Single-cell quantitative RT-PCR analysis of 95 genes listed in Supplementary Table S4 is shown. B, PCA with the set of genes significantly enriched in CD34+ cells as variable (red dots in A) and CD45dimCD34+CD38lowCD123+ single cells as observations. Distinct PCA clusters are shown with colors and numbers. Inset shows the percentage of JAM3Neg and JAM3Pos single cells in each PCA clusters (green, cluster 1; pink, cluster 2; dark gray, overlap). Absolute numbers of cells in each group are indicated. C, Violin plots showing differential expression of genes across cells in PCA cluster 1 (green) and PCA cluster 2 (pink). ACTB and CD34 expressions are shown as control. Colored dots show the median value and error bars the interquartile range. Statistical significance is provided in Supplementary Table S5.
Single cell gene expression reveals heterogeneity of CD45dimCD34+CD38lowCD123+ cells in AML patient. A, Volcano plot showing single-cell gene expression differences between CD34+ and CD34− cells from patient #4. Single-cell quantitative RT-PCR analysis of 95 genes listed in Supplementary Table S4 is shown. B, PCA with the set of genes significantly enriched in CD34+ cells as variable (red dots in A) and CD45dimCD34+CD38lowCD123+ single cells as observations. Distinct PCA clusters are shown with colors and numbers. Inset shows the percentage of JAM3Neg and JAM3Pos single cells in each PCA clusters (green, cluster 1; pink, cluster 2; dark gray, overlap). Absolute numbers of cells in each group are indicated. C, Violin plots showing differential expression of genes across cells in PCA cluster 1 (green) and PCA cluster 2 (pink). ACTB and CD34 expressions are shown as control. Colored dots show the median value and error bars the interquartile range. Statistical significance is provided in Supplementary Table S5.
JAM-C is a functional marker of leukemia-initiating cells in AML
We measured frequencies of JAM-C–expressing LSCs in 34 AML samples known for their engrafting property in NSG mice. As previously reported (25), engraftment in mice was correlated to patient's poor OS (Supplementary Fig. S4A). More interestingly, JAM-C–expressing LSC frequencies were significantly increased in engrafting samples as compared with non-engrafting ones (Fig. 4A and B), whereas no difference in the percentages of CD34+CD38low or CD34+CD38lowCD123+ was found (Supplementary Fig. S4B and S4C). Cobblestone area forming cell assays (CAFC) performed with three patient samples showed that long term culture-initiating cell (LTC-IC) activity was higher in CD45dimCD34+CD38lowCD123+CD41−JAM-CPos cells as compared with cells lacking JAM-C expression (Fig. 4C). Extreme limiting dilution analysis (26) of pooled results indicated that LTC-ICs frequencies were three times more prevalent in the JAM-C–expressing subpopulation than in the JAM-C negative compartment (1/14.9 vs. 1/49.2; Fig. 4D and Supplementary Table S6). This prompted us to compare leukemia-initiating activity of JAM-CPos and JAM-CNeg LSCs. For two patients, the tumor bulk was reconstituted in NSG mice xenografted with 350 JAM-C–expressing LSCs, whereas the presence of human cells was hardly detectable in the blood, BM, and spleen of mice engrafted with LSCs lacking JAM-C expression (Fig. 4E, not shown). The engrafted JAM-C–expressing leukemic cells displayed myeloid immuno-phenotypes comparable to the sample of origin and contained a subset of cells presenting the CD45dimCD34+CD38lowCD123+CD41−JAM-CPos phenotype similar to the primary AML sample (Supplementary Fig. S4D and S4E). In addition, we could exclude that engrafted cells were contaminated with residual HSPCs because human lymphocytes were not detected in engrafted mice. Estimation of LIC frequencies in vivo by limit dilution assay indicated that leukemic-initiating cells were at least 9.5-fold more prevalent in the CD45dimCD34+ CD38lowCD123+CD41−JAM-CPos subpopulation as compared with the JAM-CNeg one (1/62.9 vs. 1/599.6; Fig. 4F and Supplementary Table S7). Taken together, these results indicate that the leukemic-initiating activity is highly enriched in the subset of LSCs-expressing JAM-C and that genetic alterations or specific activation of signal transduction pathways in these cells are sufficient to recapitulate leukemic disease after transplantation in immunocompromised mice.
CD45dimCD34+CD38lowCD123+CD41−JAM-CPos cells are highly enriched in leukemia-initiating activity. A, Representative flow cytometry plots showing JAM-C expression by CD45dimCD34+CD38lowCD123+CD41− cells for engrafting and non-engrafting patient samples. The threshold for the engraftment capacity was defined as being at least 5% of human cells in the bone marrow after 20 weeks. B, Summary of CD45dimCD34+CD38lowCD123+CD41−JAM-C+ cell frequencies measured in engrafting and non-engrafting patient samples. *, P = 0.0024 (Mann–Whitney test); n = 34. C, Representative results obtained in CAFC assays using the indicated number of CD45dimCD34+CD38lowCD123+CD41−–sorted cells expressing or not JAM-C. Twelve wells were plated for each concentration, and three representative patients are shown. D, Log-fraction plot of the limiting dilution model fitted to the CAFC data shown in C. The slope of the line is the log-active cell fraction. The dotted lines give the 95% confidence interval. Frequencies of CAFC for cells expressing (black lines) or not JAM-C (red lines) are respectively 1/14.9 vs. 1/49.2 (see Supplementary Table S6). E, Graphs showing the percentages of chimerism (%hCD45) in the blood, bone marrow, and spleen of mice 20 weeks after transplantation of 350 CD45dimCD34+CD38lowCD123+CD41− JAM-CNeg (white bars)- or JAM-CPos-sorted cells (black bars). One representative experiment is shown. *, P = 0.0286 (Mann-Whitney test); n = 3. F, Log-fraction plot of the limiting dilution model fitted to the xenograft results obtained for patient #8. Frequencies of LICs within CD45dimCD34+CD38lowCD123+CD41− cells expressing (black lines) or not JAM-C (red lines) are respectively 1/62.9 vs. 1/599.6 (see Supplementary Table S7).
CD45dimCD34+CD38lowCD123+CD41−JAM-CPos cells are highly enriched in leukemia-initiating activity. A, Representative flow cytometry plots showing JAM-C expression by CD45dimCD34+CD38lowCD123+CD41− cells for engrafting and non-engrafting patient samples. The threshold for the engraftment capacity was defined as being at least 5% of human cells in the bone marrow after 20 weeks. B, Summary of CD45dimCD34+CD38lowCD123+CD41−JAM-C+ cell frequencies measured in engrafting and non-engrafting patient samples. *, P = 0.0024 (Mann–Whitney test); n = 34. C, Representative results obtained in CAFC assays using the indicated number of CD45dimCD34+CD38lowCD123+CD41−–sorted cells expressing or not JAM-C. Twelve wells were plated for each concentration, and three representative patients are shown. D, Log-fraction plot of the limiting dilution model fitted to the CAFC data shown in C. The slope of the line is the log-active cell fraction. The dotted lines give the 95% confidence interval. Frequencies of CAFC for cells expressing (black lines) or not JAM-C (red lines) are respectively 1/14.9 vs. 1/49.2 (see Supplementary Table S6). E, Graphs showing the percentages of chimerism (%hCD45) in the blood, bone marrow, and spleen of mice 20 weeks after transplantation of 350 CD45dimCD34+CD38lowCD123+CD41− JAM-CNeg (white bars)- or JAM-CPos-sorted cells (black bars). One representative experiment is shown. *, P = 0.0286 (Mann-Whitney test); n = 3. F, Log-fraction plot of the limiting dilution model fitted to the xenograft results obtained for patient #8. Frequencies of LICs within CD45dimCD34+CD38lowCD123+CD41− cells expressing (black lines) or not JAM-C (red lines) are respectively 1/62.9 vs. 1/599.6 (see Supplementary Table S7).
Functional characterization of leukemia-initiating activity in KG1 cell line
Because JAM-C identifies LICs within the LSC compartment in patient samples and because leukemia-initiating activity has been associated with drug resistance, IC50 of Ara-C was determined for parental KG1 cells and freshly isolated KG1/JAM-CPos or KG1/JAM-CNeg variant cells. Both variants were resistant to Ara-C treatment (IC50 = 3.447 and 2.197 μmol/L, respectively) whereas the parental cell line was sensitive (IC50 = 28 nmol/L; Fig. 5A). We thus tested if the two variants cells differed in cell cycle or cell proliferation and found a slight increase of KG1/JAM-CNeg cells in S–G2–M phase of cell cycle while percentages of KG1/JAM-CPos cells in G1 was significantly increased (Fig. 5B and Supplementary Fig. S5A and S5B). This was correlated to a significant increase of KG1/JAM-CPos cells clonogenic activity as compared with KG1/JAM-CNeg cells (Fig. 5C). Because the primary function of JAM-C is to mediate adhesion to JAM-B (27), adherence of KG1/JAM-CPos and KG1/JAM-CNeg cells to soluble JAM-B was tested. Only JAM-C expressing cells adhered to JAM-B in a JAM-C-dependent manner as demonstrated by the inhibitory effect of anti-JAM-C blocking antibody (Fig. 5D). This suggests that leukemic cells expressing JAM-C may be more easily retained in JAM-B–expressing niches. Therefore, we tested whether the leukemia-initiating activity of KG1 cells relied on JAM-C/JAM-B interaction. We found that the leukemia-initiating activity was restricted to KG1/JAM-CPos cells and was dependent on Jam-B expression within leukemic microenvironment as demonstrated by the strong decrease of KG1/JAM-CPos cells engraftment observed in Jam-b-deficient NSG mice (Fig. 5E and Supplementary Table S8). The function of JAM-B/JAM-C interaction in leukemic engraftment was confirmed using CD34+ cells purified from a patient sample and engrafted in control or Jam-b–deficient NSG mice. A two-fold decrease in leukemic cell expansion in the bone marrow of Jam-b−/− NSG mice was observed 14 weeks after engraftment (Fig. 5F). These results suggest that JAM-C–expressing leukemic cells are endowed with leukemia-initiating activity, which is further supported by JAM-B–expressing niches in the bone marrow.
Functional properties of KG1/JAM-CNeg and KG1/JAM-CPos cells. A, Representative graph of Ara-C cytotoxicity on parental KG1 (dots), KG1/JAM-CNeg (gray triangles) and KG1/JAM-CPos (black triangles). The half maximal inhibitory concentration values (IC50) deduced from dose-response curves are shown (μmol/L). B, Quantification of cell cycle states using Ki67/DAPI staining of KG1/JAM-CNeg and KG1/JAM-CPos cells as shown in Supplementary Fig. S5a. *, P < 0.05 (unpaired t test); n = 4. C, Clonogenic properties of KG1/JAM-CNeg (white bar) and JAM-CPos (black bar) cells in CFU assays. ***, P = 0.0002 (unpaired t test); n = 4. D, Adhesion assay of KG1/JAM-CNeg and KG1/JAM-CPos cells on soluble JAM-B in presence or not of JAM-C blocking antibody and control isotype. ***, P < 0.0001 (unpaired t test); n = 3. E, Engraftment properties of KG1/JAM-CNeg and KG1/JAM-CPos cells in Jam-bwt/wt or Jam-bko/ko NSG mice. Injected animals were sacrificed 21 days after graft (see Supplementary Table S8). F, Graphs showing the percentages of chimerism (%hCD45) in the bone marrow of mice 14 weeks after transplantation in Jam-bwt/wt (black bar) or Jam-bko/ko (gray bar) NSG mice of 1.3 × 105 CD34+ cells isolated from AML patient sample. *, P = 0.0265 (Mann–Whitney test); n = 3.
Functional properties of KG1/JAM-CNeg and KG1/JAM-CPos cells. A, Representative graph of Ara-C cytotoxicity on parental KG1 (dots), KG1/JAM-CNeg (gray triangles) and KG1/JAM-CPos (black triangles). The half maximal inhibitory concentration values (IC50) deduced from dose-response curves are shown (μmol/L). B, Quantification of cell cycle states using Ki67/DAPI staining of KG1/JAM-CNeg and KG1/JAM-CPos cells as shown in Supplementary Fig. S5a. *, P < 0.05 (unpaired t test); n = 4. C, Clonogenic properties of KG1/JAM-CNeg (white bar) and JAM-CPos (black bar) cells in CFU assays. ***, P = 0.0002 (unpaired t test); n = 4. D, Adhesion assay of KG1/JAM-CNeg and KG1/JAM-CPos cells on soluble JAM-B in presence or not of JAM-C blocking antibody and control isotype. ***, P < 0.0001 (unpaired t test); n = 3. E, Engraftment properties of KG1/JAM-CNeg and KG1/JAM-CPos cells in Jam-bwt/wt or Jam-bko/ko NSG mice. Injected animals were sacrificed 21 days after graft (see Supplementary Table S8). F, Graphs showing the percentages of chimerism (%hCD45) in the bone marrow of mice 14 weeks after transplantation in Jam-bwt/wt (black bar) or Jam-bko/ko (gray bar) NSG mice of 1.3 × 105 CD34+ cells isolated from AML patient sample. *, P = 0.0265 (Mann–Whitney test); n = 3.
JAM-C regulates leukemia-initiating activity through SFK-pathway activation
Because differential gene expression between KG1/JAM-CPos and KG1/JAM-CNeg variant cells pointed towards pathways involved in signal transduction, freshly sorted variant cells were lysed and immunoblotted with an anti-phosphotyrosine antibody to measure activation of multiple signaling pathways. As shown in Figure 6A, the two variant cell lines displayed similarities in phosphotyrosine profiles. However, a major differentially phosphorylated band was detected around 80 kDa (p80). This protein was 3.10 ± 0.53-fold more phosphorylated in KG1/JAM-CPos cells as compared with KG1/JAM-CNeg cells (Supplementary Fig. S6A). To identify the signaling networks responsible for this specific phosphorylation increase in JAM-C–expressing cells, cells were treated with several pharmacological inhibitors. Tyrosine phosphorylation of p80 was dramatically reduced after dasatinib, SU6656, and PP2 treatments. All these compounds are known to inhibit SRC family kinases (SFK). In contrast, inhibition of RAF/MEK/ERK pathway and VEGFR/PDGFR kinases with sorafenib did not alter p80 phosphorylation (Fig. 6B). This prompted us to analyze SFK activation in the two variant cell lines. Western blot analysis and FACS analysis showed an increased phosphorylation of SFK in KG1/JAM-CPos as compared with KG1/JAM-CNeg cells (Fig. 6C). Phosphorylation status of other signaling proteins such as ERK1/2, AKT, or STAT1/3/5 was similar in the two variants (Supplementary Fig. S6B-F). To test if p80 and SFK phosphorylations were dependent on JAM-C expression, JAM-C was knocked down in KG1/JAM-CPos cells using shRNA. Silencing JAM-C expression dramatically decreased p80 and SFK phosphorylation as compared with control (Fig. 6D). To address whether JAM-C expression was sufficient, we used a membrane permeant peptide corresponding to the last 19 aa of JAM-C intracellular domain (JAM-C-ICD) or a control peptide corresponding to the same amino acids in random sequence. Such an approach has already been used to modulate JAM protein family signaling in epithelial and neuronal cells (28, 29). We found that tyrosine phosphorylation of p80 and SFK were increased in KG1/JAM-CPos cells treated with JAM-C-ICD peptide as compared with control (Fig. 6E). No effect was observed in KG1 cells lacking JAM-C expression (Supplementary Fig. S6G). Because JAM-C, SFK, and p80 seemed to belong to the same pathway, we performed a proteomic analysis of the tyrosine phosphorylated band of 80 kDa extracted from KG1/JAM-CNeg and KG1/JAM-CPos cells treated or not with SU6656. More than twenty proteins were significantly enriched in KG1/JAM-CPos cells as compared with KG1/JAM-CNeg cells (Red dots in Supplementary Fig. S6H). Some of them, such as BCAP, UBPL2, DDX54, and K1C14 were not sensitive to SFK inhibition. Other proteins such as APC7, ARMC8, RANB9, RBP10, WDR26 were not any more phosphorylated upon SFK inhibition and represented the most likely p80 candidates (Supplementary Fig. S6I). However, all attempts to validate some of these candidate proteins by immunoprecipitation and Western blotting were unsuccessful due to the poor quality of available reagents. More interestingly, treatment of KG1/JAM-CPos cells with JAM-C-ICD peptide significantly increased clonogenicity as compared with control peptide (Fig. 6F). No effect of the peptide was detected using KG1/JAM-CNeg cells, suggesting that increased clonogenic activity and leukemia-initiating activity observed in KG1/JAM-CPos cells relies on SFK activation and p80 tyrosine phosphorylation by JAM-C. In agreement with these results, JAM-C knockdown (Fig. 6G) and SFK inhibition (Fig. 6H) markedly decreased the number of colonies formed by KG1/JAM-CPos cells.
JAM-C controls SFK activation in KG1/JAM-CPos cells. A, Top, Western blot analysis of KG1/JAM-CNeg and KG1/JAM-CPos total cell lysates using 4G10 anti-phosphotyrosine antibody. Arrow, prominent differentially tyrosine phosphorylated protein with an apparent MW of 80kDa (p80). Bottom, loading control using anti-ERK2 antibody. B, Top, Western blot analysis of total cell lysates from KG1/JAM-CPos treated with the indicated kinase inhibitor using 4G10 antiphosphotyrosine antibody. Bottom, loading control using anti-ERK2 antibody. C, Top, Western blot analysis of KG1/JAM-CNeg and KG1/JAM-CPos total cell lysates using antiphospho Src-family kinase (Tyr416); anti-ERK2 is shown as loading control. Bottom, flow cytometry profiles of p-SFK staining in KG1/JAM-CNeg (dashed lines) and KG1/JAM-CPos (solid line). D, Left, flow cytometry analysis of JAM-C expression by KG1/JAM-CPos transduced with control shRNA (solid lines) or shRNA against JAM-C (dotted and dashed lines). Right, Western blots showing p80 tyrosine phosphorylation, SFK phosphorylation, and ERK2 expression in KG1/JAM-CPos cells transduced with the indicated shRNAs. Results obtained with two different shRNA against JAM-C are shown. E, Western blot analysis of total cell lysates from KG1/JAM-CPos cells treated with the indicated membrane permeant peptide. Results showing p80 tyrosine phosphorylation, SFK phosphorylation, and ERK2 expression are shown. F, Clonogenic properties of KG1/JAM-CNeg and KG1/JAM-CPos cells treated with JAM-C ICD or random peptides. **, P = 0.0058 (unpaired t-test); n = 4. G, Clonogenic properties of KG1/JAM-CPos transduced with control shRNA or two different shRNA against JAM-C. ***, P = 0.0004 and 0.0002 (unpaired t test); n = 3. H, Clonogenic properties of KG1/JAM-CPos treated with SU6656 and PP2 kinase inhibitors. ***, P < 0.001 (paired t test); n = 4.
JAM-C controls SFK activation in KG1/JAM-CPos cells. A, Top, Western blot analysis of KG1/JAM-CNeg and KG1/JAM-CPos total cell lysates using 4G10 anti-phosphotyrosine antibody. Arrow, prominent differentially tyrosine phosphorylated protein with an apparent MW of 80kDa (p80). Bottom, loading control using anti-ERK2 antibody. B, Top, Western blot analysis of total cell lysates from KG1/JAM-CPos treated with the indicated kinase inhibitor using 4G10 antiphosphotyrosine antibody. Bottom, loading control using anti-ERK2 antibody. C, Top, Western blot analysis of KG1/JAM-CNeg and KG1/JAM-CPos total cell lysates using antiphospho Src-family kinase (Tyr416); anti-ERK2 is shown as loading control. Bottom, flow cytometry profiles of p-SFK staining in KG1/JAM-CNeg (dashed lines) and KG1/JAM-CPos (solid line). D, Left, flow cytometry analysis of JAM-C expression by KG1/JAM-CPos transduced with control shRNA (solid lines) or shRNA against JAM-C (dotted and dashed lines). Right, Western blots showing p80 tyrosine phosphorylation, SFK phosphorylation, and ERK2 expression in KG1/JAM-CPos cells transduced with the indicated shRNAs. Results obtained with two different shRNA against JAM-C are shown. E, Western blot analysis of total cell lysates from KG1/JAM-CPos cells treated with the indicated membrane permeant peptide. Results showing p80 tyrosine phosphorylation, SFK phosphorylation, and ERK2 expression are shown. F, Clonogenic properties of KG1/JAM-CNeg and KG1/JAM-CPos cells treated with JAM-C ICD or random peptides. **, P = 0.0058 (unpaired t-test); n = 4. G, Clonogenic properties of KG1/JAM-CPos transduced with control shRNA or two different shRNA against JAM-C. ***, P = 0.0004 and 0.0002 (unpaired t test); n = 3. H, Clonogenic properties of KG1/JAM-CPos treated with SU6656 and PP2 kinase inhibitors. ***, P < 0.001 (paired t test); n = 4.
To translate our results to clinical situation, patient samples were then analyzed for SFK phosphorylation. As previously reported (30), increased SFK phosphorylation was observed in cells harboring the CD45dimCD34+CD38low phenotype as compared with CD45dimCD34−CD38low cells (Supplementary Fig. S7A–S7D). More interestingly, in all tested patient samples, we found a subset of CD45dimCD34+CD38lowCD123+CD41−JAM-CPos cells presenting remarkable high levels of p-SFK (Fig. 7A–D), indicating that SFK hyper-activation is restricted to rare leukemic cells endowed with leukemia-initiating activity. Pooling results obtained for all patient samples showed that more than 40% of CD45dimCD34+CD38lowCD123+CD41−JAM-CPos cells harbored the hyper-activated p-SFK phenotype as compared with frequencies of p-SFK cells comprised between 0.98% and 4.3% in other phenotypically defined compartments (Fig. 7E). Altogether, our results show that JAM-C identifies SFK-activated leukemia-initiating cells in AML.
SFK hyperactivation in CD45dimCD34+CD38lowCD123+CD41−JAM-CPos cells from AML patient samples. A–D, Flow cytometry profiles showing SFK phosphorylation (Tyr416) in JAM-CNeg and JAM-CPos CD45dimCD34+CD38lowCD123+CD41− cellular compartments in four AML patient samples at diagnosis. E, Cytoscape representation of results obtained for samples shown in A–D. Each node represents one leukemia cell-subset. The size of the node is proportional to the percentage of p-SFK positive cells as compared with isotype control staining.
SFK hyperactivation in CD45dimCD34+CD38lowCD123+CD41−JAM-CPos cells from AML patient samples. A–D, Flow cytometry profiles showing SFK phosphorylation (Tyr416) in JAM-CNeg and JAM-CPos CD45dimCD34+CD38lowCD123+CD41− cellular compartments in four AML patient samples at diagnosis. E, Cytoscape representation of results obtained for samples shown in A–D. Each node represents one leukemia cell-subset. The size of the node is proportional to the percentage of p-SFK positive cells as compared with isotype control staining.
Discussion
Despite introduction of AML molecular classification at diagnosis as recommended by the European Leukemia Net (ELN; ref. 31), objective criteria to help clinicians forecast AML treatment response or risk of relapse remain limited. In this study, we have identified JAM-C as a LIC-specific marker within the most immature CD34+CD38lowCD123+ leukemic compartment, which is predictive of disease outcome. In contrast to other published LSC markers (32), JAM-C–expressing LSCs are systematically increased at relapse, indicating that their quantification appears as an interesting approach to quantify biomarkers related to stemness (33).
An additional new finding of our study is that CD34+CD38lowCD123+ cells expressing JAM3 at the single cell level also express a number of other genes related to adhesion and BM niche retention (34). In contrast, cells lacking JAM3 expression are more heterogeneous for adhesion-related gene expression. This is consistent with the idea that leukemia maintenance and chemotherapeutic resistance depend on adhesion and localization of a fraction of leukemic cells within BM niches. Although the latter may vary depending on the nature of the leukemia (35), several studies have demonstrated that ligand/receptor pairs involved in normal HSC localization in BM niches also play a role in LIC maintenance. Indeed, inhibition of adhesion molecules expressed by leukemic cells, such as CD44 (12) or integrin α4β1 (36), results in decreased leukemic burst in xenograft models. This has been translated into new therapeutic approaches, some of which are already at clinical trial stages (37–39). More recently, CD98, which amplifies multiple adhesive signals mediated by β1 and β3 integrin and regulates amino acid transport, has emerged as a promising new therapeutic target for AML treatment (17). In this context, JAM-C represents an attractive target for antibody-based therapy since it could also act through multiple pathways such as modulation of β1 and β3 integrin-mediated adhesion or regulation of CXCL12 secretion (40, 41). Anti-JAM-C antibodies have already been developed with success for lymphoma treatment in preclinical models (42, 43). Future studies will no doubt aim to determine if such reagents are suitable for adjuvant therapy in AML.
Using combination of CD34, CD38, CD123, and JAM-C, we demonstrate that it is possible to achieve enrichment of one LIC in less than 100 cells. Such a property does not depend on the expression of a single adhesion molecule, rather, it reflects a BM “niche pro-adhesive” phenotype of individual cells that can only be revealed by high dimensional data analytics or single-cell analyses (9, 32). Differential activation of signal transduction pathways represents an addition level of cell heterogeneity with respect to gene expression or phenotypic heterogeneity. We identify p-SFK as a signal transduction pathway controlled, at least partially, by JAM-C expression. Along this line, antibodies against JAM-C have been shown to inhibit ERK1/2 phosphorylation in Jeko-1 lymphoma cells (43), whereas JAM-C engagement with soluble JAM-B induces SFK activation in glioblastoma cells (44). Although we show that JAM-C engagement with JAM-B contributes to leukemia-initiating activity in this study, expression of JAM-C by itself appears necessary and sufficient to induce SFK activation in a JAM-B-independent manner. One possibility is that JAM-C expressed on the surface of the cells forms large cluster through cis-interaction and activates polarity complex proteins such as cdc42, which acts on downstream effectors such as SFK (45, 46). In such a model, JAM-C signaling may be the controlled through GRASP55-regulated intracellular trafficking of JAM-C or Pard3a-stabilization of JAM-C at the membrane as reported in spermatids and neurons respectively (29, 47). Alternatively, one could not exclude that JAM-C is constitutively released from leukemic cells as a soluble factor and acts in a cell autonomous autocrine manner to activate SFK. Indeed, two reports have shown that JAM-C is shed from endothelial cells upon cleavage by proteases, such as ADAM10, ADAM17, or elastase (48, 49). In this study, we have identified BACE2 as a gene encoding a protease overexpressed by JAM-C–expressing KG1 cells. This suggests that progressive loss of JAM-C expression by KG1/JAM-CPos could be due to JAM-C release by BACE2. However, we did not confirm differential BACE2 expression at the protein level, indicating that an additional level of regulation must exist if JAM-C shedding by BACE2 is the molecular mechanism regulating differential SFK activation in KG1 variant cell lines. The fact that leukemic cells presenting hyper-activation of SFK are uniquely found in CD34+CD38lowCD123+JAM-CPos compartment is consistent with previous studies that have shown increased phosphorylation of SFK in the CD34+CD38lowCD123+ compartment (30). However, when analyzed on this subset, the p-SFK cytometry profile consisted in a Gaussian distribution of the signal that was positively shifted compared with that of more mature leukemic cell compartments. Adding JAM-C to the flow-cytometry panel allows clear identification of two LSC subsets presenting two different levels of p-SFK.
Given the importance of SFK activation in drug resistance and cell adhesion, SFK blockade represents an exciting new method of treatment for AML as illustrated by several ongoing clinical trials combining Dasatinib with conventional chemotherapy (NCT02013648, NCT01876953, NCT01238211). However, quantification of drug efficacy against LIC remains difficult to assess and relies mostly on indirect primary endpoints such as disease-free survival (37, 39). In addition, patient's eligibility for these clinical trials is based on molecular stratification that is, at least partially, independent of LIC frequencies quantification. The flow-cytometry methods described in this study will substantially improve patient stratification and help detecting biological response to drugs inhibiting SFK phosphorylation. In summary, this study provides new insights into the molecular mechanisms involved in leukemia-initiating activity and identifies JAM-C as a poor prognosis marker in de novo AML.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Authors' Contributions
Conception and design: M. De Grandis, C. Montersino, A.-S. Chretien, R. Castellano, S.J.C. Mancini, N. Vey, M. Aurrand-Lions
Development of methodology: A. Sergé, A.-S. Chretien, R. Castellano, Y. Collette
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. De Grandis, F. Bardin, C. Fauriat, L. Pouyet, M.-J. Mozziconacci, Y. Collette, N. Vey, M. Aurrand-Lions
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. De Grandis, C. Fauriat, C. Zemmour, A. El-Kaoutari, A. Sergé, S. Granjeaud, A.-S. Chretien, G. Bidaut, J.-M. Boher, Y. Collette, N. Vey
Writing, review, and/or revision of the manuscript: M. De Grandis, A.-S. Chretien, J.-M. Boher, S.J.C. Mancini, N. Vey, M. Aurrand-Lions
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): F. Bardin, A.-S. Chretien
Study supervision: M. Aurrand-Lions
Acknowledgments
We thank Dr. P. Dubreuil and Dr. E. Soucié for proofreading the manuscript and providing helpful comments. We thank N. Mossadegh-Keller for helping to conduct the single cell experiments and J. Pakradouni for helping in patients' samples management. We are grateful to the flow cytometry, proteomics/mass spectrometry, and animal core facilities for providing supportive help. We thank the Canceropôle PACA and PACA region for supporting development of single cell technology.
Grant Support
This work was partly supported by grants from the ARC Foundation (PJA#20141201990 to M. Aurrand-Lions and PJA#20161204555 to S.J.C. Mancini) and Gefluc. The project has been partly supported by SIRIC program (INCa-DGOS-Inserm 6038). M. De Grandis is a postdoctoral fellow supported by Fondation de France (2013-00042638) and by PRT-K (INCa-DGOS_10923).
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.