CSF1R inhibitors exhibit antitumor activity in acute myeloid leukemia by blocking paracrine signals from support cells

Abstract

To identify new therapeutic targets in acute myeloid leukemia (AML), we performed small-molecule and small-interfering RNA (siRNA) screens of primary AML patient samples. In 23% of samples, we found sensitivity to inhibition of colony-stimulating factor 1 (CSF1) receptor (CSF1R), a receptor tyrosine kinase responsible for survival, proliferation, and differentiation of myeloid-lineage cells. Sensitivity to CSF1R inhibitor GW-2580 was found preferentially in de novo and favorable-risk patients, and resistance to GW-2580 was associated with reduced overall survival. Using flow cytometry, we discovered that CSF1R is not expressed on the majority of leukemic blasts but instead on a subpopulation of supportive cells. Comparison of CSF1R-expressing cells in AML vs healthy donors by mass cytometry revealed expression of unique cell-surface markers. The quantity of CSF1R-expressing cells correlated with GW-2580 sensitivity. Exposure of primary AML patient samples to a panel of recombinant cytokines revealed that CSF1R inhibitor sensitivity correlated with a growth response to CSF1R ligand, CSF1, and other cytokines, including hepatocyte growth factor (HGF). The addition of CSF1 increased the secretion of HGF and other cytokines in conditioned media from AML patient samples, whereas adding GW-2580 reduced their secretion. In untreated cells, HGF levels correlated significantly with GW-2580 sensitivity. Finally, recombinant HGF and HS-5-conditioned media rescued cell viability after GW-2580 treatment in AML patient samples. Our results suggest that CSF1R-expressing cells support the bulk leukemia population through the secretion of HGF and other cytokines. This study identifies CSF1R as a novel therapeutic target of AML and provides a mechanism of paracrine cytokine/growth factor signaling in this disease.

Graphical abstract

Figure 1.

Ex vivo AML patient sample screen reveals that knockdown/inhibition of CSF1R reduces leukemia cell survival in >20% of samples. (A) Schematic of screening primary AML patient samples against small-molecule inhibitors and siRNAs against the tyrosine kinome to identify new therapeutic targets. (B) siRNA tyrosine kinome screen (n = 93 kinase siRNAs) identifies CSF1R as the top “hit” in primary AML patient samples (n = 158 or 162) to significantly reduce cell viability. (C) High degree of specificity among the CSF1R-targeted small-molecule inhibitors GW-2580, ARRY-382, and JNJ-28312141, compared with other class III receptor tyrosine kinases. Data from (1) Davis et al and (2) Wright et al. (D) Strong correlation observed between GW-2580 AUC and z score of the viability from siCSF1R compared with that of other tyrosine kinase siRNAs (n = 162 patient samples). Significance determined by Spearman rank correlation. (E) siCSF1R has the strongest correlation and most significant association with GW-2580 AUC in the siRNA tyrosine kinome screen. Slope of linear regression line calculated for each siRNA as indicated in panel D was plotted against the P value, determined by significance test for linear regression. (F) Profile of sensitivity to GW-2580 across the cohort of primary AML patient samples (n = 315). The relative positions of representative dose-response curves (G-H) are indicated. (G-H) Representative dose-response curves for a (G) sensitive and (H) nonsensitive primary AML patient sample to GW-2580.

Figure 2.

Resistance to CSF1R inhibitor is associated with adverse prognostic risk gene mutations and cytogenetic abnormalities. (A) GW-2580 AUC from primary AML patient samples (n = 315) was compared for a multitude of clinical and genetic characteristics, with number of samples with evaluable data and the P value listed for each characteristic. Prognostic risk was determined using the ELN guidelines for AML (see Döhner et al). The presence/absence of translocations was determined from karyotype. Only translocations that were found in ≥2 patients were considered. Mutational data were collected by either targeted sequencing, whole-exome sequencing, or targeted polymerase chain reaction (PCR)-based methods (FLT3-ITD and NPM1). Significance was determined using either Mann-Whitney or Kruskal-Wallis tests (for categorical variables) or Spearman rank correlation (for continuous variables), and corrected for multiple comparisons if applicable. (B) GW-2580 AUC among the patient population with clinical data (n = 202 samples from 199 patients), subdivided into de novo (n = 158), secondary (n = 24), and relapsed (n = 20) AML disease presentation categories. Statistics were calculated on subdivided categories by the Kruskal-Wallis test with the Dunn multiple comparisons test. (C) Kaplan-Meier survival curve of patients with AML with both clinical and survival data (n = 173), grouped by the response of their corresponding ex vivo primary sample to GW-2580: sensitive (0-20th percentile), indeterminate (20th-80th), and resistant (80th-100th). P value obtained from the overall log-rank test.

Figure 3.

CSF1R is expressed not on the bulk leukemia population in primary AML patient samples but on a small subpopulation of supportive cells. (A) Schematic diagram of CyTOF analysis to profile CSF1R^hi cells in primary AML patient samples. (B) Percentage of CSF1R^hi cells in primary samples from patients with AML (n = 66) and healthy donors (HD) (n = 11). (C) Correlation of the proportion of CSF1R^hi cells in primary AML patient samples with the sample’s response to GW-2580. Significance determined by Spearman rank test. (D-E) Representative CyTOF plots of CSF1R expression in primary AML patient samples that show (D) sensitivity and (E) resistance to GW-2580. (F-G) Violin plots of expression intensity of other cell-surface markers in CSF1R^hi cells from (F) AML patient and (G) healthy donor samples. (H) Cell-surface marker expression (median arcsinh) in CSF1R^hi cells for AML patient and healthy donor samples. (I) t-SNE analysis of CSF1R expression in cells isolated from primary AML patient sample (n = 66). (J) t-SNE analysis with labeled CSF1R^hi and non-CSF1R^hi cells in primary AML patient samples (n = 66) and healthy donors (n = 11).

Figure 4.

HGF stimulates growth in CSF1R inhibitor–sensitive samples and its secretion is regulated by CSF1R activation. (A) Schematic of analysis connecting cytokine growth assay results (data from Carey et al) with CSF1R inhibitor sensitivity. (B) Cytokines and growth factors that increase AML cell growth are significantly associated with sensitivity to GW-2580. Data represent log₂ fold change of GW-2580 AUCs between responders and nonresponders (n = 68 primary AML patient samples and 94 cytokines/growth factors) vs the unadjusted P value, determined by Student t tests. (C) Schematic of evaluating cytokine secretion after stimulation/inhibition of CSF1R in primary AML patient samples (n = 15). (D-E) Change in cytokine levels in conditioned media of primary AML patient samples after (D) CSF1R stimulation and (E) CSF1R inhibition. Cytokine levels for each patient sample are normalized to untreated and ranked by median value. (F) Difference in rank order of cytokines from panels D and E identifies cytokine secretion profile associated with upregulated and downregulated CSF1R activity. (G) Baseline HGF levels in primary AML patient samples correlate with GW-2580 sensitivity (n = 10). Significance determined by Spearman rank correlation.

Figure 5.

Sensitivity to CSF1R inhibitors correlates with MET inhibitor sensitivity and is eliminated after external cytokine stimulation. (A-C) Strong correlation in primary AML patient samples (n = 315) between GW-2580 sensitivity and sensitivity to 3 MET inhibitors: (A) crizotinib, (B) foretinib, and (C) SGX-523. Significance determined by Spearman rank correlation. (D-G) GW-2580 dose-response curves for 4 primary AML patient samples treated with recombinant HGF (1 µg/mL), HS-5–conditioned media, or untreated. Error bars represent mean plus or minus the standard error of the mean (n = 4 replicates); nonlinear curve fitting conducted using least squares regression. Significance determined by 1-way analysis of variance on the area under each curve with the Dunn test for multiple comparisons. (H) Model of CSF1R inhibitor sensitivity in primary AML patient samples resulting from paracrine secretion of cytokines by CSF1R-expressing supportive cells.