Human KIT+ myeloid cells facilitate visceral metastasis by melanoma

Colonization of visceral organs with melanoma in humanized NSG-SGM3 mice is dependent upon human CD33+CD11b+CD117+ progenitor cells. A gene signature of KIT/CD117–expressing CD33+ subset correlates with decreased overall survival in TCGA melanoma samples and represents a novel candidate biomarker.


Introduction
Mechanisms sustaining the growth of human melanoma in distant organs remain poorly defined. This is important because metastatic dissemination remains a major clinical challenge (Eggermont et al., 2016). Melanoma can metastasize to all organs (Gupta and Brasfield, 1964;Einhorn et al., 1974;Meyer and Stolbach, 1978;Nathanson et al., 1967;Patel et al., 1978), and patients with nonpulmonary visceral metastasis, including liver, have the worst prognosis (Balch et al., 2009). Improved survival has been documented for patients with metastatic melanoma treated with B-raf proto-oncogene serine/threonine kinase-and/ or mitogen-activated protein kinase kinase-targeted therapies (Flaherty et al., 2016;Long et al., 2017) or with blocking antibodies targeting CTL-associated protein (CTLA)-4 (Hodi et al., 2010) and/ or programmed death (PD)-1 (Wolchok et al., 2017). However, a significant fraction of patients do not achieve prolonged survival even in combination therapy trials and succumb to treatmentresistant metastatic disease (Wolchok et al., 2017).
Dissemination and growth in distant organs are driven by a complex interplay between cancer cells and their microenvironment. A number of cell types of myeloid lineage have been implicated in cancer metastasis (Garner and de Visser, 2020). In the mouse, neutrophils are engaged in various steps of metastasis (Jaillon et al., 2020), including support of cancer cell proliferation and survival (Acharyya et al., 2012;Ouzounova et al., 2017;Szczerba et al., 2019), angiogenesis (Bekes et al., 2011;Nozawa et al., 2006), increased extravasation of disseminated cancer cells (Spiegel et al., 2016), and inhibition of natural killer (NK) cell-and CTL-mediated cancer cell clearance (Coffelt et al., 2015;Spiegel et al., 2016). Immature myeloid cells that have not completed differentiation and are able to exert suppressive effects on adaptive immunity (i.e., myeloid-derived suppressor cells) promote invasion, angiogenesis, and metastasis formation (Datta et al., 2019). Furthermore, tumor-associated macrophages promote development of metastatic disease by means of supporting tumor growth in target tissues and escape from NK and T cells (Etzerodt et al., 2019;Lewis et al., 2016;Mantovani et al., 2017;Ruffell and Coussens, 2015). Thus, myeloid cells could serve as putative therapeutic targets. Their remarkable heterogeneity makes an in-depth characterization a major goal, where in vivo studies would be most informative. However, differences in myeloid cells and their downstream signaling pathways between humans and mice (Hagai et al., 2018;Kanazawa, 2007;Mestas and Hughes, 2004;Williams et al., 2010) make it uncertain to directly extrapolate mouse in vivo data into humans. Here, using a transplantable model of human melanoma in humanized mice, we find that CD117 + CD11b + CD33 + myeloid cells support melanoma growth in distant organs.

Results and discussion
Metastatic melanoma tumors in patients and in humanized mice are infiltrated with CD33 + myeloid cells To define the landscape of leukocytes in metastatic melanoma, we analyzed transcriptional profiles of metastatic melanoma tumors from 14 patients with RNA sequencing (RNA-seq; Table  S1). CIBERSORT, which estimates the fraction of various leukocyte RNA (Newman et al., 2015), revealed that nearly half of the leukocyte transcripts originated from myeloid cells (Fig. 1 A). Modular analysis (Banchereau et al., 2016) revealed several lymphoid modules including B cells (36 genes), T cells (107 genes), and cytotoxic/NK cells (59 genes) as well as IFN modules (138 genes), which together clustered patient samples into two groups (hot and cold tumors; Fig. 1 B and Table S2). Inflammation (876 genes) and myeloid (78 genes) modules were spread across all samples (Fig. 1 B). These findings were confirmed using The Cancer Genome Atlas (TCGA) melanoma cohort (Cancer Genome Atlas Network, 2015) from primary (n = 66) and metastatic (n = 264) melanoma tumors from different patients (Fig. S1 A). There, CIBERSORT revealed significantly higher monocyte/macrophage-related transcripts in metastatic versus primary tumors (Fig. 1 C). Metastatic sites also showed a significantly higher expression of pan-myeloid marker CD33 (sialic acid binding Ig-like lectin 3; P = 2.3e-07; Fig. 1 D). Immunofluorescence staining revealed the presence of CD33 + myeloid cells in close proximity to melanoma cells in patient tumors (Fig. S1 B).
hNSG-SGM3 mice support distant organ colonization with melanoma We then implanted the two humanized mouse strains with 10 7 Me275 human melanoma cells s.c. (Fig. 2 A ;Rongvaux et al., 2014). Upon necropsy before any tumor-induced mortality, hNSG-SGM3 mice displayed a vastly greater number of macroscopic tumors in the spleen and liver than hNSG mice (Fig. 2 B; and Fig. S2, A-C), while implantation site s.c. tumors (primary tumors) grew at similar rates in both strains (Fig. 2 C). Immunofluorescence staining for melanoma-associated proteins, melanoma antigen recognized by T cells 1 (MART-1), and glycoprotein 100 (gp100) confirmed the presence of melanoma cells and tumor formation in distant visceral organs (Fig. 2 D and Fig. S2 D). Tumors were imaged in vivo or ex vivo with an in vivo imaging system (IVIS) to detect luciferase-labeled Me275 cells (Fig. 2, E and F). Luciferase signal was present in multiple visceral organs (liver,spleen,pancreas,stomach,kidney,lungs,and bone;Fig. 2 E). Kinetic experiments following the appearance of macroscopic tumors as well as luminescence signal reflecting tumor growth in vivo revealed that Me275 cells require at least 42 d for visceral tumors to appear (Fig. 2, F and G).
To determine if human leukocytes were involved, we compared Me275 tumor growth in NSG-SGM3 mice with or without a human immune system. Only sporadic macroscopic tumors in visceral organs could be found upon necropsy (Fig. 2 H), and no MART-1/gp-100 staining could be detected in the liver harvested from nonhumanized mice (Fig. 2 I). In contrast, distant organs were colonized with Me275 cells in hNSG-SGM3 mice (Fig. 2, H and I). In both cases, Me275 cells grew at the primary site at similar kinetics (Fig. S2 E). Thus, human leukocytes are critical for distant organ colonization by melanoma cells. This was further confirmed by adoptive cell transfer (ACT) experiments where hCD45 + leukocytes were purified from the liver and spleen of tumor-naive hNSG-SGM3 mice and adoptively transferred (10 7 cells) into nonirradiated tumor-naive nonhumanized NSG-SGM3 mice or NSG mice. Both cohorts were implanted with Me275 cells s.c. immediately thereafter (Fig. 2 J). NSG-SGM3 mice adoptively transferred with hCD45 + leukocytes showed significantly higher numbers of melanoma tumors in the liver than controls without ACT ( Fig. 2 K). In contrast, NSG mice adoptively transferred with hCD45 + cells purified from hNSG-SGM3 mice developed only sporadic liver tumors ( Fig. 2 K). Thus, the capacity of hCD45 + leukocytes to support melanoma growth in distant organs is dependent on at least one of the host human cytokines SCF, GM-CSF, and/or IL-3.
Distant organ colonization is dependent upon hCD33 + myeloid cells Tissue examination revealed close proximity of hCD33 + cells and melanoma cells in the liver and spleen of hNSG-SGM3 mice ( Fig. 3 A). The transcriptional profiles of hCD33 + cells purified from the spleen and liver of hNSG-SGM3 mice were established using RNA-seq, and the analysis was focused on the myeloid gene set selected by Nanostring. The Venn diagram analysis revealed a 78% overlap in myeloid gene expression between melanoma metastases from patients and hCD33 + cells from mice ( Fig. 3 B). The top networks were driven by the expression of STAT3, STAT1, and NFKB1, identical to those observed in melanoma patients (Fig. 3 Table S6, and Table S7). To establish whether hCD33 + myeloid cells are involved in tumor formation, hCD45 + cells isolated from the liver and spleen of hNSG-SGM3 mice were subdivided using magnetic beads into hCD33 + and hCD33 neg cell fractions (Fig. 3 D) and used in ACT experiments as described above. Recipient mice that received hCD33 + cells showed melanoma tumors in the spleen and in the liver, while those that received hCD33 neg cells did not (Fig. 3 E; with no impact on the growth of the primary tumors as shown in Fig. S2 G).
Single-cell RNA-seq (scRNA-seq) of hCD33 + cells from the liver/spleen of hNSG-SGM3 mice (Fig. 4 F; Fig. S3, A-C; and Table S8) revealed that KIT + cells, which coexpress ITGAM (CD11b) and IL3RA but not CSF2RA (Fig. 4 G), are composed of four clusters (Fig. 4 F, clusters 2, 3, 5, and 7; and Fig. S3, B and C), including two clusters of cells expressing transcripts coding for mast cell proteolytic enzymes (TPSAB1, TPSB2) and leukotriene catabolism (HPGD, HPGDS, LTC4S; clusters 2 and 3); one cluster of mature mast cells with high expression of high-affinity IgE receptor (FCER1A) and CCR3 (cluster 7); and one cluster of dividing progenitor cells expressing cell cycle and DNA synthesis genes (cluster 5). In line with this, tissue analysis revealed differences between CD117 and tryptase expression in both experimental metastatic tumors in hNSG-SGM3 mice and in metastasis from melanoma patients (Fig. S3, D-F). As the depletion of FCER1A-expressing cells did not impact melanoma growth in distant organs (Fig. S3 I), we conclude that prometastatic activity is distant from mature mast cells and is linked with progenitor cells.
We then stratified the cells into two groups based on the expression level of the KIT gene (Fig. 4 H) to establish the transcriptome of KIT + cells. The differentially expressed genes (DEGs) between the two groups of cells were computed using the Wilcoxon rank sum test. All the genes that had an absolute log 2 fold-change value >0.2 and a false discovery rate of <5% were selected as KIT signature genes (n = 221; Table S9). The prognostic significance was evaluated using the log 2 foldchange weighted mean expression of the KIT signature genes in samples in TCGA (Cancer Genome Atlas Network, 2015). Since the score is a measure of the specific cell type, the samples were then stratified into two groups based on the median of the score, and the survival difference in the two groups was visualized using Kaplan-Meier plots (Fig. 4 I). The Kaplan-Meier analysis revealed that the high KIT score was associated with significantly lower survival probability (P < 0.0005; hazard ratio = 1.94; 95% confidence interval = 1.33, 2.82).
Thus, hCD33 + CD11b + CD117 + myeloid cells facilitate distant organ colonization by human melanoma after subcutaneous implantation. The molecular mechanisms regulating the ability of these cells to promote tumor growth in our model remain to be identified. Importantly, despite the presence of endogenous murine myeloid cells, the distant organ colonization is sporadic in the absence of human myeloid cells. This suggests that the molecular pathways governing myeloid cell-dependent melanoma growth are restricted in the absence of human cells. Our studies herein suggest a potential novel prognostic biomarker and downstream effector molecule(s) that might represent a therapeutic target.
(C) CIBERSORT analysis of RNA-seq data from TCGA primary (n = 66) and metastatic (Met) melanoma (n = 264) tumors. Values are mean percentage. Two-way ANOVA with Bonferroni's multiple comparisons test. ****, P < 0.0001; **, P < 0.01. (D) CD33 expression in TCGA primary and metastatic melanoma tumors. P = 2.3 × 10 −7 Wilcoxon test. The error bar is the SD. (E) hCD33 + cells from the spleen of three hNSG (blue) and three hNSG-SGM3 (orange) mice were gated and subjected to t-distributed stochastic neighbor embedding (tSNE) reduction. Indicated markers were color-mapped from blue (low density) to red (high density) into the tSNE map. (F) tSNE plots of hCD33 + cells from the liver as analyzed in E. (G) Bulk RNA-seq of hCD33 + cells enriched and pooled from the spleen and liver of hNSG and hNSG-SGM3 mice. DEGs were illustrated in the dot plot using log 2 fold change as the x axis and TPM of hNSG (blue) or hNSG-SGM3 (orange) as the y axis with criteria of logTPM expression ≥1 and absolute log2 fold change ≥1 (dotted lines). (H) Heatmap comparing IPA on the DEGs in CD33 + cells from hNSG and hNSG-SGM3 mice for immune canonical pathway analysis. Eos, eosinophils; DCs, dendritic cells; MCs, mast cells; gd T, γΔ T cells; Mono/Mac, monocyte/macrophage; PC, plasma cells; PMN, polymorphonuclear leukocytes. fMLPN, formyl-L-methionyl-L-leucyl-phenylalanine.

Materials and methods
Cell lines Melanoma cancer cell line Me275 (Research Resource Identifier [RRID]:CVCL_S597), which was established from surgically excised melanoma metastases from patient LAU50, was provided by Pedro Romero at the Ludwig Institute for Cancer Research at the University of Lausanne (Lausanne, Switzerland). Tumor cells were cultured in complete RPMI (RPMI 1640, n = 12-13 mice per strain, two-tailed Mann-Whitney test. (I) Localization of hCD45 (red), MART-1/gp100 (green), and DAPI (blue) in the liver from one mouse per strain from H. Scale bar = 1 mm for the whole section and 60 µm for the selected zoom-ins. (J) Outline of the experiment. 10 7 hCD45 + cells enriched from naive hNSG-SGM3 mice were transferred i.v. into NSG-SGM3 or NSG mice and subsequently implanted with 10 7 Me275 cells s.c. on the same day. (K) Macroscopic tumors in the liver at 8 wk after ACT as in J. n = 5 mice per recipient group, one-way ANOVA test. ****, P < 0.0001; ***, P < 0.001; **, P < 0.01. p, photons; sr, steradiance.  25 mM Hepes, 1 mM sodium pyruvate, 1% nonessential amino acid, 1% penicillin-streptomycin, and 2 mM L-glutamine) supplemented with 10% FBS at 37°C with 5% CO 2 atmosphere and authenticated using Short Tandem Repeat profiling analysis by the American Type Culture Collection. The mycoplasma test was performed regularly, and cells were negative for mycoplasma before each experiment.

Tumor model
Tumor cells were injected s.c. into the flank of the mice. Tumor size was monitored every 7 d with a caliper. Tumor volume (ellipsoid) was calculated as follows: (short diameter) 2 × long diameter/2. Alternatively, luciferase-labeled melanoma cells were injected i.v. into the mice. Mice were killed, and the macroscopic metastases were identified and scored in various organs including lymph nodes (axillary and brachial), livers, spleen, kidneys, and lungs.

In vivo imaging
Before mice were anesthetized with Isoflurane, an aqueous solution of luciferin (150 mg/kg i.p.) was injected 10 min before imaging with IVIS (PerkinElmer). The animals were placed into the light-tight chamber of the charge-coupled device camera system, and the photons emitted from the luciferase-expressing cells within the animal were quantified using Living Image (PerkinElmer). To image dissected organs, mice were first injected i.p. with luciferin (150 mg/kg) for 10 min and quickly killed to remove each organ. Organs were imaged in 12-well culture dishes with PBS containing 300 µg/ml luciferin.

RNA-seq
Total RNA was isolated from snap-frozen metastatic melanoma tissues (Table S1; Cooperative Human Tissue Network, Pennsylvania) and CD33 + cells from the spleen and livers of hNSG-SGM3 and hNSG mice using the RNA isolation kit following the manufacturer's protocol (Qiagen). Total RNA isolated was run on a Qubit (Thermo Fisher Scientific) and a Bioanalyzer 2100 Nano Chip (Agilent Technologies) to check RNA quantity and quality. Sequencing libraries were prepared using KAPA Stranded mRNA-seq kit (Roche) according to the manufacturer's protocol. First, poly-A RNA was isolated from 300 ng total RNA using oligo-dT magnetic beads. Purified RNA was then fragmented at 85°C for 6 min, targeting fragment range 250-300 bp. Fragmented RNA was reverse transcribed with an incubation of 25°C for 10 min and 42°C for 15 min and an inactivation step at 70°C for 15 min. This was followed by second strand synthesis at 16°C for 60 min. Double-stranded cDNA fragments were purified using Ampure XP beads (Beckman Coulter), then A-tailed and ligated with Illumina adapters. Adapter-ligated DNA was purified using AMPure XP beads and followed by 10 cycles of PCR amplification. The final library was cleaned up using AMPure XP beads. Quantification of libraries was performed using real-time quantitative PCR (Thermo Fisher Scientific). Sequencing was performed on an Illumina NextSeq 500 platform generating single-end reads of 75 bp. All primary analysis of RNA-seq was processed using CASAVA pipeline (Illumina, v1.8.2). Sequences were aligned with Bowtie 2 (Kim et al., 2013), and counts were generated with RSEM (Anders et al., 2015) using the annotations from Ensembl GRCh37 (Harrow et al., 2012). The files from alignment result were converted to BAM format using SAMtools (Li et al., 2009). Raw counts were normalized to log 2 transformed transcript per million (TPM) or fragments per kilobase of transcript per million mapped reads (FPKM; log 2 (FPKM + 1)). CIBERSORT was used to estimate the proportions of diverse immune cell types using the genes that define the signature expression of the immune cell types. We used the default 22 cell types (LM22) provided (Newman et al., 2015). For modular analysis, a set of 260 transcriptional modules was used as a preexisting framework (Banchereau et al., 2016). Module-level activity score was calculated by R package gene set variation analysis (GSVA) score from RNA-seq data (Hänzelmann et al., 2013). Ingenuity pathway analysis (IPA; Qiagen) was applied to reveal transcriptional networks, and ClueGO was used to illustrate biological interpretation of genes (Bindea et al., 2009).

scRNA-seq
Enriched hCD33 + cells from liver and spleen of hNSG-SGM3 mice were resuspended in PBS containing 0.04% BSA, the cell numbers were counted on the Contess II automated cell counter (Thermo Fisher Scientific), and ∼12,000 cells were loaded per channel on Chromium microfluidic chips (10x Genomics). Singlecell capture, barcoding, and library preparation were performed using the 10x Chromium version 2 chemistry according to the manufacturer's protocol (10x Genomics). The quality of cDNA and libraries was checked on an Agilent 4200 TapeStation, quantified by KAPA quantitative PCR, and sequenced on a HiSeq 4000 (Illumina) to an average depth of 50,000 reads per cell. We quantified gene expression counts from raw sequencing data using Cell Ranger v2.2 with GRCh38. Datasets from liver and spleen of two independent experiments were normalized using Harmony (Korsunsky et al., 2019).

Database and statistical analysis
Statistical analysis was performed in Prism 8 (GraphPad). Figure legends denote P values as follows: ****, P < 0.0001; ***, P < 0.001; **, P < 0.01; and *, P < 0.05. Comparisons between any two groups were analyzed using the Mann-Whitney test or two-tailed t test, and comparisons between any three or more groups were analyzed by ANOVA as indicated in the respective legends. The Kaplan-Meier curves of melanoma patient data were generated using RNA-seq data from the TCGA-skin cutaneous melanoma project (Cancer Genome Atlas Network, 2015). The DEGs between the two groups of cells were computed using the Wilcoxon rank sum test. The statistical significance of the difference in the Kaplan-Meier survival plot was computed using the log-rank test using R package.
Online supplemental material Fig. S1 shows human myeloid cells in metastatic melanoma tumors and hNSG-SGM3 mice. Fig. S2 shows that hNSG-SGM3 mice promote melanoma growth in distant organs via hCD33 + cells. Fig. S3 shows the expression of hCD117 in hNSG-SGM3 mice and metastatic melanoma tumor. Table S1 shows the list of melanoma patient tumors used in the study. Table S2 shows the list of genes in annotated immune modules. Table S3 shows the kinetics of human engraftment in the blood of hNSG and hNSG-SGM3 mice. Table S4 shows human engraftment in the tissues of hNSG and hNSG-SGM3 mice. Table S5 shows DEGs of bulk RNA-seq on hCD33 + cells enriched from the spleen and liver of hNSG and hNSG-SGM3 mice. Table S6 shows IPA on upstream regulator for the myeloid genes expressed in CD33 + cells from hNSG-SGM3 mice. Table S7 shows IPA on upstream regulator for the myeloid genes expressed in human melanoma tumors. Table S8 shows marker genes specific for each cluster in scRNA-seq of hCD33 + cells from hNSG-SGM3 mice. Table S9 shows KIT signature genes identified from scRNA-seq of hCD33 + cells from hNSG-SGM3 mice. Table S10 shows the list of antibodies used in the study.
Disclosures: C.I. Yu reported a patent to humanized mouse model for cancer metastasis pending. P. Metang reported "other" from University of Texas Southwestern Medical Center outside the submitted work. J. Banchereau reported grants from Merck during the conduct of the study; grants from Sanofi, personal fees from Cue Biopharma, personal fees from Neovacs, personal fees from Ascend Pharma, and personal fees from Georgiamune outside the submitted work; in addition, J. Banchereau had a patent to humanized mouse model for cancer metastasis pending. K. Palucka reported grants from Merck, "other" from Merck, personal fees from Cue Biopharma, and personal fees from Sobi outside the submitted work; in addition, K. Palucka had a patent on humanized mice to study metastasis pending. No other disclosures were reported.