Molecular Medicine Israel

Monitoring melanoma patients on treatment reveals a distinct macrophage population driving targeted therapy resistance


  • Targeted therapy resistance is associated with high expression of POSTN in melanoma
  • POSTN induces a targeted therapy resistance (TTR)-associated macrophage phenotype
  • POSTN promotes therapy resistance in a mouse melanoma model
  • TTR macrophages orchestrate melanoma resistance formation via CD44 signaling


Resistance to targeted therapy remains a major clinical challenge in melanoma. To uncover resistance mechanisms, we perform single-cell RNA sequencing on fine-needle aspirates from resistant and responding tumors of patients undergoing BRAFi/MEKi treatment. Among the genes most prominently expressed in resistant tumors is POSTN, predicted to signal to a macrophage population associated with targeted therapy resistance (TTR). Accordingly, tumors from patients with fast disease progression after therapy exhibit high POSTN expression levels and high numbers of TTR macrophages. POSTN polarizes human macrophages toward a TTR phenotype and promotes resistance to targeted therapy in a melanoma mouse model, which is associated with a phenotype change in intratumoral macrophages. Finally, polarized TTR macrophages directly protect human melanoma cells from MEKi-induced killing via CD44 receptor expression on melanoma cells. Thus, interfering with the protective activity of TTR macrophages may offer a strategy to overcome resistance to targeted therapy in melanoma.


Despite the groundbreaking outcomes achieved in the understanding and treatment of metastatic melanoma, it remains one of the most aggressive cancers worldwide. More than 50% of melanomas carry mutations in the mitogen-activated protein kinase (MAPK) pathway, with oncogenic alterations in BRAF and NRAS being the most prevalent. Therefore, substantial efforts have been made to develop selective BRAF inhibitors in combination with MEK inhibitors. Although these drugs are associated with pronounced clinical benefits, approximately 50% of the responders develop acquired resistance after a median of 6–12 months. Therefore, early molecular characterization of high-risk stage II, III, and IV tumors in melanoma patients has become a research priority to define the factors driving resistance.

Applying next-generation sequencing techniques, many genetic and non-genetic mechanisms of adaptive resistance to BRAF and MEK inhibitors have been identified. Tsoi et al.10 demonstrated that the regulation of melanoma resistance proceeds via stepwise de-differentiation of four distinct cell subtypes: undifferentiated, neural crest-like, transitory, and melanocytic cells. These data revealed the high plasticity of melanoma cells and indicated that different phenotypic states could contribute to intratumoral melanoma heterogeneity and therapeutic response. The existence of specific, intrinsically therapy-resistant melanoma subpopulations suggests new potential vulnerabilities that could be targeted, for instance in the context of combinatorial treatment strategies. Indeed, recent studies have demonstrated that in cell culture or in patient-derived xenograft (PDX) models, targeting de-differentiated melanoma cells could overcome resistance induced by BRAFi/MEKi or immunotherapy-associated cytokines.

To understand how molecular features of drug resistance are acquired during targeted therapy, we acquired longitudinal tumor cell collections by fine-needle aspiration (FNA) from four consenting melanoma patients at baseline and at different time points following BRAFi/MEKi treatment, followed by single-cell RNA sequencing (scRNA-seq). Gene expression analysis revealed POSTN as the factor most prominently associated with resistance to therapy. Together with the analysis of further patient cohorts and functional assays, our study revealed that high POSTN expression was associated with infiltration and polarization of a targeted therapy resistance (TTR)-associated macrophage population that mediates resistance formation.


Screening for targeted therapy resistance mechanisms using FNA biopsies from patients on treatment

To perform longitudinal monitoring of melanoma patients on treatment and to investigate how the cellular landscape dynamically changes in response to targeted therapy, we applied a longitudinal FNA cellular collection method—cohort 1 (Figure 1A). Applying scRNA-seq technology, we analyzed FNAs from four consenting melanoma patients before dabrafenib and trametinib (BRAFi/MEKi) treatment and at different time points following treatment (Figure 1B). Two tumors from patients 1 and 2, respectively, turned out to be therapy resistant. Tumor 3 from patient 3 demonstrated a complete response, while tumor 4 from patient 4 responded in the beginning without reaching a complete response and progressed later. All patients had different treatment histories before receiving BRAFi/MEKi therapy (Table S1).

The scRNA-seq analysis revealed 31 cell clusters in our samples (Figure S1A). The identified cell types were CD8+ T cells, CD4+ T cells, NK cells, B cells, macrophages, and melanoma cells (Figure 1C). Malignant cells were inferred to have copy number aberrations compared to normal cell types (Figure S1B). Notably, no stromal cells (fibroblasts and endothelial cells) were detected in these samples. Longitudinal analysis of the individual biopsies revealed that the cell-type composition gradually changed during the treatment course (Figure 1D), where the greatest changes involved infiltration of immune cells.

Melanoma cell populations of BRAFi/MEKi-resistant tumors display high POSTN and MDK expression in patients

We first focused on malignant cells from the four melanoma tumors. A total of 45,916 malignant cells were compared between the responders and resistant groups to identify transcriptomic differences. The malignant cells from the different patients clustered separately, revealing a high degree of patient-specific heterogeneity (Figure S1C). To investigate whether the transcriptomic profiles of malignant cells in resistant tumors overlap with resistance-related cellular phenotypes previously identified in cell culture and in PDX model systems, we applied AddModuleScore from Seurat using melanoma signatures from Hoek,14 Tirosh,8 Tsoi,10 Rambow,9 and Verfaillie.15 Although some cells displayed high positive enrichment scores with known melanocytic/proliferative signatures, there was a lack of enrichment for known drug-resistant signatures (Figure S1D).

To identify genes potentially associated with drug resistance, we performed differential expression analyses between the responding (tumor 3) and resistant (tumors 1 and 2) malignant cells. Of the top 10 genes overexpressed in the resistant cells, three were secreted factors: SPP1MDK, and POSTN (Figures 2A and S2A), which were also expressed in other cells of the FNA samples (Figures 2B and S2B). However, SPP1 was present only in the malignant cells of tumors 1 and 4 (Figures S2B and S2C), suggesting that SPP1 in malignant cells is not generally involved in TTR. In contrast, MDK and POSTN were prominently expressed in malignant cells at different time points of treatment in resistant tumors 1 and 2, while their expression was generally low in the responding tumor 3 (Figure S2C). Intriguingly, expression of MDK and POSTN increased with time in biopsies from tumor 4, correlating with the switch from therapeutic response to resistance formation in this patient (Figure S2C).

To confirm that high POSTN and MDK expression is associated with targeted therapy resistance, we performed an RNAscope assay for POSTN and MDK on melanoma tumors in a separate patient cohort (cohort 2, Figure 2C; Table S1) where patients were grouped according to fast progression versus slow progression disease after start of targeted therapy. To perform an unbiased statistical analysis, categorization of the patients in this cohort into fast vs. slow progression disease was done by determining the median progression-free survival (PFS) (136.5 days). The percentage of POSTN+ cells was significantly higher in four fast-progressing tumors compared to five slow-progressing tumors (p = 0.016, Figures 2D and 2E). Likewise, fast-progressing tumors displayed more prominent POSTN (also termed periostin) protein expression than slow-progressing tumors, as shown by immunohistochemical staining of tumor sections (Figure S2D). MDK expression was also notably higher in four fast-progressing tumors compared to four slow-progressing tumors (Figures 2F and 2G), although the difference was not significant (p = 0.19).

In various tumor types, MDK and POSTN have been found to be strongly expressed in cancer-associated fibroblasts (CAFs).16,17 Since our FNA technology failed to capture CAFs (Figure 1C), we investigated POSTN and MDK expression in the tumor microenvironment of targeted therapy-resistant melanoma biopsies using the Tumor Profiler dataset.18 In addition to their expression in therapy-resistant malignant cells, POSTN and MDK were expressed in several other cell populations, notably in fibroblasts of resistant tumors (Figures 2H and 2I). Thus, distinct cell types within a tumor can apparently act as a source for POSTN and MDK, raising the question about the identity of their effector cells.

POSTN or MDK are unable to directly induce resistance against MEKi-driven cell death in melanoma cells

Potential effector cells of POSTN or MDK in resistant tumors could be the malignant cells themselves. To test whether POSTN or MDK could directly affect the responsiveness of melanoma cells to MAPK inhibitors, we first identified primary melanoma cell lines displaying either high or low endogenous POSTN and MDK expression (M150325, M131205 and M150543, M161201, respectively; Figures S3A and S3B). In the two POSTNhigh cell lines, POSTN expression could be efficiently downregulated by transfection with POSTN siRNA (Figure S3C). However, both inhibition of POSTN in POSTNhigh cell lines and reconstitution of POSTN levels in the POSTNlow cell lines by adding recombinant POSTN protein did not alter MEKi-induced pERK downregulation (Figures S3D and S3E). Accordingly, annexin V staining did not reveal a difference in the level of MEKi-induced cell death upon manipulation of POSTN levels (Figures S3F and S3G). Likewise, adding recombinant MDK to the MDKlow cell lines did also not stabilize pERK or increase cell survival upon MEKi treatment (Figure S3H). Thus, neither POSTN nor MDK altered the susceptibility of melanoma cells to MEKi-induced cell death.

High expression of POSTN in resistant tumors is associated with infiltration of a specific macrophage population

As POSTN and MDK were unable to directly drive resistance in melanoma cells, we examined the cell populations that could be affected by POSTN or MDK. CellChat Explorer analysis of our scRNA-seq data from cohort 1 predicted that both POSTN and MDK mostly signal to macrophages, given the high expression of the POSTN receptors ITGB5, ITGAV, and ITGB3 (Figures 3A and S4A) and of several MDK receptors, such as LRP1 and NOTCH2 (Figure S4B), respectively, on these cells.

We then reclustered the macrophage population by unsupervised hierarchical clustering with the top 100 highly variable genes and found four distinct clusters (Figure 3B). Cluster 1 was notably different from other clusters and exhibited high expression of CD48MNDA, and FCN1 (Figure 3C), i.e., markers associated with a pro-inflammatory phenotype,19,20,21 suggesting that cluster 1 represented anti-tumorigenic macrophages likely associated with targeted therapy sensitivity (TTS). In contrast, clusters 2, 3, and 4 demonstrated common upregulation of several markers characteristic for a pro-tumorigenic phenotype, such as CD63GPNMBMCR1MSR1CD163, and LIPA22,23,24 (Figure 3C). Moreover, each of these clusters displayed high expression of a specific set of other phenotypic genes, such as SPP1CCL2MARCO, and LGALS3 in cluster 2, SPP1 and C1QC19,25 in cluster 3, and LGALS3MARCO, and C1QC in cluster 4. Given their apparent pro-tumorigenic phenotype, we grouped clusters 2, 3, and 4 into one macrophage population predicted to represent (TTR)-associated macrophages. Among the markers significantly upregulated in TTR compared to TTS macrophages were SPP1CD63CD163LIPAMSR1MRC1GPNMBCCL2MARCO, and LGALS3 (Figure 3D). Enrichment scores for both M1 and M2 macrophage signatures26,27 in TTR and TTS macrophages were low (<0.5), indicating that the TTS and TTR macrophage phenotypes do not belong to these classical categories (Figures S4C and S4D).

Cell-type proportion analysis in our tumors demonstrated that TTR macrophages were indeed present at high levels in resistant tumors 1 and 2 (Figure 3E). In contrast, in responding tumor 3, the proportion of TTR macrophages decreased, while the levels of TTS macrophages increased during treatment. In tumor 4, the proportion of TTR macrophages increased in accordance with the timeline of resistance formation, with predominant presence of TTR macrophages at late time points of treatment.

CellChat Explorer analysis using malignant cells as the sender and the TTR and TTS macrophage populations, respectively, as receivers, demonstrated that TTR macrophages expressed significantly higher levels of POSTN and MDK receptors compared to TTS macrophages (Figures S5A and S5B). Furthermore, CellChat Explorer analysis also predicted that the POSTN pathway was used by malignant cells to signal exclusively to TTR macrophages (Figure S5C).

Next, we validated the presence of TTR macrophages in a separate set of four melanoma patients (cohort 3) using paired biopsies collected before the start of BRAFi/MEKi therapy and at the resistant stage (Figure 3F; Table S1). Multiplex immunofluorescence using five specific TTR markers (CD163, GPNMB, LIPA, MSR1, and CD63) and the general macrophage marker CD68 identified one macrophage cluster (high CD68 and CD163), which could be further subdivided into three clusters, one with high CD68, one with high CD163, and one highly expressing all five TTR markers (Figures S5D–S5F). Consistent with the scRNA-seq data, these TTR macrophages were already present at the baseline (Figure 3G). Intriguingly, however, the proportion of TTR macrophages significantly increased at relapse (Figure 3H). To compare TTR macrophage infiltration between fast- and slow-progressing tumors from cohort 2 (Figure 3I), we performed an additional multiplex immunohistochemistry analysis and found a significant increase in the number of TTR macrophages in fast-progressing in comparison to slow-progressing tumors (Figure 3JJ and 3K). Taken together, resistance to BRAFi/MEKi in melanoma appears to be strongly correlated with increased TTR macrophage infiltration.

POSTN expression polarizes macrophages toward a TTR phenotype and counteracts targeted therapy in vivo

The results above raise the possibility that POSTN or MDK expression in melanoma mediate emergence of TTR macrophages and, as a consequence, resistance formation. To examine a potential association of these factors with the presence and function of TTR macrophages, we first assessed whether POSTN or MDK affects macrophage polarization. To this end, we differentiated macrophages from buffy coat-isolated monocytes and exposed them to recombinant POSTN or MDK for 3 days (Figure 4A), followed by expression analysis of the 10 markers used previously to characterize TTR macrophages on patient biopsies in situ (Figure 3D). Importantly, macrophages treated with POSTN displayed significant upregulation of most TTR macrophage markers, whereas MDK failed to polarize macrophages to a TTR phenotype (Figures 4B and 4C)….

Sign up for our Newsletter