AMBRA1 levels predict resistance to MAPK inhibitors in melanoma

To explore the possible role of AMBRA1 in melanoma response to MAPKi treatment, we assessed AMBRA1 expression in two cohorts of metastatic melanoma patients [GSE50509 (30) and GSE65185 (31)], pretreatment (baseline) and after treatment (resistant) with MAPKi. This analysis revealed a downregulation of AMBRA1 upon therapy resistance in ~32% of the cases across both cohorts (Fig. 1A). Consistently, AMBRA1 expression was down-regulated in BRAFi posttreatment samples in a cohort of melanoma patient-derived xenografts (PDXs) [GSE129127 (32)] (Fig. 1B) and in an established PDX model (MEL006) of minimal residual disease (33) (SI Appendix, Fig. S1A). By using a 25% cut-off, we categorized baseline samples from the GSE50509 and GSE65185 cohorts into AMBRA1^HIGH (upper quartile) and AMBRA1^LOW (lower quartile) groups (Fig. 1C). Notably, the significant reduction in AMBRA1 expression in MAPKi-resistant samples was observed solely in the AMBRA1^HIGH baseline group (Fig. 1D).

We next employed human melanoma cell lines ranked for AMBRA1 protein expression, among which we selected FM-93/2 (AMBRA1^HIGH) and M24 (AMBRA1^LOW) as representative cell lines for our initial investigation of resistance development (SI Appendix, Fig. S1B). Following chronic exposure to BRAFi, we observed a significant reduction in AMBRA1 protein levels in all FM-93/2-derived BRAFi-resistant cell lines (R1 to R4) compared to the sensitive line (S) (Fig. 1E). Conversely, no major differences in AMBRA1 expression were observed in R1-R4 vs. S in the M24 cell line (Fig. 1E). Extending the analysis to other AMBRA1^LOW (FM-55/M2, Mel-5392, and OCM-3; SI Appendix, Fig. S1C) and AMBRA1^HIGH (M17, M88, and Ma-Mel-51; SI Appendix, Fig. S1D) cell lines, we observed a ~67% reduction in the AMBRA1 expression ratio across all resistant and sensitive cells (Fig. 1 F and G). Evaluation of AMBRA1 expression in paired resistant and sensitive cell lines mirrored the cohort results (Fig. 1 A and D), with significant differences for AMBRA1 downregulation observed only in the AMBRA1^HIGH group (Fig. 1H). Notably, AMBRA1^LOW cells required shorter time for the establishment of resistant cell lines, when compared with AMBRA1^HIGH cells (SI Appendix, Fig. S1E). Chronic exposure to BRAFi in murine YUMM1.7 and YUMM1.1 melanoma cell lines also resulted in a significant loss of Ambra1 expression (SI Appendix, Fig. S1F). Following a 7-d depletion of BRAFi, AMBRA1 protein levels were not rescued in either FM-93/2 (SI Appendix, Fig. S1G) or YUMM1.7 (SI Appendix, Fig. S1H) BRAFi-resistant cells.

Collectively, these findings indicate a widespread downregulation of AMBRA1 expression upon the acquisition of resistance to MAPKi therapy in melanoma.

We next sought to assess the correlation between AMBRA1 protein expression levels and response to MAPKi in melanoma cell lines. By utilizing the primary PRISM database, BRAF-mutated cell lines (V600E or V600K; SI Appendix, Fig. S2A) with different AMBRA1 levels (SI Appendix, Fig. S2B) were screened for individual MAPKi sensitivity (Fig. 2A). Results were first visualized using a heatmap (SI Appendix, Fig. S2C) and then represented as a linear model fitted between the median drug scores and AMBRA1 protein levels for each cell line (SI Appendix, Fig. S2D). Overall, this comprehensive analysis indicates that AMBRA1 expression negatively correlates with response to MAPKi (Fig. 2B). The MEKi cobimetinib exhibited the strongest negative correlation with AMBRA1 expression levels (R² = 0.42 and P = 0.0027) (Fig. 2B).

Extending this analysis in vitro using the panel of BRAF^V600E-mutated AMBRA1^HIGH and AMBRA1^LOW melanoma cell lines, we evaluated cell viability following acute treatment with increasing doses of BRAFi or MEKi (Fig. 2 C and D). The resistance index to either drug, calculated as the ratio between percentage of survival of MAPKi-treated and -untreated cells, revealed that AMBRA1^LOW cells exhibited a higher resistance to both BRAFi and MEKi compared with AMBRA1^HIGH cells (Fig. 2E). Further analysis determining EC₅₀ values for both drugs in each cell line (Fig. 2F) and plotting them against AMBRA1 protein levels (SI Appendix, Fig. S1B) confirmed a negative correlation with AMBRA1 expression for both BRAFi (r = −0.9513; P = 0.0003) (Fig. 2G) and MEKi (r = −0.8816; P = 0.0038) (Fig. 2H).

Altogether, these results suggest that low levels of AMBRA1 are associated with BRAFi/MEKi resistance.

Next, we genetically manipulated AMBRA1 in melanoma cells to investigate its impact on MAPKi response. Transfection with a plasmid encoding AMBRA1 in the AMBRA1^LOW M24 cells followed by BRAFi treatment (Fig. 3A) and in FM-93/2- (SI Appendix, Fig. S3A) and M17-derived (SI Appendix, Fig. S3B) BRAFi-resistant cells demonstrated a significant resensitization to BRAFi (Fig. 3 B and C and SI Appendix, Fig. S3 A and B). Conversely, AMBRA1^HIGH FM-93/2 cells transfected with a siRNA targeting AMBRA1 (siAMBRA1) (Fig. 3D) exhibited increased resistance to BRAFi compared to the control (siScr) (Fig. 3 E and F). Remarkably, while BRAFi effectively inhibited extracellular signal-regulated kinase (ERK) phosphorylation (pERK1/2-T202/Y204), AMBRA1 modulation did not influence phosphorylated ERK levels in either vehicle- or BRAFi-treated cells (Fig. 3 A and D), indicating that ERK is not involved in this signal transduction mechanism. Further experiments in BRAF^V600E-mutated SK-Mel-5 cells, silenced for AMBRA1 using an additional siRNA (siAMBRA1 #2), revealed an elevated resistance index for both BRAFi and MEKi treatments (Fig. 3 G and H and SI Appendix, Fig. S3 C and D).

To validate our findings ex vivo, we employed Ambra1 wild-type [BPA-derived melanoma cell (Bdmc^+/+)] or knock-out (Bdmc^−/−) primary murine melanoma cells (SI Appendix, Fig. S3 E and F). This analysis confirmed a higher resistance of Bdmc^−/− cells to both BRAFi (Fig. 3I) and MEKi (SI Appendix, Fig. S3G). In vivo validation involved subcutaneously injecting C57Bl/6 mice with either Bdmc^+/+ (sBPA^+/+) or Bdmc^−/− (sBPA^−/−) cells, followed by oral treatment with BRAFi (30 mg/kg) or a control solution (Veh) for 21 d (SI Appendix, Fig. S3 H and I). Tumor measurements over time revealed reduced therapy response to BRAFi in sBPA^−/− vs. sBPA^+/+ mice (P < 0.0001), evident in the tumor growth kinetics (Fig. 3J), representative images (Fig. 3K) and weight of tumors (Fig. 3L) at the time of collection.

Overall, these findings indicate that loss of AMBRA1 confers a higher resistance to MAPKi in vitro and in vivo.

A prevalent mechanism driving resistance to MAPKi in advanced melanoma is phenotype switching, a dynamic process characterized by transcriptomic features associated with invasiveness and dedifferentiation (16–19, 34–36), including expression of genes within the “Neural Crest Stem Cell-like” (NCSC-like) signature (37). Our prior transcriptomic data demonstrated an upregulation of NCSC-like-related genes in Ambra1 KO tumors (BPA^−/−) (Supplementary information in ref. 25 and SI Appendix, Fig. S4A). To further investigate this aspect, publicly available data from melanoma patients (TCGA-SKCM, n = 473) and cell lines [CCLE (Cancer Cell Line Encyclopedia), n = 49] were analyzed and samples were ranked into AMBRA1^HIGH and AMBRA1^LOW groups, based on AMBRA1 messenger RNA (mRNA) and protein expression levels, respectively (Fig. 4A). Gene Set Enrichment Analysis (GSEA) indicated a significant enrichment for “Undifferentiated” and NCSC-like gene sets (37) in the AMBRA1^LOW groups, while the “Melanocytic” gene set (37) was enriched in the AMBRA1^HIGH groups (Fig. 4A). These results indicate that AMBRA1^LOW tumors exhibit genetic features of dedifferentiation. A significant enrichment for the Undifferentiated signature was observed in the AMBRA1^LOW MAPKi-resistant tumors (PROG) derived from the AMBRA1^HIGH pretreatment group (PRE) in both the GSE65185 (Fig. 4B) and GSE50509 cohorts (SI Appendix, Fig. S4B) as well as in posttreatment samples from the GSE129127 PDX platform (SI Appendix, Fig. S4C).

We further characterized this aspect by selecting key regulator genes for these signatures, namely NGFR and AXL (for the Undifferentiated and NCSC-like signatures), and MITF (for the Melanocytic state). NGFR and AXL expression negatively correlated, whereas MITF expression positively correlated with AMBRA1 at mRNA or protein level, respectively, in both the TGCA-SKCM dataset (n = 443) (Fig. 4C) and in the AMBRA1^HIGH/AMBRA1^LOW melanoma cell panel (Fig. 4D).

To investigate the link between AMBRA1-dependent resistance to BRAFi and phenotype switching, we analyzed the expression of NGFR/AXL/MITF upon acquired resistance in AMBRA1^HIGH FM-93/2 cells. Remarkably, chronic exposure of AMBRA1^HIGH FM-93/2 cell line to BRAFi induced a switch from an NGFR^LOW/AXL^LOW/MITF^HIGH profile (characterizing sensitive cells) to an NGFR^HIGH/AXL^HIGH/MITF^LOW profile (which instead typifies R1-R4 BRAFi-resistant cells) (Fig. 4E and SI Appendix, Fig. S4 D–F). By contrast, AMBRA1^LOW M24 resistant cells show similar genetic features of the sensitive counterpart (Fig. 4E and SI Appendix, Fig. S4 D–F). These results reinforce the notion that AMBRA1^LOW cells possess a dedifferentiated genetic state. As a matter of fact, AMBRA1^LOW M24 cell line reconstitution with AMBRA1 (myc-AMBRA1) led to a significant reversal of gene expression toward a more differentiated state (NGFR^LOW /AXL^LOW/ MITF^HIGH) (Fig. 4F), which was accompanied by restored sensitivity to BRAFi (Fig. 3 A–C). Consistent with previous data showing increased metastatic potential of melanoma cells upon AMBRA1 deficiency (25), GSEAs also established an association between AMBRA1^LOW MAPKi-resistant tumors and an enriched invasive signature (38) in all cohorts analyzed (Fig. 4B and SI Appendix, Fig. S4 B and C). Accordingly, BRAFi-resistant cells of both murine (Fig. 4G) and human (SI Appendix, Fig. S4G) origin displayed higher cell migration capacity. Reconstitution experiments with an AMBRA1-encoding plasmid reduced relative cell migration, proving a direct link between AMBRA1 and invasiveness of BRAFi-resistant cells (SI Appendix, Fig. S4H).

Altogether, these results indicate that low expression levels of AMBRA1 are indicative of a phenotype switch toward a more invasive/dedifferentiated and BRAFi-resistant state (SI Appendix, Fig. S4I).

The NCSC-like state has been associated with activation of the FAK1 pathway in melanoma cell subclones exhibiting resistance to MAPKi (14). Our findings reveal that AMBRA1^LOW cells display a NCSC-like gene expression state (Fig. 4), along with resistance to MAPKi (Fig. 2). Consistent with our previous study demonstrating hyperactivated FAK1 signaling upon AMBRA1 deficiency in melanoma (25), all FM-93/2-derived BRAFi-resistant cell lines displaying features of the NCSC-like state (Fig. 4E and SI Appendix, Fig. S4 D–F) exhibited elevated levels of phosphorylated FAK-Y397 (pFAK-Y397), when compared to the sensitive line (Fig. 5 A and B).

To further support this evidence, we investigated FAK1 activation following acute BRAFi treatment in AMBRA1-silenced SK-Mel-5 cells (Fig. 5C and SI Appendix, Fig. S5A). In this setting, higher levels of pFAK-Y397 and of its downstream target pSRC-Y416 were observed (Fig. 5 C and D and SI Appendix, Fig. S5B), while AMBRA1 silencing did not affect the phosphorylation status of pERK1/2-T202/Y204 (Fig. 5C and SI Appendix, Fig. S5C), consistent with previous observations (Fig. 3 A and D). Similar significant results were obtained in BRAFi-treated syngeneic sBPA^−/− vs. sBPA^+/+ mice (Fig. 5 E and F and SI Appendix, Fig. S5D).

We next aimed to establish a direct link between AMBRA1, FAK1 activation, and therapy resistance by conducting cell viability assays. Transient transfection with a plasmid encoding a mutant form of AMBRA1 (AMBRA1^P170S) (Fig. 5 G and H) mimicking the AMBRA1-deficient FAK1-related phenotype (characterized by elevated pFAK-Y397) (28, 29), significantly increased the resistance index to BRAFi in both FM-93/2 (Fig. 5 I and J) and M24 (SI Appendix, Fig. S5 E–G) cells, compared to AMBRA1^WT-expressing cells. In a similar fashion, we silenced FAK1 (siFAK1) in FM-93/2 cells and re-expressed either the WT (FAK1^WT) or the P876A/P882A mutant (FAK1^AA) form of FAK1, which abrogates the AMBRA1/FAK1 interaction mimicking the AMBRA1-deficient-like phenotype of FAK1 activation (Fig. 5G) (25). Analogous to results described for the AMBRA1^P170S mutant, BRAFi-treated FAK1^AA-expressing cells (Fig. 5K) showcased higher resistance to BRAFi than the control (FAK^WT) cells (Fig. 5 L and M).

Moreover, FM-93/2-derived BRAFi-resistant cells displayed a reduction of the active form of the bona fide autophagy marker LC3 (LC3-II) compared to sensitive cells (Fig. 5A and SI Appendix, Fig. S5H). To evaluate the involvement of the proautophagic function of AMBRA1 (20, 21) in BRAFi resistance, we performed reconstitution experiments of a mutant form impairing AMBRA1 autophagy function (AMBRA1^LIRaa) in M24 cells followed by BRAFi treatment (SI Appendix, Fig. S5I). Our results do not support AMBRA1-mediated regulation of autophagy as an additional mechanism contributing to therapy resistance to BRAFi (SI Appendix, Fig. S5 J and K).

Taken together, these results indicate that AMBRA1-dependent activation of FAK1 signaling in melanoma is responsible for increased resistance to BRAFi.

To address the therapeutic relevance of our findings, we tested the efficacy of pharmacological inhibition of FAK1, either alone or in combination with BRAFi, for melanoma treatment. To assess this, we subjected AMBRA1^HIGH FM-93/2 and AMBRA1^LOW M24 cells to FAKi or a combination of FAKi and BRAFi (Fig. 6A) and evaluated cell viability using the ratio between FAKi treatment and its corresponding control cases (Materials and Methods) (SI Appendix, Fig. S6A). Consistent with earlier findings (25), AMBRA1^LOW M24 cells exhibited sensitivity to FAKi, further enhanced upon combined therapy with BRAFi (Fig. 6B and SI Appendix, Fig. S6A). Conversely, AMBRA1^HIGH FM-93/2 cells demonstrated unresponsiveness to FAKi, with no additional effect induced by BRAFi cotreatment (Fig. 6B and SI Appendix, Fig. S6A). To validate the AMBRA1 dependence of this response, we employed propidium iodide (PI) staining followed by FACS analyses in SK-Mel-5 silenced for AMBRA1 (siAMBRA1), obtaining results in line with AMBRA1^LOW M24 and AMBRA1^HIGH FM-93/2 cells (SI Appendix, Fig. S6 B and C).

Subsequently, we extended our investigation to Ambra1 WT and KO cells (Bdmc^+/+ and Bdmc^−/−) by employing two different FAKi (PF-56227 and defactinib) (SI Appendix, Fig. S6 D and E). In either case, the combined treatment with BRAFi elicited only a marginal additional response in Bdmc^−/− cells compared to FAKi treatment alone, mirroring the response of Bdmc^+/+ cells to BRAFi (Fig. 6C and SI Appendix, Fig. S6F). Similar outcomes were obtained in a syngeneic model of mice orally administered a combined BRAFi (30 mg/kg) and FAKi (50 mg/kg) therapy (SI Appendix, Fig. S6 G and H).

Subsequent testing of FAKi efficacy in AMBRA1^LOW FM-93/2- (Fig. 6D), M17- (SI Appendix, Fig. S6I), YUMM1.7- (SI Appendix, Fig. S6J), and YUMM1.1- (SI Appendix, Fig. S6K) derived BRAFi-resistant cells demonstrated a significantly increased response to FAKi treatment compared to respective controls, unlike the AMBRA1^HIGH sensitive cell line.

Altogether, these results underscore the potential of FAKi utilization to overcome melanoma resistance to BRAFi in tumors exhibiting low AMBRA1 expression.

Given the strong correlation between AMBRA1 expression and response to BRAFi in both clinical and experimental contexts, we speculated whether the propensity of some melanomas to develop BRAFi resistance could be attributed to intratumor heterogeneity of AMBRA1 expression. Indeed, from a first analysis by immunofluorescence, we identified a small cell population exhibiting lower AMBRA1 signal in the AMBRA1^HIGH FM-93/2 cells (SI Appendix, Fig. S7A). To further investigate this, we generated single-cell-derived subclones from the human FM-93/2 and the murine YUMM1.7 cell lines, which both express high levels of AMBRA1, and performed a retrospective analysis of AMBRA1 expression in these subclones following exposure to BRAFi. The initial assessment unveiled a highly variable response to BRAFi among the subclones of both cell lines (Fig. 6 E and F). Subsequently, we identified the top BRAFi-sensitive and -resistant subclones based on a 20% (FM-93/2) or 15% (YUMM1.7) cut-off on the BRAFi resistance index. RT-qPCR and western blot analyses revealed that subclones with the highest BRAFi resistance index exhibited genetic features of phenotype switching (SI Appendix, Fig. S7 B and C) and reduced AMBRA1 levels in both the FM-93/2 (clone 1.D12; 19% response to BRAFi; Fig. 6G) and YUMM1.7 (clones 2.B10, 1.D7, and 2.G5; 19%, 10%, and 9.5% response to BRAFi, respectively; Fig. 6H) models. This supports the hypothesis of a subclonal heterogeneous AMBRA1 expression in melanoma prior to treatment.

Furthermore, elevated levels of pFAK-Y397 were detected in the top therapy-resistant subclones (Fig. 6 G and H), showing a strong inverse correlation with AMBRA1 expression levels in FM-93/2 (r = −0.7794; P = 0.0028; Fig. 6I) and YUMM1.7 (r = −0.5137; P = 0.0086; Fig. 6J) subclones. Phosphorylation of Src (pSrc-Y416) was also detected in the least BRAFi-sensitive YUMM1.7-derived subclones (Fig. 6H and SI Appendix, Fig. S7D), confirming full activation of the FAK1 signaling pathway. Pearson’s correlation analysis confirmed the interdependency among response to BRAFi, Ambra1 expression, and pFak-Y397 and pSrc-Y416 levels (Fig. 6K).

To establish a direct link between AMBRA1 expression and FAK1 activation in the top BRAFi-resistant subclones, we reconstituted the FM-93/2 AMBRA1^LOW/pFAK-Y397^HIGH 1.D12 subclone with either WT AMBRA1 (myc-AMBRA1^WT) or the single-point mutant P170S (myc-AMBRA1^P170S) (Fig. 7A). Expression of AMBRA1^WT induced a complete rescue of FAK1 signaling, whereas the AMBRA1^P170S-expressing cells exhibited phospho-activation levels of FAK1 and SRC similar to control cells (Fig. 7 A and B and SI Appendix, Fig. S7E), indicating AMBRA1-dependent activation of FAK1 signaling in BRAFi-resistant subclones. Moreover, analysis of cell viability showed that ectopic expression of AMBRA1^P170S restored cellular resistance to BRAFi (Fig. 7C and SI Appendix, Fig. S7F).

Consistent with these results, the FM-93/2-derived AMBRA1^LOW/ pFAK-Y397^HIGH 1.D12 subclone (SI Appendix, Fig. S7G) displayed high sensitivity to FAKi alone, which was further enhanced in cotreatment with BRAFi (Fig. 7 D and E). Conversely, the AMBRA1^HIGH/ pFAK-Y397^LOW 2.A10 subclone (SI Appendix, Fig. S7G) exhibited limited response to FAK1 inhibition, whether used as a single treatment or in combination with BRAFi (Fig. 7 D and E). Cells from the 1.D12 subclone re-expressing AMBRA1^WT also corroborated the AMBRA1 dependence of such response (Fig. 7F), whereas AMBRA1^P170S-expressing cells exhibited either a higher response to FAKi or a response to combined therapy comparable to control cells (β-Gal) (Fig. 7F).

These findings strongly suggest that the expression levels of AMBRA1 in the tumor are potentially predictive of whether treatment with FAKi should be administered as mono- or combined therapy to overcome BRAFi resistance in melanoma. To validate this hypothesis, we performed colony formation assays in AMBRA1^LOW M24 and AMBRA1^HIGH FM-93/2 cells upon FAKi treatment as monotherapy or in combination with BRAFi. FAKi monotherapy effectively prevented colony formation in AMBRA1^LOW M24 cells, with combined therapy with BRAFi showing no additional effects (Fig. 7G). Conversely, FAKi monotherapy exerted only limited effects on colony formation in the AMBRA1^HIGH FM-93/2 cells (Fig. 7G). In this cell line, resistant colonies to BRAFi eventually emerged upon prolonged treatment with BRAFi, which ability to grow was overcome upon combined treatment with FAKi (Fig. 7G).

Overall, these data strongly suggest that AMBRA1 expression levels in the tumor serve as a predictor for FAKi mono- or BRAFi-combined therapy to overcome therapy resistance in melanoma.

The human BRAF^V600E-mutated FM-93/2 [ESTDAB (European Searchable Tumor Line Database)-033], M17 (ESTDAB-039), M88 (ESTDAB-135), Ma-Mel-51 (ESTDAB-196), FM-55/M2 (ESTDAB-013), M24 (ESTDAB-043), Mel-5392 (ESTDAB-114), and OCM-3 (ESTDAB-129) cell lines were already in house and cultivated in RPMI 1640 Medium (ThermoFisher Scientific; cat#61870-010). Apart from Ma-Mel-51 and FM-55/M2, all cell lines were characterized in a previous study for expression levels of AMBRA1 (25). BRAF mutational status was previously determined (43). The human melanoma cell line SK-Mel-5 from ATCC® (Manassas) was cultivated in Advanced Minimum Essential Media (MEM) (ThermoFisher Scientific; cat#12492-021) and BRAF mutational status determined using the Cellosaurus Database. Immortalized murine melanoma cells YUMM1.7 and YUMM1.1 were previously described (44) and cultivated in DMEM (ThermoFisher Scientific; cat#31966-021). Primary Bdmc cells were generated from either BPA^+/+ (Bdmc^+/+) or BPA^−/− (Bdmc^−/−) mice, as previously described (25) (SI Appendix, Fig. S3E) and cultivated in RPMI. Cell culture media were supplemented with 2 mM GlutaMAX™ (ThermoFisher Scientific; cat#35050-038), 100 U/mL P/S (Penicillin-Streptomycin, ThermoFisher Scientific; cat#15140-122), 1% MEM nonessential amino acids (YUMM1.1; ThermoFisher Scientific; cat#11140-050) and 7% (YUMM1.7), 20% (Bdmc), or 10% (all remaining cells) FBS (ThermoFisher Scientific; cat#10270-106) and cultivated at 37 °C in a 5% CO₂ atmosphere. During the experiments, cells were counted using Dual-Chamber Cell Counting Slides (Bio-Rad Laboratories; cat#1450011) on a TC10™ Automated Cell Counter (Bio-Rad Laboratories) and plated at a density of 1 × 10⁵ cells/mL, unless otherwise indicated.

The reverse method was employed for siRNA transfection. Silencing was performed for 96 to 120 h, unless otherwise indicated. Sequences and final concentrations for custom-designed siRNAs are in SI Appendix, Table S1. DNA constructs for myc-flagged-AMBRA1 (myc-AMBRA1^WT/^P170S) (29), AMBRA1 (AMBRA1^WT/^LIRaa) (45), and HA-tagged-FAK1 (FAK1^WT/^AA) (25) were originated as described in the relevant references. Plasmids encoding for FAK1^WT/^AA were modified to specifically introduce silent mutations within their sequence targeted by siFAK1 and restrict the silencing effect on endogenous FAK1 only. Myc-ß-Gal-, ß-Gal, and HA-expressing plasmids were used as negative controls. All transfections were performed using Lipofectamine™ 2000 Transfection Reagent (ThermoFisher Scientific; cat#11668-019), according to the instructions.

The BRAFi vemurafenib (Selleckchem, cat#S1267) and dabrafenib (Selleckchem, cat#S2807) were respectively used for human and murine cell lines. The MEKi trametinib (Selleckchem, cat#S2673) was used for both human and murine cell lines. The FAKi defactinib (Selleckchem, cat#S7654) was used for both human and murine cell lines, while PF-562271 (Selleckchem, cat#S2890) only for murine cells. All drugs were dissolved in dimethyl sulfoxide (DMSO), used as vehicle in the relevant control cases. For single-dose treatments, timing and dosage were chosen either following dose-dependent experiments or to obtain a treatment efficacy ≥50% in the desired treatment group.

Resistant cell lines to BRAFi were generated from pools of cells to maintain heterogeneity. 1 × 10⁶ cells were separately plated and each resistant cell line (R + number, e.g., R1, R2) independently originated from its parental after exposure to 1 µM BRAFi in growth medium (conditioned medium). A control cell line (S) was kept during the selection process and treated with DMSO. Cells were defined resistant when they started to growth exponentially. Resistant cells are cultivated in conditioned medium, unless otherwise indicated.

Human FM-93/2 and murine YUMM1.7 cell lines were washed in PBS and detached with trypsin. Cells were centrifuged at 800 × g for 5 min, resuspended in growth medium, filtered using a 40 µm Falcon® Cell Strainer (Corning; cat#431750), diluted to a concentration of 1,000 cells/mL, single-cell sorted using a BD FACSMelody™ Cell Sorter (BD Biosciences) machine and plated in 96-well plates. 60% of clones survived and cells were allowed to fully recover before any treatment or further analysis.

5,000 cells were plated in triplicates in 96-well plates and treated after 24 h. Percentage of cell viability and EC₅₀ values were determined by the Cell Counting Kit-8 (Dojindo; cat#CK04-11) as previously described (25). Resistance indexes to BRAFi and MEKi (MAPKi) were calculated as the ratio between percentage of survival of MAPKi-treated and -untreated cells. Response to FAKi in cotreatment experiments was calculated as the reverse ratio between a FAKi treatment case and its relevant control, as follows: [(BRAFi+FAKi)/BRAFi]⁻¹ and (FAKi/Vehicle)⁻¹.

First, 16 cells/cm² were plated in 6-well plates and treated after 24 h with BRAFi (vemurafenib 250 nM) and/or FAKi (defactinib 500 nM) for 3 wk. Then, cells were fixed and stained with 0.05% (w/v) Crystal Violet in 20% MeOH. Plates were imaged and colonies counted manually. The percentage of surviving colonies was calculated as the ratio between the number of colonies in the treated groups vs. the control case.

1 × 10⁴ M17 and YUMM1.7 (S, R1, and R2) cells were seeded in Nunc™ Polycarbonate Cell Culture Inserts (ThermoFisher Scientific; cat#140629) as previously described (25). Cells were cultured for 24 h and nuclei of cells migrated to the bottom counterstained with 1 μg/mL Hoechst 33342 (ThermoFisher Scientific; cat#H3570) for 10 min at RT. Images were captured with a Celigo Image Cytometer (Nexcelom Bioscience). Twenty-five separate fields were acquired, and nuclei counted using the Celigo Image Cytometer Analysis Software. Fluorescence images were adjusted for brightness, contrast, and color balance using Fiji analysis software.

A total of 3 × 10⁵ FM-93/2 S, R1, and R2 cells were seeded in duplicates in Ibidi 8-wells (Ibidi 80826, ibiTreat), incubated for 24 h, fixated in −20 °C MeOH for 10 min, washed in PBS, blocked in 1% bovine serum albumine (BSA) in PBS for 30 min and incubated with anti-AMBRA1 primary antibody (1:100) in a humidity chamber at 4 °C O/N. Secondary antibody was added 1:1,000 in 1% BSA and incubated in a dark box at RT for 1 h. Nuclei were counterstained with DAPI (Invitrogen™, cat#R37606). Samples were imaged with a Nikon Eclipse Ti2 Widefield using a 20× air objective and images analyzed using Nikon’s analysis software. Analysis was made based on maximum intensity projections of Z-stacks. The mean object intensity from the maximum projection of AMBRA1 signal was then measured in each segmentation and visualized in a plot. Further details are provided in SI Appendix.

At the experimental endpoint, both adherent and in-suspension cells were collected and centrifuged at 2,000 × g for 5 min at RT, followed by a wash with PBS. Cell pellets were resuspended in PI staining solution (50 µg/mL PI, 0.1% sodium citrate, 0.1% Triton X-100, 200 µg/mL RNAse) for 30 min at 4 °C in the dark. Fluorescence intensity was analyzed in logarithmic scale using a BD FACSVerse™ Cell Analyzer (BD Biosciences) machine. Raw data were analyzed with the FlowJo v.10.6.1 software and dead cells expressed as fold change of sub-G1 cells vs. control case.

Subcutaneous injection of either Bdmc^+/+ (sBPA^+/+ mice) or Bdmc^−/− (sBPA^−/− mice) cells (2 × 10⁶/mouse) was performed in female C57Bl/6 N mice (Taconic Biosciences A/S; 10 to 12 wk of age) as previously described (25). When tumors were measurable, mice were randomly divided into subgroups and treated. Drugs were dissolved in DMSO, freshly diluted in Hydroxypropyl methylcellulose and 0.2% Tween^®80 (HPMT) at a final dosage of 30 mg/kg dabrafenib and 50 mg/kg PF-562271 and P.O. administered for 21 d, including (Fig. 3 J–L) or excluding (SI Appendix, Fig. S6H) weekends. Control mice were administered a HPMT+DMSO solution. Tumors were measured using a digital caliper in width and length and volumes (V) determined using the formula V = (a × b²)/2, where a > b. Alternatively, tumor weights were measured when the mice were killed.

Cells were washed in PBS, collected, and centrifuged at 800 × g for 5 min at RT. Cell lines and tumor samples were processed and total protein lysates obtained as previously described (25). For detection of pFAK-Y397, cells were directly disrupted in cell plates after plates were washed and fully dried of PBS. Samples were separated by sodium dodecyl-sulfate polyacrylamide gel electrophoresis (SDS-PAGE) using Criterion™ TGX Stain-Free™ Precast Gels (Bio-Rad Laboratories; cat#5678084 and cat#5678085) and blotted onto 0.2 µm polyvinylidene difluoride (PVDF) membranes (Bio-Rad Laboratories; cat#1704157) using a Trans-Blot® Turbo™ Transfer System (Bio-Rad Laboratories). Primary antibodies used are listed in SI Appendix, Table S2. PVDF membranes were incubated with the ECL™ Prime Western Blotting Detection Reagent (Amersham; cat#RPN2236) prior to detection with a ChemiDoc™ MP System (Bio-Rad Laboratories) provided with the Image Lab 6.0.1 Software (Bio-Rad Laboratories). Proteins on gels were visualized and an image captured before gels were blotted (Stain-Free activation). Densitometry analyses were performed using ImageJ version 1.52.q.

Total RNA isolation and RT from both cell and tumor samples were performed as previously described (25). Expression levels of the mRNAs of interest were measured using the PowerUp™ SYBR™ Green Master Mix (ThermoFisher Scientific; cat#A25742) on a ViiA 7 Real-Time PCR System v1.3 (Applied Biosystems). Reactions were run as triplicates and data normalized on the internal housekeeping L34. Custom-designed primer pairs are listed in SI Appendix, Table S3 and were tested through Primer-BLAST prior to purchase (TAG Copenhagen A/S).

Arbitrary cut-offs were applied to define AMBRA1^HIGH/^LOW subgroups in The Cancer Genome Atlas skin cutaneous melanoma (TCGA-SKCM) (mRNA) and the CCLE (protein) as previously described (25). Sample values for AMBRA1 in the GSE50509 (ID = ILMN_1662681) and in the GSE65185 (Gene ID: 55626) cohorts were downloaded to subgroup pretreatment samples in AMBRA1^HIGH/^LOW and the Log₂ of the AMBRA1 expression ratio between MAPKi-treated and matching pretreatment tumors graphed as waterfall plot. For PDXs, AMBRA1 expression was analyzed in a subset of matched PDXs before (PRE) and after (PROG) MAPKi treatment from GSE129127 (32). For single-cell quantification of AMBRA1 positive cells in melanoma PDX during different phases upon MAPKi treatment (MEL006), data were downloaded from GSE116237 (33) and analyzed as previously described (46). All GSEAs were performed as previously described (25) on the Melanocytic, Undifferentiated, NCSC-like by Tsoi et al. (37) and Invasive by Hoek et al. (38) gene signatures. For the correlation analysis between AMBRA1 and NGFR, AXL, and MITF expression, RSEM normalized mRNA data were downloaded from TCGA-SKCM samples (n = 448) through R studio using “TCGAbiolink” package (47). Pearson and Spearman correlations were calculated using Prism 9. Further details are provided in SI Appendix.

To evaluate the differential cytotoxic action of a panel of BRAFi and MEKi on melanoma cell lines, the primary PRISM database (48) was utilized. The list of BRAFis and MEKis was retrieved from the Drug Repurposing Hub (49). Melanoma cell lines were classified based on their BRAF mutation status using ExPASy Bioinformatics Resource Portal. To test the link between the drug scores and the protein levels of AMBRA1, protein expression of screened cell lines was retrieved from CCLE (50). The individual drug scores on each cell line were visualized in a heatmap with cell lines sorted by AMBRA1 protein expression. To determine the association between the genes of interest and drug scores, a linear model was fitted between the median drug scores in a cell line and the protein expression of this cell line. Further details are provided in SI Appendix.

Statistical analyses were performed with GraphPad Prism9. Significance was designated as follows: *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; ns = not significant. Data are shown as average ±SEM or SD, as indicated in the relevant figure legends. Further details are provided in SI Appendix.

L.D.L. and D.D.Z. designed research; L.D.L., C. Pagliuca, A.K., S.R., C.T., C. Pecorari, R.T., and P.B. performed research; C.D., M.P.P., P.B., E.H., F.C., M.F.G., P.L., P.G., and D.D.Z. contributed new reagents/analytic tools; L.D.L., A.K., S.R., C.T., C. Pecorari, R.T., J.R.B., P.B., R.B., C.B., T.S., and D.D.Z. analyzed data; and L.D.L., G.F., P.G., and D.D.Z. wrote the paper.

P.L. is Chief Scientific Officer for AMLo Biosciences Ltd.