Metastatic cells are preferentially vulnerable to lysosomal inhibition

Significance

We show that there is a functional reciprocal relationship between lysosome activity and metastasis that allows chloroquine (CQ) and other inhibitors of lysosome function, such as bafilomycin A₁, to preferentially kill human metastatic bladder cancer cells by targeting autophagy-independent lysosome functions. In addition, CQ treatment of bladder cancer cells and subsequent acquisition of resistance to this therapy lead to altered gene expression programs that drive a less aggressive and metastatic phenotype via up-regulation of ID4 (inhibitor of DNA binding 4). Clinically, this work provides a conceptual foundation for using ID4 expression as a predictive and prognostic biomarker of CQ sensitivity and metastasis in patients with bladder cancer.

Abstract

Molecular alterations that confer phenotypic advantages to tumors can also expose specific therapeutic vulnerabilities. To search for potential treatments that would selectively affect metastatic cells, we examined the sensitivity of lineage-related human bladder cancer cell lines with different lung colonization abilities to chloroquine (CQ) or bafilomycin A₁, which are inhibitors of lysosome function and autophagy. Both CQ and bafilomycin A₁ were more cytotoxic in vitro to highly metastatic cells compared with their less metastatic counterparts. Genetic inactivation of macroautophagy regulators and lysosomal proteins indicated that this was due to greater reliance on the lysosome but not upon macroautophagy. To identify the mechanism underlying these effects, we generated cells resistant to CQ in vitro. Surprisingly, selection for in vitro CQ resistance was sufficient to alter gene expression patterns such that unsupervised cluster analysis of whole-transcriptome data indicated that selection for CQ resistance alone created tumor cells that were more similar to the poorly metastatic parental cells from which the metastatic cells were derived; importantly, these tumor cells also had diminished metastatic ability in vivo. These effects were mediated in part by differential expression of the transcriptional regulator ID4 (inhibitor of DNA binding 4); depletion of ID4 both promoted in vitro CQ sensitivity and restored lung colonization and metastasis of CQ-resistant cells. These data demonstrate that selection for metastasis ability confers selective vulnerability to lysosomal inhibitors and identify ID4 as a potential biomarker for the use of lysosomal inhibitors to reduce metastasis in patients.

Metastatic disease is the primary cause of death in cancer (1). Metastatic cancer cells colonize and survive environmental stresses at distant sites because of molecular changes acquired during tumor evolution (1–3). For example, metastatic cells are more resistant to cell death and have increased prosurvival signaling (4). Autophagy is a prosurvival process that facilitates cancer cell survival under metabolic and environmental stresses and has been implicated in multiple cellular functions important for metastasis, suggesting that autophagy inhibition might be a useful strategy for treating metastatic disease (5, 6).

Two inhibitors of autophagy, chloroquine (CQ), a well-tolerated Food and Drug Administration-approved antimalarial drug (7), and its derivative, hydroxychloroquine (HCQ) (8), have been evaluated in patients with advanced cancer and are the only autophagy inhibitors currently available for use in people (9, 10). Chloroquine is a weak base with hydrophobic characteristics that diffuses into lysosomes, where it becomes protonated and trapped, thus leading to a rise in lysosomal pH. These lysosomes can no longer fuse with autophagosomes, thus blocking autophagy (11). Other, more potent lysosomal inhibitors (12, 13) are also under development. Initial phase I trials suggest that CQ/HCQ therapy can block autophagy in human tumors and may have clinical benefit (14–19). However, since CQ and HCQ can have autophagy-independent antitumor effects (7, 20, 21), it is unclear whether their clinical benefit is primarily due to their ability to inhibit autophagy, more general lysosomal function, or lysosome-independent effects. For example, Piao et al. (22) reported that two genes, LDH1A1 and HLTF, can regulate HCQ sensitivity and resistance, respectively, without directly affecting either lysosome function or autophagy.

To evaluate the importance of autophagy in driving colonization of the lung (3), we used a well-established model of lineage-related human bladder cancer cell lines with differing metastatic propensity (23, 24). We found that cells with greater metastatic propensity displayed more in vitro cytotoxicity to two mechanistically different lysosomal inhibitors, CQ and bafilomycin A₁ (BafA₁), and to knockdown of the lysosomal protein LAMP2. However, the metastatic and nonmetastatic cells were equally sensitive to depletion of different autophagy regulators. Together, these data suggest that metastatic cells are more dependent on lysosomal function but not autophagy itself. To test this hypothesis, we generated CQ-resistant metastatic cells in vitro and found that these cells overexpress the ID4 (inhibitor of DNA binding 4, a dominant-negative helix–loop–helix protein) gene. Conversely, depletion of ID4 in poorly metastatic cells was sufficient to promote resistance to CQ and BafA₁. Moreover, CQ-resistant cells have reduced metastatic ability, and this can be largely restored by ID4 depletion. These data suggest that there is an intimate reciprocal relationship between metastatic ability and lysosome function and that targeting the lysosome itself, rather than autophagy specifically, may provide an effective therapy in patients with metastasis.

Results

Metastatic Bladder Cancer Cells Are More Sensitive to Chloroquine and Bafilomycin.

Since autophagy helps cells survive environmental stress, we reasoned that metastatic cells might rely on autophagy for growth and survival at distant sites. To test this, we used two well-established lineage-related human bladder cancer cell line models derived from T24t cells (25, 26). The FL and SLT series were derived by serial passaging of T24t to the lung and the liver, respectively, to generate metastatic cell lines with progressively increased metastatic potential (SI Appendix, Fig. S1) (23, 24).

We treated FL3 and T24t cells with chloroquine or bafilomycin A₁, which inhibit late stages of autophagy by preventing lysosome acidification. We then assessed loss of cell viability by propidium iodide (PI) permeability over 48 h using INCUCYTE imaging. The more metastatic FL3 cells had substantially decreased cell viability in response to these inhibitors compared with the T24t parental cells (Fig. 1A and Movies S1–S3). Additional viability and cell-death assays [MTS tetrazolium compound reduction and lactate dehydrogenase (LDH) release assays, respectively] confirmed that FL3 cells were more sensitive to the lysosomal inhibitors than T24t (Fig. 1 B–E). We then tested the entire metastatic series and found each independently derived, more metastatic cell line was more sensitive to these inhibitors than the poorly metastatic T24t cells from which they were derived (SI Appendix, Fig. S2). Thus, in these closely matched cells, in vitro cytotoxicity to CQ and BafA₁ correlates with increased metastatic potential.

Fig. 1.

Differences in chloroquine and bafilomycin A₁ toxicity in low metastatic T24t and its high metastatic FL3 derivative. (A) Phase-contrast images overlapped with propidium iodide fluorescence (red), with PI uptake as a measure of loss of viability. T24t and FL3 cells are shown after treatment with or without 30 μM CQ or 5 nM BafA₁ for 48 h (photos represent the last frames in Movies S1–S3). (B–E) T24t and FL3 cells were treated with the indicated concentrations of CQ (B and C) or BafA₁ (D and E) for 48 h; cell viability was analyzed by MTS (B and D) and cell death was analyzed by LDH release (C and E). n = 10 for CQ experiments and n = 7 for BafA₁ experiments. (F and G) mCherry-GFP-LC3 T24t and FL3 cells (as shown in SI Appendix, Fig. S2B) were plated in a six-well plate. The next day, cells were treated with 5 nM BafA₁ in the presence or absence of starvation media (EBSS). Cells were washed and grown in full media for 6 d, and then stained with crystal violet (F) and quantitated (G). n = 3. [All bars indicate mean ± SEM; *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001; n.s., nonsignificant (P > 0.2).]

Autophagic Flux Does Not Correlate with Sensitivity to CQ or BafA₁.

We next compared autophagic flux in the various cell lines (27). Sensitivity to lysosomal inhibitors was not correlated with the amount of autophagic flux as measured by LC3 Western assays and tandem-mCherry-EGFP-LC3 flux measurements (SI Appendix, Fig. S3). To test if the lysosomal inhibitors were capable of inhibiting a well-established autophagy-specific function, T24t and FL3 cells were put under starvation stress in Earle’s balanced salt solution (EBSS). BafA₁ sensitized T24t and FL3 cells equally to cell death. In contrast, more cytotoxicity was observed in FL3 compared with T24t cells in normal growth medium conditions (Fig. 1 F and G). This suggests that both metastatic and poorly metastatic cells have a similar requirement for autophagy and harbor functional lysosomes under stress conditions but that lysosomal inhibition selectively causes increased cytotoxicity in the metastatic cells under nutrient-replete conditions.

Lysosomal, Rather than Autophagy, Dependency Is Correlated with Metastatic Potential.

We next tested whether differential sensitivity to lysosome inhibitors reflected an enhanced requirement for macroautophagy in metastatic cells. Using an inducible dominant-negative ATG4 mutant construct (ATG4DN) (28), we eliminated production of LC3II under basal conditions or in the presence of CQ (SI Appendix, Fig. S4A) in nutrient-replete or starvation conditions (SI Appendix, Fig. S4 B and C). However, this had no effect on survival and clonogenic growth of either T24t or FL3 cells (SI Appendix, Fig. S4D). To test the requirement for autophagy, we used shRNAs that target VPS34 (PIK3C3), ATG5, and ATG7, which are all required for autophagosome formation, as well as LAMP2, which is a lysosomal protein needed for autophagosome–lysosome fusion (Fig. 2A). Knockdown of VPS34, ATG5, or ATG7 in T24t and FL3 cells led to a reduction in cell number and increased cell death, but the magnitude of these effects was similar across the lines (Fig. 2 B–G and SI Appendix, Fig. S5A). In contrast, knockdown of LAMP2 (with pan-LAMP2 shRNA) demonstrated preferential cytotoxicity in FL3 cells as measured by both MTS and LDH assays (Fig. 3 A and B). Importantly, a LAMP2 isoform A-specific shRNA had no effect on viability of either cell type. Since isoform A of LAMP2, but not other splice variants, is necessary for chaperone-mediated autophagy (CMA), this indicates CMA is not responsible for these effects (Fig. 3 A and B).

Fig. 2.

T24t and FL3 cells do not differ in their response to inhibition of autophagy with ATG5, ATG7, or VPS34 knockdown. (A) Western blotting showing knockdown of ATG5, ATG7, VPS34, or LAMP2 in T24t or FL3 cells. (B–G) T24t or FL3 cells in 48-well plates were infected with two different concentrations of lentivirus (200- or 400-μL aliquots) encoding nonsilencing (NS), ATG5, ATG7, or VPS34 shRNAs and allowed to grow for 6 d; cell viability was analyzed by MTS (B, D, and F) and cell death was analyzed by LDH release (C, E, and G). n = 4 for all experiments. [All bars indicate mean ± SEM; n.s., nonsignificant (P > 0.2).] ACTB, actin B protein.

Fig. 3.

LAMP2 knockdown leads to differential cytotoxicity in T24t and FL3, while knockdown of the LAMP2A isoform involved in chaperone-mediated autophagy has no cytotoxicity in either cell line. (A and B) T24t or FL3 cells (1,000) in 48-well plates were infected with two different concentrations of lentivirus (200- or 400-μL aliquots) encoding nonsilencing pan-LAMP2 or LAMP2A shRNAs and allowed to grow for 6 d; cell viability was analyzed by MTS (A) and cell death was analyzed by LDH release (B). n = 5. (C) T24t and FL3 GFP-NLS cells in 48-well plates were infected with lentivirus encoding nonsilencing ATG5, ATG7, VPS34, or LAMP2 (pan-variant) shRNAs and followed by INCUCYTE imaging of phase, green, and red channels in the presence of propidium iodide. Cell death of the knockdown T24t and FL3 GFP-NLS cells in the above INCUCYTE experiments is shown as measured at the 140-h time point by PI permeability. Data are shown as the number of PI-positive cells per the total number of green cells. Data for each bar represent four technical replicates. [All bars indicate mean ± SEM; *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001; n.s., nonsignificant (P > 0.15).]

To confirm these results, we monitored long-term cell viability of individual cells in a population of GFP-NLS–tagged T24t and FL3 cells using propidium iodide staining during INCUCYTE imaging. Consistent with the MTS assays, all of the autophagy-targeted shRNAs reduced proliferation of both T24t and FL3 cells. ATG5 or ATG7 shRNAs caused minimal cytotoxicity in either cell line and, while VPS34 shRNA caused more toxicity, this was equivalent in both T24t and FL3 cells (Fig. 3C). In contrast, LAMP2 depletion led to preferential reduction in cell viability of metastatic cells compared with the poorly metastatic cells (Fig. 3C). Together, these data indicate that although the metastatic and poorly metastatic cells are equally sensitive to macroautophagy inhibition per se, metastasis propensity is associated with more sensitivity to blockade of lysosomal function.

ID4 Expression Is Associated with CQ Resistance in Bladder Cancer Cells.

To understand the molecular basis of the dependence of metastatic cells on lysosome function, we derived CQ-resistant versions of the metastatic cells by growing FL3 cells in progressively higher concentrations and exposure times to CQ over several months (Fig. 4A). Three CQ-resistant FL3 derivatives, C1AZ, D1AZ, and D1BZ1, were developed. All three lines were also resistant to BafA₁ (Fig. 4 B–E) compared with the FL3 cells, and their level of resistance was similar to that of the parental, poorly metastatic T24t cells from which FL3 cells were derived. However, the CQ-resistant derivatives were not generally resistant to other cell-death inducers such as etoposide (Fig. 4 F and G) or cell detachment (SI Appendix, Fig. S5B).

Fig. 4.

Generation and characterization of CQ-resistant cells. (A) FL3 cells were selected for CQ resistance by growing for several months in varying doses of CQ and varying lengths of times as illustrated. Independently derived CQ-resistant polyclonal cell lines were designated C1AZ, D1AZ, and D1BZ1. (B–F) T24t, FL3, and the FL3-derived CQ-resistant cell lines C1AZ, D1AZ, and D1BZ1 were treated with the indicated concentrations of CQ (B and C) for 48 h, BafA₁ (D and E) for 72 h, or etoposide (F and G) for 72 h. Cell viability was analyzed by MTS (B, D, and F) and cell death was analyzed by LDH release (C, E, and G). (All bars indicate mean ± SEM; *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, FL3 versus T24t, C1AZ, D1AZ, and D1BZ1; ^#P ≤ 0.05, FL3 vs. C1AZ, D1AZ, and D1BZ1 only.)

Using microarrays, we examined differentially expressed genes in the CQ-resistant lines relative to FL3 cells. A Venn diagram and accompanying gene lists indicate significant overlap across cell lines in genes whose expression is higher in the derivatives compared with that in FL3 cells (SI Appendix, Fig. S6). Genes whose expression is lower than in FL3 cells are shown in SI Appendix, Fig. S7. The top 10 statistically significant up-regulated and down-regulated genes in the CQ-resistant cells ranked in terms of high fold changes across multiple cell lines are shown in SI Appendix, Tables S1 and S2, respectively. Interestingly, the expression of LDH1A1 and HLTF, reported to mediate HCQ sensitivity and resistance in some other cancer cell lines (22), was not significantly different across our cell lines that were all derived from the same parental line but selected for differences in metastasis and then separately selected for resistance to CQ. Next, we examined the correlation between the expression levels of the top 20 genes with CQ and BafA₁ sensitivity using the NCI-60 cancer cell line panel (https://discover.nci.nih.gov/cellminer) (29, 30) (SI Appendix, Fig. S8 and Tables S1 and S2 and Datasets S1 and S2). Among our top candidate genes, we found that ID4 was highly expressed in all CQ-resistant cells compared with the CQ-sensitive FL3 cells (a 4- to 15-fold increase), and was also reduced in the metastatic derivative cells compared with T24t (SI Appendix, Fig. S9). ID4 expression was also significantly negatively correlated with sensitivity to both CQ and BafA₁ in the NCI-60 cancer cell line panel (Pearson’s correlation coefficient −0.48 for CQ and −0.40 for BafA₁; P < 0.05; population size 47/60 for CQ and 60/60 for BafA₁). Because no other candidate genes were significantly correlated with both CQ and BafA₁ sensitivity, we focused on ID4 for further study. Finally, it is notable that since there are no bladder cancer cell lines in the NCI-60 panel, this finding suggests that ID4 expression is associated with CQ resistance across cancer types.

ID4 Expression Promotes Resistance to CQ and BafA₁.

To determine if ID4 expression regulates CQ sensitivity, we depleted ID4 in three cell lines that were less sensitive to CQ/BafA₁: the parental T24t cells and the CQ-resistant derived C1AZ and D1BZ1 (SI Appendix, Fig. S10A). Reduction in ID4 expression increased the cytotoxicity of CQ and BafA₁ in all three lines (Fig. 5 A–F and SI Appendix, Fig. S10 B–G) but had no direct effect on the autophagy/lysosome pathway, as we observed no change in basal or starvation-induced autophagy in the T24t cells with ID4 knockdown (SI Appendix, Fig. S11). This suggests that ID4 does not directly control lysosomal processes but is involved in repressing the increased need of metastatic cells for lysosomal function.

Fig. 5.

Decreased ID4 expression leads to sensitivity to lysosomal inhibitors CQ and BafA₁, while increased ID4 expression leads to less aggressive bladder cancer. (A–F) T24t- and FL3-derived CQ-resistant cell lines D1BZ1 and C1AZ were infected with lentivirus encoding ID4 or nonsilencing shRNA. Cells were then plated in 48-well plates and treated with the indicated concentrations of CQ (A, C, and E) for 48 h or BafA₁ (B, D, and F) for 72 h and cell death was analyzed by LDH release. n = 4 for CQ and BafA₁ experiments. (All bars indicate mean ± SEM; *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001.) (G) Kaplan–Meier survival curve for TCGA BLCA patients. Overall survival in days is shown for patients with a copy-number gain of ID4 (red) or patients diploid for ID4 (black). (H) RSEM normalized ID4 mRNA expression values from TCGA BLCA patients (muscle invasive) subset into groups according to both copy number and expression. The ID4 loss group was defined as a GISTIC score less than 0 and ID4 mRNA expression in the lower quantile (<25%) of all ID4 expression values. The ID4 wild-type group was defined as a GISTIC score of 0 and ID4 mRNA expression between the middle quantiles (25 to 75%) of expression values. The ID4 gain group was defined as a GISTIC score greater than 0 and ID4 mRNA expression in the upper quantile (>75%) of all ID4 expression values. (I) Kaplan–Meier survival curve for TCGA BLCA patients (muscle invasive) with both a copy-number gain of ID4 and high expression of ID4 (red) or patients diploid for ID4 with wild-type ID4 expression levels (black).

ID4 Amplification/Expression Is Associated with Better Prognosis in Bladder Cancer Patients.

We next examined data from The Cancer Genome Atlas (TCGA) to determine which tumor characteristics determine ID4 expression in human cancers. We found several types of cancer had homozygous or heterozygous deletion of ID4 while in others it was amplified (SI Appendix, Fig. S12 A and B). Interestingly, amplification of ID4 in bladder urothelial carcinoma (BLCA) tumors occurred at a rate of 39.7% (162/408 tumors), while the average copy-number gain for another gene on chromosome 6 was only 16.2% (a significance of P = 0.03; SI Appendix, Fig. S12C). Amplification of ID4 was associated with improved patient survival in TCGA database (Fig. 5G; P = 0.012), and this was also true in ovarian cancers and uveal melanomas in TCGA datasets (SI Appendix, Fig. S12 D and E). ID4 mRNA correlated with ID4 gain or loss in bladder cancer (Fig. 5H), and when stratifying patients by both ID4 gain and mRNA expression, high expression was associated with better survival (P = 0.033; Fig. 5I). High ID4 expression (as measured by RNA sequencing; RNA-seq) was also associated with better survival of bladder cancer patients in two additional patient cohorts (SI Appendix, Fig. S13). Together, the cell line and human clinical studies presented here strongly suggest that more aggressive cancers have reduced expression of ID4 (SI Appendix, Fig. S8C). These data also led us to hypothesize that reduced ID4 expression confers both metastatic competence and increased sensitivity to CQ, BafA₁, and lysosomal inhibitors.

Selection for CQ Resistance Leads to Reversal of Gene Expression Changes and a Reduction in Metastatic Potential.

ID4 expression is low in metastatic cells, but not in the poorly metastatic T24t or in the CQ-resistant C1AZ, D1AZ, and D1BZ1 cell lines (SI Appendix, Fig. S9). To determine if this was part of more extensive overall changes in gene expression patterns related to CQ sensitivity, we evaluated the transcriptome of all of the cells. Analysis of these data indicated that acquisition of CQ resistance by metastatic cells was associated with an overall change in gene expression pattern so that they were closer to the original poorly metastatic T24t cells from which FL3 was derived by in vivo selection for metastatic ability, rather than to the FL3 cells themselves from which they were directly derived (Fig. 6A). Principal component analysis further confirmed this relationship (SI Appendix, Fig. S14A). This suggests that there is a causal reciprocal relationship between CQ sensitivity and metastatic propensity.

Fig. 6.

Selection for CQ resistance leads to a reversal of the metastatic phenotype. (A) Dendrogram analysis of cell lines hierarchically clustered by expression of all microarray probes. (B) Nine weeks after tail vein injection of equal amounts of the FL3 cell line or its CQ-resistant derivatives D1AZ, C1AZ, or D1BZ1, mouse lungs were harvested and genomic DNA was isolated. The graph shows the amount of metastatic human tumor burden in the lung as assessed by qPCR with a human-specific primer. (C) As a measure of cell line aggressiveness, the graph shows the overall survival of mice after tail vein injection of equal amounts of the FL3 cell line or its CQ-resistant derivatives D1AZ, C1AZ, or D1BZ1. (^§P = 0.12, *P = 0.029.)

To test this hypothesis, we injected FL3 cells, or the CQ-resistant FL3 derivatives D1AZ, C1AZ, and D1BZ1, into the tail veins of female athymic NCr nu/nu mice and examined metastatic colonization of the lungs over 2 mo. Human tumor burden in the lung was assessed by quantitative real-time PCR with a human-specific 12p primer set. This revealed that all three CQ-resistant cell lines were less metastatic than the parental FL3 cell line (Fig. 6B). Moreover, the D1AZ and D1BZ1 lines were associated with better overall survival [death end points, defined by our Institutional Animal Care and Use Committee (IACUC); P = 0.029; Fig. 6C], while the C1AZ line had a strong trend toward better overall survival (P = 0.12; Fig. 6C). In addition, on termination of the experiment, we noticed two FL3-injected mice had large nonlung tumors but did not yet have signs of distress, suggesting the overall survival difference would have been even more profound if we had continued the experiment. Only one of the mice injected with CQ-resistant derivatives, a C1AZ mouse, had to be killed due to signs of distress. We conclude from these data that selection for CQ resistance selects for gene expression patterns that favor less aggressive tumors and that selection for CQ resistance also reverses the metastatic phenotype of bladder cancer cells.

ID4 Expression Inhibits the Metastatic Potential of CQ-Resistant Cell Lines.

To test if these effects were due to ID4, we designed an in vivo model to test the relative metastatic fitness in the CQ-resistant cells when ID4 expression was reduced (Fig. 7A). To simulate the heterogeneity present in tumors, we used a pooled polyclonal population of the parental and three CQ-resistant cell lines. Nonsilencing shRNA or ID4 shRNA were introduced into each of the four lines represented in the polyclonal population, creating a mix of eight different groups within the tumor cell population (Fig. 7A, Left and SI Appendix, Fig. S14B). Each shRNA was labeled with a barcode to allow genetic identification of the relative levels of each cell type (SI Appendix, Fig. S14 C and D), and the parental FL3 cells, which have minimal expression of ID4, were labeled with distinct barcodes for nonsilencing and ID4 shRNAs. Equal numbers of cells from the four groups (parental FL3 with nonsilencing shRNA, parental FL3 with ID4 shRNA, CQ-resistant pool with nonsilencing shRNA, and CQ-resistant pool with ID4 shRNA) were pooled and injected into the tail veins of nude mice. After 90 d, mice were killed; autopsies were carried out, and complete lungs and any visual extrapulmonary metastatic tumors were isolated. shRNA vector DNA sequences were amplified in all tissues collected, and next-generation sequencing was used to identify the relative amount (“count”) of each shRNA vector. A similar sequencing analysis was carried out on the preinjection pool of cells (Fig. 7B). Determination of the percentage change in each shRNA vector read count from the original preinjection pool compared with that in the lungs was used as a measure of the relative metastatic fitness. As expected, the representation of the metastatic parental FL3 cells (irrespective of whether they had nonsilencing or ID4 shRNA) increased in the lungs of most (70%) mice (Fig. 7C), while representation of the CQ-resistant, nonsilencing pool was increased in the lungs of only a single mouse. In stark contrast, ID4 knockdown in the CQ-resistant pool led to increased representation of these tumor cells in the lungs in over one-third of mice (Fig. 7B). Additionally, when examining the nine nonpulmonary metastatic tumors, two tumors were composed predominantly of CQ-resistant cells with ID4 knockdown, while no tumors were isolated from CQ-resistant cells harboring nonsilencing shRNA (Fig. 7D). This experiment extends and confirms the previous experiment showing that selection for CQ resistance reduces metastatic fitness of the FL3 cells and demonstrates that ID4 knockdown alone is sufficient to reverse this phenotype and thus increase metastatic fitness. This demonstrates a cause-and-effect relationship between ID4 expression, CQ resistance, and lung metastasis. Taken together, these data indicate that ID4 is a metastasis suppressor in human bladder cancer cells and that its increased expression in CQ-resistant cells contributes to making these cells less metastatic (Fig. 6 B and C).

Fig. 7.

Knockdown of ID4 leads to greater metastatic fitness in CQ-resistant cell lines. (A) Schematic of the experiment to address metastatic fitness upon ID4 modulation in parental FL3 cells and CQ-resistant cell lines. Lentiviral vectors encoding ID4 and nonsilencing shRNAs were modified with different barcodes so as to identify the parental or resistant cells. Cells were pooled with 250,000 cells from each NSsh- and ID4sh-expressing cell line. Pooled cells were injected into the tail vein of nude mice (four separate experiments of 10 mice were injected; n = 4). After 90 d, or as required, mice were killed and the lungs of mice and visible nonpulmonary metastatic tumors were isolated. Genomic DNA was purified from each lung or tumor sample and the vector sequences were amplified with unique secondary barcodes for each lung or tumor. Next-generation sequencing was performed to identify the percentage of total reads that corresponded to each barcode as a measurement of the relative quantity of each cell line within a given lung or tumor. This percentage was compared with the percentage of reads in the original preinjection cell pool to determine whether the relative percentage increased. (B) Each mouse was scored as to whether a given cell population increased its overall percentage of reads compared with the preinjection cell pool as a measure of higher metastatic fitness. For each cell type (indicated by different colors), filled circles in this panel indicate when the relative percentage of cells (measured by the overall NGS reads) increased in the lungs of a given mouse with respect to its original percentage of the initial preinjection population. (C) The graph shows the relative metastatic fitness for each cell type as denoted by the average percentage of mice that scored positively (ones that increased their representation as scored in B) for that cell type over the four injection groups (n = 4). [All bars indicate mean ± SEM; *P = 0.018; n.s., nonsignificant (P = 0.56); P value determined by Wilcoxon rank-sum test.] (D) The figure shows the anatomic position of nonpulmonary metastatic tumors. Filled circles in this panel indicate the cell line dominating each tumor. (Pink, parental FL3 with nonsilencing shRNA; red, parental FL3 with ID4 shRNA; blue, CQ-resistant cell line with ID4 shRNA. There were no metastatic tumors dominated by a CQ-resistant cell line with nonsilencing shRNA.)

Discussion

We have previously shown that some cancer cells are particularly dependent on autophagy for their growth and viability even under unstressed conditions, while others have no enhanced requirement for autophagy (31, 32). Autophagy has also been shown to affect multiple cellular processes important for metastasis. For example, it promotes resistance to anoikis, modulates epithelial-to-mesenchymal transition, promotes tumor cell motility and subsequent invasion, influences cancer stem cell renewal/differentiation, and maintains tumor cell dormancy (6). However, while previous studies have measured the impact of blocking autophagy on metastasis, there are no studies that have examined if cancer cells that have already gained metastatic potential differ in their autophagy dependence compared with nonmetastatic counterparts. Here we evaluated this relationship.

Due to multiple influences of autophagy on metastatic processes, we hypothesized that as a cancer cell becomes more metastatic, it would more likely be dependent on autophagy for growth and survival. However, this hypothesis was incorrect; both the high metastatic and the low metastatic cells had equivalent sensitivity to knockdown of essential autophagy regulators, indicating that autophagy was similarly required for growth/survival of these cell lines. Our model is a late-stage model that preferentially reflects the ability of cells to survive, seed, and grow in target organs after direct injection into the bloodstream. These models were used because there are no human bladder cancer cell models that can be used to study spontaneous metastasis to the lungs. Therefore, it is possible that differences in autophagy dependence might be observed in cells with alterations in earlier steps in the metastatic process such as invasion or extravasation.

The more important conclusion of our work is that although autophagy itself was similarly important, there was a substantial difference in the sensitivity to inhibition of lysosome function when highly metastatic cells were compared with poorly metastatic cells. Moreover, we identified an inverse reciprocal relationship between gene expression patterns promoting metastasis and those promoting CQ/BafA₁ resistance such that selection for metastasis bestows sensitivity to CQ/BafA₁, and then selection for CQ/BafA₁ resistance in the metastatic cells returns the cells to a poorly metastatic phenotype. Inhibitor of DNA binding 4 is expressed in CQ-resistant/poorly metastatic cells and is highly down-regulated in metastatic/CQ-sensitive cells. Moreover, knockdown of ID4 imparts sensitivity to CQ and BafA₁ and also confers increased metastatic capability in the CQ-resistant cells, indicating that it is sufficient to explain a large part of the phenotypic differences observed between highly metastatic and poorly metastatic bladder cancer cells.

ID4 is proposed to act as a dominant-negative transcription factor (i.e., it binds to basic helix–loop–helix transcription factors but lacks a DNA-binding motif, and thus prevents transcription mediated by these factors); it likely acts by regulating transcription of a number of other genes in conferring lysosomal dependency and metastatic capability. ID4 promoter methylation is associated with unfavorable recurrence-free survival and an increased risk of lymph node metastasis in breast cancer patients (33, 34). Consistent with this, we found ID4 expression is associated with less aggressive bladder cancers in human patients and, importantly, reduction in ID4 expression led to increased metastatic potential in our metastatic fitness model. The therapeutic implication of this association is that more aggressive and metastatic cancers, which have lost ID4 expression, may be more amenable to therapeutic treatment with CQ. The use of CQ and HCQ as autophagy inhibitors is being investigated in many clinical trials (9, 10). Although some of the clinical studies involve patients selected for particular mutations such as the BRAF mutation in melanoma (e.g., NCT02257424), most of these studies do not attempt to select patient-based tumor characteristics that would be expected to confer sensitivity to CQ, HCQ, and/or autophagy inhibition. Our work suggests that in bladder cancer, and perhaps other tumor types as well, ID4 expression might provide such a marker. Our work also has implications for targeting autophagy and suggests that doing so by targeting the lysosome may have additional benefits compared with targeting earlier steps in the autophagy process—by targeting both autophagy-dependent tumor cell survival/growth, as we saw in all of the cells tested here, while also targeting the lysosome dependency that we see only in more metastatic cells. Additionally, the reciprocal relationship that we identified when we selected for resistance to CQ implies that tumor cells that find a way to circumvent this dependency may consequently become less aggressive by gaining additional ID4 expression.

It is not clear what non–macroautophagy-dependent lysosomal functions confer lysosome dependency on the metastatic cells. We ruled out chaperone-mediated autophagy (CMA) since LAMP2A is required for CMA (35), and knockdown of the LAMP2A splice variant alone has no effect on the growth and viability of the metastatic cells (Fig. 3 A and B). It is known that upon oncogenic transformation, cancer cells undergo changes in their lysosomes, including their content and activity, and differences in the membrane composition of lysosomes and in subcellular localization (36, 37). Invasive and metastatic cells have particularly large numbers of lysosomes located at the periphery of the cell instead of in their typical juxtanuclear region. These peripheral lysosomes have recently been implicated in the regulation and degradation of focal adhesion proteins, the degradation of Rho proteins, including RhoA, and the degradation of extracellular matrix upon exocytosis; these processes regulate cellular migration and invasion of metastatic cells (37, 38). It may be that these processes are involved in the connection between metastasis and additional sensitivity to lysosomal inhibitors. Some of the observed lysosome-dependency phenotype could possibly be linked to vulnerability of the lysosomal membrane to cationic amphiphilic drugs (39) and altered lysosomal sphingomyelinase activity (37, 39). Further determination of the mechanism by which metastatic cells like those tested here become particularly dependent on the lysosome may identify other ways to capitalize on this susceptibility that could be used together with drugs like chloroquine to target metastatic cells.

Methods

Reagents.

Chemicals, antibodies, and other reagents were obtained from sources as detailed in SI Appendix, Methods. The ATG4B doxycycline-inducible dominant-negative construct (iA4DN) was cloned into the acceptor lentiviral plasmid pCW57.1 Tet RE inducible system as indicated in SI Appendix, Methods.

Cell Culture.

T24t and the derived lung and liver metastatic series have previously been described (23, 24). All cell lines were authenticated by short tandem repeat profiling as matching T24. All cells were grown in DMEM/F12 medium (Gibco) supplemented with 5% FBS and penicillin-streptomycin.

Real-Time PCR.

Quantitative PCR reactions were performed using standard methodologies and normalized to SDHA as housekeeping gene control. For details, see SI Appendix, Methods.

Western Blots.

Cells were lysed with modified RIPA buffer (150 mM NaCl, 1% Nonidet P-40, 0.5% sodium deoxycholate, 0.2% SDS, 50 mM Tris⋅HCl, 5 mM EDTA, 500 mM NaF) containing protease inhibitor mixture (Roche). Protein lysates were sonicated briefly, resolved by SDS/PAGE, analyzed by Western blot, and visualized by enhanced chemiluminescence.

MTS and LDH Release Viability/Cytotoxicity Assays.

Cell proliferation/viability was measured using the MTS assay (Promega kit), while cell death/cytotoxicity was measured by LDH release (G-Biosciences). For details, see SI Appendix, Methods.

Clonogenic Assays.

For clonogenic assays, small numbers of cells (typically 10,000) were plated in six-well plates. Cells were treated for 24 to 48 h with CQ or bafilomycin A₁ and then rinsed and allowed to recover and grow for a number of days (typically 6 d); cells were then fixed (10% methanol, 10% acetic acid) and stained with crystal violet (BD). Plates were scanned and stain was solubilized with 30% acetic acid, and absorbance was measured at 540 nm. Measurement of cell growth/viability was normalized to untreated control cells.

Microarray Analysis.

cDNA probes made from RNA from T24t, FL3, C1AZ, D1AZ, and D1BZ1 cell lines were hybridized to the Affymetrix GeneChip PrimeView Human Gene Expression Array using standard Affymetrix protocols and analyzed by Partek. For details, see SI Appendix, Methods.

Anoikis Experiments.

Cells were plated on poly-HEMA–coated plates and cell death was analyzed by PI uptake/exclusion. For details, see SI Appendix, Methods.

Derivation of CQ-Resistant FL3 Cells.

FL3 cells were selected for CQ resistance by growing for several months in varying doses of CQ (lowest concentration, 30 μM; highest concentration, 100 μM) for varying times (shortest time, 48 h; longest time, 72 h) as illustrated in Fig. 4A. For details, see SI Appendix, Methods.

INCUCYTE Imaging.

Quantitative live-cell imaging to assess growth and viability was performed using an IncuCyte Zoom imaging system (Essen BioScience). For details, see SI Appendix, Methods.

TCGA Bladder Urothelial Carcinoma Data.

Gene expression and copy-number data were downloaded from TCGA data portal (April 2016) for 408 bladder urothelial tumors. Copy-number data were downloaded from TCGA in the form of GISTIC (40) scores for each gene, discretized to values of −2/−1/0/1/2 to represent homozygous loss, heterozygous loss, diploid, heterozygous gain, and homozygous gain, respectively. Overall survival data for TCGA BLCA dataset were downloaded from cBioPortal (41, 42) (April 2016). Gene expression data were downloaded from TCGA in the form of gene-level RSEM (43) normalized RNA-seq counts.

Survival Analysis.

Kaplan–Meier curves for Fig. 5 G and I and SI Appendix, Fig. S12 D and E were generated with copy-number and gene expression data downloaded from TCGA. We used a GISTIC (40) score of greater than zero to identify samples which had an ID4 copy-number gain and a score of zero to identify samples diploid for ID4. Kaplan–Meier curves were generated in R (44) with the “survival” package (45, 46).

Kaplan–Meier curves for SI Appendix, Fig. S13 were generated from Gene Expression Omnibus datasets GSE32894 and GSE13507 (both represent data from Illumina expression bead chip arrays). The curves were generated using the R2 Genomics Analysis and Visualization Platform software hosted by the Academic Medical Center in Amsterdam (r2.amc.nl). The Illumina Probe ID used for ID4 is ILMN_ 1721758. The expression cutoff modus in both datasets was set at the median for the given dataset.

Cluster Dendrogram Analysis.

Affymetrix CEL files from the GeneChip Human PrimeView arrays for each cell line were normalized with robust multiarray average (RMA) (47) in the “affy” package (48), and hierarchical clustering was performed with the “stats” package (44) using all 49,495 probes.

Copy-Number Frequency Simulations.

To determine the significance of ID4 copy-number gain or loss frequency in each cancer type, we calculated the frequency of copy-number gain or loss for all other genes on chromosome 6. We compared the frequencies of gain or loss of ID4 with the distribution of frequencies obtained for other chromosome 6 genes to obtain a P value.

Statistics.

Unless otherwise indicated, statistical analysis was performed using the unpaired two-tailed Student t test. Statistical analysis of metastatic fitness was performed using a Wilcoxon rank-sum test. Hazard analysis of P values for the Kaplan–Meier curve in Fig. 6C was done using a log-rank test.

shRNA Lentiviral Transduction.

shRNA pLKO.1 vectors were acquired from the Functional Genomics Facility at the University of Colorado Cancer Center. Lentiviruses were prepared largely according to protocols published by the RNAi Consortium (https://portals.broadinstitute.org/gpp/public/resources/protocols). For details, see SI Appendix, Methods.

Measurement of Autophagic Flux by Ratiometric Flow Cytometry.

Cells stably expressing mCherry-GFP-LC3 were used for flow cytometric analysis as previously described (49). For details, see SI Appendix, Methods.

Experimental Metastasis Assay.

Five-week-old female athymic NCr nu/nu mice (Charles River), maintained in accordance with the University of Colorado Denver and IACUC guidelines, were inoculated with FL3 cells or chloroquine-resistant FL3 derivatives D1AZ, C1AZ, and D1BZ1. Mice were injected with 1 × 10⁶ cells via the tail vein, with 10 mice per group. Nine weeks after injection (or as necessary under protocol guidelines due to animal distress or morbidity), mice were killed and human tumor burden in the lung was assessed by quantitative real-time PCR with a human-specific 12p TaqMan probe and primer set using 0.5 µg of genomic DNA isolated from both lungs (50).

Metastatic Fitness Assay.

Multiplex shRNA analysis of pooled populations.

To deconvolute pooled populations of cells treated with the same shRNA (shNS and shID4; see SI Appendix, Methods), we removed the endogenous canonical loop sequence (CTCGAG) and replaced it with eight unique 8-bp barcodes that could be read out using next-generation sequencing (NGS) using annealing oligomer mixtures (SI Appendix, Table S3). For details, see SI Appendix, Methods.

Metastatic competition of chloroquine-resistant cell lines.

Four different experimental groups of 5-wk-old female athymic NCr nu/nu mice (Charles River), maintained in accordance with the University of Colorado Denver and IACUC guidelines, were inoculated with FL3, D1AZ, C1AZ, and D1BZ1 cells stably expressing barcoded shNS or shID4 constructs. All cell lines were trypsinized, counted, and pooled at an equal number of cells. Mice were injected with 2 × 10⁶ cells via the tail vein (i.v.). Mice were euthanized when they met IACUC criteria for euthanasia or a terminal time point of 90 d post tumor challenge. Lungs and grossly visible ectopic tumors were isolated and flash-frozen in liquid nitrogen.

NGS library preparation and sequencing.

Genomic DNA was isolated from each sample (i.e., lungs or ectopic tumor) using the Gentra Puregene Kit (Qiagen). Sequences were then amplified with custom dual-indexed primers (51), and each sample was quantified by Qubit dsDNA HS Assay Kit (Invitrogen) and KAPA Library Quant Kit for Illumina-based sequencers (Kapa Biosystems) before equimolar pooling for NGS. For details, see SI Appendix, Methods.