Induction of Lysosome Membrane Permeabilization as a Therapeutic Strategy to Target Pancreatic Cancer Stem Cells

Despite significant efforts to improve pancreatic ductal adenocarcinoma (PDAC) clinical outcomes, overall survival remains dismal. The poor response to current therapies is partly due to the existence of pancreatic cancer stem cells (PaCSCs), which are efficient drivers of PDAC tumorigenesis, metastasis and relapse. To find new therapeutic agents that could efficiently kill PaCSCs, we screened a chemical library of 680 compounds for candidate small molecules with anti-CSC activity, and identified two compounds of a specific chemical series with potent activity in vitro and in vivo against patient-derived xenograft (PDX) cultures. The anti-CSC mechanism of action of this specific chemical series was found to rely on induction of lysosomal membrane permeabilization (LMP), which is likely associated with the increased lysosomal mass observed in PaCSCs. Using the well characterized LMP-inducer siramesine as a tool molecule, we show elimination of the PaCSC population in mice implanted with tumors from two PDX models. Collectively, our approach identified lysosomal disruption as a promising anti-CSC therapeutic strategy for PDAC.


Compound 1 Impairs the Self-Renewal and Tumor Initiating Capacity of PDAC Cells by Depleting the CSC Compartment
Having observed a preferential cytotoxic effect on human PDAC CSCs, we next asked whether Cpd 1 affects key CSCs properties and biological functions. First, we confirmed by flow cytometry that Cpd 1 preferentially targets the CSC compartment by analyzing the percentage of bulk and CD133+ cells undergoing apoptosis 24 and 48 h post treatment (Figure 2A-C). Indeed, when CD133+ cells were analyzed separately, we observed that Cpd 1 induced significantly more apoptosis in the PDAC2 CD133+ population compared to the bulk population 24 and 48 h post treatment ( Figure 2C).

Compound 1 Impairs the Self-Renewal and Tumor Initiating Capacity of PDAC Cells by Depleting the CSC Compartment
Having observed a preferential cytotoxic effect on human PDAC CSCs, we next asked whether Cpd 1 affects key CSCs properties and biological functions. First, we confirmed by flow cytometry that Cpd 1 preferentially targets the CSC compartment by analyzing the percentage of bulk and CD133+ cells undergoing apoptosis 24 and 48 h post treatment (Figure 2A-C). Indeed, when CD133+ cells were analyzed separately, we observed that Cpd 1 induced significantly more apoptosis in the PDAC2 CD133+ population compared to the bulk population 24 and 48 h post treatment ( Figure 2C). We next compared the decline in CD133-positive or autofluorescent-positive [27] CSC populations in cultures treated with Cpd 1 or gemcitabine, the standard of care chemotherapeutic for PDAC and to which CSCs are inherently resistant [28]. Adherent cultures of PDAC2 or PDAC4 were treated for 3 days with gemcitabine or Cpd 1 at 2.5 µM or 1.25 µM ( Figure 3A-D). We first assessed We next compared the decline in CD133-positive or autofluorescent-positive [27] CSC populations in cultures treated with Cpd 1 or gemcitabine, the standard of care chemotherapeutic for PDAC and to which CSCs are inherently resistant [28]. Adherent cultures of PDAC2 or PDAC4 were treated for 3 days with gemcitabine or Cpd 1 at 2.5 µM or 1.25 µM ( Figure 3A-D). We first assessed the capacity of each treatment to induce apoptosis ( Figure 3A) and modulate the percentage of CD133-positive or autofluorescent-positive CSCs ( Figure 3B) in PDAC4 cells by flow cytometric analysis. While all the capacity of each treatment to induce apoptosis ( Figure 3A) and modulate the percentage of CD133-positive or autofluorescent-positive CSCs ( Figure 3B) in PDAC4 cells by flow cytometric analysis. While all three treatments induced apoptosis, treatment with gemcitabine preferentially targeted the non-CSC (CD133-negative or autofluorescent-negative) population evident by the increase in the percentage CD133-positive or autofluorescent-positive cells following treatment. In contrast, treatment with both doses of Cpd 1 led to both an increase in apoptosis and a significant decrease in the percentage of CD133-positive or autofluorescent-positive cells, indicating again selective elimination of PDAC CSCs ( Figure 3A,B). We then assessed the effect of gemcitabine or Cpd 1 on self-renewal by assessing sphere-forming capacity and, as expected, found that while gemcitabine either did not affect sphere formation in the case of PDAC2 or increased it in the case of PDAC4 ( Figure 3C,D), Cpd 1 decreased sphere forming capacity at both doses tested and in both PDAC lines ( Figure 3C,D). Finally, in order to determine if Cpd 1 impacts tumor initiating capacity, we treated sphere-derived PDAC2 cells for 3 days with either gemcitabine or Cpd 1 at 1.25 µM, and subsequently subcutaneously injected non-treated and treated cells in vivo. Four weeks post injection we observed that while cells pre-treated with gemcitabine gave rise to larger tumors, Cpd 1-treated cells exhibited a complete inhibition in tumor formation ( Figure 3E,F). This reduced tumor-initiating capacity correlates with the inhibitory effects of Cpd 1 on CSCs seen in vitro ( Figure 3A-D).

Characterization and Development of Chemical Probes to Study the Mechanism of Action of Series 1 Compounds
In order to more fully understand the mechanism by which Series 1 hits impact CSC viability, self-renewal and tumor initiating capacity, we first carried out a structure-activity relationship (SAR) analysis, comparing the activity of Cpd 1 against a series of structural analogs ( Figure S2) including hit Cpd 2. After performing dose-response viability assays on PDAC1 spheres, we observed that the imidazo[1,2-b]pyridazine scaffold of Cpds 1 and 2 could be replaced by imidazo[1,2-a]pyrazine bicycle (Cpd 17). Importantly, we observed that the activity of this class of compounds depended on the presence of a basic nitrogen residue at C-6 position of the imidazo[1,2-b]pyridazine moiety as revealed by the lack of activity of Cpds 19 and 21. Noteworthily, all compounds in this series with a basic profile (pKa > 7) retained activity with EC50′s below 10 µM whereas non-basic compounds (pKa < 7) had no activity, even at 50 µM ( Figure S2). Thus, as an example, Cpd 16 bearing a piperidine secondary amine moiety, showed good potency (EC50 = 3.26 µM) but its corresponding analog Cpd 15, where the basicity was removed by the protection of the N-atom with a tert-butyl carbamate group, completely lost activity. The non-basic Cpds 13 and 18 with a hydroxyl group in the 6-position fragment instead of a basic amine were also inactive. Similar behavior was observed for Cpd 19 substituted with a non-basic tetrahydropyranyl ring. Furthermore, the non-basic cyclohexyl derivative 21, an analog of Cpd 16 where the basic NH moiety of piperidine ring was replaced with a CH2-group was also inactive, reinforcing the importance of a basic center (basicity) for the activity of this class of compounds.
Next, we initiated target engagement and target deconvolution tasks using label-free techniques and chemical probes, respectively, in order to identify the molecular target(s) responsible for the observed anti-CSC phenotype of Series 1 compounds. Cpd 2 was selected as a starting Cpd for both of these activities primarily due to its easier synthetic accessibility. Before proceeding, however, we first tested the anti-CSC activity of Cpd 2. We observed that Cpd 2, like Cpd 1, reduced the percentage of CD133+ cells and tumor growth in nude mice ( Figure S3 and Figure 3E,F), highlighting its anti-CSC properties. Next, we performed a kinome screen to determine if Cpd 2 interacted with kinases,

Characterization and Development of Chemical Probes to Study the Mechanism of Action of Series 1 Compounds
In order to more fully understand the mechanism by which Series 1 hits impact CSC viability, self-renewal and tumor initiating capacity, we first carried out a structure-activity relationship (SAR) analysis, comparing the activity of Cpd 1 against a series of structural analogs ( Figure S2) including hit Cpd 2. After performing dose-response viability assays on PDAC1 spheres, we observed that the imidazo[1,2-b]pyridazine scaffold of Cpds 1 and 2 could be replaced by imidazo[1,2-a]pyrazine bicycle (Cpd 17). Importantly, we observed that the activity of this class of compounds depended on the presence of a basic nitrogen residue at C-6 position of the imidazo[1,2-b]pyridazine moiety as revealed by the lack of activity of Cpds 19 and 21. Noteworthily, all compounds in this series with a basic profile (pKa > 7) retained activity with EC 50 s below 10 µM whereas non-basic compounds (pKa < 7) had no activity, even at 50 µM ( Figure S2). Thus, as an example, Cpd 16 bearing a piperidine secondary amine moiety, showed good potency (EC 50 = 3.26 µM) but its corresponding analog Cpd 15, where the basicity was removed by the protection of the N-atom with a tert-butyl carbamate group, completely lost activity. The non-basic Cpds 13 and 18 with a hydroxyl group in the 6-position fragment instead of a basic amine were also inactive. Similar behavior was observed for Cpd 19 substituted with a non-basic tetrahydropyranyl ring. Furthermore, the non-basic cyclohexyl derivative 21, an analog of Cpd 16 where the basic NH moiety of piperidine ring was replaced with a CH 2 -group was also inactive, reinforcing the importance of a basic center (basicity) for the activity of this class of compounds.
Next, we initiated target engagement and target deconvolution tasks using label-free techniques and chemical probes, respectively, in order to identify the molecular target(s) responsible for the observed anti-CSC phenotype of Series 1 compounds. Cpd 2 was selected as a starting Cpd for both of these activities primarily due to its easier synthetic accessibility. Before proceeding, however, we first tested the anti-CSC activity of Cpd 2. We observed that Cpd 2, like Cpd 1, reduced the percentage of CD133+ cells and tumor growth in nude mice ( Figure S3 and Figure 3E,F), highlighting its anti-CSC properties. Next, we performed a kinome screen to determine if Cpd 2 interacted with kinases, some of which are known to be involved in CSC biology and if targeted can have potential anti-CSC activity. The in vitro binding activity of Cpd 2 against a panel of 468 kinases (KINOMEscan ® , San Diego, CA, USA) at 1 µM concentration was assessed. The compound turned out to be highly selective, showing a selectivity index S(35) of 0.01. Only 4 kinases were significantly bound: HASPIN (30% signal versus control), the non-receptor tyrosine-protein kinase 2 (TYK2) (12%), serine/threonine-protein kinase pim-3 (PIM-3) (18%) and PIM-1 (5%). After dose response characterization using kinase orthogonal assays, the compound showed significant activity for PIM-1,-3 and HASPIN (IC 50 = 98, 129 and 915 nM, respectively) and very weak inhibition of TYK2 (IC 50 = 65 µM) ( Figure S4A and Table S1). However, we discarded PIM-1,-3 and HASPIN as the relevant targets of Cpd 2 against CSCs because compound Cpd 1 is not a PIM-1,-3 nor a HASPIN inhibitor (IC 50 > 10 µM for all; Figure S4B). These results pointed to a non-kinase target as responsible for the anti-CSC effects of Series 1 compounds.
Next, a chemical proteomics approach based on the design and use of Photo-Affinity Labeling (PAL) chemical probes derived from Cpd 2 was attempted. To this end, we designed an irreversible PAL chemical probe-Cpd 25-based on Cpd 2 and its non-basic amide inactive analogue Cpd 26. A so-called "minimalist linker" with a photoactivable diaziridine moiety and an alkyne reporter tag was incorporated into Cpd 2. The diazirine group of 25 would allow, after incubation with cells and short UV irradiation, the generation of highly reactive carbene species, which would react covalently with those proteins bound or very proximal to the probe. On the other hand, the terminal alkyne group of 25 could be used for subsequent conjugation to suitable reporters (rhodamine-N 3 or biotin-N 3 ) using bio-orthogonal click chemistry. These conjugated molecules were then used to determine the localization of the probe in cells by imaging techniques and in pull-down experiments to identify probe-linked cellular protein targets [29]. The PAL probe 25 was synthesized following standard procedures ( Figure 4A). Next, the effect of Cpds 2 and 25 (and the inactive Cpd 26) on PDAC4 CSC viability was assessed side by side, and the activity of probe 25 was comparable or even better than that of Cpd 2 (EC 50 = 2.62 and 7.17 µM, respectively, Figure 4B), whereas, as expected, Cpd 26 was inactive. , the non-receptor tyrosine-protein kinase 2 (TYK2) (12%), serine/threonine-protein kinase pim-3 (PIM-3) (18%) and PIM-1 (5%). After dose response characterization using kinase orthogonal assays, the compound showed significant activity for PIM-1,-3 and HASPIN (IC50 = 98, 129 and 915 nM, respectively) and very weak inhibition of TYK2 (IC50 = 65 µM) ( Figure S4A and Table S1). However, we discarded PIM-1,-3 and HASPIN as the relevant targets of Cpd 2 against CSCs because compound Cpd 1 is not a PIM-1,-3 nor a HASPIN inhibitor (IC50 > 10 µM for all; Figure S4 B). These results pointed to a non-kinase target as responsible for the anti-CSC effects of Series 1 compounds. Next, a chemical proteomics approach based on the design and use of Photo-Affinity Labeling (PAL) chemical probes derived from Cpd 2 was attempted. To this end, we designed an irreversible PAL chemical probe-Cpd 25-based on Cpd 2 and its non-basic amide inactive analogue Cpd 26. A so-called "minimalist linker" with a photoactivable diaziridine moiety and an alkyne reporter tag was incorporated into Cpd 2. The diazirine group of 25 would allow, after incubation with cells and short UV irradiation, the generation of highly reactive carbene species, which would react covalently with those proteins bound or very proximal to the probe. On the other hand, the terminal alkyne group of 25 could be used for subsequent conjugation to suitable reporters (rhodamine-N3 or biotin-N3) using bio-orthogonal click chemistry. These conjugated molecules were then used to determine the localization of the probe in cells by imaging techniques and in pull-down experiments to identify probe-linked cellular protein targets [29]. The PAL probe 25 was synthesized following standard procedures ( Figure 4A). Next, the effect of Cpds 2 and 25 (and the inactive Cpd 26) on PDAC4 CSC viability was assessed side by side, and the activity of probe 25 was comparable or even better than that of Cpd 2 (EC50 = 2.62 and 7.17 µM, respectively, Figure 4B), whereas, as expected, Cpd 26 was inactive.  Probe 25 was then used to determine intracellular localization. Specifically, 7-day-old PDAC spheres re-plated as adherent for 24 h were treated with Cpd 25, exposed to UV to crosslink the prob to its potential putative targets and/or proximal proteins, and subsequently conjugated to rhodamin via click chemistry to visualize the probe. Confocal microscopy imaging revealed that Cpd 2 concentrated in vesicular structures reminiscent of lysosomes. Co-staining of cells for the lysosome associated membrane protein 2 (LAMP2), a marker of lysosomes, confirmed that Cpd 25 concentrate close to LAMP2-positive structures, appearing to localize within lysosomes ( Figure 4C,D).

Compound 2 Induces Lysosomal Membrane Permeabilization
Given that the localization of Cpd 2 to lysosomes appeared to be required for its activity, w considered that it might be inducing lysosomal membrane permeabilization (LMP) as a mechanism to reduce CSC viability. It is known that LDCD-mediated cell death can be produced b lysosomotropic drugs. The accumulation of these agents in the lysosome is dependent on the lipophilicity and basicity, both of which are essential for their interaction with the lysosoma membrane, protonation in the acid lysosomal environment, and subsequent retention withi lysosomes [30]. As described above, the anti-CSC activity of Cpd 2 also depends on its basicity (pK = 10.01) and its lipophilicity (cLogP = 4.04), which falls within the accepted range for lysosomotropi agents. The same behavior was observed for other active analogues in this series ( Figure S2 Therefore, to test this hypothesis, we first confirmed that the effect of Cpd 2 on PDAC2 CSC surviva was partially prevented (≈ 3-fold) by the addition of Bafilomycin-A ( Figure 5A), a V-ATPase inhibito which increases the pH of the lysosome (less acidic) thereby, presumably reducing the accumulatio of Cpd 2 in lysosomes.
Next, we explored whether Cpd 2 could induce canonical characteristics of LMP by using standard assay involving loading cells with acridine orange (AO), a fluorescence dye used to stai acidic vacuoles. Re-plated PDAC2-derived sphere cultures were loaded with AO and subsequentl treated with either Cpd 2 or negative control Cpd 21, and AO fluorescence was visualized b fluorescence microscopy. As expected, untreated cells showed a vesicular pattern of AO staining i the far-red channel, reflecting accumulation in lysosomes, and a diffuse cytosolic and nuclear stainin in the green channel, reflecting AO's known ability to bind nucleic acids ( Figure 5B). Treatment o cells with 5 µM of Cpd 2 greatly reduced lysosomal staining in the far-red channel within just 2 h an significantly increased the overall signal in the green channel, a phenomenon indicative of LM ( Figure 5B). As expected, cells treated with the non-active Cpd 21 had the same pattern of staining i Probe 25 was then used to determine intracellular localization. Specifically, 7-day-old PDAC2 spheres re-plated as adherent for 24 h were treated with Cpd 25, exposed to UV to crosslink the probe to its potential putative targets and/or proximal proteins, and subsequently conjugated to rhodamine via click chemistry to visualize the probe. Confocal microscopy imaging revealed that Cpd 25 concentrated in vesicular structures reminiscent of lysosomes. Co-staining of cells for the lysosome-associated membrane protein 2 (LAMP2), a marker of lysosomes, confirmed that Cpd 25 concentrated close to LAMP2-positive structures, appearing to localize within lysosomes ( Figure 4C,D).

Compound 2 Induces Lysosomal Membrane Permeabilization
Given that the localization of Cpd 2 to lysosomes appeared to be required for its activity, we considered that it might be inducing lysosomal membrane permeabilization (LMP) as a mechanism to reduce CSC viability. It is known that LDCD-mediated cell death can be produced by lysosomotropic drugs. The accumulation of these agents in the lysosome is dependent on their lipophilicity and basicity, both of which are essential for their interaction with the lysosomal membrane, protonation in the acid lysosomal environment, and subsequent retention within lysosomes [30]. As described above, the anti-CSC activity of Cpd 2 also depends on its basicity (pKa = 10.01) and its lipophilicity (cLogP = 4.04), which falls within the accepted range for lysosomotropic agents. The same behavior was observed for other active analogues in this series ( Figure S2). Therefore, to test this hypothesis, we first confirmed that the effect of Cpd 2 on PDAC2 CSC survival was partially prevented (≈ 3-fold) by the addition of Bafilomycin-A ( Figure 5A), a V-ATPase inhibitor, which increases the pH of the lysosome (less acidic) thereby, presumably reducing the accumulation of Cpd 2 in lysosomes.
Next, we explored whether Cpd 2 could induce canonical characteristics of LMP by using a standard assay involving loading cells with acridine orange (AO), a fluorescence dye used to stain acidic vacuoles. Re-plated PDAC2-derived sphere cultures were loaded with AO and subsequently treated with either Cpd 2 or negative control Cpd 21, and AO fluorescence was visualized by fluorescence microscopy. As expected, untreated cells showed a vesicular pattern of AO staining in the far-red channel, reflecting accumulation in lysosomes, and a diffuse cytosolic and nuclear staining in the green channel, reflecting AO's known ability to bind nucleic acids ( Figure 5B). Treatment of cells with 5 µM of Cpd 2 greatly reduced lysosomal staining in the far-red channel within just 2 h and significantly increased the overall signal in the green channel, a phenomenon indicative of LMP ( Figure 5B). As expected, cells treated with the non-active Cpd 21 had the same pattern of staining in both the red and green channels as untreated cells, even at a dose of 10 µM, indicating that the activity of Cpd 2 against CSC viability and lysosomal integrity were linked. Flow cytometric analysis performed on PDAC2 cells from a parallel experiment quantitatively confirmed these results at 6 h ( Figure S5A) and 24 h ( Figure 5C) post treatment, with Cpd 2 decreasing overall AO mean far-red fluorescence intensity (MFI) in treated cells and with Cpd 21 having no effect on MFI. Similar affects were observed with PDAC1, PDAC3 and PDAC4 cells with Cpd 2 at 5 µM at 24 h ( Figure S5B). We additionally assessed the effect of Cpd 2 on cytosolic cysteine cathepsin activity, another standard assay for monitoring the release of lysosomal contents after LMP. Consistent with the microscopy-and cytometry-based results, treatment of re-plated PDAC2-derived spheres with Cpd 2 increased cytosolic cysteine cathepsin activity, while Cpd 21 had no effect ( Figure 5D).
Cancers 2020, 12, 1790 10 of 25 both the red and green channels as untreated cells, even at a dose of 10 µM, indicating that the activity of Cpd 2 against CSC viability and lysosomal integrity were linked. Flow cytometric analysis performed on PDAC2 cells from a parallel experiment quantitatively confirmed these results at 6 h ( Figure S5) and 24 h ( Figure 5C) post treatment, with Cpd 2 decreasing overall AO mean far-red fluorescence intensity (MFI) in Cpd 2-treated cells and with Cpd 21 having no effect on MFI. Similar affects were observed with PDAC1, PDAC3 and PDAC4 cells. We additionally assessed the effect of Cpd 2 on cytosolic cysteine cathepsin activity, another standard assay for monitoring the release of lysosomal contents after LMP. Consistent with the microscopy-and cytometry-based results, treatment of re-plated PDAC2-derived spheres with Cpd 2 increased cytosolic cysteine cathepsin activity, while Cpd 21 had no effect ( Figure 5D).

Broad Spectrum of LMP-Inducing Agents Target CSCs
In accordance with evidence indicating that glioblastoma and leukemic CSCs present an enhanced susceptibility to LMP-inducing agents [19,20], well-established LMP-inducers mefloquine and siramesine also affected PDAC2 CSC viability with varying potencies ( Figure 6A). We next assessed the effect of siramesine (SIR) alongside Cpd 2 on the self-renewal/sphere-forming capacity of PDAC1 and PDAC3 cultures, and observed that both compounds decreased sphere forming capacity at both concentrations tested ( Figure 6B). Like Cpd 2, we confirmed that the anti-CSC effect of siramesine is associated with LMP activity ( Figure 6C).
Based on the sum of these results, we speculated whether pancreatic CSCs might have an increased lysosomal mass compared to non-CSCs, which might account for their increased sensitivity to LMP-inducing agents. Indeed, altered lysosomal biology has been reported in CSCs from other tumor types such as breast and embryonal rhabdomyosarcoma [31,32]. To explore this possibility, we loaded re-plated PDAC2-derived spheres and PDAC2 adherent cell populations with AO and observed a clear increase in lysosomal mass in re-plated PDAC2-derived spheres compared to adherent cells, as evidence by an increased AO signal in the far-red channel relative to the green ( Figure 6D). This observation was validated by flow cytometry (Figure 6E), using AO and lysotracker. Moreover, we took advantage of our previously published RNA sequencing data (ArrayExpress: E-MTAB-3808 [33]) to confirm whether CSC-enriched sphere cultures over express a previously reported lysosomal biogenesis gene signature [34] compared to their adherent counterpart cultures. As expected, CSC-enriched sphere cultures showed transcriptional upregulation ( Figure S6A) and significant enrichment ( Figure  S6B) of this gene signature set. Together, these data suggest that Cpd 2 is a novel LMP-inducing agent.

LMP Inhibits Pancreatic CSC-Mediated Tumorigenesis and Disease Recurrence In Vivo
Lastly, we wanted to translate these findings to human patient-derived xenografts (PDX) in vivo. Initial in vivo PK studies of Cpd 2 in BALB/c mice showed high clearance (>100% hepatic blood flow) and low levels in plasma after intravenous administration pointing to high rate of drug elimination ( Figure S7A,B). Given the poor PK properties of Cpd 2, the well-known orally bioavailable LMP-inducer siramesine was selected as a tool compound to test the proof of concept that LMP-inducing agents can impact PDAC CSC biology in vivo. It is important to note the good pharmacokinetics of siramesine, non-toxic effects in humans and the well-established mechanism of action as an LMP [35,36]. Moreover, siramesine and Cpd 2 showed, in our hands, comparable anti-CSC activity in vitro ( Figure 6B). To this end, we implanted nude mice with PDAC1 or PDAC3 PDXs. Once tumors reached 500 mm 3 , mice were randomized to one of the following four treatment groups: Diluent control; Gemcitabine (GEM) (biweekly 125 mg/kg i.p.); siramesine (SIR) (30 mg/kg oral gavage); or GEM + SIR ( Figure S7C). Interestingly, for both utilized PDX models, no significant differences were observed for vehicle-treated and SIR single treatment; however, tumors treated with GEM or a combination of GEM + SIR were significantly reduced compared to control tumors and single treatment tumors ( Figure 7A). Importantly, flow cytometry analysis of dissociated tumors at day 23 post treatment showed differences in the percentage of cells expressing the CSC markers autofluorescence, CD133 or CD133/CXCR4 following treatment. While not significant, for PDAC1 there was a marked increase in autofluorescent-positive cells with GEM treatment, indicating an enrichment in this population  On the contrary, both the CD133+ or CD133+/CXCR4+ CSC populations significantly decreased only in the combination therapy group ( Figure 7B). A similar trend was observed with PDAC3 ( Figure 7B). Since our analyses of the percentage of CSCs in tumors following treatment revealed that only combination therapy significantly reduced the CSC population, we injected equal numbers of On the contrary, both the CD133+ or CD133+/CXCR4+ CSC populations significantly decreased only in the combination therapy group ( Figure 7B). A similar trend was observed with PDAC3 ( Figure 7B). Since our analyses of the percentage of CSCs in tumors following treatment revealed that only combination therapy significantly reduced the CSC population, we injected equal numbers of cells from dissociated GEM-treated and GEM+SIR-treated tumors into recipient mice to establish secondary tumors as a means of measuring tumor relapse potential ( Figure 7C). Eight weeks post injection, cells derived from tumors treated with GEM showed enhanced tumor take and increased CSC frequency (1 in 272 for PDAC1 and 1 in 102 for PDAC3) compared to cells derived from tumors treated with GEM+SIR (1 in 3879 for PDAC1 and 1 in 8290 for PDAC3) ( Figure 7D). Taken together, these data demonstrate that while GEM alone can reduce tumor size, the addition of a CSC-specific therapy (i.e., SIR) is necessary to eliminate the CSC population and subsequent tumor relapse potential.

Discussion
From a clinical perspective, eradication of the CSC compartment should lead to tumor elimination; however, targeting CSCs has proven difficult and technically challenging, and to date few anti-CSC therapies exists. Unfortunately, and as discussed above, screening platforms to discover new anti-CSC therapies have been hampered by the lack of in vitro systems that are sufficiently enriched in CSCs to facilitate the identification of "true" anti-CSC compounds. Consequently, the use of surrogate models of CSCs are becoming increasingly more common. For example, human pluripotent stem cell-and mESC-based platforms have been shown to faithfully reproduce properties of somatic CSCs [37][38][39][40]. Based on these observations and the fact that in general mESCs share many properties with CSCs, including self-renewal and expression of pluripotency-associated genes [8], we used mESCs as a surrogate system for CSCs in our small-scale screen to identify a unique class of small-molecules with potent anti-pancreatic CSC properties.
Using this approach, two classes of compounds were identified in the initial screen using mESCs. Series 2 contained PI3K and/or mTOR kinase activity inhibitors, and while this series was not further explored in this study, their identification confirmed the effectiveness of the screen as PI3K/mTOR signaling is important for CSC biology [21,22], and inhibitors of this pathway are currently being developed to target CSCs [23][24][25][26]. Series 1 contained 5 structurally similar imidazo[1,2-b]pyridazine-3-carboxylic acid (3-trifluoromethoxy-phenyl) amide derivatives with a basic nitrogen residue at C-6 position of the imidazo[1,2-b]pyridazine or imidazo[1,2-a]pyrazine scaffolds. Importantly, not only were Series 1 compounds effective against mESCs, but they faithfully recapitulated their activity in primary human PDAC cultures, derived from 4 different PDX models, showing potent and specific anti-CSC activity. Specifically, we demonstrated that the identified compounds preferentially targeted pancreatic CSCs over non-CSCs as Series 1 compounds were more effective in cultures enriched in CSCs (e.g., 3D spheres or 2D re-plated spheres) and cells expressing known pancreatic CSC markers, such as CD133 or autofluorescence, decreased in percentage and preferentially entered into apoptosis upon treatment with low micromolar doses. The loss of the CSC compartment in vitro translated into an ablation or significant reduction in the in vivo tumorigenic potential of cells pre-treated with these compounds, specifically with Cpd 1 or Cpd 2, respectively. The complete opposite was seen with Gemcitabine, which we have previously shown enriches for CSCs [28,41].
Having discarded their interaction with kinases via a KINOMEscan, Photo-Affinity Labeling (PAL) chemical probes, with a photoactivable diaziridine moiety and an alkyne reporter tag, were utilized to identify proteins bound or proximal to the probe or to visualize the probe via fluorescence microscopy, respectively, as a means of elucidating the anti-CSC mechanism of action of these compounds. This approximation showed that the Series 1 compounds had a clear affinity for the lysosome. In addition, we determined that the presence of the basic nitrogen residue at C-6 position was necessary for the activity of Series 1 compounds, as non-basic derivatives (pKa < 7) showed no anti-CSC activity. For example, the anti-CSC activity of Cpd 2 depended on its basicity (pKa = 10.01) and its lipophilicity (cLogP = 4.04), both of which fall within the accepted range for LMP inducers. One of the best studied LMP inducers is the σ 2 receptor agonist, siramesine While initially designed to treat anxiety and depression [42], it was shown to be a more potent LDCD inducer, destabilizing the lysosomal membrane to promote the release of cathepsins resulting in cathepsin-mediated death in tumor cells [43,44]. In 2017, Jensen et al., tested the anti-CSC effects of siramesine on glioblastoma in vitro and in vivo and showed that siramesine had potent effects on glioma cell lines and patient-derived spheroids cultures enriched in CSCs in vitro [20]. Another recently described LMP inducer is mefloquine, a quinoline used for the treatment and prevention of malaria [45], but which was shown to induce LDCD in human acute myeloid leukemia (AML) cells including AML CSCs in vitro and in vivo [19]. It is important to note that the aforementioned studies represent the only two studies to date demonstrating that siramesine or mefloquine can target CSCs by inducing LDCD. While other studies have shown that siramesine can impact breast, lung, cervix, prostate and connective tissue tumor cell lines, the effect on CSCs was not explored in these studies [43,44,46].
The accumulation of LMPs in the lysosome is dependent on their lipophilicity and basicity (cLogP > 2 and 6.5 < pKa < 11) [30]. Specifically, the lipophilicity of these agents is essential to facilitate their interaction with the membranes of and entry into lysosomes and their basic center(s) are crucial for their retention in the lysosome following their protonation by the acidic environment of the organelle (pH: 4.5-5.0) leading to their subsequent accumulation. Based on the similarities in basicity and lipophilicity with siramesine and mefloquine and the localization of our compounds with the lysosome in PDAC CSCs, we hypothesized and experimentally confirmed via numerous approaches that our Series 1 compounds were also acting via an LCDC-based mechanism of action. Moreover, when compared to the known LMP inducer siramesine, Cpd 2 showed equivalent anti-CSC activity, confirming that the Series 1 compounds function by inducing LCDC. The aforementioned points have several implications. Firstly, using an independent and unrelated screening approach, we also identified LMP inducers as potent inhibitors of CSCs. As mentioned above, only two other studies to date have identified that LMP inducers can target CSCs of glioblastoma [20] and AML [19]. Importantly, our study expands this observation to PDAC CSCs and highlights that LMP inducers may represent a potential anti-CSC therapy for many tumor types (both solid and non-solid). Secondly, siramesine showed equivalent anti-CSC activity as Cpd 2 in vitro. Furthermore, when siramesine was used together with gemcitabine in two PDX models, combination treatment efficiently and significantly reduced the CSC compartment, which was validated by flow cytometry and in serial tumorigenesis assays where the CSC frequency was reduced to 1 in 8290. While gemcitabine alone equally reduced PDAC1 and PDAC3 tumor growth in first generation tumors, flow cytometry analysis of these tumors confirmed an enrichment rather than a reduction in the CSC compartment, which resulted in tumor relapse upon serial passage. Siramesine alone had no effect on tumor size, but did reduce the percentage of CD133 + and CD133 + /CXCR4 + in PDAC3 tumors; however, this effect was not sufficient to reduce tumor growth at the dosage used (30 mg/kg). Thus, in the case of LMP inducers, higher doses or targeting both the CSC and the more differentiated bulk tumor cells (e.g., with gemcitabine) would likely be necessary to achieve sustained tumor clearance. It is interesting that Jensen et al. similarly showed that siramesine alone was ineffective in vivo (with 100 mg/kg) using implanted invasive glioblastoma spheroids or gliobastomas PDXs [20]; however, the authors did not use siramesine in combination with standard chemotherapeutics, nor did they examine the CSC compartment or perform serial transplantation analysis. Thus, the ability of siramesine to effectively target glioblastoma CSCs is still unresolved and would benefit from combination therapy-based studies. Lastly, while we did not measure lysosomal size, we did show that PDAC CSCs have increase lysosomal mass. The latter we attribute to increased lysosomal biogenesis, which was confirmed by RNAseq analysis using a previously published lysosomal biogenesis gene signature set [33]. Using this same signature, Sukhai et al. demonstrated a similar phenotype for primitive and progenitor AML CSCs, which they successfully targeted with mefloquine [19]. Taken together, these observations indicate a plausible mechanism to explain the observed differential and specific activity of the LMP inducers siramesine or mefloquine at the level of CSCs, at least for AML and PDAC, respectively. It would not be surprising if increased lysosomal mass is a conserved hallmark of CSCs across different tumor entities, which is necessary for their biological activity and "stem"-ness. Thus, targeting the lysosome may represent a true Achilles' heel for CSCs in general and should be further explored.

CNIO-640 Chemical Library
The Experimental Therapeutics Program at the CNIO, ETP-CNIO, owns a chemical library of about 50,000 single compounds built as a result of the consolidation of several sub-libraries selected attending to different criteria such as chemical diversity, kinase-targeted focus, potential to disrupt protein-protein interactions, and the presence of low molecular weight compounds to facilitate fragment-based drug discovery. The drug-likeness of the whole library was also ensured by the application of filters such as "rule of five" [47]. The compounds were selected from commercial origin as well as from internally newly designed and synthesized chemical matter. Representative libraries of the whole 50,000 library with smaller sizes were defined after clustering, based on similarity analysis, and selection of representative compounds from each cluster. A 640-compound library, subject of the current detailed screening campaign, is the minimum size set of compounds representing the chemical ETP-CNIO collection.

Primary Human Pancreatic Cancer Cells
The PDAC patient-derived xenograft (PDX) Panc354 (referred to hereafter as PDAC3) was obtained from Dr. Manuel Hidalgo under a Material Transfer Agreement with the Spanish National Cancer Centre (CNIO), Madrid, Spain (Reference no. I409181220BSMH). PDAC1, PDAC2 and PDAC4 PDXs were established from primary patient resected tumors. Samples were provided by the BioBank Hospital Ramón y Cajal-IRYCIS (PT13/0010/0002), integrated in the Spanish National Biobanks Network and they were processed following standard operating procedures with the appropriate approval of the Ethical and Scientific Committees (Control no. No. Control: DE-BIOB-73 AC65, RG.BIOB-96 REV, and CI-BIOB-08-10-01 rev), with informed consent and according to Declaration of Helsinki principles. Primary patient tumors were subcutaneously implanted in immunocompromised female 6to 8-week-old NU-Foxn1nu nude mice (Envigo, Barcelona, Spain) and subsequently passage in vivo to establish PDX PDAC1, PDAC2 and PDAC4. Primary PDX-derived in vitro cultures were established as previously detailed [48]. Briefly, xenografts were minced, enzymatically digested with collagenase (Stem Cell Technologies) for 60 min at 37 • C and after centrifugation for 5 min at 1800 rpm, the cell pellets were resuspended and cultured in RPMI

Sphere Formation Assay
Pancreatic CSC spheres were generated as previously described [48] in ultra-low attachment plates (Corning) using CSC media: serum-free DMEM/F12 medium . All cell culture was carried out at 37 • C in a 5% CO 2 humidified incubator. To quantify spheres of > 40 µm, 1 mL of sample volume was analyzed with a CASY®Cell Counter (Roche Applied Sciences, Mannheim, Germany). The CASY®Cell Counter measuring principle is based on a capillary particle counter with pulse area analysis, that permits the determination of cell count, cell concentration, cell volume (peak), and average cell diameter, specifically diameters of 40-80 µm, 80-120 µm and >120 µm.

Drug Screening and Cell Viability Assays
Murine embryonic stem cells (mESCs) were plated at 20,000 cells per well in gelatinized 96 well plates. Compounds from the 640 compound ETP library were added by a Beckman FX 96 tip to achieve a final concentration of 5 µM in each well, and each compound was tested in duplicate. Cells were treated for a total of 30 h prior to assessing cell viability using CellTiter 96 ® AQueous One Solution Cell Proliferation Assay (MTS) assay according to manufacturer's instructions (G3582, Promega, Madison, WI, USA). Dose response curves were carried out on mESCs under the same conditions using indicated compounds. To test the viability of CSCs, 10,000 PDAC cells were plated in spheroid conditions and immediately treated with indicated compounds and concentrations for the duration of the sphere formation assay. For adherent non-CSCs, 20,000 cells were plated in wells of standard tissue culture-treated plates (Corning, Corning, NY, USA) and treated the following day with indicated compounds and concentrations. For re-plated spheres ( Figure S1B), 7-day-old spheres were filtered using a 40 µM filter (Fisher Scientific, Hampton, NH, USA), washed once with 1× PBS, and incubated with trypsin 0.05% EDTA at 37 • C (Thermo Fisher Scientific, Waltham, MA, USA) to achieve a single cell suspension (approximately 10-30 min, depending on the PDAC culture). Trypsinized cells were counted and 20,000 cells were plated in wells of standard tissue culture-treated plates (Corning) and treated the following day with indicated compounds and concentrations.

Flow Cytometry
Cells were analyzed with a 4-laser Attune NxT Acoustic Cytometer (ThermoFisher Scientific). Cells and digested tumors (as described in [41]) were resuspended in FLOW buffer (1X PBS; 3 mM EDTA (v/v); 3% FBS (v/v)), and the following fluorescently-tagged antibodies were used to label cells for 30 min at

Pharmacokinetic Studies (PK)
Female BABL/c mice, 10 weeks old were used (n = 3 per time point). Compound 2 was formulated in 0.9% NaCl solution for both oral and intravenous (i.v.) injections. Plasma samples were collected following a single i.v. or oral administration of 8 mg/kg at 0.08, 0.16, 0.25, 0.5, 1, 4, 8 and 24 h. The extraction of Cpd 2 from plasma was achieved by solid phase extraction followed by high performance liquid chromatography/tandem mass spectrometry (Agilent 1100, Applied Biosystems API2000) analysis. The amount of inhibitor and the internal standard in each mouse plasma sample were quantified based on calibration curves generated using standards of known concentrations of compound. This assay method was sufficiently accurate for the quantification of Cpd 2 with a limit of detection (LOD) and a low limit of quantification (LLOQ) of 5 ng/mL. Pharmacokinetic parameters were estimated using WinNonlin Version 5.2 software (Pharsight Corp., Mountain View, CA, USA), by fitting both the experimental i.v and oral data to a bicompartimental model.
Mice were housed according to institutional guidelines and all experiments were performed in compliance with the institutional guidelines for the welfare of experimental animals approved by the Instituto de Salud Carlos III Ethics Committee (CBA12_2014-v3) and La Comunidad de Madrid (PROEX 53/14), and in accordance with the guidelines for Ethical Conduct in the Care and Use of Animals as stated in The International Guiding Principles for Biomedical Research involving Animals, developed by the Council for International Organizations of Medical Sciences (CIOMS).

KinomeScan
The KINOMEscan TM screening platform employs an active site-directed competition binding assay to quantitatively measure interactions between test compound and 468 human kinases and disease relevant mutant variants. These assays do not require ATP and thereby report thermodynamic interaction affinities. The assay was performed at DiscoveRx Corporation.

Synthesis of Affinity Chemical Probes
General Procedure; Chemicals were purchased from Aldrich Chemical Company Ltd., Apollo Scientific Ltd. and TCI Europe N.V. Unless otherwise stated, commercial chemicals, and solvents were used without further purification. Anhydrous dichloromethane (DCM), dioxane, acetonitrile, and TEA (Tris base, acetic acid and EDTA) were purchased from Aldrich in Sure SealTM bottle and kept under nitrogen. Proton ( 1 H) NMR spectra were recorded on Bruker Avance II 300 using DMSO-d6 as solvent. Chemical shifts are expressed in parts per million (ppm) (δ relative to residual solvent peak for 1 H). residue purified by automated chromatography in SiO 2 (DCM/MeOH up to 9:1) giving a residue which was purified by HPLC to yield 4 mg of desired compound 26 (22% Yield). LC-MS (ESI) RT

Kinase Assays
PIM1/3 kinase activities were measured as previously described in Martínez-Gonzalez et al. [49]. HASPIN activity was measured using the commercial ADP GLO assay (Promega), a homogeneous assay measuring ADP accumulation, as a universal product of kinase activity. The assay was done following general manufacturer recommendations and adapting protein to optimal conditions. Kinase buffer was 15 mM HEPES ((4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid), pH 7.4, 20 mM NaCl, 1 mM egtazic acid (EGTA), 0.02% Tween-20, 10 mM MgCl 2 and 0.1 mg/mL BGG (bovine γ-globulin). The assay to measure autophosphorylation of HASPIN (ProQinase, Freiburg, Germany) was done at 150 µM ATP. In order to calculate the IC 50 of described compounds, serial 1:5 dilutions were tested. Luminescence counts were read in a Victor instrument (Perkin Elmer) Values were plotted against inhibitor concentration and fit to a sigmoid dose-response curve with activity base from IDBS software. TYK2 kinase assay was performed at ProQinase.

Immunofluorescence with Click Chemistry Probes
Seven-day-old spheres were disaggregated with trypsin and re-plated in chamber slides. One day after plating, cells were treated with 2.5 µM Cpd 25 for 30 min, then media was removed and replaced with ice cold PBS. Cells were exposed to 366 nM UV for 30 min using a hand-held lamp then washed, fixed, permeabilized and washed once again in PBS containing 1% BSA. Click it chemistry was performed using Click-it reaction cocktail from Invitrogen's Click-it Cell Reaction Buffer Kit (Cat. no. C10269) containing 5 µM Rhodamine-azide for 2 h. Cell were subsequently stained with LAMP2 mouse monoclonal antibody from Abcam and anti-mouse IgG Alexa Fluor 488 and with DAPI. Preparations were mounted and visualized by confocal microscopy (Leica TCS SP5 Confocal Microscope, Wetzlar, Germany).

Acridine Orange Lysosomal Integrity Assay
Re-plated spheres were loaded with 1 µg/mL acridine orange (A1301, Thermo Fisher Scientific) for 15 min then washed 3 times with PBS. Cells were treated with indicated concentrations of compounds for 2 h. Relative intensities of acridine orange binding to nucleic acids versus that contained in lysosomes was observed with an upright fluorescent microscope or via flow cytometry as detailed above. To visualize lysosomal mass in spheres versus adherent cells, acridine orange was added at 1 µg/mL for 15 min and fluorescent signal was observed in absence of compound treatment via flow cytometry as detailed above or with an upright fluorescent microscope (Nikon Eclipse Ti microscope with NIS Elements BR Imaging Software, Amsterdam, Netherlands).

Measurement of Cytosolic Cathepsin Activity
Spheres were re-plated, and then treated the following day with the indicated concentration of compounds for 6 h in CSC media. Cells were subsequently placed on ice and extracted with either 25 µg/mL or 250 µg/mL of digitonin in 250 mM sucrose, 20 mM HEPES, 10 mM KCl, 1.5 mM MgCl 2 , 1 mM EDTA, 1 mM EGTA, 1 mM pefablock, pH 7.5. The extracts were taken and spun down briefly to eliminate floating cells. Cathepsin activity was measured in a 96 well format by measuring cleavage of zFR-AMC (Z-Phe-Arg-AMC. Z: N-carbobenzyloxy; 7-Amino-4-methylcoumarin), as previously described [50].

RNA Sequencing
Illumina RNA sequencing of five primary PDX-derived PDAC cultures grown as either adherent (Non-CSCs) or anchorage-independent spheres (CSCs), in duplicate, was previously performed [33], and raw data was deposited in ArrayExpress accession number: E-MTAB-3808. Differential expression of genes across the different conditions was calculated with Cuffdiff. Heatmaps showing gene expression levels (FPKM, Fragments per kilobase of transcript per million mapped fragments) for the different samples were drawn with Morpheus, (https://software.broadinstitute.org/morpheus) for a subset of selected genes involved in Lysosome Biogenesis [34]. This signature set was uploaded to the Gene Set Enrichment Analysis (GSEA) module of the Genepattern suite from the Broad Institute and false discovery rate (FDR) < 25% was considered statistically significant.

Statistical Analyses
Results are presented as means ± standard error of the mean (SEM) unless stated otherwise. An unpaired two-tailed t test was performed to determine significance. p values < 0.05 were considered statistically significant. All analyses were performed using GraphPad Prism version 6.0 (San Diego, California USA). CSC frequencies and significance were determined using the on-line extreme-limiting dilution assays (ELDA) program (http://bioinf.wehi.edu.au/software/elda/) [51].

Data Availability
Unique identifiers for publicly available datasets are indicated, a list of figures that have associated raw data can be provided, and there are no restrictions on data availability.

Conclusions
In conclusion, while CSCs represent the root of most tumors, they are not without their weaknesses. In this study, we screened a chemical library to identify candidate anti-CSC molecules and identified several compounds, belonging to the same series, with potent activity against PaCSCs in several PDX culture models. Interestingly, these compounds acted on PaCSCs via induction of LMP, highlighting a potentially important role for lysosomes in PaCSC biology. The latter is supported by the fact that PaCSCs have increased lysosomal mass, a phenotype that makes them more susceptible to LMP inducers and one that appears to be conserved across CSCs of different tumor entities. While our lead compounds are still far from in vivo application, we show that the well characterize LMP-inducer siramesine can efficiently target and eliminate the PaCSC population in mice implanted with pancreatic PDX tumors, demonstrating the potential clinical application of LMP inducers as a promising anti-CSC therapeutic strategy for PDAC. Thus, inducers of LMP, in general, should be explored as possible therapeutics for CSC-driven tumors.