Background
Pancreatic ductal adenocarcinoma (PDAC) is one of the most lethal cancers worldwide [
1], with a 5-year overall-survival reached in only 10% of patients [
2]. This poor prognosis is due to a lack of early biomarkers, limited therapeutic options, and inherent or acquired chemoresistance [
3,
4]. Currently, surgical resection is the only curative option for patients diagnosed with PDAC but < 20% are diagnosed at an early stage and therefore eligible for surgery. In all other cases (i.e., advanced PDAC), chemotherapy using FOLFIRINOX or gemcitabine plus nab-paclitaxel are the only treatment options [
5,
6]. Unfortunately, these chemotherapy regimens increase survival up to 13 months at most, largely due to development of chemoresistance. So far, immunotherapy has not been successful for PDAC patients and new therapies under investigation targeting the tumor or the tumor microenvironment have not reached the clinic [
7‐
9]. Therefore, identifying strategies to combat PDAC resistance to currently used chemotherapies is of crucial importance.
Gemcitabine is a cytotoxic DNA-intercalating drug, which arrests aberrant cell proliferation. Chemoresistance to gemcitabine in PDAC has been extensively studied and reported to be multifactorial [
10]. On the other hand, paclitaxel is a microtubule-stabilizing drug that impedes cell division leading to replication errors and cell death. Paclitaxel also potentiates gemcitabine efficacy by increasing intratumor uptake and inhibiting its inactivation by catabolizing enzymes [
10,
11]. Mechanisms underlying paclitaxel resistance in PDAC are poorly understood [
12,
13]. Three studies have investigated paclitaxel resistance in PDAC, reporting that it involves metabolic adaptation [
14], sustained c-MYC activation [
15], and expression of orexin receptor type 1 [
16].
ATP-binding cassette (ABC) transporters are responsible for active transport of many substrates, including cytotoxic drugs, across the cell membrane towards the extracellular space. ABC transporters are therefore known as multidrug resistance pumps [
13,
17]. The ABC family consists of 49 members, among which ABCB1 (also known as MDR1 or P-glycoprotein, P-gp) has been extensively studied in cancer. Overexpression of ABCB1 can mediate paclitaxel resistance in different tumor types, including colorectal, lung, ovarian and breast [
18‐
21]. Surprisingly, in PDAC, ABC transporters, including ABCB1, have been associated with gemcitabine resistance [
13,
22‐
26] but not with paclitaxel resistance.
In the present study, we have generated three independent paclitaxel- and gemcitabine resistant PDAC models and found that ABCB1 is amplified in paclitaxel resistant, but not in gemcitabine resistant PDAC cells. We show that pharmaceutical or genetic inhibition of ABCB1 effectively restores paclitaxel sensitivity in the resistant cell lines. We show that ABCB1 is expressed heterogeneously in PDAC patients. Moreover, as clinical trials of currently available ABCB1 inhibitors have not proven successful due to lack of efficacy or toxicity [
21,
27], we screened a kinase-inhibitor (KI) library for KIs that attenuate efflux of ABCB1 substrates, thereby overcoming paclitaxel resistance. We identified several novel KIs that can be further (pre) clinically explored as therapeutic strategies in combination with paclitaxel to overcome paclitaxel resistance in PDAC and other cancers.
Methods
Materials and cell culture
Three PDAC cell lines were used: Patu-T (mesenchymal phenotype), kindly provided by Dr. Irma van Die (Amsterdam UMC, Amsterdam, The Netherland), Suit-2.028 (epithelial phenotype) and Suit-2.007 (mesenchymal phenotype), kindly provided by Dr. Adam Frampton (Imperial College London, London, UK). Patu-T were maintained in DMEM, supplemented with 10% heat-inactivated bovine fetal serum and 1% penicillin/streptomycin, while both Suit-2 cell lines were cultured in RPMI supplemented as described above. All cells were kept in humidified atmosphere of 5% CO2 and 95% air at 37 °C, subcultured twice a week, tested monthly for mycoplasma contamination by MycoAlert Mycoplasma Detection Kit (Westburg, Leusden, The Netherlands) and cell identity was verified by short tandem repeats (STR) profiling.
Gemcitabine was kindly provided by Eli Lilly Corporation (Indianapolis, IN, USA) and dissolved in sterile water. Paclitaxel and verapamil were obtained from Sigma (T7402 and V4629, Sigma-Aldrich, St. Luis, MO, USA).
Generation of resistant cell lines
To establish gemcitabine-resistant (GR) and paclitaxel-resistant (PR) cell lines, concentrations causing 50% reduction in cell growth (IC50) were determined in parental cells. Cells were then exposed to the respective IC50 of the drug and grown for at least 2 weeks with the drug until reaching 80% confluency. After acquiring resistance, the drug concentration was doubled (2 × IC50) and cells were cultured until they could grow to confluence. The process was repeated with stepwise increasing drug concentrations until the maximum tolerated concentration was reached after 6–12 months. Parental cells never exposed to the drug were cultured in parallel with the resistant cells. To determine stable resistance, PR and GR cells were grown in drug-free medium and baseline growth and resistance to the maximum tolerated concentration was analyzed by SRB assay at regular intervals for up to 2 months. The resistance factor was calculated as the ratio of the IC50 of resistant versus IC50 of parental cells. For IC50 > 12 μM, the resistance factor was calculated using the maximum drug concentration used in the SRB assay. Batches of resistant cells used in experiments were maintained in drug-free medium ≤ 2 months.
Immunohistochemistry
Expression of ABCB1 in PDAC patients was evaluated by immunohistochemistry (IHC) in paraffin-embedded tumor specimens from 32 PDAC patients who underwent resection and were treated with gemcitabine (1,000 mg/m2) plus nab-paclitaxel (125 mg/m2) as first-line therapy. All specimens were obtained after patient’s written consent approved by the Ethics Committee of “Area Vasta Emilia Nord” (protocol code 12003—17/03/2021). Tissue sections were stained overnight with rabbit anti-human ABCB1 (E1Y7S, mAb #13978; Cell Signaling Technology; dilution 1:400). Sections were reviewed independently by two researchers blinded to clinical data, who scored the immunostaining on the basis of staining intensities and number of stained cells as “low” or “high”. Overall survival (OS) was calculated from the date of pathologic diagnosis (i.e., the date of surgery/biopsy) to the date of death. OS curves were constructed using Kaplan–Meier method, and differences were analyzed using log-rank test with SPSS v.25 statistical software (IBM).
SRB assay
For sulforhodamine B (SRB) assay, cells were seeded in 96-well flat bottom plates at a density of 3000–4000 cells/well. After 72 h of drug exposure, plates were fixed with 50% TCA, and incubated with 0.4% SRB at room temperature avoiding light. Plates were washed with 1% acetic acid to remove unbound SRB and air dried. 10 mM Tris was used to extract protein-bound dye and optical density was measured with a BioTek Synergy HT plate reader (SN 269140, BioTek Instruments Inc.) at 490 and 540 nm. IC50 was determined through interpolation in GraphpadPrism (version 9.0, Intuitive Software for Science, USA).
RT-qPCR and DNA-qPCR
For RT-qPCR, RNA was isolated with an RNEasy Plus Mini kit (QIAGEN, Cat. 74136). 800 ng RNA was used to generate cDNA with the Thermo Scientific RevertAid H Minus First Strand cDNA Synthesis Kit (Thermo Fisher Scientific, Waltham, MA, USA). For DNA-qPCR, genomic DNA was isolated with the GenElute™Mammalian Genomic DNA Miniprep kit, following manufacturer instructions (Cat. GIN350, Sigma-Aldrich, St. Luis, MO, USA). RNAse A solution (provided with the kit, 1:10 dilution) was used to obtain RNA-free DNA. 7.5 ng of genomic DNA was used as template. qPCR was performed in triplicate using the PowerUp™ SYBR™ Green Master Mix (Thermo Fisher Scientific, Waltham, MA, USA) in a QuantStudioTM 6 Flex Real-Time PCR system (Applied Biosystems®, ThermoFisher Scientific, Waltham, MA, USA). Primers were from Sigma-Aldrich (St. Louis, MO, USA) and were directed against exon-exon boundaries for RT-qPCR or directed against intron–exon boundaries or introns for DNA-qPCR (Supplemental Table S
1). Relative mRNA expression and relative DNA amount was calculated using the 2
−(ΔΔCt) method with
ACTB and
GAPDH as reference genes.
Western blot
Cells were lysed with RIPA buffer supplemented with 1% protease/phosphatase inhibitor cocktail (PIC, Sigma-Aldrich, St. Luis, MO, USA), 40 μg lysates were separated by SDS–polyacrylamide gel electrophoresis and transferred to PVDF membranes. Membranes were incubated overnight at 4 °C with rabbit-anti-human ABCB1 (E1Y7B; mAb #13,342; Cell signaling Technology; dilution 1:1000) and mouse-anti-human β-actin (sc-47778; Santa Cruz Biotechnology, Santa Cruz, CA, USA; dilution 1:1000) antibodies, followed by incubation with HRP-conjugated anti-rabbit (#7074; Cell Signaling Technology; dilution 1:2000) and Alexa Fluor® 647-conjugated anti-mouse (115–605-146; Jackson ImmunoResearch, Bio-Connect, Huissen, The Netherlands; dilution 1:1000) secondary antibodies for 1 h at room temperature. Signals were detected using enhanced chemiluminescence (ECL) and fluorescence readouts.
Hoechst exclusion assay
Cells were seeded at a density of 7000 cells/well in 96-well flat bottom black imaging plates (#655090; Greiner Bio-One™). Cells were allowed to attach for 8 h and then treated with 1 μM of the selected KI, 10 μM verapamil, or DMSO as negative control. Each condition was tested in triplicate wells. 24 h after seeding, 1 μg/mL Hoechst33342 was added followed by an additional 2-h incubation. Subsequently, all wells were aspirated and received fresh medium including the respective inhibitors, without Hoechst33342 and plates were placed in a Nikon Eclipse Ti confocal microscope equipped with an automated stage, temperature and CO
2-controlled incubator for live imaging and a Plan Apo ×20/0.75 NA objective (Nikon Instruments Inc., Melville, NY, USA). Total intensity of nuclear Hoechst33342 was calculated with CellProfiler [
28] after watershed segmentation in FIJI-ImageJ [
29]. The intensity values of three images from three replicate wells were averaged for each condition. Values of experimental groups were normalized to those of the DMSO control group.
Cells were reverse transfected with 50 nM SMARTpool siGENOME siRNAs (Dharmacon) using INTERFERin transfection reagent (Polyplus; 409–50). A mixture of siRNAs targeting all kinases in the human genome, diluted to a total concentration of 50 nM with a concentration for each individual siRNA ~ 0.05 nM was used as control (siKINASEpool). Medium was refreshed after 24 h. For SRB assays, 3000 cells/well were seeded in triplicate wells in 96-well flat bottom plates. For RT-qPCR, 120,000 cells/well were seeded in duplicate wells in 24-well plates. At 48 h and 72 h post-transfection, cells in 24-well plates were processed for RT-qPCR and cells in 96-well plates were incubated for an additional 72 h in presence of DMSO or paclitaxel and subsequently processed for SRB assay.
Cells in the exponential growth phase (70% confluent) were treated with colcemid (KaryoMax, #15212012, Gibco™) for 1-2 h at a final concentration of 0.1 μg/ml. Cells were then detached by trypsinization, collected, and treated with a hypotonic solution (75 mM KCl) for 15 min. Next, cells were fixed in Carnoy’s fixative (3:1 Methanol:Glacial acetic acid), washed three times and resuspended in 200 μL of Carnoy’s solution. Finally, metaphase chromosomes were prepared by dropping the cell suspension onto glass slides and mounted with ProLong™ Diamond Antifade mountant containing DAPI (Invitrogen, P36966, Waltham, MA, USA). Chromosomes and extrachromosomal DNA (ecDNA) were visualized with a Nikon Eclipse Ti2 confocal microscope, with a 60X objective and a 2 × digital magnification.
Kinase inhibitor screening
The L1200 library from Selleckchem® (Munich, Germany) was used, containing 760 KIs that were dissolved in DMSO or water at a concentration of 10 mM or 1 mM. 3000 cells/well were seeded in 96-well flat bottom plates. After 24 h cells were treated with DMSO only (0.1%), 1 μM KI and DMSO, or 1 μM KI in combination with 0.1 μM paclitaxel. 10 μM verapamil was used as a positive control. After 72 h, cells were fixed and analyzed using an SRB-assay. The KI library was screened in single technical replicates, and the experiment was repeated in two biological replicates. Rescreening of selected KIs with SRB assay was performed in duplicate technical replicates and the experiment was repeated in three biological replicates.
Bottom-up proteomics sample preparation
For bottom-up proteomics, peptides were prepared by lysis of cells with a 5% SDS solution and Roche cOmplete™ Mini EDTA-free Protease Inhibitor Cocktail (Merck Darmstadt, Germany), followed by sonication (10 min, every 30 s) (Bioruptor Pico Diagenode, Belgium). Next, protein lysates were quantified using a modified Pierce Micro BCA assay (Termo Fisher Scientifc Rockford, IL), and 5 μg of proteins were used for peptide digestion. Prior to digestion, proteins were reduced with 20 mM dithiothreitol (Merck Darmstadt, Germany) at 45 °C for 30 min, alkylated with 40 mM iodoacetamide (Merck Darmstadt, Germany) for 30 min at room temperature in the dark and acidified with 2.5% phosphoric acid. Finally, proteins were diluted with 90% methanol/100 mM triethylammonium bicarbonate (TEAB) (Merck Darmstadt, Germany) for efficient trapping in Micro S-Trap columns (ProtiFI, Farmingdale, NY, USA). Digestion was performed in the S-Trap overnight at 37 °C using Trypsin/Lys-C Mix Mass Spec Grade (Promega, Walldorf, Germany), followed by elution in 50 mM TEAB, 0.2% formic acid (FA) and 50% acetonitrile (ACN) (Merck Darmstadt, Germany). Eluted peptides were dry-evaporated and resuspended in 10% FA solution for subsequent tandem mass spectrometry analysis (LC–MS/MS), as described in
Supplementary Methods. The MS proteomics data have been deposited to the ProteomeXchange Consortium via PRIDE (accession number PXD040930).
RNA-seq
Total RNA was extracted from cells using the MiRVAna kit (Ambion, Thermo Fisher Scientific). Library preparation was performed using the Illumina TruSeq Stranded total RNA Library Prep gold Kit (20020598, Illumina Inc., San Diego, USA) and Agencount AMPure XP beads (Beckman Coulter, Brea, USA). Library concentration was determined using a Qubit dsDNA BR kit (Thermo Scientific), and the size distribution was examined with an Agilent Bioanalyzer. Libraries were paired-end sequenced (2 × 75 bp) on a NextSeq500 (Illumina). BclToFastq was used for the preprocessing of the raw data (trimming and filtering), then FASTQ files were checked for read quality and adapters were removed with Trimmomatic. The resulting reads were then mapped to the human reference genome (GRCh38) with STAR mapping tool (version 2.5.3a) and gene counts extracted with HTSeq. Raw RNA-sequencing data have been deposited on GEO database under accession number GSE228106.
Differential expression analysis
Differential expression analysis was performed with R package Deseq2 (version 1.22.2) for RNA-seq data and R package Limma (version 3.38.3) for proteomics data. In all datasets, black and white cases were allowed retaining the 0 for both parental and resistant cells. For RNA-seq data, only genes having a total sample count > 10 were retained. Volcano plots were generated using R package EnhancedVolcano, principal component analysis and sample correlation analyses were performed with plotPCA function of DeSeq2 R package and pheatmap R package (version 1.0.12). Finally, the R-package ggvenn was used to count the genes significantly upregulated in common among the cell lines and between RNA-seq and proteomics datasets.
Statistical analysis
Experiments were performed at least 3 times and data are expressed as mean ± SD of 3 experiments performed in triplicate, unless otherwise specified. To compare between two groups a two-tailed unpaired Student’s t-test was used. For multiple groups comparisons an ordinary one-way ANOVA multiple comparison test with Dunnet’s post-hoc test was used, unless otherwise specified in figure legends. Statistical significance was set at p < 0.05 and is indicated by *, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p < 0.0001.
Discussion
Chemoresistance is a major hurdle in the treatment of PDAC patients and understanding how to revert resistance to available treatments is of crucial importance. Gemcitabine resistance has been extensively investigated in PDAC, while little research has been performed on paclitaxel resistance [
12,
13]. We find that ABCB1 overexpression is a shared response to continued exposure to paclitaxel in three independent PDAC models and is not associated with gemcitabine resistance in those models. ABCB1 is involved in multidrug resistance of many solid cancers [
27,
40‐
42]. Taxols, in particular paclitaxel, are among the substrates of this transporter. However, the role of ABCB1 in resistance to paclitaxel has not been addressed in pancreatic cancer.
Interestingly, previous studies using pancreatic and other cancer cell lines have shown that ABCB1 is involved in gemcitabine resistance [
13,
22‐
26]. Our data do not support such a role: gemcitabine exposure did not induce ABCB1 expression (besides some increase at the mRNA level in some instances which was not mirrored by enhanced protein levels). Moreover, the induction of ABCB1 in paclitaxel resistant PDAC cells did not lead to cross-resistance to gemcitabine and inhibition of ABCB1 through verapamil did not alter GR cells sensitivity to gemcitabine. The aforementioned studies largely focused on HNF1A or PLK1 mediated gemcitabine resistance mechanisms that involved ABCB1 [
22,
23,
25] while Chen et al. did observe increased ABCB1 expression in SW1990 cells treated with gemcitabine [
24]. Notably, a different study in fact reported increased sensitivity to gemcitabine in a panel of cancer cell lines overexpressing ABCB1 [
43], and we find a similar trend in some of our models. Taken together, there is no direct evidence involving ABCB1 in gemcitabine resistance and our study indeed argues against such a mechanism in PDAC cells.
Previous studies showed that ABCB1 overexpression can be caused by gene locus amplification [
44]. Indeed, the expression of 4 genes belonging to locus 7q21.12 (
ABCB4,
ADAM22,
TP53TG1 and
SRI) is also increased in PR cells. We discriminate between de-repression and amplification by DNA-qPCR, further confirming locus amplification as the underlying mechanism. Gene amplification and chemoresistance have been linked to the presence of ecDNA [
33]. The continuous exposure to a drug like paclitaxel affecting the cell cycle could lead to genomic instability and therefore to the generation of ecDNA fragments [
45] but we did not find evidence for this. The fact that DNA-qPCR fold-change values were similar for the tested genes in the 7q21.12 locus, while RT-qPCR fold-change values for the same genes differed considerably, suggests that additional mechanisms, on top of locus amplification, may regulate paclitaxel-induced ABCB1 overexpression.
One such mechanism we considered, involves SRI/Sorcin.
SRI is located on the same gene locus as
ABCB1 and is often co-amplified in multidrug-resistant cancers [
32,
35]. In our experiments, SRI was the only candidate specifically induced by paclitaxel along with ABCB1 in the transcriptomics and proteomics datasets for Patu-T PR and Suit-2.028 PR. SRI encodes Sorcin, a calcium-binding protein that has been associated with increased tumor aggressiveness [
35,
46] and can induce ABCB1 expression in leukemia [
36]. In particular, Sorcin activates Protein Kinase A (PKA)-CREB1 signaling leading to activation of the ABCB1 promoter at cAMP-Response elements (CRE). Our results argue against this mechanism in PDAC paclitaxel resistance: SRI knockdown did not affect ABCB1 expression and failed to sensitize Patu-T PR cells to paclitaxel. It is possible that ABCB1 regulation is different in PDAC cells as compared to leukemic cells or the gradual increase in ABCB1 and SRI caused by paclitaxel differs from engineered SRI overexpression, as used by Yamagishi et colleagues [
36]. Moreover, the impact of sorcin on ABCB1 transcription is modulated by the presence of other mechanisms regulating calcium ion homeostasis, which may vary between cell types [
47].
Our findings in a small patient cohort show that ABCB1 is expressed in PDAC patients that after resection received at least one cycle of gemcitabine + nab-paclitaxel treatment. We observe a trend towards a correlation of ABCB1 expression with poor survival, although in this small cohort it was not statistically significant. It will be interesting to assess a larger cohort and compare to patients that have received a different therapy regimen such as FOLFIRINOX. Nevertheless, this indicates ABCB1 is expressed and may represent a target for chemosensitization to paclitaxel in PDAC patients. We confirmed its role as a candidate target for attenuating paclitaxel resistance in PDAC using gene silencing and the ABCB1 inhibitor, verapamil. Unfortunately, verapamil or other ABCB1 inhibitors have not performed well in the clinic. Reasons for the failure of ABCB1 inhibitors in clinical trials include lack of efficacy or dose-limiting toxicity [
21,
27].
Our KI screen identified novel candidate strategies to sensitize PDAC cells to paclitaxel without affecting growth in the absence of the chemotherapy. We found two compounds (i.e., apatinib and SGI-1776 free base) that have already been reported as ABCB1 inhibitors [
38,
39]. Interestingly, several compounds that were identified as ABCB1 inhibitors in different cancers failed to sensitize PDAC cells in our screen (i.e., erlotinib [
48], imatinib [
49], and nilotinib [
50]). Whether this reflects different inhibitory mechanisms or differences in potency is currently unknown, but it underscores the need to test each drug in the appropriate cancer type. We also identified KIs such as nazartinib, naquotinib, and derazantinib, that have not been previously implicated in PDAC paclitaxel resistance or ABCB1 inhibition. Using a Hoechst efflux assay, we confirmed these KIs act by inhibiting ABCB1. For some KIs and in some PR models, inhibition of ABCB1 expression was observed. However, this never reduced it to the nearly absent levels observed in CTR cells. Moreover, this effect did not correlate with the efficacy of the KIs in attenuating Hoechst efflux or cell proliferation in the presence of paclitaxel. This indicates that these KIs mainly act by suppressing ABCB1 efflux activity and it points to candidate strategies for combination therapies for chemosensitization. Interestingly, nazartinib, naquotinib, and derazantinib have already passed phase I clinical trials [NCT02108964 [
51], NCT02500927 [
52], NCT03230318 [
53]], suggesting that their safety profile is acceptable. Apatinib mono treatment showed in vivo tumor growth arrest in PDAC xenograft models [
54] and synergized with paclitaxel in gastric cancer murine models [
55,
56]. Apatinib also reverted breast cancer multidrug resistance in vivo [
38] and it is being tested in phase I and II clinical trials in combination with different chemotherapeutic agents, including paclitaxel [NCT02697838 [
57]]. Results from these trials will provide more information on the clinical relevance and feasibility of combining KIs with ABCB1 substrates to improve patient response to therapy.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.