Purpose:

Gemcitabine is most commonly used for pancreatic cancer. However, the molecular features and mechanisms of the frequently occurring resistance remain unclear. This work aims at exploring the molecular features of gemcitabine resistance and identifying candidate biomarkers and combinatorial targets for the treatment.

Experimental Design:

In this study, we established 66 patient-derived xenografts (PDXs) on the basis of clinical pancreatic cancer specimens and treated them with gemcitabine. We generated multiomics data (including whole-exome sequencing, RNA sequencing, miRNA sequencing, and DNA methylation array) of 15 drug-sensitive and 13 -resistant PDXs before and after the gemcitabine treatment. We performed integrative computational analysis to identify the molecular networks related to gemcitabine intrinsic and acquired resistance. Then, short hairpin RNA–based high-content screening was implemented to validate the function of the deregulated genes.

Results:

The comprehensive multiomics analysis and functional experiment revealed that MRPS5 and GSPT1 had strong effects on cell proliferation, and CD55 and DHTKD1 contributed to gemcitabine resistance in pancreatic cancer cells. Moreover, we found miR-135a-5p was significantly associated with the prognosis of patients with pancreatic cancer and could be a candidate biomarker to predict gemcitabine response. Comparing the molecular features before and after the treatment, we found that PI3K-Akt, p53, and hypoxia-inducible factor-1 pathways were significantly altered in multiple patients, providing candidate target pathways for reducing the acquired resistance.

Conclusions:

This integrative genomic study systematically investigated the predictive markers and molecular mechanisms of chemoresistance in pancreatic cancer and provides potential therapy targets for overcoming gemcitabine resistance.

Translational Relevance

Chemoresistance is one of the main reasons for poor prognosis of patients with pancreatic cancer. However, the mechanism of chemoresistance is still unclear, and there is a lack of effective treatment strategies to overcome chemoresistance. Here, we established 66 patient-derived xenografts (PDXs) to recapitulate the features of patients with pancreatic cancer. The multiomics analysis of gemcitabine-treated PDXs provided a comprehensive insight of the intrinsic and acquired drug resistance, and several candidate molecular markers and combinatorial targets were identified for gemcitabine resistance in pancreatic cancer.

Pancreatic cancer is one of the most lethal cancers with 5-year overall survival rate of less than 9% (1). It is predicted to be the second most common cause of cancer-related deaths by 2030 (2). Although many novel therapies, such as targeted drugs and immunotherapy, were effective in other cancer types, they failed to benefit patients with pancreatic cancer (3). Currently, the chemotherapeutic drug, gemcitabine, a deoxycytidine analogue inhibiting DNA replication and thereby arresting tumor growth, is still the most common choice for patients with pancreatic cancer. But its effectiveness is limited (4). Most of the treated patients changed from initially sensitive to resistant to gemcitabine treatment within few weeks (5). Therefore, it is of great importance to explore the biological mechanisms and biomarkers of gemcitabine resistance for improving pancreatic cancer therapy.

Chemoresistance broadly includes intrinsic and acquired resistance (5). Intrinsic resistance means that chemotherapy is ineffective from the start of treatment due to patient genetic factors or tumor microenvironment, such as the hypovascularized, dense tumor stroma in pancreatic cancer, which has been postulated to create a physical barrier for drug delivery of gemcitabine (5, 6), or low expression of hENT1 (7) and high expression of ATP-binding cassette (ABC)-transporter family (8), which are related to the transport and excretion of chemotherapeutic drugs, respectively. Whereas acquired resistance develops only after a certain time of exposure of tumor cells to anticancer drugs, due to genetic or epigenetic alterations in the cancer cells (6, 9). For example, the epithelial-to-mesenchymal transition of cancer cells during chemotherapy is related to acquired resistance (10). There are also many other molecular factors thought to be associated with chemotherapeutic resistance, such as the deregulation of miRNAs (11), long noncoding RNA (12), etc. But most of these studies are based on cell lines, which poorly reflect the heterogeneity and the microenvironment of pancreatic cancer, making them hard to be translated into clinics (13).

Recently, the application of patient-derived xenograft (PDX) models, which retain the genetic information, pathologic characteristics, and histologic structure of primary tumors, enabled a good in vivo model to evaluate the tumor heterogeneity and evolution, and to study drug responses more reliably than traditional cell lines (14, 15). For example, Izumchenko and colleagues (16) screened 92 PDXs established from various solid tumors administered clinically and correlated patient outcomes. They observed a significant association between drug responses in the patients and their corresponding PDXs in 87% of the therapeutic outcomes. Many studies tried to subtype pancreatic cancer by large-scale genomic profiling to identify candidate molecular mechanisms of drug resistance (17, 18). But it is difficult to examine the dynamic changes during the drug treatment from clinical samples. Therefore, PDXs are regarded as a promising model for identifying chemo-resistant biomarkers and mechanisms.

In this study, we successfully constructed 66 pancreatic cancer PDX mouse models. The PDX tumors before and after gemcitabine treatment were jointly profiled by whole-exome sequencing (WES), RNA sequencing (RNA-seq), miRNA sequencing (miRNA-seq), and DNA methylation array. The following integrative multiomics analysis and functional validation experiment revealed that MRPS5 and GSPT1 had strong effects on tumor cell proliferation, and CD55 and DHTKD1 contributed to gemcitabine resistance. Moreover, we found miR-135a-5p was significantly associated with the prognosis of patients with pancreatic cancer and was identified as a candidate molecular marker to predict gemcitabine response in clinics. Comparing the molecular profiles before and after gemcitabine treatment, we found that PI3K-Akt, p53, and hypoxia-inducible factor-1 (HIF1) pathways were significantly altered in multiple patients, which provide candidate target pathways for reducing the acquired resistance.

Cells lines and collection of pancreatic cancer specimens

All studies involving human tissue were performed with approval from the Institutional Ethical Review Boards of Peking Union Medical College Hospital (Beijing, P.R. China) with written informed consent obtained from patients and in accordance with the Declaration of Helsinki. The human pancreatic cancer cell lines, MIA PaCa-2, Panc-1, and BxPC-3, were purchased from the ATCC from 2016 to 2019 and cultured in DMEM or RPMI1640 with 10% FBS (HyClone) in a 5% CO2 cell culture incubator at 37°C. All cell lines were authenticated using high-resolution small tandem repeats profiling at Department of General Surgery Laboratory (Peking Union Medical College Hospital, Beijing, P.R. China). The mycoplasma of cells was examined by MycoSensor PCR Assay Kit (Beyotime, catalog No. C0301S; last tested in 2019). Cells were grown for 15 passages, and then replaced with fresh stocks. The clinical pancreatic cancer specimens were obtained from patients who received surgery in Peking Union Medical College Hospital (Beijing, P.R. China) and employed for the construction of a tissue microarray (TMA) and PDX models.

Generation of PDX models

Tumor tissues from patients with pancreatic cancer were collected in culture medium and kept on wet ice for engraftment within 24 hours after resection. The cancer tissues were carefully divided into three parts by using a surgical blade. One was flash frozen and stored at −80°C for genomic profiling, and one was fixed in 10% neutral-buffered formalin and paraffin embedded for histopathologic analysis. The third part was cut into size of 1 to 3 mm and implanted subcutaneously into the flank region of SCID female mice (4–5 weeks old, purchased from Shanghai Lidi Biotechnology Co., Ltd). Successfully engrafted tumor models were then passaged and banked after three passages in athymic nude (nu/nu) mice (4–5 weeks old, purchased from Shanghai Lidi Biotechnology Co., Ltd). These models were housed in SPF barrier facilities.

The treatment of gemcitabine on PDX models

The frozen tumors of PDX models were collected and quickly resuscitated in 37°C; water, and were then cut into small pieces at a diameter of 1 to 3 mm and inoculated subcutaneously in SCID mice. When the tumors grew to 500 to 800 mm3, the tumors were removed under sterile conditions and placed in Hank's Balanced Salt Solution. The tumors were cut into sizes of 1 to 3 mm and passed on to 25 Nu/Nu mice for drug efficacy. Twelve mice were randomly divided into two groups, six mice in each group, when the volume of the tumor reached 100 to 300 mm3. One was gemcitabine group administrated with gemcitabine, 60 mg/kg, i.p., every 4 days × 6 for 3 weeks, and another group was control group given physiologic saline the same way. The daily behavior of animals was monitored every day after administration. Weight and tumor volume of mice were measured every 3 days. The tumor volume was calculated using the formula (a × b × b)/2, a represents the major tumor axis and b represents the minor tumor axis. The tumor growth inhibition (TGI%) rate was calculated using the formula [1 − (Ti − T0)/(Ci − C0)] ×100. Ti and Ci were the average tumor volume on the day of investigation in the gemcitabine group and control group, respectively. T0 and C0 were the average tumor volume on day 0 in the gemcitabine group and control group, respectively. According to the value of TGI%, PDX models were divided into four groups. The models with TGI% >100, 80 to 100, 50 to 80, and <50 were divided into “sensitive,” “partial sensitive,” “partial resistant,” and “resistant” groups, respectively.

IHC analysis

CK7, CK19, MUC1, and Ki-67 (all from Proteintech Group) antibodies were used to measure CK7, CK19, MUC1, and Ki-67 protein level in primary pancreatic cancer tissues and PDXs by IHC staining as described previously (19). Two experienced pathologists assessed the result independently. The staining intensity was graded from 0 to 3 (negative, low, medium, and high, respectively). The staining extent (positive percentage) was scored from 0% to 100%, and the intensity score × percentage score ×100 was used as the final staining score.

ISH

We detected the miR-135a-5p in pancreatic cancer TMA by ISH. The digoxigenin-labeled miR-135a-5p detection probe (YD00611609-BEG) and the microRNA ISH Kit (339450) were purchased from Qiagen, and DAB Kit (D8001) was purchased from Sigma. The procedures were followed according to the manufacturer's protocol. Tissue sections were incubated with miR-135a-5p probe at 38°C overnight, treated for 1 hour at 37°C by biotinylated mouse antidigoxigenin the next day, and stained for 20 to 30 minutes by DAB. Two experienced pathologists assessed the result independently. The final staining score was calculated as the above IHC analysis.

High-content screening

For each target gene, we designed three short hairpin RNA (shRNA) sequences with different targets and constructed them into lentivirus plasmids. All shRNA sequences are provided in Supplementary Table S1. Then, these three plasmids were mixed in equal proportion for lentivirus packaging, which contains GFP gene sequence for fluorescence detection. The cancer cell suspension was inoculated into 96-well with 2,000 cells per well. When the cell fusion reached about 20% to 30%, the appropriate amount of virus was added according to multiplicity of infection value. After 2 to 3 days of infection, we observed the expression of GFP in the reporter gene of lentivirus. The cells whose fluorescence rate reached 70% to 90% were cultured until the fusion degree reached 70% to 90% and collected for further experiments. Cells infected with different lentivirus were inoculated into 96-well with 2,000 cells per well and given IC50 concentration gemcitabine and control treatment, respectively, with three repetitive wells in each group. Then, Celigo System (Nexcelom) was used to take photos and calculate the number of cells with green fluorescence in each scanned plate. The data were plotted and the cell proliferation curve for 5 days and the cell inhibition curve for 3 days were drawn. The experiment was repeated three times.

WES and data preprocessing

Exome capture was performed using the SureSelect Human All Exon V5 (Agilent Technologies) according to the manufacturer's instructions. Sequencing libraries were constructed on Illumina HiSeq 2500 in 2 × 125 bp paired-end runs. Output from Illumina software was processed by a robust preprocess pipeline to yield BAM files containing well-calibrated, aligned human reads with mouse reads removed. The preprocess pipeline includes concatenating the human (GRCh37) and murine (GRCm38) genome, aligning reads to the new reference genome using Burrows-Wheeler Aligner (BWA v0.7.5a), marking of duplicated reads using picard, base recalibration via Genome Analysis Toolkit (GATK v4.0.8.1), and separating the species-specific reads. The uniquely mapped reads were separated on the basis of which genome they are derived. For the reads that only had two mapped loci and uniquely mapped in each genome, we separated them according to the alignment score. All reads with identical alignment score or more than two mapped loci on both genomes were discarded. Germline mutation calling was performed using HaplotypeCaller. Finally, the variants that reported as pancreatic cancer mutational hotspots in Catalogue of Somatic Mutations in Cancer were used for subsequent analysis.

RNA-seq and data preprocessing

RNA library was constructed and sequenced using Illumina HiSeq 2000/2500 platform with paired-end reads in accordance with the manufacturer's instructions. RNA-seq reads' quality was first quantified using FastQC v0.11.5 (https://www.bioinformatics.babraham.ac.uk/projects/fastqc), and then reads were trimmed using Trimmomatic v0.36 (20). Output of trimmomatic was processed by a robust preprocess pipeline to yield BAM files containing well-calibrated, aligned human reads with mouse reads removed. The preprocess pipeline includes concatenating the human (GRCh37) and murine (GRCm38) genome, aligning reads to the new reference genome using STAR v2.6.0a (21), separating the species-specific reads, and obtaining the reads count using featureCounts (22).

miRNA-seq and data preprocessing

miRNA library was constructed and sequenced using Illumina Hiseq 2000/2500 platform with single-end reads in accordance with the manufacturer's instructions. Reads were quantified using FastQC v0.11.5, and then trimmed using Trimmomatic v0.36, aligned to human miRNA referenced in miRBase v20 (23) using bowtie2 v2.3.2 (24), and read counts were quantified with samtools v0.1.19.

RNA-seq data analysis

We calculated transcripts per million (TPM) to normalize gene expression and only genes with TPM > 0.3 were kept for further analysis. A differential gene expression analysis between resistant and sensitive PDXs was performed using Bioconductor package DESeq2 (v1.22.2) (25) with default parameters. To assess the changes of mRNAs before and after the treatment, we first calculated generalized fold change (GFOLD) score for each gene in each PDX sample. The mRNAs with absolute value of GFOLD score > 1 were considered as significantly changed for the treatment. Only mRNAs altered in more than 40% PDX samples were kept for subsequent analysis. Protein-protein interaction network was constructed using Bioconductor package STRINGdb (v1.16.0) (26). Package pathview (v1.16.7) (27) was used to perform enrichment analysis.

miRNA-seq data analysis

We calculated TPM to normalize read counts of miRNA. Only miRNAs with TPM greater than 1.0 were kept for further analysis. The differential expression analysis on miRNA was performed using DESeq2. In addition, the miRNAs with Padj < 0.05 were considered significantly differentially expressed between the two groups. To assess the changes of miRNAs before and after the treatment, we first calculated GFOLD score for each miRNA in each PDX sample. The miRNAs with absolute value of GFOLD score > 1.0 were considered as significantly changed for the treatment. Only miRNAs altered in more than 40% PDX samples were kept for subsequent analysis. The miRNA target genes were predicted using miRTarBase, miRWalk, and TargetScan. Then, the Spearman correlations between miRNA and its target genes were calculated. The cutoff for significant correlation was set as FDR < 0.05. Cytoscape was used to visualize the miRNA regulatory network.

DNA methylation array and data analysis

DNA methylation data were generated using Illumina Human Methylation450K BeadChip before and after gemcitabine treatment, respectively. We used Bioconductor package ChAMP v2.12.4 (28) to extract methylation score β for each probe, normalized and corrected across samples and probes, and found differentially methylated positions (DMP) between resistant and sensitive groups. To evaluate the change of methylation during treatment, we calculated the differential β value at the same position before and after treatment for each sample. We considered that methylation position with absolute value of differential β > 0.1 had significantly changed during treatment. Only methylation positions altered in more than 40% PDX samples was kept for subsequent analysis.

Statistical analysis

Statistical comparisons between treatment groups were done using Student t test (two-tailed) and analyzed using one-way ANOVA. All statistical tests were performed using the Statistical Program for Social Sciences (SPSS16.0 for Windows) and were two-sided. Differences were considered statistically significant at P < 0.05. Significance levels by P value: ****, P < 0.0001; ***, P < 0.001; **, P < 0.01; * P < 0.05.

Data availability

The DNA methylation array data are submitted to NCBI Gene Expression Omnibus database (GSE165764). All the sequencing data are submitted to both NCBI SRA database (SRP303224) and GSA database (CRA002096).

Histologic and genomic characterization of PDXs

To analyze the biomarkers and mechanism of gemcitabine resistance, we established a large cohort of PDX mouse models (Fig. 1A). From August 2012 to August 2014, a total of 136 pancreatic cancer patients' tissues were used to establish PDX models and 66 of them were successful (success rate 49%). Among these PDXs, 63 were derived from the primary tumor tissues and the remaining three from metastases. The clinicopathologic features showed that majority of the patients had aggressive phenotypes, such as poor differentiation and advanced tumor–node–metastasis stage (Supplementary Table S2). The established PDXs demonstrated high similarities with the original tumors according to the histologic features, particularly concerning the tumor cell morphology, tubule formation, and associated stroma (Fig. 1B). The putative pancreatic cancer epithelial biomarkers (CK7 and CK19) and clinical biomarkers (Ki67 and MUC1) were also consistently retained along the different passages (Fig. 1B).

Figure 1.

Histologic and molecular characterization of PDXs. A, Flow chart of PDX construction, gemcitabine pharmacodynamics experiment, sensitivity grouping, and multiomics detection. B, Hematoxylin and eosin (H&E) staining (left) and IHC staining of CK7, CK19, Ki67, and MUC1 in clinical cancer tissues and PDX tumors tissues from passage 1 to passage 4 (magnification, 10 ×; scale bars, 100 μm; right). C, Summary of mutations in the 28 PDX samples. The mutation types (right) of KRAS, TP53, CDKN2A, and SMAD4 together with their frequencies (left) in our dataset were presented. D and E, The mutant allele distributions of KRAS and TP53. F, The mutation percentages of five common signaling pathways in pancreatic cancer in our PDX models and another two public datasets TCGA and UTSW.

Figure 1.

Histologic and molecular characterization of PDXs. A, Flow chart of PDX construction, gemcitabine pharmacodynamics experiment, sensitivity grouping, and multiomics detection. B, Hematoxylin and eosin (H&E) staining (left) and IHC staining of CK7, CK19, Ki67, and MUC1 in clinical cancer tissues and PDX tumors tissues from passage 1 to passage 4 (magnification, 10 ×; scale bars, 100 μm; right). C, Summary of mutations in the 28 PDX samples. The mutation types (right) of KRAS, TP53, CDKN2A, and SMAD4 together with their frequencies (left) in our dataset were presented. D and E, The mutant allele distributions of KRAS and TP53. F, The mutation percentages of five common signaling pathways in pancreatic cancer in our PDX models and another two public datasets TCGA and UTSW.

Close modal

To further validate the models, five PDXs and the matched primary patient tumors were characterized genetically by examining somatic mutation hotspots from WES data. After removing the reads mapped to mouse genome based on a previous study (29), we analyzed the mutational hotspots between primary tumor and PDX tissues with high confidence. Three of the five models exhibited high concordance (88.73%–88.96%) with primary tumors after three passages. The other two PDX models retained a modest consistency (62.17%–64.83%) and a few novel somatic mutations were observed after five passages (Supplementary Fig. S1; Supplementary Table S3). As reported in the previous studies, the disappearance of primary mutations and the emergence of new variants are commonly observed and may be caused by the natural selection of the host environment, and it was frequently observed that more passages generally had a lower consistency, suggesting that PDX models with close passage should be better selected for pharmacodynamics study to ensure the consistency with clinical results as much as possible.

On the basis of the 28 sequenced PDXs (28/66 PDXs were resistant or sensitive to gemcitabine treatment, see details in the following sections), we confirmed the models can recapitulate the molecular features of pancreatic cancer by analyzing the mutational spectrum of PDXs (Supplementary Table S4). Most PDX samples harbored high-frequency mutation genes reported in other pancreatic cancer–related databases, such as KRAS and TP53 (Fig. 1C). KRAS was mutated in 28 of 28 PDXs (100%) with 14 G12D mutations (50%), 11 G12V (39%), 2 G12R (7%), and 1 Q61L (4%; Fig. 1D). TP53 mutations were detected in 21 of 28 PDXs (75%; Fig. 1C). In 11 of 28 (39%) cases, TP53P72R was observed in all the mutated models and one model harbored multiple TP53 mutations (Fig. 1E). We also compared the mutational spectrum of five most altered signaling pathways in the studied PDX models and in another two clinical cohorts from The Cancer Genome Atlas (TCGA) and UT Southwestern (UTSW; Fig. 1F; Supplementary Table S5). The mutation rates of RAS and p53 signaling pathways were extremely high in all the three datasets, and the Notch and TGFβ signaling pathways showed high concordance between PDXs and UTSW dataset than TCGA dataset. Interestingly, the mutation rate of DNA damage pathway in PDX was much higher than that the other two datasets. It is likely that the tumor was subjected to selection pressure after being transplanted into mice. Generally, PDX models have high concordant mutational spectrum with large-scale clinical pancreatic cancer datasets.

Pharmacodynamics experiment of gemcitabine in PDX models

To study drug resistance, the 66 established PDXs were treated with gemcitabine for 3 weeks. The drug responses were evaluated by TGI% (Supplementary Fig. S2). As shown in Fig. 2A, these PDX models displayed a range of TGI% to gemcitabine from −16.2 to 170.06 and were classified into four groups: sensitive group (n = 16), partial sensitive group (n = 19), partial resistant group (n = 18), and resistant group (n = 13) by TGI% and the median TGI% of these four groups was 120.71, 92.44, 67.35, and 36.23, respectively (Fig. 2B and C; Supplementary Table S6). To determine whether these subgroups corresponded with individual patient treatment responses, retrospective clinical follow-up information was collected and six patients with distant metastasis when diagnosed and one patient who died within 1 month after surgery were excluded. We integrated the patients corresponding to the sensitive group and partial sensitive group PDXs into one group, namely high TGI% group, and integrated the patients corresponding to the partial resistant group and resistant group into another group, namely low TGI% group. The Kaplan–Meier survival analysis showed a statistically significant difference in median overall survival between high TGI% and low TGI% groups (54 vs. 16 months; P = 0.0158; Fig. 2D). The clinicopathologic factor analysis showed poor N staging was correlated with low TGI% (P = 0.001; Supplementary Table S7). Moreover, univariate and multivariate analysis indicated low TGI% was an independent risk factor for pancreatic cancer prognosis (P = 0.011 and 0.038, respectively; Supplementary Table S8). These results demonstrated that PDXs may serve as a predictive model for response to gemcitabine therapy and as a marker for predicting prognosis of patients.

Figure 2.

Pharmacodynamics of gemcitabine in PDX models. A, TGI% of 66 PDXs treated with gemcitabine. B, The four groups with different sensitives to gemcitabine defined by TGI%. Green represents sensitive group, purple for the partly sensitive group, brown for the partly resistant group, and red for the resistant group. C, Representative tumor growth curves (top) and TGI% curves (bottom) of PDXs in the four groups. D, The Kaplan–Meier survival analysis between patients with high TGI% (sensitive and partly sensitive groups) and low TGI% (resistant and partly resistant groups) in PDXs.

Figure 2.

Pharmacodynamics of gemcitabine in PDX models. A, TGI% of 66 PDXs treated with gemcitabine. B, The four groups with different sensitives to gemcitabine defined by TGI%. Green represents sensitive group, purple for the partly sensitive group, brown for the partly resistant group, and red for the resistant group. C, Representative tumor growth curves (top) and TGI% curves (bottom) of PDXs in the four groups. D, The Kaplan–Meier survival analysis between patients with high TGI% (sensitive and partly sensitive groups) and low TGI% (resistant and partly resistant groups) in PDXs.

Close modal

Molecular features of gemcitabine intrinsic resistance in pancreatic cancer

To clarify the molecular characteristics of drug-resistant and -sensitive models, multiomics techniques (including WES, RNA-seq, miRNA-seq, and DNA methylation array) were used to examine the molecular features of PDX tissues from sensitive (n = 15) and resistant (n = 13) groups. The WES analysis showed no correlation between drug responses and the highly mutated genes, such as KRAS, TP53, CDKN2A, SMAD4, LRP1B, and KMT2C (Supplementary Fig. S3).

The RNA analysis identified 109 differentially expressed genes (DEmRNA), of which 62 mRNAs were upregulated and 47 mRNAs were downregulated in the resistant group compared with the sensitive group (Fig. 3A; Supplementary Table S9). Several genes have been reported related to chemoresistance, such as SGK1 (30), STC1 (31), CD55 (32), and INPP4B (33). Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis showed that ABC-transporters, amino sugar and nucleotide sugar metabolism, and AGE-RAGE signaling pathway in diabetic complications were the top three enriched pathways, however, with only two or three genes differentially expressed in each pathway (Supplementary Fig. S4; Supplementary Table S9). We then analyzed the enrichment of DEmRNAs in literature-supported gemcitabine-related pathways and found that several genes of p53 signaling pathway were downregulated and genes related to DNA repair pathway were upregulated in the resistant group (Fig. 3B). The p53 signaling pathway is involved in the development of many cancers. Activation of p53 signaling pathway promotes tumor cell apoptosis, which may increase drug sensitivity. It is known that gemcitabine can disrupt the proliferation of tumor cells by blocking the replication of DNA. Activation of the DNA repair pathway inhibits the death of tumor cells induced by gemcitabine, and may cause development of drug resistance.

Figure 3.

Molecular features of gemcitabine intrinsic resistance in pancreatic cancer. A, Heatmaps of DEmRNAs, DEmiRNAs, and DEDNAmethy between the sensitive and resistant groups before the treatment. B, The enrichment of DEmRNAs in p53 signaling pathway and DNA repair pathway. The y-axis indicates the expression profile (log TPM) of these genes in the resistant and sensitive groups. The statistical significance levels of the differences between the two groups are marked at the top. C, The resistance-related regulatory network constructed according to the DEmRNAs (circular) and DEmiRNAs (diamond). The color of nodes indicates the log fold change between the resistant group and sensitive group (red for positive and blue for negative). The size of nodes indicates the number of neighbors (defined by Spearman correlation). The edges indicates the correlations (red for positive and blue for negative) between DEmRNAs and their target genes.

Figure 3.

Molecular features of gemcitabine intrinsic resistance in pancreatic cancer. A, Heatmaps of DEmRNAs, DEmiRNAs, and DEDNAmethy between the sensitive and resistant groups before the treatment. B, The enrichment of DEmRNAs in p53 signaling pathway and DNA repair pathway. The y-axis indicates the expression profile (log TPM) of these genes in the resistant and sensitive groups. The statistical significance levels of the differences between the two groups are marked at the top. C, The resistance-related regulatory network constructed according to the DEmRNAs (circular) and DEmiRNAs (diamond). The color of nodes indicates the log fold change between the resistant group and sensitive group (red for positive and blue for negative). The size of nodes indicates the number of neighbors (defined by Spearman correlation). The edges indicates the correlations (red for positive and blue for negative) between DEmRNAs and their target genes.

Close modal

The miRNA analysis identified 29 differentially expressed miRNAs (DEmiRNA), 24 upregulated and five downregulated, by comparing the resistant with the sensitive group (Fig. 3A; Supplementary Table S9). Then, we calculated the Spearman correlations between DEmiRNAs and their target genes (derived from miRTarBase, miRWalk, and TargetScan) in DEmRNAs and selected the correlated pairs to construct a resistance-related miRNA-centered regulatory network (Fig. 3C; Supplementary Table S9). In this network, miR-299-5p, miR-381-3p, miR-934, miR18a-5p, miR127-5p, and miR-411-5p and SGK1 and STC1 were hub nodes with many correlations, which will provide important clues for further exploring the mechanism of chemoresistance in pancreatic cancer.

To further explore the role of these DEmRNAs, we detected the basic expression levels of 62 upregulated genes in the resistant-related network in three pancreatic cancer cell lines, MIA PaCa-2, PANC-1, and BxPC-3. The results showed that 43, 48, and 53 genes had high expression level in MIA PaCa-2, PANC-1, and BxPC-3, respectively (Supplementary Table S10). Then they were knocked down by lentivirus system and high-content screening (HCS) was used to detect their effect on proliferation and gemcitabine sensitivity in pancreatic cancer cells. Results showed that the knockdown of 30.2% (13/43), 35.4% (17/48), and 35.8% (19/53) genes could inhibit the proliferation of MIA PaCa-2, PANC-1, and BxPC-3 significantly, respectively (Fig. 4A–C). Among them, the knockdown of MRPS5 and GSPT1, which have been reported to promote cell proliferation in other cancer types (34), exhibited inhibitory effects in all the three cell lines (Fig. 4D), and the knockdown of 27.9% (12/43), 14.6% (7/48), and 28.3% (15/53) genes was able to significantly increase the sensitivity of MIA PaCa-2, PANC-1, and BxPC-3 to gemcitabine, respectively (Fig. 4E–G). Among these genes, the knockdown of CD55 and DHTKD1 increased chemosensitivity in all the three cell lines (Fig. 4H). CD55 was reported to promote self-renewal and cisplatin resistance in endometrioid tumors (32). DHTKD1 is a mitochondrial protein encoding gene and involved in energy production in mitochondria and amino acid metabolism (35, 36). But their effects on chemoresistance of pancreatic cancer remain unclear and deserve future exploration.

Figure 4.

Functional validations of DEmRNAs mediating gemcitabine resistance. A–C, The HCS results of proliferation rates when upregulated mRNAs in the resistant group were knocked down by lentivirus in pancreatic cancer cell lines MIA PaCa-2, PANC-1, and BxPC-3, respectively. D, The Venn diagram of the genes that significantly affect cell proliferation in the three cell lines. E–G, The HCS results of gemcitabine inhibition rates when upregulated mRNAs in the resistant group were knocked down by lentivirus in the cell lines. H, The Venn diagram of the genes that significantly affect cell chemosensitivity. I, The expression level of miR-135a-5p, miR-376a-5p, and miR-493-5p in the resistant and sensitive groups' PDXs tissues validated by qRT-PCR. J and K, The Kaplan–Meier survival analysis between high and low miR-135a-5p group in all patients with pancreatic cancer and in the subgroup with gemcitabine treatment. *, P < 0.05; **, P < 0.01; ***, P < 0.001.

Figure 4.

Functional validations of DEmRNAs mediating gemcitabine resistance. A–C, The HCS results of proliferation rates when upregulated mRNAs in the resistant group were knocked down by lentivirus in pancreatic cancer cell lines MIA PaCa-2, PANC-1, and BxPC-3, respectively. D, The Venn diagram of the genes that significantly affect cell proliferation in the three cell lines. E–G, The HCS results of gemcitabine inhibition rates when upregulated mRNAs in the resistant group were knocked down by lentivirus in the cell lines. H, The Venn diagram of the genes that significantly affect cell chemosensitivity. I, The expression level of miR-135a-5p, miR-376a-5p, and miR-493-5p in the resistant and sensitive groups' PDXs tissues validated by qRT-PCR. J and K, The Kaplan–Meier survival analysis between high and low miR-135a-5p group in all patients with pancreatic cancer and in the subgroup with gemcitabine treatment. *, P < 0.05; **, P < 0.01; ***, P < 0.001.

Close modal

miRNAs are regarded as promising novel biomarkers. We validated three miRNAs which were upregulated in the resistant groups by qRT-PCR (Fig. 4I). Among them, the expression of miR-135a-5p was the most abundant. Then, we detected its expression on a TMA containing 174 surgical tumor tissues from patients with pancreatic cancer by IHS histochemistry. The Kaplan–Meier survival analysis showed that high level of miR-135a-5p significantly indicated poor overall survival (19 vs. 24 months; P = 0.0365; Fig. 4J). Moreover, high level of miR-135a-5p also significantly indicated poor overall survival in the subgroup receiving gemcitabine treatment after surgery (P =0.006; Fig. 4K). Univariate and multivariate analysis showed that it was an independent risk factor for the prognosis of patients with pancreatic cancer (Table 1). Therefore, miR-135a-5p may serve as a biomarker for gemcitabine response and prognosis of pancreatic cancer.

Table 1.

Univariate and multivariate analyses of patients' prognosis factors.

UnivariateMultivariate
VariablesnHR (95% CI)P valueHR (95% CI)P value
Gender   0.238   
 Male 96    
 Female 78 0.785 (0.526–1.173)    
Age   0.860   
 <60 77    
 ≥60 97 0.964 (0.644–1.444)    
Smoke   0.091   
 No 64    
 Yes 110 0.705 (0.470–1.058)    
Drink   0.197   
 No 33    
 Yes 141 0.727 (0.448–1.180)    
Diabetes   0.880   
 No 30    
 Yes 144 0.96 (0.565–1.631)    
CA199 (before surgery)   0.573  0.997 
 Normal 30   
 High 119 1.182 (0.662–2.110)  0.999 (0.457–2.182)  
Location   0.967  0.507 
 Head 102   
 Body/tail 68 1.009 (0.667–1.525)  0.813 (0.441–1.499)  
Differential degree   0.001  0.012 
 Low 59   
 High/moderate 115 0.504 (0.333–0.761)  0.458 (0.25–0.84)  
Tumor staging   0.484  0.120 
 T1 10   
 T2/T3 163 1.381 (0.560–3.405)  5.319 (0.647–43.750)  
Lymph node staging   0.047  0.519 
 N0 74   
 N1/N2 94 1.533 (1.006–2.334)  1.203 (0.685–2.112)  
R state   0.539  0.676 
 R0 122   
 R1 14 1.262 (0.601–2.651)  0.794 (0.270–2.338)  
miR-135a-5p level   0.043  0.038 
 Low 128   
 High 46 1.593 (1.015–2.499)  1.889 (1.037–3.442)  
UnivariateMultivariate
VariablesnHR (95% CI)P valueHR (95% CI)P value
Gender   0.238   
 Male 96    
 Female 78 0.785 (0.526–1.173)    
Age   0.860   
 <60 77    
 ≥60 97 0.964 (0.644–1.444)    
Smoke   0.091   
 No 64    
 Yes 110 0.705 (0.470–1.058)    
Drink   0.197   
 No 33    
 Yes 141 0.727 (0.448–1.180)    
Diabetes   0.880   
 No 30    
 Yes 144 0.96 (0.565–1.631)    
CA199 (before surgery)   0.573  0.997 
 Normal 30   
 High 119 1.182 (0.662–2.110)  0.999 (0.457–2.182)  
Location   0.967  0.507 
 Head 102   
 Body/tail 68 1.009 (0.667–1.525)  0.813 (0.441–1.499)  
Differential degree   0.001  0.012 
 Low 59   
 High/moderate 115 0.504 (0.333–0.761)  0.458 (0.25–0.84)  
Tumor staging   0.484  0.120 
 T1 10   
 T2/T3 163 1.381 (0.560–3.405)  5.319 (0.647–43.750)  
Lymph node staging   0.047  0.519 
 N0 74   
 N1/N2 94 1.533 (1.006–2.334)  1.203 (0.685–2.112)  
R state   0.539  0.676 
 R0 122   
 R1 14 1.262 (0.601–2.651)  0.794 (0.270–2.338)  
miR-135a-5p level   0.043  0.038 
 Low 128   
 High 46 1.593 (1.015–2.499)  1.889 (1.037–3.442)  

Note: P < 0.05 indicates the statistical significance of differences.

Abbreviation: CI, confidence interval.

Molecular signatures and pathways altered after gemcitabine treatment

Molecular alteration during chemotherapy may contribute to acquired chemoresistances. To explore the molecular signatures of PDXs during gemcitabine therapy, the multiomics data for each sample before and after treatment were compared. GFOLD was used to evaluate the change of RNA and miRNA (37). We identified 156 DEmRNAs frequently increased or decreased after gemcitabine treatment (Fig. 5A). These genes were enriched in complement and coagulation cascades, p53 signaling pathway, and HIF1 signaling pathway (Fig. 5B). To uncover the interactions between these genes, a gene network was constructed on the basis of STRING database (Supplementary Fig. S5). A subnetwork consisting of GDF15, H2AFX, RRM2, CDKN1A, CDT1, and HISTH3H, which were upregulated after gemcitabine treatment, may mediate drug response by regulating DNA replication and DNA repair (Fig. 5C). GDF15 has been reported to suppress proapoptotic activity of macrophage which leads to tumor progression (38). CDKN1A, a cyclin-dependent kinase inhibitor regulating the cell cycle in the G1-phase, plays important role in DNA replication and DNA damage repair in S-phase (39). CDT1, which assembles at replication origins, can form prereplicative complexes during the G1-phase of the cell cycle to involve DNA replication (40). H2AFX, HIST1H3H, and HIST1H1D are members of histone family and play a part in transcription regulation, DNA repair, DNA replication, and chromosomal stability (41, 42), and overexpression of RRM2 can also cause gemcitabine resistance (43). These results suggested that gemcitabine responses are affected by the deregulation of DNA replication and DNA repair pathways after the drug treatment.

Figure 5.

Molecular signatures and pathways altered after gemcitabine treatment. A and B, The expression heatmap and the enriched KEGG pathways for DEmRNAs. C, DNA replication- and DNA repair–related protein-protein interaction network of DEmRNAs. D and E, The expression heatmap of DEmiRNAs and the enriched KEGG pathways of DEmiRNAs' target genes. F, The DEmiRNA-DEmRNA regulatory network. The color of each node indicates the differential expression direction of the gene or miRNA after the treatment (red for upregulated and blue for downregulated). The edges indicate the Spearman correlations (red for positive and blue for negative) between DEmRNAs and their target genes. G and H, The methylation heatmap and the enriched KEGG pathways for DEDNAmethy site–associated genes.

Figure 5.

Molecular signatures and pathways altered after gemcitabine treatment. A and B, The expression heatmap and the enriched KEGG pathways for DEmRNAs. C, DNA replication- and DNA repair–related protein-protein interaction network of DEmRNAs. D and E, The expression heatmap of DEmiRNAs and the enriched KEGG pathways of DEmiRNAs' target genes. F, The DEmiRNA-DEmRNA regulatory network. The color of each node indicates the differential expression direction of the gene or miRNA after the treatment (red for upregulated and blue for downregulated). The edges indicate the Spearman correlations (red for positive and blue for negative) between DEmRNAs and their target genes. G and H, The methylation heatmap and the enriched KEGG pathways for DEDNAmethy site–associated genes.

Close modal

We also performed the same analysis on miRNA expressions (Fig. 5D). Interestingly, a few DEmiRNAs changed in opposite directions of their known functions in the resistant and sensitive groups, such as miR-214-3p, miR-214-5p, miR-199a-3p, miR-411-3p, and miR-381-3p (Fig. 5D). This may be due to the different dominant cell populations in the drug-resistant and -sensitive groups. For example, the miR-214-3p and miR-214-5p, the known tumor suppressors (44), their downregulation in the resistant group after the treatment may be caused by the reduced cancer cell populations with high expression of miR-214, which are sensitive to gemcitabine. The predicted target genes of DEmiRNAs were enriched in PI3K-Akt, p53, HIF1, and some other cancer-related pathways by KEGG analysis (Fig. 5E). Notably, the DEmRNAs after gemcitabine treatment were also enriched in p53 and HIF1 signaling pathways. Therefore, we focused on genes in PI3K-Akt, p53, and HIF1 signaling pathway to construct DEmiRNA-DEmRNA regulatory network (Fig. 5F). CDKN1A, RRM2, EGLN3, and PDK1 are regulated by various miRNAs and may contribute to the acquired resistance.

To identify DMPs, we calculated the differential β value at the same probe before and after treatment for each sample. Differential β value greater than 0.1 meant hypermethylation, and less than −0.1 meant hypomethylation (Fig. 5G). DMPs found in more than 40% PDX samples were used for enrichment analysis. There were four DMP changes in opposite directions in the resistant and sensitive groups, one was hypermethylated in the sensitive group and the others were hypomethylated (Fig. 5G). KEGG pathway analysis of genes in DMPs revealed an enrichment of signaling pathways similar to the results of RNA and miRNA analyses, like MAPK, PI3K-Akt, Ras, and Wnt signaling pathways (Fig. 5H), indicating these pathways are involved in acquired gemcitabine resistance.

Previously, researchers attempted to use pancreatic cancer molecular subtypes to predict chemoresistance. We measured the gene signatures from Collisson et al. in our PDX models (Fig. 6A; ref. 45). On the basis of the expressions before the treatment, 71.4% (20/28) PDXs belonged to classical subtype, 28.6% (8/28) PDXs belonged to quasi-mesenchymal subtype (QM-PDA), and no exocrine-like subtype (Fig. 6B; Supplementary Table S11). Collisson et al. reported that QM-PDA subtype was more sensitive to gemcitabine than classical subtype, but we found no significant difference between different subtypes of our PDX models (Fig. 6B). Then, we analyzed the molecular subtypes after the treatment. We found 22 of 28 PDXs kept the same subtypes and six PDXs changed after the treatment. Four PDXs changed from classical type to QM-PDA type, one from QM-PDA type to classical type, and one from QM-PDA type to exocrine type (Fig. 6B; Supplementary Table S11). Notably, all the six PDXs with subtype transition were in the sensitive group. This subtype change indicated that sensitive cells have been suppressed by gemcitabine and the remaining resistant cells became the major populations. Then, we also tested the gene signatures reported by Bailey et al. (Fig. 6C; ref. 17). The subtypes in our PDXs had no correlation with gemcitabine response, either (Fig. 6D). After the treatment, only one model changed from pancreatic progenitor subtype to squamous subtype, and the others remained unchanged (Fig. 6D; Supplementary Table S12). These results demonstrated that known subtyping methods may not be suitable for evaluating the pharmacodynamics of pancreatic cancer.

Figure 6.

PDX subtypes based on gene signatures and their correlations with gemcitabine responses. A, The subtypes distribution of PDX models in the sensitive and resistant groups before and after gemcitabine treatment according to the classification method of Collison et al. B, Subtype changes of PDX model in the sensitive and resistant group after gemcitabine treatment according to the classification method of Collison et al. C, The subtypes distribution of PDX models in the sensitive and resistant group before and after gemcitabine treatment according to the classification method of Bailey et al. D, Subtype changes of PDX model in the sensitive and resistant group after gemcitabine treatment according to the classification method of Bailey et al. E, The result of gemcitabine sensitivity prediction signature of Tiriac et al. applied into the RNA-seq data of our PDXs. F, The comparation of sensitive scores between sensitive and resistant groups. G, The comparation of sensitive scores before (labeled as pre) and after (post) gemcitabine treatment. ns, not significant.

Figure 6.

PDX subtypes based on gene signatures and their correlations with gemcitabine responses. A, The subtypes distribution of PDX models in the sensitive and resistant groups before and after gemcitabine treatment according to the classification method of Collison et al. B, Subtype changes of PDX model in the sensitive and resistant group after gemcitabine treatment according to the classification method of Collison et al. C, The subtypes distribution of PDX models in the sensitive and resistant group before and after gemcitabine treatment according to the classification method of Bailey et al. D, Subtype changes of PDX model in the sensitive and resistant group after gemcitabine treatment according to the classification method of Bailey et al. E, The result of gemcitabine sensitivity prediction signature of Tiriac et al. applied into the RNA-seq data of our PDXs. F, The comparation of sensitive scores between sensitive and resistant groups. G, The comparation of sensitive scores before (labeled as pre) and after (post) gemcitabine treatment. ns, not significant.

Close modal

Tiriac et al. constructed gemcitabine sensitivity signatures by correlating pancreatic cancer organoid transcriptional profiles with the pharmacotyping results (46). We applied this signature into the RNA-seq data of our PDXs (Fig. 6E; Supplementary Table S13). The results showed that the sensitive scores were higher in the gemcitabine-sensitive group than that in the resistant group, but without significant difference (Fig. 6F). However, the sensitive scores were significantly reduced after gemcitabine treatment (Fig. 6G), which is consistent with the fact that the resistance of PDX models increases after gemcitabine treatment. These results showed that gemcitabine sensitivity signature screened by organoid was consistent with that of our PDXs to a certain degree.

Because of lack of early markers and symptoms, most of the patients with pancreatic cancer are diagnosed at advanced stages. The current standard of care for advanced pancreatic cancer is chemotherapy, such as gemcitabine. However, the clinical response is still limited. Overcoming gemcitabine resistance is crucial to improve the treatment of pancreatic cancer.

In this study, we generated 66 PDXs and verified that the histologic and molecular features of primary tumors are basically conserved through serial engraftment in PDXs. After gemcitabine treatment, we found that some PDXs responded well to the treatment, while others responded poorly. When analyzing the clinical data of these PDXs, we found that the prognosis of patients with higher TGI% in corresponding PDXs was better than that of patients with lower TGI% in corresponding PDXs, indicating that PDXs could effectively replicate the clinical outcomes of patients. There are some studies consistent with our results (14, 47), and Izumchenko et al. found a significant association between drug responses in patient with a variety of solid tumors and corresponding PDXs in 87% of the therapeutic outcomes (16). Of course, the sensitivity of gemcitabine is also related to pharmacokinetics or pharmacodynamics parameters in PDXs. One of the limitations of this study is that we did not detect the pharmacokinetics or pharmacodynamics of gemcitabine in these PDXs.

To analyze molecular features related to intrinsic and acquired resistance, we obtained WES, RNA-seq, miRNA-seq, and DNA methylation data of these PDXs before and after gemcitabine treatment. The WES analysis found no correlation between the drug responses and the mutations of frequently mutated genes, including KRAS, TP53, CDKN2A, SMAD4, LRP1B, and KMT2C. This result is consistent with previous studies that gene mutations confer few differences in gemcitabine responses (14, 48, 49). On the basis of the candidates identified by the omics data, lentivirus-based HCS methods validated that the knockdown of MRPS5 and GSPT1 significantly inhibited cell proliferation and CD55 and DHTKD1 strongly regulated gemcitabine resistance. MiR-135a-5p upregulated in the resistant group was verified to be significantly associated with the prognosis of patients with pancreatic cancer in our cohort and low expression of miR-135a-5p has a better overall survival in gemcitabine-treated subgroup, indicating its potential to predict the gemcitabine response. These results demonstrated the PDXs are great models to reveal the mechanisms and biomarkers of gemcitabine resistance in pancreatic cancer.

For the acquired gemcitabine resistance, we found that there were few genetic changes of PDX models after treatment, which again suggests that mutations may not contribute to the gemcitabine resistance in pancreatic cancer. However, the mRNA and miRNA expressions and DNA methylations changed significantly after the treatment. Interestingly, the functional enrichment analysis showed that DNA replication, DNA repair, p53, HIF1, and PI3K-Akt signaling pathways changed significantly after gemcitabine treatment, providing therapeutic targets to reduce gemcitabine resistance. It has been reported that knockdown of replication stress response–related genes, which could sense and protect the DNA damage or replication blocks induced by gemcitabine, can increase the effect of gemcitabine (50). Recently, many studies have tried to predict the anticancer drug responses by molecular subtyping. However, we found that several existing subtyping methods could not predict gemcitabine responses in our PDX models (17, 45), indicating that we may need to focus more on individualized treatment for patients with pancreatic cancer (49).

In conclusion, we established PDX models of pancreatic cancer to recapitulate molecular features of patients with pancreatic cancer. The multiomics analysis of PDXs before and after gemcitabine treatment provided systematic sight into the intrinsic and acquired drug resistance. These results provide important candidates to discover novel biomarkers and develop combinatorial treatment strategies for pancreatic cancer.

No disclosures were reported.

G. Yang: Data curation, formal analysis, validation, visualization, methodology, writing–original draft. W. Guan: Data curation, formal analysis, investigation, visualization, methodology, writing–original draft. Z. Cao: Resources, data curation, validation, methodology. W. Guo: Data curation, software, formal analysis, visualization, methodology, writing–original draft. G. Xiong: Data curation, validation, methodology. F. Zhao: Validation. M. Feng: Validation. J. Qiu: Validation, methodology. Y. Liu: Validation. M.Q. Zhang: Conceptualization. L. You: Conceptualization, resources, investigation, methodology. T. Zhang: Conceptualization, resources, funding acquisition, project administration. Y. Zhao: Conceptualization, resources, funding acquisition. J. Gu: Conceptualization, supervision, funding acquisition, investigation, project administration, writing–review and editing.

The authors acknowledge the contribution of all the investigators at the participating study sites. For this study, J. Gu was supported by the National Natural Science Foundation of China (NSFC; grant Nos. 61721003 and 61922047) and BNRIST Young Innovation Fund (grant No. BNR2020RC01009). L. You, T. Zhang and Y. Zhao were supported by NSFC (grant Nos. 81772639, 81974376, 81672960, and 81972258), Key Projects of International Scientific and Technological Innovation Cooperation Between Chinese and Italian Governments (grant No. 2018YFE0118600), and the Chinese Academy Medical Science Innovation Fund for Medical Sciences (grant No. 2016-I2M-1-001).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Siegel
RL
,
Miller
KD
,
Jemal
A
. 
Cancer statistics, 2020
.
CA Cancer J Clin
2020
;
70
:
7
30
.
2.
Rahib
L
,
Smith
BD
,
Aizenberg
R
,
Rosenzweig
AB
,
Fleshman
JM
,
Matrisian
LM
. 
Projecting cancer incidence and deaths to 2030: the unexpected burden of thyroid, liver, and pancreas cancers in the United States
.
Cancer Res
2014
;
74
:
2913
21
.
3.
Pu
N
,
Lou
W
,
Yu
J
. 
PD-1 immunotherapy in pancreatic cancer: current status
.
J Pancreatol
2019
;
2
:
6
10
.
4.
Maeda
S
,
Unno
M
,
Yu
J
. 
Adjuvant and neoadjuvant therapy for pancreatic cancer
.
J Pancreatol
2019
;
2
:
100
6
.
5.
Binenbaum
Y
,
Na'ara
S
,
Gil
Z
. 
Gemcitabine resistance in pancreatic ductal adenocarcinoma
.
Drug Resist Updat
2015
;
23
:
55
68
.
6.
Amrutkar
M
,
Gladhaug
IP
. 
Pancreatic cancer chemoresistance to gemcitabine
.
Cancers
2017
;
9
:
157
.
7.
Greenhalf
W
,
Ghaneh
P
,
Neoptolemos
JP
,
Palmer
DH
,
Cox
TF
,
Lamb
RF
, et al
Pancreatic cancer hENT1 expression and survival from gemcitabine in patients from the ESPAC-3 trial
.
J Natl Cancer Inst
2014
;
106
:
djt347
.
8.
Xu
M
,
Li
L
,
Liu
Z
,
Jiao
Z
,
Xu
P
,
Kong
X
, et al
ABCB2 (TAP1) as the downstream target of SHH signaling enhances pancreatic ductal adenocarcinoma drug resistance
.
Cancer Lett
2013
;
333
:
152
8
.
9.
Noll
EM
,
Eisen
C
,
Stenzinger
A
,
Espinet
E
,
Muckenhuber
A
,
Klein
C
, et al
CYP3A5 mediates basal and acquired therapy resistance in different subtypes of pancreatic ductal adenocarcinoma
.
Nat Med
2016
;
22
:
278
87
.
10.
Wang
Z
,
Li
Y
,
Kong
D
,
Banerjee
S
,
Ahmad
A
,
Azmi
AS
, et al
Acquisition of epithelial-mesenchymal transition phenotype of gemcitabine-resistant pancreatic cancer cells is linked with activation of the notch signaling pathway
.
Cancer Res
2009
;
69
:
2400
7
.
11.
Xiong
G
,
Feng
M
,
Yang
G
,
Zheng
S
,
Song
X
,
Cao
Z
, et al
The underlying mechanisms of non-coding RNAs in the chemoresistance of pancreatic cancer
.
Cancer Lett
2017
;
397
:
94
102
.
12.
Xiong
G
,
Liu
C
,
Yang
G
,
Feng
M
,
Xu
J
,
Zhao
F
, et al
Long noncoding RNA GSTM3TV2 upregulates LAT2 and OLR1 by competitively sponging let-7 to promote gemcitabine resistance in pancreatic cancer
.
J Hematol Oncol
2019
;
12
:
97
.
13.
Choi
M
,
Bien
H
,
Mofunanya
A
,
Powers
S
. 
Challenges in Ras therapeutics in pancreatic cancer
.
Semin Cancer Biol
2019
;
54
:
101
8
.
14.
Gao
H
,
Korn
JM
,
Ferretti
S
,
Monahan
JE
,
Wang
Y
,
Singh
M
, et al
High-throughput screening using patient-derived tumor xenografts to predict clinical trial drug response
.
Nat Med
2015
;
21
:
1318
25
.
15.
Farge
T
,
Saland
E
,
de Toni
F
,
Aroua
N
,
Hosseini
M
,
Perry
R
, et al
Chemotherapy-resistant human acute myeloid leukemia cells are not enriched for leukemic stem cells but require oxidative metabolism
.
Cancer Discov
2017
;
7
:
716
35
.
16.
Izumchenko
E
,
Paz
K
,
Ciznadija
D
,
Sloma
I
,
Katz
A
,
Vasquez-Dunddel
D
, et al
Patient-derived xenografts effectively capture responses to oncology therapy in a heterogeneous cohort of patients with solid tumors
.
Ann Oncol
2017
;
28
:
2595
605
.
17.
Bailey
P
,
Chang
DK
,
Nones
K
,
Johns
AL
,
Patch
AM
,
Gingras
MC
, et al
Genomic analyses identify molecular subtypes of pancreatic cancer
.
Nature
2016
;
531
:
47
52
.
18.
Waddell
N
,
Pajic
M
,
Patch
AM
,
Chang
DK
,
Kassahn
KS
,
Bailey
P
, et al
Whole genomes redefine the mutational landscape of pancreatic cancer
.
Nature
2015
;
518
:
495
501
.
19.
Du
Y
,
Liu
Z
,
You
L
,
Hou
P
,
Ren
X
,
Jiao
T
, et al
Pancreatic cancer progression relies upon mutant p53-induced oncogenic signaling mediated by NOP14
.
Cancer Res
2017
;
77
:
2661
73
.
20.
Bolger
AM
,
Lohse
M
,
Usadel
B
. 
Trimmomatic: a flexible trimmer for Illumina sequence data
.
Bioinformatics
2014
;
30
:
2114
20
.
21.
Dobin
A
,
Davis
CA
,
Schlesinger
F
,
Drenkow
J
,
Zaleski
C
,
Jha
S
, et al
STAR: ultrafast universal RNA-seq aligner
.
Bioinformatics
2013
;
29
:
15
21
.
22.
Liao
Y
,
Smyth
GK
,
Shi
W
. 
featureCounts: an efficient general purpose program for assigning sequence reads to genomic features
.
Bioinformatics
2014
;
30
:
923
30
.
23.
Griffiths-Jones
S
,
Saini
HK
,
van Dongen
S
,
Enright
AJ.
miRBase: tools for microRNA genomics
.
Nucleic Acids Res
2008
;
36
:
D154
8
.
24.
Langmead
B
,
Salzberg
SL
. 
Fast gapped-read alignment with Bowtie 2
.
Nat Methods
2012
;
9
:
357
9
.
25.
Love
MI
,
Huber
W
,
Anders
S
. 
Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2
.
Genome Biol
2014
;
15
:
550
.
26.
Franceschini
A
,
Szklarczyk
D
,
Frankild
S
,
Kuhn
M
,
Simonovic
M
,
Roth
A
, et al
STRING v9.1: protein-protein interaction networks, with increased coverage and integration
.
Nucleic Acids Res
2013
;
41
:
D808
15
.
27.
Luo
W
,
Brouwer
C
. 
Pathview: an R/Bioconductor package for pathway-based data integration and visualization
.
Bioinformatics
2013
;
29
:
1830
1
.
28.
Morris
TJ
,
Butcher
LM
,
Feber
A
,
Teschendorff
AE
,
Chakravarthy
AR
,
Wojdacz
TK
, et al
ChAMP: 450k chip analysis methylation pipeline
.
Bioinformatics
2014
;
30
:
428
30
.
29.
Blum
Y
,
de Reynies
A
,
Dusetti
N
,
Iovanna
J
,
Marisa
L
,
Nicolle
R
. 
SMAP: exploiting high-throughput sequencing data of patient derived xenografts
.
bioRxiv
2018
. Available from: https://www.biorxiv.org/content/10.1101/440008v1.
30.
Zhu
R
,
Yang
G
,
Cao
Z
,
Shen
K
,
Zheng
L
,
Xiao
J
, et al
The prospect of serum and glucocorticoid-inducible kinase 1 (SGK1) in cancer therapy: a rising star
.
Ther Adv Med Oncol
2020
;
12
:
1758835920940946
.
31.
Zhao
F
,
Yang
G
,
Feng
M
,
Cao
Z
,
Liu
Y
,
Qiu
J
, et al
Expression, function and clinical application of stanniocalcin-1 in cancer
.
J Cell Mol Med
2020
;
24
:
7686
96
.
32.
Saygin
C
,
Wiechert
A
,
Rao
VS
,
Alluri
R
,
Connor
E
,
Thiagarajan
PS
, et al
CD55 regulates self-renewal and cisplatin resistance in endometrioid tumors
.
J Exp Med
2017
;
214
:
2715
32
.
33.
Rijal
S
,
Fleming
S
,
Cummings
N
,
Rynkiewicz
NK
,
Ooms
LM
,
Nguyen
NY
, et al
Inositol polyphosphate 4-phosphatase II (INPP4B) is associated with chemoresistance and poor outcome in AML
.
Blood
2015
;
125
:
2815
24
.
34.
Wei
Z
,
Jia
J
,
Heng
G
,
Xu
H
,
Shan
J
,
Wang
G
, et al
Sirtuin-1/mitochondrial ribosomal protein S5 axis enhances the metabolic flexibility of liver cancer stem cells
.
Hepatology
2019
;
70
:
1197
213
.
35.
Xu
W
,
Zhu
H
,
Gu
M
,
Luo
Q
,
Ding
J
,
Yao
Y
, et al
DHTKD1 is essential for mitochondrial biogenesis and function maintenance
.
FEBS Lett
2013
;
587
:
3587
92
.
36.
Xu
WY
,
Zhu
H
,
Shen
Y
,
Wan
YH
,
Tu
XD
,
Wu
WT
, et al
DHTKD1 deficiency causes Charcot-Marie tooth disease in mice
.
Mol Cell Biol
2018
;
38
:
e00085
18
.
37.
Feng
J
,
Meyer
CA
,
Wang
Q
,
Liu
JS
,
Shirley Liu
X
,
Zhang
Y
. 
GFOLD: a generalized fold change for ranking differentially expressed genes from RNA-seq data
.
Bioinformatics
2012
;
28
:
2782
8
.
38.
Ratnam
NM
,
Peterson
JM
,
Talbert
EE
,
Ladner
KJ
,
Rajasekera
PV
,
Schmidt
CR
, et al
NF-kappaB regulates GDF-15 to suppress macrophage surveillance during early tumor development
.
J Clin Invest
2017
;
127
:
3796
809
.
39.
Koster
R
,
di Pietro
A
,
Timmer-Bosscha
H
,
Gibcus
JH
,
van den Berg
A
,
Suurmeijer
AJ
, et al
Cytoplasmic p21 expression levels determine cisplatin resistance in human testicular cancer
.
J Clin Invest
2010
;
120
:
3594
605
.
40.
Petropoulos
M
,
Champeris Tsaniras
S
,
Taraviras
S
,
Lygerou
Z
. 
Replication licensing aberrations, replication stress, and genomic instability
.
Trends Biochem Sci
2019
;
44
:
752
64
.
41.
Piquet
S
,
Le Parc
F
,
Bai
SK
,
Chevallier
O
,
Adam
S
,
Polo
SE
. 
The histone chaperone FACT coordinates H2A.X-dependent signaling and repair of DNA damage
.
Mol Cell
2018
;
72
:
888
901
.
42.
Tiberi
G
,
Pekowska
A
,
Oudin
C
,
Ivey
A
,
Autret
A
,
Prebet
T
, et al
PcG methylation of the HIST1 cluster defines an epigenetic marker of acute myeloid leukemia
.
Leukemia
2015
;
29
:
1202
6
.
43.
Jones
RJ
,
Baladandayuthapani
V
,
Neelapu
S
,
Fayad
LE
,
Romaguera
JE
,
Wang
M
, et al
HDM-2 inhibition suppresses expression of ribonucleotide reductase subunit M2, and synergistically enhances gemcitabine-induced cytotoxicity in mantle cell lymphoma
.
Blood
2011
;
118
:
4140
9
.
44.
Penna
E
,
Orso
F
,
Taverna
D
. 
miR-214 as a key hub that controls cancer networks: small player, multiple functions
.
J Invest Dermatol
2015
;
135
:
960
9
.
45.
Collisson
EA
,
Sadanandam
A
,
Olson
P
,
Gibb
WJ
,
Truitt
M
,
Gu
S
, et al
Subtypes of pancreatic ductal adenocarcinoma and their differing responses to therapy
.
Nat Med
2011
;
17
:
500
3
.
46.
Tiriac
H
,
Belleau
P
,
Engle
DD
,
Plenker
D
,
Deschenes
A
,
Somerville
TDD
, et al
Organoid profiling identifies common responders to chemotherapy in pancreatic cancer
.
Cancer Discov
2018
;
8
:
1112
29
.
47.
Nicolle
D
,
Fabre
M
,
Simon-Coma
M
,
Gorse
A
,
Kappler
R
,
Nonell
L
, et al
Patient-derived mouse xenografts from pediatric liver cancer predict tumor recurrence and advise clinical management
.
Hepatology
2016
;
64
:
1121
35
.
48.
Witkiewicz
AK
,
McMillan
EA
,
Balaji
U
,
Baek
G
,
Lin
WC
,
Mansour
J
, et al
Whole-exome sequencing of pancreatic cancer defines genetic diversity and therapeutic targets
.
Nat Commun
2015
;
6
:
6744
.
49.
Witkiewicz
AK
,
Balaji
U
,
Eslinger
C
,
McMillan
E
,
Conway
W
,
Posner
B
, et al
Integrated patient-derived models delineate individualized therapeutic vulnerabilities of pancreatic cancer
.
Cell Rep
2016
;
16
:
2017
31
.
50.
Smith
SC
,
Petrova
AV
,
Madden
MZ
,
Wang
H
,
Pan
Y
,
Warren
MD
, et al
A gemcitabine sensitivity screen identifies a role for NEK9 in the replication stress response
.
Nucleic Acids Res
2014
;
42
:
11517
27
.