Reversible chemoresistance of pancreatic cancer grown as spheroids

Abstract

Better in vitro models are needed to identify active drugs to treat pancreatic adenocarcinoma (PAC) patients. We used 3D hanging drop cultures to produce spheroids from five PAC cell lines and tested nine FDA-approved drugs in clinical use. All PAC cell lines in 2D culture were sensitive to three drugs (gemcitabine, docetaxel, and nab-paclitaxel), however most PAC (4/5) 3D spheroids acquired profound chemoresistance even at 10 μM. In contrast, spheroids retained sensitivity to the investigational drug triptolide, which induced apoptosis. The acquired chemoresistance was also transiently retained when cells were placed back into 2D culture and six genes potentially associated with chemoresistance were identified by microarray and confirmed using quantitative RT-PCR. We demonstrate the additive effect of gemcitabine and erlotinib, from the 12 different combinations of nine drugs tested. This comprehensive study shows spheroids as a useful multicellular model of PAC for drug screening and elucidating the mechanism of chemoresistance.

Introduction

Pancreatic adenocarcinoma (PAC) is the fourth leading cause of cancer death in the United States, and a major cause of cancer death worldwide. The 5-year survival rate for patients with PAC had been less than 5% from 1950-2000, and only in recent decades has increased to about 12% overall ¹. The poor prognosis for patients with PAC directly results from the fact that patients are typically diagnosed late, commonly with widespread metastases, and their cancers commonly become resistant to the vast majority of chemotherapeutic agents. Patients without documented metastatic disease are surgical candidates and exhibit a much higher 5 year-survival of ~20%, however, many surgically resected patients subsequently recur due to undetected micro-metastatic disease at the time of surgery ². Historically, gemcitabine had been the standard-of-care first-line systemic treatment for PAC, although only minimal and transient responses were obtained in most patients ³. Combination chemotherapeutic strategies prolong survival relative to gemcitabine alone, albeit only modestly. For instance, the combination of gemcitabine and nab-paclitaxel prolonged median survival to 8.5 months in patients with metastatic disease (vs. 6.7 months for gemcitabine alone) ⁴, and the four drug combination FOLFIRINOX (folinic acid, 5-fluorouracil, irinotecan and oxaliplatin) prolonged median survival to 11.1 months (vs. 6.8 months for gemcitabine alone) ⁵, although FOLFIRINOX is associated with greater toxicity than the combination of gemcitabine and nab-paclitaxel. A number of drugs have been assessed for their efficacy in treating PAC patients. Among these, our previous findings demonstrated the potent efficacy of triptolide against all 34 PAC cell lines, mostly in the single digit nanomolar concentrations ⁶. A triptolide analogue called minnelide has been developed that is water-soluble and less toxic, which has been evaluated in a phase I trial in patients with advanced GI cancers ⁷, and a phase II trial completed in 2019 for PAC patients (NCT03117920).

In addition to their notable chemoresistance, PAC are also notoriously stromal cell rich, and the non-neoplastic desmoplastic reaction can account for up to 90% of the total cellularity of the tumour. The desmoplastic reaction results from extensive proliferation of alpha-smooth muscle actin- and vimentin-positive cancer associated fibroblasts (CAFs) that produce exuberant extracellular matrix ⁸, which may be a barrier to efficient drug delivery. For instance, Olive et al. reported that stromal depletion by pharmacological inhibition of sonic hedgehog (Shh) signalling cascade led to superior intratumoural gemcitabine delivery, improved therapeutic response, and median survival in genetically engineered mouse model ⁹. However, Özdemir et al. demonstrated that prolonged CAF depletion by genetically ablating Shh signalling did not improve gemcitabine delivery and in fact reduced survival in a murine model ¹⁰. CAFs and PAC interactions appear to be more complex than initially expected, and recently, Duluc C et al. reported that cytokines from the CAFs may confer chemoprotection to PAC cells ¹¹. Therefore, it is necessary to establish physiological models in vitro to elucidate direct interactions between CAFs and PAC cells.

Most chemotherapeutic drugs have been studied in 2D cultures. However, these flat cultures with cell-to-plastic interactions are an inadequate representation of the tumour microenvironment, perhaps one reason why only ~ 5% of investigational cancer chemotherapies tested in 2D cultures show activity in patients ¹². Cancer cells can be grown in 3D, preserving cell-to-cell and cell-to-extracellular matrix contacts ¹³. It has been shown that gene expression profiles in 3D culture, in contrast to 2D cultures, more accurately reflect in vivo expression profiles ¹⁴. A number of studies have demonstrated that some drugs active in 2D culture are not similarly active in an in vivo environment, and spheroids had increasingly been considered to be an important experimental model to establish efficacy and prioritize drugs to test in xenograft models ¹⁵. In fact, legislation passed in December 2022 in the US allowing the FDA to not require the animal model testing for new drug approvals ¹⁶. Organoid models mimicking patient tissues has become one of the alternatives, although the comparable quality of evaluation as animal models are still required.

Cancer cells in spheroids grow in close proximity to their neighbours and are forced to interact more with them with direct cell-cell surface contact, more closely resembling growth of cells within a cancer. Spheroids can be formed by a variety of methods, including matrigel, pellet, spinner, liquid overlay, hanging drop cultures, etc ¹⁷. Hanging drop culture, in use for more than century, allows cells to settle and concentrate at the bottom of drops by gravity-promoted aggregation ¹⁸. A variety of cell types can form using this method and heterotypic spheroids also have been made ¹⁷.

In the present study, we demonstrate remarkable chemoresistance to three FDA approved drugs when PAC cells are grown in 3D, although we show that the drug is clearly able to penetrate these structures. In contrast, chemosensitivity was retained to the experimental drug triptolide. Co-plating PAC and CAF cells consistently produced CAF centres and PAC shells. The spheroid model of PAC is an ideal system for drug screening and to determine the mechanism of chemoresistance.

Materials and Methods

Cell Culture

Thirteen PAC cell lines, Panc 215, Panc10.05, TS0111, Panc 8.13, Panc 198, Panc 247, Panc 2.5, Panc 3.014 (derived from surgically resected PAC tissue at Johns Hopkins Hospital)^{19, 20}, A6L, A32-1, A10.7, A2.1(derived from rapid autopsy of a PAC patient as part of the Johns Hopkins Gastrointestinal Cancer Rapid Medical Donation Program)²¹, PK-9 (derived from a biopsy specimen of the head of the pancreas from PAC patients and kindly provided by Dr. Akira Horii of Tohoku University, Sendai, Japan)²², and three CAF cell lines, CAF19, CAF35, and CAF39 (derived from surgically resected PAC tissue at Johns Hopkins Hospital) ²³, were grown in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 10% foetal bovine serum (Thermo Fisher Scientific, Waltham, MA), 1×Antibiotic Antimycotic Solution (ABAM, Sigma, St. Louis, MO), and 5 μg/ml of Plasmocin^™ (InvivoGen, San Diego, CA). CAF cell lines were grown from surgically resected primary PACs and demonstrated to have normal chromosomes using SNP microarrays. Human Pancreatic Duct Epithelial Cell (HPDE) ²⁴, which is immortalized human pancreatic duct epithelial cell line generously provided by Dr. Ming-Sound Tsao (University of Toronto, Toronto, ON, Canada), was maintained in Keratinocyte-SFM media supplemented by epidermal growth factor and bovine pituitary extract (Thermo Fisher Scientific). Mycoplasma contamination was checked monthly using MycoAlert^® Mycoplasma Detection Kit (LONZA, Basel, Switzerland). DNA fingerprinting to confirm cell line identity was performed using the Profiler^® or Identifier^® microsatellite kits (Thermo Fisher Scientific).

Fluorescent cell line labelling

PACs were green fluorescent protein (GFP) labelled using recombinant lentiviruses as previously described ²⁵. CAF cells were labelled with mCherry using recombinant lentivirus (pLVX-IRES-mCherry, Clontech Laboratories Inc., Mountain View, CA) followed by purification using fluorescent activated cell sorting (DakoCytomation MoFlo, Johns Hopkins Bloomberg School of Public Health).

Formation and Analysis of Spheroids

To form spheroids, cells were subcultured, counted, and plated into 384 well Perfecta3D Hanging Drop Plates (3D Biomatrix, Ann Arbor, MI, USA) at 500-3000 cells/30 μl. Mixed spheroids were formed by serially diluting A6L and CAF35 cells at: 10:0, 9:1, 3:1, 1:1,1:3, 1:9, and 0:10 cell:cell ratios and a total of 6000 cells. Genomic DNA was extracted from four spheroids and the ratio of the two different types of cells calculated using the informative marker D5S1358. Mixed spheroids were imaged by using fluorescent microscopy (Nikon, Tokyo, Japan). To determine the volume of individual spheroids, the major and minor axes of the spheroid were measured using a calibrated ocular micrometre in a phase microscope, and volumes calculated (3/4)*π*major axis*minor axis²). To test the propensity of CAF cells to occupy the centre of PAC+CAF mixed spheroids, GFP-labelled PAC cells were plated and incubated for 72 hours to preform spheroids followed by the addition of mCherry-labelled CAF cells. After 48 hours incubation, spheroids were imaged by using fluorescent microscopy (Nikon).

Immunohistochemistry

Spheroids were fixed with 4% paraformaldehyde at 4 °C overnight. Thin layers of solid agarose containing spheroids were made and embedded in paraffin. Five μm sections were labelled with anti-vimentin (Ventana, Tucson, AZ; Cat# 790-2917, predilute), anti-cytokeratin 19 (Agilent Technology, Santa Clara, CA; Cat# M0772, 1:100), and proliferation cell number antigen (PCNA) Ki67 (Ventana; Cat# 790-4286, predilute).

Chemotherapeutic agents

Gemcitabine was purchased from BIOTANG Inc (Waltham, MA); Docetaxel, triptolide, capecitabine, leucovorin, 5- fluorouracil (5-FU), irinotecan, oxaliplatin, mitomycin C, and cisplatin from Sigma; Erlotinib from Selleck Chemicals (Houston, TX); and nab-paclitaxel (Abraxane) from celgene (Summit, NJ).

Chemosensitivity Testing

Cells from the same passage of the same flask were simultaneously plated in both standard 96 well plates (2D) and 384 hanging drop plates (3D). On day 2, culture media containing drug were added in six replicates wells to achieve the final concentration of 100 nM or 10 μM of each drug. Spheroids were lysed using spheroid lysis buffer (Scivax, Kawasaki, Japan), and ATP quantified using CellTiter-Glo^® (Promega, Madison, WI). 10 μM was chosen because this is the higher than the maximum plasma concentration (Cmax) that can be achieved for the vast majority of drugs, and 100 nM was chosen to be conservatively below the Cmax.

With triptolide, we treated spheroids at 0 nM, 1 nM, 3 nM, 10 nM, 30 nM, 100 nM. 300 nM, 1 μM, 3 μM, 10 μM. Live and dead cells were stained with calcein AM and EthD-III, respectively, using the Live/Dead Cell Staining Kit II (Promokine, Heidelberg, Germany) according to manufacturer’s instructions, and the half maximal inhibitory concentration (IC₅₀) values determined using four-parameter logarithmic with GraphPad Prism 6 (GraphPad Software, Inc., La Jolla, CA).

Chemotherapy combinations, FOLFIRINOX (folinic acid, 5-FU, irinotecan, and oxaliplatin), GTX (gemcitabine, docetaxel, and capecitabine), and gemcitabine plus docetaxel, nab-paclitaxel, or erlotinib regime, were treated at 10 μM for each drug.

Time Course of Chemosensitivity Restoration

A6L, Panc 10.05 and TS0111 cells were cultured in either 2D or 3D for two days, trypsinised, filtered using a 40 μm cell strainer, plated, and immediately treated with either 100 nM gemcitabine or docetaxel (Day 0). Alternatively, cells were recovered in T25 flasks for 24 hours, or 4 days, trypsinised and treated with drugs (Day 1 or Day 4) and incubated for three days, and cell viability was assessed by CellTiter-Glo^® (Promega) as described above.

Intracellular docetaxel and gemcitabine concentrations

Gemcitabine or docetaxel was added to PAC cells cultured in 2D or 3D culture at the final concentration of 10 μM. Cells were dissociated using trypsin after eight hours drug incubation. Cells were washed with PBS twice and protein amount determined with Pierce BCA Protein Assay Kit (Thermo Fisher Scientific). Docetaxel or gemcitabine were extracted from cells after lysis with 1:1 (v/v) acetonitrile: water by the scrape method. The supernatant was further diluted with 1:1 (v/v) acetonitrile: water containing the respective stable-label internal standard. Docetaxel concentrations were quantitated using a validated analytical chromatography consisting of ultra-performance liquid chromatography (UPLC) with mass spectrometric detection ²⁶. Chromatographic separation of gemcitabine was achieved with a Agilent Zorbax XDB^®, C18 column (50mm×2.1mm, 3.5μm) and isocratic elution with a water: acetonitrile with 0.1% formic acid (15:85, v/v) mobile phase at a flow rate of 0.2 mL/min over a 3 min total analytical run time. An AB Sciex 5500 triple quadrupole mass spectrometer operated in positive electrospray ionization mode was used for the detection of docetaxel, docetaxel-d9, gemcitabine and gemicitabine-¹³C, ¹⁵N₂. Docetaxel and gemcitabine were quantitated over the assay range of 0.5-1,000ng/mL with dilutions of up to 1:10.

RNA extraction and oligonucleotide microarray

PAC cells were cultured in 2D or 3D for 48 hours. Cells were harvested and washed with PBS twice. PAC cells cultured in 3D were also transferred to 2D culture and maintained for four days (3D back to 2D). RNA Extraction and oligonucleotide microarray were conducted by Johns Hopkins Medical Institution (JHMI) Deep Sequencing and Microarray Core (Dr. Haiping Hao, director). Briefly, total RNA was extracted by using RNeasy Mini kit (QIAGEN), and expression detected with the GeneChip Human Genome U133 Plus 2.0 Array (Affymetrix, Santa Clara, CA) on PAC cells cultured in 2D, 3D, 3D back to 2D, or PAC patient derived xenografts in duplicate. Total RNA was converted to labelled, fragmented cRNA, and arrays were hybridised for 16 hours at 45ºC, 60 rpm. All arrays were scanned in the Affymetrix GeneChip Scanner 3000 and raw analysis performed with Affymetrix Command Console and Expression Console.

Microarray Analysis

All analyses were performed within the R statistical software suite ²⁷, using both standard package (https://www.R-project.org, last accessed 9/8/17) and customized routines. Affymetrix cell files were pre-processed using the frozen robust RMA protocol ²⁸ . Substantial cell-line-specific effects were evident after fRMA, so an additional row-centring normalization was performed by subtracting the mean expression level for each probe, calculated within cell-line but across experimental conditions. Unsupervised cluster analysis was performed with agglomerative hierarchical clustering with Euclidean distance and Ward’s method, based on the expression of the most differentially expressed genes. Moderated empirical Bayes Linear Models as implemented in the limma package ²⁹ were used to identify differentially expressed genes. A Wilcoxon-Mann-Whitney rank sum test was applied to MsigDB Hallmark gene sets ³⁰ to identify significantly differentially expressed pathways.

A single xenograft was hybridized to the hgu133v2 array, in duplicate. As the xenograft represented a single condition, it was not possible to apply the row-centring procedure described above to the xenograft samples. Accordingly we adapted Geman’s rank-based top-scoring-pair (TSP) approach ³¹ to put the gene expression patterns of the xenografts into the 2D vs. 3D context. The TSP approach identified pairs of genes wherein Gene A tended to be more highly expressed than Gene B in one condition, while the relationship flipped in the other condition. Gene pairs were scored according to how well they separate samples and ties broken by comparing the mean expression of the two genes in each sample. More than 150,000 gene pairs (~.01% of all possible gene pairs) could be used to perfectly separate 2D samples from 3D. The top 100 of these were used for graphical representations.

Statistical Analysis

Unpaired t-test corrected by Holm-Sidak method for multiple comparisons correction was used to analyse the cell viability differences in each group of 2D and 3D culture using GraphPad Prism 6 (GraphPad Software, Inc). Mean difference in each group is more than 20% survival and a p value of 0.05 was used as a threshold of significance.

Results

PAC cell lines form live spheroids in hanging drop culture

We tested thirteen established PAC cell lines, three CAF cell lines isolated from PAC tumours, and the human pancreatic duct epithelial (HPDE) cells for their ability to spontaneously form spheroids in hanging drop cultures. The majority of PAC cell lines, A6L, Panc 215, Panc10.05, A10.7, TS0111, Panc 8.13, Panc 198, and Panc 3.014 (8/13=62%), all three CAF cell lines formed a single spheroid per well within three days, while HPDE cells formed multiple smaller spheroids in a well within the three days (Figure 2A-F). On the other hand, the other five PAC cell lines did not form spheroids but rather grew as loosely associated cells(Figure 2G and H). This is presumably due to the lower expression of adhesion molecules such as E-cadherin ³².

Figure 2.

PAC cells chemoresponses in 2D and 3D Culture. Percent survival of PAC, HPDE, and CAF cells cultured in 2D or 3D are displayed after treatment with gemcitabine (gem), docetaxel (dtx), nab-paclitaxel (Nab-Ptx), or triptolide (tri) at the concentration of 100 nM or 10 μM (Mean ± SD). All growth assays were done in six replicates. *: p<0.05, **: p<0.01, ***: p<0.001

As expected, the volume of the spheroids increased as a function of the number of cells plated (Figure S1A). Next, we wished to confirm the viability of spheroids that formed in these hanging drop cultures. When spheroids were stained with ‘Live-Dead’ stain, the calcein-AM (live stain) was cleaved by cell esterase and exhibited green fluorescence (Figure S1B). Spheroids also produced ATP in proportion to the number of cells plated (Figure S1E). Moreover, some cells in PAC spheroids replicated as demonstrated by positive proliferating cell nuclear antigen (PCNA, Ki67) immunohistochemistry (Figure S1F). Finally, when transferred to a 96-well tissue culture plates, spheroids attached to them and cells migrated laterally on the side (Figure S1C) or across the bottom (Figure S1D) of the plates.

PAC cells acquired chemoresistance in 3D culture

To test whether chemosensitivity would be affected in 3D culture, we selected five PAC cell lines for further analysis, because they possess unique mutation profiles across the genes KRAS, TP53, SMAD4, CDKN2A, and MLL3 (Table S1), under the presumption that we might gain insight into which gene mutations contributed to chemoresistance. We tested those cell lines with the nine FDA approved drugs in current clinical use. The PAC cells were treated with the drugs for 3 days following their plating and attachment in 2D culture (on day 2). Once spheroids had formed after 2 days, they were subjected to a 3-day drug treatment regimen. The spheroids were 150-250 μm at the time the drugs were added. The growth of all five PAC cell lines was inhibited by approximately 50% or more with the addition of gemcitabine or docetaxel at 100 nM in 2D culture (Figure 2). Nab-paclitaxel also demonstrated the same growth suppression across all PAC cell lines at 10 μM. However, we found that most spheroids acquired profound resistance to these three drugs in hanging drop culture, including at doses as high as 10 μM, compared with sister cultures (same cell line, same passage, and same day of subculture) of PAC cell lines grown in standard 2D culture (Figure 2). Only A10.7 did not demonstrate acquired resistance to these three drugs in 3D culture. On the other hand, most PAC cells were already resistant in 2D with the other six drugs at 100 nM, and PAC spheroids similarly exhibited resistance as those three drugs especially with 5-FU at 10 μM (Figure S2). After demonstrating gemcitabine and docetaxel resistance at 10 μM in 3D, spheroids were digested to single cells, plated back in 2D culture, and they grew and expanded (data not shown). As for three drugs (gemcitabine, docetaxel, and nab-paclitaxel), HPDE was already resistant to docetaxel and nab-paclitaxel even at 10 μM in 2D culture, although it gained resistance to gemcitabine in 3D culture. However, this effect was far less consistent with two other CAF cell lines, except CAF19 treated with gemcitabine.

We also used the experimental drug triptolide, where 34 out of 34 PAC cell lines in 2D culture were previously shown to be sensitive at a doses as low as 20 nM ^{6, 33}. In contrast to general chemoresistance of spheroids to the nine drugs, they retained chemosensitivity to triptolide in 3D with similar IC₅₀ values below 25 nM (Figure 2 and Figure S3). These findings were confirmed by direct observation where live/dead staining showed that triptolide treatment caused A6L spheroid toxicity and cell death at doses of 10 nM and above (Figure S4A-C), in contrast to gemcitabine and docetaxel (Figure S4D-I). We also confirmed that triptolide induced apoptosis in 3D in a dose dependent manner, based on the production of caspase 3/7 in PAC spheroids (Figure S5), consistent with the previous observations of others in 2D cultures ^{34, 35}.

Acquired chemoresistance was transiently retained

We wished to determine whether the chemoresistance had become a permanent property of the malignant cells in spheroids or whether it was a transient feature directly related to the environment in which the cells were grown at a given time (2D or 3D). A6L, Panc 10.05 and TS0111 spheroids grown in hanging drop cultures were subcultured and placed back into 2D culture and monitored for their response to these drugs over time. A6L and Panc 10.05 showed a transient chemoresistance to gemcitabine and docetaxel when initially placed back into 2D culture (Figure 3A-D), but they reacquired chemosensitivity to both gemcitabine and docetaxel within 4 days and 1 day, respectively. With the TS0111 cell line, the cells lost chemoresistance immediately on plating into 2D culture (Figure 3E-F).

Figure 3.

Chemosensitivity of PAC cells when cultured back to 2D from 3D. Chemosensitivity of control PAC cells grown in 2D culture (‘2D → 2D’) compared to those placed in 2D culture after spheroids growth (‘3D → 2D’) at Day 0, 1 or 4. Percent survival of Panc 10.05 cells (A, B) and TS0111 cells (C, D) with gemcitabine (A, C) or docetaxel (B, D) (Mean ± SD). All growth assays were done in six replicates. *: p<0.05, **: p<0.01, ***: p<0.001

Drugs accumulated in PAC spheroids

The structure of PAC spheroids was clearly more complex than PAC cells in 2D culture ³⁶. To determine whether the acquired chemoresistance was simply due to decreased drug penetration (or increased efflux), we measured gemcitabine and docetaxel concentrations within PAC cells grown in either 2D or 3D cultures. Gemcitabine was taken up and accumulated similarly in PAC cells in 2D and 3D culture (Figure 4A). In three out of four cell lines, docetaxel accumulated intracellularly to a greater degree when cells were grown as spheroids compared to 2D culture (Figure 4B). This essentially excluded the possibility that the measured chemoresistance in 3D was simply due to lack of uptake.

Figure 4.

Drug accumulation in 4 different PAC cells in 2D and 3D culture. Gemcitabine (A) and docetaxel (B) were added and incubated after eight hours incubation at the concentration of 10 μM (Mean ± SD). Each sample was tested as 3 biologic replicates. *: p<0.05, **: p<0.01, ***: p<0.001

Gene expression changes associated with growth in 3D culture

Because the observed chemoresistance was reversed when cells were replated back into 2D culture, we hypothesized that changes in gene expression might be mediating the effect. Accordingly, to determine the genes responsible for the acquired chemoresistance in 3D culture, we attempted to identify those that were differentially expressed in 2D vs. 3D culture. We grew cells in both conditions, harvested RNA and determined expression using Affymetrix U133 microarrays. Considering 2D growth as the baseline state, we identified a set of 1343 genes whose expression was upregulated when cells were grown in 3D and 1642 whose expression was downregulated. We also measured expression in cells that were grown in 3D, but returned to 2D growth for 4 days prior to RNA harvest. We compared expression in 2D, 3D and after reverting 3D back to 2D growth, and further excluded genes whose expression were altered in the rare chemosensitive A10.7 spheroids. As a result, 51 upregulated and 105 downregulated candidate chemoresistance genes were obtained (Tables S2 and S3). Those genes were subject to pathway analysis; however, no pathway was found that was clearly associated with chemoresistance. We then compared this set of 156 genes by gene ontology annotation (www.geneontology.org) and literature searches. We were able to identify a core set of nine genes, including six with upregulated and three with downregulated expression in 3D (Table S3). We then performed real-time quantitative (q) RT-PCR on eight of these transcripts (Figure S4), although we eliminated cyclin B1 because the qRT- PCR assay performed poorly. From this analysis, we identified candidate genes if their transcripts changed in the same direction as that identified in the microarray data and if these changes were seen in at least two out of three cell lines. This resulted in four upregulated genes (NFE2, VEGFA, NDRG1 and MUC20) and two downregulated genes (DUSP6 and CDK2).

Gene expression patterns between 2D culture, 3D culture and matched xenograft

We were interested in determining whether cancer cells growing as 3D spheroids in culture would mostly closely resemble expression of the same cells in 2D culture or as 3D xenografts. First, we compared gene expression patterns of four cell lines in 2D and 3D cultures, and identified the most significantly and consistently altered 100 genes (See Materials and Methods). For one of these PAC cell lines, we also had a patient derived xenograft (PDX) from the original patient’s cancer, A6L ³⁷. Next, we compared the expression of the 100 identified genes in A6L PDX, noting that the tissue had never been cultured in vitro (Figure 5). The A6L PDX gene expression clearly more closely resembles A6L and the other three PAC cell lines when grown in 3D culture than any cell lines in 2D culture including A6L itself. In this limited analysis, we found that PAC cells grown as 3D organoids had gene expression profiles closer to the cells grown as 3D xenografts, consistent with the findings of others ^{14, 38}.

Figure 5.

Heatmap of genes differentially expressed in 2D vs. 3D. Cluster analysis based on 100 pairs of genes whose expression best separates cells grown in 2D culture from those grown in 3D culture. Blue and yellow cells in the heatmap indicate which of the two genes in the pair is more highly expressed. Only after identifying the 100 genes, the xenograft sample was analysed. PDX: Patient-derived xenograft. Each sample was tested in duplicate (r: biologic replicates). For gene names, see supplemental tables S2,S3.

Combination chemotherapy

Three PAC cell lines, A6L, Panc 215, and Panc 10.05 showed chemoresistance to all nine agents even at 10 μM in 3D culture as shown above (Figure 2, Figure S2). Next, to assess the efficacy of combination therapies, we tested these three PAC cell lines with twelve combination therapies of nine drugs, including gemcitabine/nab-paclitaxel, gemcitabine/erlotinib, GTX (gemcitabine, docetaxel, capecitabine), and FOLFIRINOX (5-FU, leucovorin, irinotecan, oxaliplatin). These therapies were based on gemcitabine or 5-FU, which both had been the first line therapy for PAC patients ^{3, 39}. Therefore, we tested additive effects on top of gemcitabine and 5-FU at 10 μM. Although most combinations of the drugs did not show any additive effect, the gemcitabine/erlotinib combination treatment showed an additive effect on all three PAC cell lines in 3D culture, although formal analysis was not performed (Figure 6A-C, arrows). Interestingly, only Panc 215 exhibited an additive effect of 5-FU with more than two drugs of any of the three other drugs in the FOLFIRINOX (leucovorin, irinotecan, oxaliplatin, or FOLFIRINOX itself) in 3D culture (Figure 6D-F, arrowheads), while 5-FU only with leucovorin and any other drugs in FOLFIRINOX exhibited an additive effect in 2D culture (Figure 6E).

Figure 6.

Chemoresponses to chemotherapy combinations in 2D or 3D culture. Gemcitabine based chemotherapy (A-C) and 5-FC based chemotherapy (D-G) were evaluated in three PAC cell lines (Mean ± SD). Arrows indicate additive effects by gemcitabine+erlotinib. Arrowheads indicate additive effects of 5-fluorouracil with any of the three other drugs in the FOLFIRINOX (leucovorin, irinotecan, oxaliplatin, or FOLFIRINOX itself). All growth assays were done in six replicates. Gemcitabine (Gem), Nab-paclitaxel (Nab-Ptx), Erlotinib (Erl), F: 5-fluorouracil, L: leucovorin, I: irinotecan, O: oxaliplatin. *: p<0.05, **: p<0.01, ***: p<0.001

Mixed spheroids form consistent structures

Because PAC is a notoriously stromal cell rich cancer and widely chemorefractory, it is natural to hypothesize a relationship between two ⁴⁰. As there are some reports suggesting that stromal cells might contribute to chemoresistance ^{11, 41–43}, we wished to model this interaction. A6L+CAF35 cells were plated in equal amounts (3,000 cells each) and grown for three days. The mixed spheroids were fixed, sectioned, stained with haematoxylin and eosin (HE), and immunostained for vimentin (vim) and cytokeratin 19 (ck19) (Figure 7A-C). HE staining showed a clear distinction between a core and shell areas (Figure 7A, yellow arrow heads). To determine cell identity, immunohistochemistry demonstrated that the vimentin positive CAF35 cells were localized in the centre of the spheroids (Figure 7B), whilst the cytokeratin 19 positive PAC cells surrounded them as a shell (Figure 7C). The localization of the two cell types was further confirmed by using fluorescence labelled PAC (EGFP) and CAF cells (mCherry, Figure S7A-B). This pattern of localization was consistent and also observed with other different PAC and CAF cell combinations (Figure S7C-L). Notably, the PK-9 PAC cell line did not form spheroids by themselves (Figure 1H), but when plated in the presence of CAFs cells, PK-9+CAFs cells formed tightly aggregated spheroids around the CAFs (Figure S7I-L).

Figure 7.

The structure of mixed spheroids. A6L+CAF35 mixed spheroids were fixed, sectioned, and stained for haematoxylin and eosin (A, HE), vimentin (B, Vim), and cytokeratin 19 (C, Ck19). Thin shell of PAC cells (300 cells) surrounds CAF spheroids (3,000 cells) (D). A6L percentage in the mixed spheroids as determined by quantification of microsatellite DNA after plating A6L:CAF35 cells at varying ratios (3 biologic replicates, Mean ± SD) (E). Individual CAF35 cells (mCherry) were added to preformed A6L spheroids (EGFP). Note that CAF35 spheroids are found in the centres of preformed A6L spheroids (F). Scale bars: 100 μm. A6L+CAF35 mixed spheroids cultured for 7 days (G) or 34 days (H). Arrowheads in G indicate early basolateral polarization. Arrows in H indicate cells with basolateral polarization. Scale bars: 50 μm. Human PAC tissue (I). Scale bars: 20 μm.

Figure 1.

Spheroids cultured in hanging drop plates. PAC cells, A6L, Panc 215, and A10.7 formed compact and single spheroids (A-C). Human pancreatic duct epithelial (HPDE) cells formed multiple spheroids (D). Cancer associated fibroblasts (CAFs), CAF19 (E) and CAF35 (E), both formed compact and single spheroids. PAC cells, Panc 247 (G) and PK-9 (H), did not form spheroids. Spheroids were photographed by phase microscopy. Scale bar: 100 μm.

We also attempted to form thin shells of PAC cells by co-plating a fixed number of CAF cells (3,000) with only 100 or 300 PAC cells. The 300 PAC cells formed thin layers around the large CAF spheroids (Figure 7D). To investigate the degree to which we could control the final composition of mixed spheroids, we plated A6L and CAF35 in hanging drop cultures at a series of different ratios. Using a microsatellite marker that distinguishes the two cell types (derived from two different patients), we confirmed that the percentage of A6L cells in mixed spheroids directly increased as a function of the ratio of A6L:CAF35 cells plated as expected (Figure 7E), demonstrating the ability to control the ratio of PAC: CAFs in the resultant spheroids.

We hypothesized that CAF cells might possess higher homophilic affinity or higher density of cell adhesion molecules than PAC cells, possibly explaining why the CAFs would end up occupying the central position in these mixed spheroids ⁴⁴. If this simply explained the localization of the two cell populations, we hypothesized that adding individual CAF cells to preformed PAC spheroids should result in a second, separate pure CAF spheroid distinct from the preformed cancer spheroid. Surprisingly, we discovered that in these cultures the CAF cells still migrated to occupy the central core position (Figure 7F). In longer term mixed cultures, CAF cells die and the PAC cells demonstrate some basolateral polarization (day 7, Figure 7G; day 34, Figure 7H), reminiscent of primary PAC (Figure 7I).

We tested the spheroids for replication by immunolabeling for PCNA and found most of the replicating cells were located in the shell region (Figure S8A). Few CAF cells were positive in PCNA staining, but CAF cells in the mixed spheroid abundantly produced collagen type I and fibronectin (Figure S8B), indicating that CAF cells promoted vigorous extracellular matrix (ECM) production in mixed spheroids. The level of expression in the CAF component of the mixed spheroid was not obviously increased compared to that in the pure CAF spheroids (Figure S8H-I). The extent of cell replication and ECM expression were reflected in each component (Figure S8A-I).

CAF cells did not confer obvious chemoprotection to one PAC cell line

As discussed, PACs are unique in their prominent stromal cell component and their widely chemorefractory nature. Several reports have suggested that CAFs might confer chemoprotection to PAC cells ^41–43. First, we tested if gemcitabine chemoresistance would be acquired by conditioned media (CM) from two CAFs cell lines in 2D culture, but CM from CAFs compared to drug alone (MOCK) showed virtually no difference in three PAC cell lines grown in 2D (Figure S9A). As presented above, the majority of PAC cell lines grown in 3D demonstrate chemoresistance, whilst only A10.7 cells retained chemosensitivity in 3D, allowing the opportunity to examine the potential effect of CAF on it. We formed spheroids from a mixture of A10.7 and CAF19 (gemcitabine resistance CAF cell line in 3D culture). ATP quantity from pure A10.7 spheroid was much higher than that from pure CAF19 spheroids (14.2 times) (Figure S9B). Therefore, we hypothesized that ATP quantification of the mixed spheroid from both A10.7 and CAF19 cells would reflect mostly the A10.7 PAC cells. The ATP quantity from pure A10.7 spheroids and A10.7+CAF19 mixed spheroids were essentially identical in the absence of drug (Figure S9C). We therefore compared survival of pure PAC spheroids, pure CAF spheroids, and mixed spheroids. The A10.7 + CAF19 mixed spheroids showed the similar killing curves to the pure A10.7 spheroids in response to gemcitabine (Figure S9D). These results suggest that CAF cells do not confer obvious chemoprotection to PAC cells in culture, albeit only in this one unusual chemosensitive line.

Discussion

We conducted comprehensive chemosensitivity tests in 3D culture with nine FDA-approved drugs that are currently used for treatment of PAC patients in the United States. We previously demonstrated that gemcitabine sensitively killed 31 out of 34 PAC cell lines in 2D culture with a median IC₅₀ of 2.8 nM ⁶. However, four out of five PAC cell lines became profoundly resistant in 3D, not only to gemcitabine, but also to docetaxel and nab-paclitaxel. A similar finding with gemcitabine has been previously reported ⁴⁵. The chemoresistance of PAC spheroids to gemcitabine raised the possibility that spheroid chemosensitivity more accurately represented drug responses in vivo, where most patients, despite initial responses, eventually develop resistance.

We sought the mechanism by which most PAC spheroids acquired chemoresistance. First, we considered the possibility that drug penetration was prevented by the complex three-dimensional structure such as the existence of tight junctions ^{36, 46}, or that overexpression of drug efflux pumps in cells grown in 3D decreased intracellular drug accumulation ⁴⁷. However, both drugs accumulated in PAC spheroids as well as, or greater than those cultured in 2D. Additionally, two out of three PACs (A6L and Panc 10.05) transiently retained acquired the chemoresistance to both gemcitabine and docetaxel after they were cultured back to 2D from 3D. Therefore, we hypothesized that the resistance was due to changes in transcription that are growth format dependent (see below).

Along these lines, we identified 51 with increased transcripts in 3D and 105 with decreased transcripts in 3D. Using gene-ontology, we compared our results to those genes associated with chemoresistance in cancer, and confirmed four up-regulated and two down-regulated genes whose timing matched the acquisition of the chemoresistance phenotype. It is also possible that the mechanism of resistance was not manifest in changes in transcription, rather through some other signalling mechanisms. Identifying which of these genes mediates the chemoresistance is important, since it could theoretically restore potency of the three drugs, or offer a new drug target for reversal of chemoresistance in PAC.

The chemoresistance that we found was not universal. Although PAC cells grown in hanging drop culture exhibited chemoresistance to most drugs, we found that triptolide sensitivity was retained in 3D culture with the low nanomolar IC₅₀s (<25 nM), and cell death was mediated by apoptosis, especially the outer most layer of cells. Considering Liu et al.’s finding that C_max in rat plasma is greater than 27 nM ⁴⁸, triptolide derivatives may prove to be excellent candidates to treat PAC patients. Triptolide also showed some toxicity to the control CAF cells, however Chugh R et al. demonstrated that a triptolide derivative, minnelide, demonstrated advantages over gemcitabine in an orthotopic PAC mouse model ⁴⁹. A phase II clinical trial using minnelide for metastatic PAC patients (NCT03117920) was completed in 2019, and a phase I trial in GI malignancies was recently published ⁷.

The chemoresistance phenotype in PAC might be related to the remarkable stromal rich environment. Several reports have claimed that CAF cells might confer chemoprotection to PAC cells ^{11, 41–43}. For instance, conditioned media from CAFs conferred gemcitabine chemoresistance to PAC cells via interleukin-6 ¹¹. However, we were not able to demonstrate CAF conditioned media conferring chemoprotection to PAC cells. As only one cell line grown as spheroids that retained sensitivity to gemcitabine, we used PAC+CAFs mixed spheroids to determine if CAF cells would confer chemoprotection to these PAC cells, because mixed spheroids should suitably model the direct contact and closer communication with CAFs and their possible influence on PAC cell chemosensitivity. However, we did not observe acquired gemcitabine resistance to PAC cells. Özdemir and colleagues reported that stromal depletion did not improve gemcitabine efficiency and even reduced survival in a murine model ¹⁰. Moreover, clinical trials with hedgehog inhibitors have failed to show efficacy in patients ⁵⁰. Tumour-stroma interactions in PAC appear to be considerably complex, and the functional role of CAFs in PAC microenvironment remains to be fully elucidated to provide a better therapeutic strategy.

To circumvent acquired chemoresistance, most PAC patients are treated with combination chemotherapy. Chemoresponses to these combination therapies were also evaluated using pure PAC spheroids, because CAFs did not alter chemosensitivity of PAC cells grown in 3D culture. We demonstrated for the first time that gemcitabine plus erlotinib exhibits an additive effect in all three tested PAC cells in a 3D culture, and in fact the synergistic effect of these agents has previously been shown in 2D culture and PAC xenografts^{51, 52}. In fact, recent clinical trial showed modest improvement in prolonging the survival of PAC patients, compared to gemcitabine alone (6.24 months v 5.91 months) ⁵³. Only Panc 215 exhibited efficacy of 5-FU based chemotherapy combinations including FOLFIRINOX. FOLFIRINOX is one of the established chemotherapy combinations, however it can be toxic and PAC patients who do not respond to this therapy are commonly discontinued ⁵⁴. It is possible that one could pre-screen PAC patients for pharmacotyping using organoid culture to identify patients who would receive benefit from this therapy, a concept promoted by Dr. Burkhart, and others ^{55, 56}. We have also been successful at establishing organoids from primary cancers using the methods of Clevers and colleagues⁵⁶.

In summary, we report substantial chemoresistance of PAC cells when grown in 3D culture based on a comprehensive evaluation, and this may explain why these drugs are either not effective or only transiently effective in PAC patients. This model should be useful to identify more active drugs and to determine the precise mechanism of chemoresistance.

Biographies

Yoshihisa Matsushita, Ph.D.

Yoshihisa Matsushita received Ph.D. in integrated medicine from Tokyo Women’s Medical University. He completed his post-doctoral fellowship in Dr. Eshleman’s lab. He is currently a high-school teacher at Wakayama Prefectural Koyo High School in Japan.

Alexis Norris, Ph.D.

Alexis Norris received her Ph.D. in Dr. Eshleman’s laboratory cellular and molecular medicine programme from Johns Hopkins University. She completed a post-doctoral fellowship in the laboratory of Jonathan Pevsner in bioinformatics. Specializing in cancer genomics and bioinformatics, she is currently a bioinformatician at the FDA.

Yi Zhong, Ph.D.

Yi Zhong received Ph.D. in pathology from Kyoto University. Specializing in pathology, he is currently a research assistant at Hackensack Meridian Health CDI.

Asma Begum, Ph.D.

Asma Begum received Ph.D. in cancer biology from Tokyo Dental and Medical University. Specializing in cancer biology, she is currently a scientist at Leidos Biomedical Research, Inc.

Hong Liang

Hong Liang received Bachelor of Science from Jianghan University. She is currently a research assistant in Dr. Eshleman’s laboratory at the Johns Hopkins University School of Medicine.

Marija Debeljak

Marija Debeljak received Bachelor of Science from Millersville University of Pennsylvania. She is currently a Ph.D. candidate at Johns Hopkins University School of Medicine.

Nicole Anders

Nicole Anders received Master of Science in pharmaceutical science from Campbell University. Specializing in bioprocessing and analytical chemistry, she is currently a global scientist at Takeda Pharmaceuticals, Inc.

Michael Goggins, M.D.

He received M.D. from University of Dublin Trinity College. He is currently a professor of pathology, medicine, and oncology at the Johns Hopkins University School of Medicine.

Zesheen A. Rasheed, M.D., Ph.D.

Zesheen A. Rasheed received M.D. and Ph.D. Specializing in medical oncology, he is currently a global clinical head, oncology R&D at AstraZeneca.

Ralph H. Hruban, M.D.

Ralph H. Hruban received M.D. from Johns Hopkins University. He is currently a director of the Sol Goldman Pancreatic Cancer Research Centre, and director of pathology, and professor of pathology and oncology at Johns Hopkins University School of Medicine

Christopher L. Wolfgang, M.D.

He received M.D. and Ph.D. from Temple University. Dr. Christopher L. Wolfgang is currently a professor of surgery, pathology and oncology and director of pancreatic surgery chief, hepatobiliary and pancreatic surgery at NYU.

Elizabeth D. Thompson, M.D., Ph.D.

Elizabeth D. Thompson received M.D. and Ph.D. from University of Virginia. Specializing in cancer immunology, she is currently an assistant professor of pathology at Johns Hopkins University.

Michelle A. Rudek, Pharm.D., Ph.D.

She received Pharm.D. and Ph.D. from Virginia Commonwealth University. She was a professor of oncology at Johns Hopkins University and director of the Sidney Kimmel Comprehensive Cancer Centre analytical pharmacology core. She is recently deceased.

Jun O. Liu, Ph.D.

Jun O. Liu received Ph.D. in biochemistry from Massachusetts Institute of Technology. He is currently a professor of pharmacology and molecular science at Johns Hopkins University.

Leslie Cope, Ph.D.

Leslie Cope received his Ph.D. from Johns Hopkins University. Specializing in biostatistics, he is currently an associate professor of oncology at Johns Hopkins University.

James R. Eshleman, M.D., Ph.D.

James R. Eshleman received M.D. and Ph.D. from the University of Pennsylvania. He is a professor of pathology and oncology, and associate director of Molecular Diagnostics Laboratory at Johns Hopkins University.