Genome-wide CRISPR screen reveals PSMA6 to be an essential gene in pancreatic cancer cells

Fetal bovine serum was purchased from HyClone (Logan, UT). Cell culture reagents, fluorescent secondary antibodies, and RNAiMAX transfection reagent were purchased from Invitrogen (Carlsbad, CA). All siRNAs (custom cherry-picked libraries) were purchased from Dharmacon (Lafayette, CO). PSMA6 and 18S TaqMan probes were purchased from Thermo Fisher Scientific (Waltham, MA).

All cell lines were maintained in a humidified incubator at 37 °C in 5% CO₂. PANC-1, hTert HPNE, Mia PaCa-2, HPAF-II, and AsPC-1 cells were purchased from ATCC and used experimentally within five passages. All cell lines were maintained according to ATCC recommendations, and ATCC authenticated the cell lines by short tandem repeat (STR) DNA profiling. The cells were verified to be mycoplasma-free by using the MycoProbe Mycoplasma Detection kit (R&D Systems, Minneapolis, MN). Cas9 stable cell lines were made by virally transducing cells with LentiCAS9-Blast (Addgene, Cambridge, MA; cat. # 52962) [15] and selecting with 8 μg/mL of blasticidin for 5 days. Expression was verified by Western blot analysis (Additional file 1b).

Stable Cas9-expressing PANC-1 cells were transduced with the CRISPR lentiviral library at an experimentally established MOI of 0.3 in the presence of 4 μg/mL of polybrene overnight. The cells were selected with 2 μg/mL of puromycin for 9 days, at which point 1 × 10⁸ cells were collected and frozen for genomic DNA isolation. A further 1 × 10⁸ cells were grown in the presence of 100 nM gemcitabine for 6 days, after which the cells were frozen for genomic DNA isolation. Sequencing was performed on the Illumina HiSeq 2500 platform (100 bp SE), and raw FASTQ files were deconvoluted by barcode and trimmed of excess nucleotides by using custom scripts on the St. Jude Children’s Research Hospital high-performance computing facility. The resulting amplicons were then analyzed with MAGeCK-VISPR [16].

Genomic DNA was extracted with QIAamp Blood Maxi kit (Qiagen, cat. # 51192) in accordance with the manufacturer’s protocol. Using a nested PCR program, we generated barcoded amplicons containing the integrated gRNA sequences. Briefly, 10 separate 100-μL redundant reactions were performed, each containing 5 μg of DNA, Premix Ex Taq HS (TaKaRa, cat. # RR030A), and 6 μL of a 10 μM solution of each primer (F1 and R1) (Additional file 2). The first round of the PCR amplification program was as follows: step 1, 95 °C for 1 min; step 2, 95 °C for 30 s; step 3, 55 °C for 30 s; and step 4, 72 °C for 30 s; with steps 2–4 being repeated 15 times. Then, 5 μL of the PCR product was used to seed the second round of PCR, along with Premix Ex Taq HS, 6 μL of the R2 primer, and 6 μL of a 10 μM solution of the F2 primer in a staggered mixture that contained the Illumina adapters and a barcode to identify the sample after sequencing analysis (Additional file 2). The second-round PCR program was as follows: step 1, 95 °C for 1 min; step 2, 95 °C for 30 s; step 3, 63 °C for 30 s; and step 4, 72 °C for 30 s; with steps 2–4 being repeated 17 times.

RNA was extracted with a Maxwell RSC simplyRNA Tissue Kit and a Maxwell RSC Instrument (Promega, Madison, WI). The RNA concentration was measured with a NanoDrop 8000 UV-Vis Spectrophotometer (Thermo Fisher Scientific). A SuperScript VILO cDNA Synthesis Kit (Life Technologies, Carlsbad, CA) was used to synthesize cDNA according to the manufacturer’s protocol. To determine mRNA expression, Applied Biosystems TaqMan assays (20×), Fast Advanced Master Mix (Life Technologies), and an Applied Biosystems 7900HT Fast Real-Time PCR System (Life Technologies) were used in accordance with the TaqMan Fast protocol. Gene expression was normalized to the 18S rRNA housekeeping gene, which did not vary in its expression during the growth of the cell lines. Each experiment was performed at least three times, and all samples were analyzed in triplicate.

Lentivirus was generated in HEK293T cells (ATCC, Manassas, VA) in 225-cm² flasks. Briefly, 22.2 μg of a CRISPR pooled gRNA library (human sgRNA library Brunello in lentiGuide-Puro) transfer vector (Addgene, cat. #73178) [17], 16.7 μg of psPAX2 plasmid (Addgene, cat. # 12260), and 11 μg of pMD2.G plasmid (Addgene, cat. # 12259) were combined with Lipofectamine 3000 (Thermo Fisher Scientific) in accordance with the manufacturer’s protocol, and the mixture was used to transfect the cells. The virus-containing medium was collected 48 h after transfection and centrifuged at 500×g for 5 min to remove cells and debris. The supernatant containing the virus was then filtered with a 0.45-μM PES filter and frozen at − 80 °C. Viral transduction was accomplished by adding 150 μL of virus-containing medium per 1 × 10⁶ cells (titer determined experimentally, MOI = 0.3) to 225-cm² flasks of PANC-1 cells at 75% cellular confluence, along with 4 μg/mL polybrene (Sigma-Aldrich), for 16 h. The virus-containing medium was then replaced with fresh growth medium.

To determine the stage of apoptotic cell death in control and treated PANC-1 and Mia PaCa-2 cells, we performed flow cytometric analysis on PANC-1 and Mia PaCa-2 cells grown in vitro, using the PE Annexin V Apoptosis Detection Kit I (BD Biosciences, San Jose, CA) in accordance with the manufacturer’s protocol. Briefly, cells were washed twice with cold PBS and resuspended in 1× Binding Buffer at a concentration of 2 × 10⁶ cells/mL. Aliquots of 200 μL of the solution (containing 4 × 10⁵ cells) were transferred to 5-mL round-bottom tubes, then 5 μL of PE Annexin V and 5 μL of 7-AAD cell viability dye were added to the tubes. The cells were gently vortexed and incubated for 15 min. at room temperature while protected from light. Next, 400 μL of 1× Binding Buffer was added to each tube and the samples were immediately analyzed on a custom-configured BD Fortessa cytometry analyzer using FACSDiva software (Becton-Dickinson, San Jose, CA). Data were analyzed using FlowJo software (TreeStar, Ashland, OR). All experiments were performed with at least three biological replicates, and at least 200,000 events were collected per sample.

Three individual pools of Tet-on shRNA stable PANC-1 cells were generated after lentiviral transduction of early passage PANC-1 cells with SMARTvector (hEF1a) inducible PSMA6 shRNA plasmids (Dharmacon 1255-01EG5687shRNA sequences: TAGAGTCCTAACCACTTCG, GATCTGGAAACTAACGAC, ACAGGTAAGTGGCATCACG). PANC-1 cells were selected with puromycin (2 μg/ml) for 3 days then analyzed for knockdown efficiency and stored with Bambanker Serum Free Freezing Media (Wako Chemicals #302–14,681) in liquid nitrogen vapor phase for future studies. Knockdown efficiency was tested 3 days post doxycycline treatment and mRNA of PSMA6 was compared to the same stable cell line not treated with doxycycline, ACAGGTAAGTGGCATCACG sequence had a > 80% knockdown of PSMA6 (Fig. 4f) and was used for subsequent studies. All inducible stable cell lines were maintained in tetracycline screened fetal bovine serum (Hyclone, Logan UT).

Alpha tubulin and PSMA6 antibodies were purchased from Cell Signaling Technologies (Boston, MA). Briefly, PANC-1 cells were treated with 25 nM siControl (non-targeting) or siPSMA6 for 72 h, lysed with RIPA buffer, and supernatant was collected for gel electrophoresis. PVDF membrane was probed with alpha tubulin and PSMA6 antibodies and imaged on a Li-Cor FC and bands were analyzed with Li-Cor Odyssey software.

All genes that correlated to depleted sgRNAs from the negative-selection (drop-out) CRISPR screen were filtered at a maximum P-value of 0.05. The scores of the remaining 1073 genes were transformed such that the highest value was represented as 1 in order to assign weight. All statistically significant genes were verified to have a weight greater than 0, and the list and corresponding weights were loaded into Enrichr [18] for downstream analysis. Enrichment analysis of pathways were obtained by selecting the Reactome Pathways 2016 and EMBL GO Biological Process databases. For transcription factor enrichment analysis, the eXpression2Kinases tool [19] was used with all default settings. The cBioPortal tool [20, 21] was used to measure expression of PSMA6 across available patient samples from the TCGA Research Network: http://​cancergenome.​nih.​gov/​. Kaplan-Meier survival plot was generated using Kaplan-Meier plotter using Pan-cancer RNAseq dataset [22].

MAGeCK-VISPR [16] was used to rank and sort gRNAs by P-value and/or FDR (Additional file 3). Data from at least three independent replicated experiments were quantitatively analyzed by two-way ANOVA with the Sidak multiple comparisons test or by Student’s 2-tailed t-test, using GraphPad Prism 7.0 software, as indicated. All data are represented as the mean ± SD.

We conducted a negative-selection (drop-out) genome-wide CRISPR screen to uncover novel essential genes and/or genes that might sensitize pancreatic cancer cells to the current frontline chemotherapeutic agent, gemcitabine. We transduced the Brunello CRISPR library (on day 0) and allowed cell outgrowth for a further 9 days under puromycin selection. The stable cells were then treated with 100 nM (the approximate IC₁₀) [23‐25] of gemcitabine (on day 9) for an additional 6 days (to day 15) (Fig. 1a; Additional file 1a). Next-generation sequencing (NGS) was performed on 100-bp amplicons with upwards of 8 × 10⁷ reads, of which approximately 90% were mapped to the gRNA library (Fig. 1b). Our sequencing also revealed an obvious change in the representation of gRNAs as measured by the GINI index [26], which is a measurement of inequality (Fig. 1c), and demonstrated a selection event. After performing NGS, we identified drop-out hits by using MAGeCK-VISPR software; a complete list of hits can be found in Additional file 3. The hits were prioritized by eliminating previously identified essential genes that had been shown to be critical in most cell lines by Hart et al. [10]. We hypothesized that those previously identified essential genes would not be of interest because of their lack of specificity for PDAC cell lines. The resulting top 100 (approximately) hits were then screened with four pooled siRNAs per gene in PANC-1 and hTert HPNE cells. HPNE cells are an intermediary cell line formed during acinar-to-ductal metaplasia that we used as our noncancerous pancreas cell line control. The siRNAs that showed preferential essentiality were deconvoluted and further tested in the AsPC-1, Mia PaCa-2, and HPAF-II cell lines (Fig. 1d).

Interestingly, a Gene Ontology (GO) analysis and a pathway analysis from the Reactome database revealed cell-cycle genes to be significantly enriched among the top depleted sgRNAs in our screen (Additional file 4a–d). Additionally, a transcription factor enrichment analysis revealed that MYC is an upstream transcription factor for the genes identified in the negative-selection screen (Additional file 4e).

We conducted a siRNA screen with a pool of four individual siRNAs in an effort to validate the top 100 (approximately) hits from the genome wide CRISPR screen and to identify hits that selectively affected PANC-1 cells as compared to HPNE cells, which are a model of noncancerous pancreatic tissue. In both pooled siRNA screens, we monitored the cells for a short period (66 h post transfection) to look for the most potent essential genes. Longer, 5-day siRNA screens for these same gene hits revealed that almost all the siRNAs targeting these genes elicit some degree of growth defect (Additional file 5a). The pooled siRNA screen in PANC-1 cells revealed several siRNAs, including those targeting ARHGEF12, CCDC136, CRNN, FOXD1, NUDT19, PSMA6, STOML2, TSNARE1, and USP22, that significantly impede growth (Fig. 2a). Similarly, the pooled siRNA screen in HPNE cells revealed several potent siRNAs, some of which are unique to HPNE, including those targeting MEN1 and CRNN (Fig. 2b). Interestingly, HPNE cells appear to be sensitive to the non-targeting siRNA alone, most likely because one or more of the individual non-targeting siRNAs within the pool has off-target toxic effects within these cells. However, we were most interested in the targets that have little or no phenotypic effect in HPNE cells but are highly potent in PANC-1 cells. Given this criterion, we identified CCDC136 and PSMA6 as potential selective targets (Fig. 2b).

To confirm our observations with siRNA-mediated knockdown, we deconvoluted all pooled siRNAs that showed any potent effect in PANC-1 cells by testing the four individual siRNAs independently. The pool of four individual siRNAs was deconvoluted to validate the phenotype and to limit the probability of the phenotype’s being caused by an off-target effect. For additional validation, we tested whether our observations were cell-line specific. To this end, we measured the resulting growth characteristics with live-cell imaging and quantification of four individual siRNAs per target in three different PDAC cell lines: HPAF-II (Fig. 3a), Mia PaCa-2 (Fig. 3b), and AsPC-1 (Fig. 3c). We clearly showed that there are cell line specific on-target effects; for example, the siRNA targeting CCDC136 is very potent in PANC-1 cells but has limited or no effect in other tested pancreatic cancer cell lines, which includes HPAF-II, Mia PaCa-2, and AsPC-1 cells. Thus, CCDC136 knockdown may exploit a unique molecular mechanism in PANC-1 cells that is not present in all models. It is also noteworthy that the effects of siRNA knockdown of SCAP, ARHGEF12, and OR10G9 on cell growth vary among these cell lines, whereas the siRNA targeting MEN1 was toxic in all the tested cell lines, including HPNE (Figs. 2a and b, 3a–c). The siRNA targeting PSMA6 (siPSMA6) showed similar toxicity siRNA targeting MEN1, but noncancerous HPNE cells were moderately resistant to it (Figs. 2b, 3a–c, Additional file 5b). Movies illustrating the phenotype observed in HPNE and PANC-1 cells can be found in Additional file 6.

We hypothesized that PSMA6 inhibition resulted in cellular apoptosis because the PSMA6 protein was a critical member of the proteasome. To test this, we ran flow cytometry on PANC-1 cells after 72 h of treatment with 25 nM siPSMA6 (Fig. 4a). We identified a clear shift in late apoptosis, with a subpopulation of siPSMA6-treated PANC-1 and Mia PaCa-2 cells exhibiting a significant upward shift in late apoptotic events when compared to cells treated with siControl (non-targeting siRNA) (Fig. 4b and c; Additional file 7). We confirmed PSMA6 protein knockdown with a western blot (Fig. 4d and e).

To perform long-term spheroid assays, we made PANC-1 cells that stably expressed shRNA against PSMA6 (shPSMA6), using the tet-on system inducible by doxycycline (Dox) to circumvent the limitation of transient siRNA. We validated the shPSMA6-stable cells and were able to achieve knockdown of PSMA6 with greater than 80% efficiency (Fig. 4f). Using the tet-on shPSMA6 cells, we conducted 10-day spheroid assays in ultra-low–attachment round-bottom plates and monitored the spheroids with imaging followed by a terminal viability assay (CellTiter-Glo). We found that the cells with silenced PSMA6 were significantly smaller and visually less dense (Fig. 4g); furthermore, total cell viability, as measured by the CellTiter-Glo luminescence, was decreased by approximately 60% in cells where PSMA6 expression was blocked (Fig. 4h). As expected, the Cancer Genome Atlas shows PSMA6 is expressed in human PDAC samples and has a relatively low mutation rate (Additional file 8a). And in support of our spheroid assays, we generated a Kaplan-Meier survival curve using KM plotter [22] which shows PDAC patients with high expression of PSMA6 having a significantly shorter overall survival rate (P = 0.0009) (Additional file 8b).

After verifying the necessity of PSMA6 for PANC-1 and Mia PaCa-2 cell survival, we next set out to interrogate the susceptibility of PANC-1 and Mia PaCa-2 cells to disruption of the broader biological pathways that involve PSMA6 expression. Because PSMA6 is a critical member of the proteasome, we hypothesized that PANC-1 and Mia PaCa-2 cells would be sensitive to therapeutic inhibition of the proteasome. To that end, we treated both cell lines with bortezomib, a current FDA-approved proteasome inhibitor that binds the catalytic site of the 26S proteasome and has also been shown to interact with the β subunits of the 20S proteasome [27]. Bortezomib was chosen in an effort to phenocopy PSMA6 knockdown, with the caveat that PSMA6 is just a member of the much larger proteasome complex that is inhibited by bortezomib. After treatment with bortezomib we found that it significantly increased cellular death. In fact, sub–1 nM concentrations decreased PANC-1 cell viability by approximately 90% after a 5-day treatment (Fig. 5a). We also treated PANC-1 and Mia PaCa-2 cells with 1 nM bortezomib for 48 h, then stained the cells with 7-AAD and annexin V and performed flow cytometry, revealing that bortezomib treatment significantly and rapidly induced late apoptosis (Fig. 5b and c; Additional file 9).