PDX-derived organoids model in vivo drug response and secrete biomarkers

Patient-derived organoid models are proving to be a powerful platform for both basic and translational studies. Here we conduct a methodical analysis of pancreatic ductal adenocarcinoma (PDAC) tumor organoid drug response in paired patient-derived xenograft (PDX) and PDX-derived organoid (PXO) models grown under WNT-free culture conditions. We report a specific relationship between area under the curve value of organoid drug dose response and in vivo tumor growth, irrespective of the drug treatment. In addition, we analyzed the glycome of PDX and PXO models and demonstrate that PXOs recapitulate the in vivo glycan landscape. In addition, we identify a core set of 57 N-glycans detected in all 10 models that represent 50%–94% of the relative abundance of all N-glycans detected in each of the models. Last, we developed a secreted biomarker discovery pipeline using media supernatant of organoid cultures and identified potentially new extracellular vesicle (EV) protein markers. We validated our findings using plasma samples from patients with PDAC, benign gastrointestinal diseases, and chronic pancreatitis and discovered that 4 EV proteins are potential circulating biomarkers for PDAC. Thus, we demonstrate the utility of organoid cultures to not only model in vivo drug responses but also serve as a powerful platform for discovering clinically actionable serologic biomarkers.

QPCT U2AF2   Table S2 Information of patients in EV study. PDAC, pancreatic ductal adenocarcinoma; GERD: Gastroesophageal reflux disease; N/A: not available; IBS: irritable bowel syndrome. Both CA19-9 levels and National comprehensive cancer network (NCCN) stage of the PDAC are shown.

Methods
Drug Treatment Assay: Established organoid cultures were collected and digested as above. For organoids hard to dissociate for single cells, TrypLE was used in place of Accutase. Cells were diluted in organoid growth media at the density of 50,000 cells/ml and 100 ul of the suspension was added into each well of 96 well precoated with matrigel. After 4 days of growth, media were replaced with fresh media, and drugs were dispensed using a Tecan D300e digital dispenser. Cell death was measured after 4 days using CytoTox Glo (Promega).

Morphological and Histological Analysis:
Organoids were plated at a density of 25,000 cells/well and images were taken every day for 12 days. About 200 images were obtained for each line. The images were analyzed for changes using the organoseg software program 11 . Briefly, raw images were segmented and analyzed using inbuilt parameters such as area, perimeter, and eccentricity. The data were plotted as box plots using Prism. To generate organoid tissue sections, they were grown in chamber slides and fixed in 4% PFA for 2 hours followed by incubation with Hematoxylin solution for 10 minutes and washed twice with water. The organoids were scraped and sandwiched between two layers of Histogel (Sigma) using a cryomold and transferred to a tissue cassette followed by fixation in 10% formalin.

Establishment of xenografts:
Four to six-week-old Foxn1/Nu mice were purchased from Taconic and utilized for these studies. All animal work carried out was approved by the BIDMC Institutional Animal Use and Care Committee (IACUC) and animals were maintained in accordance with guidelines of the American Association of Laboratory Animal Care. To initiate propagation, cryopreserved xenografts were rapidly thawed cut into ~3 x 3 x 3 mm fragments, and subsequently implanted subcutaneously in cohorts of 10 mice per PDX line studied, with one small fragments in each mouse. When tumors reached a size of 1500 mm3 they were excised for cohort expansion, cut into ~3 x 3 x 3 mm fragments, and transplanted to the final cohort of mice to be treated with relevant therapeutic agents, with bilateral subcutaneously implantation of fragments in each mouse.
Treatment protocol: Xenografts from experimental PDX cohorts were grown to a size of 200-250 mm 3 , at which time mice were randomized and enrolled on the study. The dose and schedule of treatments were described in the supplemental materials (Fig.S3A). Mice were treated for 28 days and monitored daily for signs of toxicity, with weights and tumor measurements taken three times per week. Tumor length and width were measured using a digital caliper and the tumor volumes estimated using the following formula: tumor volume = [length x width 2 ] / 2 30 . Relative tumor growth inhibition (TGI) was calculated by the relative tumor growth of treated mice divided by the relative tumor growth of control mice (T/C). Experiments were terminated on day 28. For evaluation of drug responses in PDX models, the RECIST calculation from Jackson Lab was used. Briefly, the percentage of tumor volume change was calculated as (Ve-Vs)/Vs*100, Ve was tumor volume at the endpoint and Vs was tumor volume at the start of treatments. The categories of responses were defined as the following: where Vm is the minimal volume change, and Va is the average volume change.
Genomic concordance between PDX and PXO: Variant calls on whole exome sequencing data were made using a standard pipeline. The sequence reads were aligned in hg19 using BWA (v.0.7.8), recalibrated, and identified variants using Picard-tools/GATK v3.0. Copy number analysis on PDX-WES data was done using CNVkit and variants were filtered and annotated using Annovar for gene annotation, mutation consequences, prediction of deleteriousness, and allele frequency. SNV concordance between tumor-organoid pairs was determined from the overlap of variant calls and variant allelic fractions. For each SNV called in PDX models, we used BCFTools-v1.9 to read evidence for the SNV present in WES-organoids after filtering for base quality and mapping quality to be considered concordant. Visualization of DNA aberrations is displayed using Oncoprint/ComplexHeatmap -R package.

RECIST Category Best Response: Vm Average Response: Va
Complete Response (CR) < -95% < -40% Partial Response (PR) < -50% <-20% Stable Disease (SD) < 35% < 30% Progressive Disease (PD) Anything else Drug response concordance between PDX and PXO: AUC values from PXO drug-response curve is classified into 4 classes using Jenks natural breaks classification. We used the implementation of Jenks algorithm by BAMMtools, an R package (Rabosky, et al, 2014). The optimal number of breaks was estimated the number of breaks that would give >95% goodness of variance fir (GVF).