Tumorgrafts as in vivo surrogates for women with ovarian cancer

Abstract

Purpose: Ovarian cancer has a high recurrence and mortality rate. A barrier to improved outcomes includes a lack of accurate models for preclinical testing of novel therapeutics.

Experimental design: Clinically relevant, patient-derived tumorgraft models were generated from sequential patients and the first 168 engrafted models are described. Fresh ovarian, primary peritoneal, and fallopian tube carcinomas were collected at the time of debulking surgery and injected intraperitoneally into severe combined immunodeficient mice.

Results: Tumorgrafts demonstrated a 74% engraftment rate with microscopic fidelity of primary tumor characteristics. Low-passage tumorgrafts also showed comparable genomic aberrations with the corresponding primary tumor and exhibit gene set enrichment of multiple ovarian cancer molecular subtypes, similar to patient tumors. Importantly, each of these tumorgraft models is annotated with clinical data and for those that have been tested, response to platinum chemotherapy correlates with the source patient.

Conclusions: Presented herein is the largest known living tumor bank of patient-derived, ovarian tumorgraft models that can be applied to the development of personalized cancer treatment.

Figure 1

Clinical correlation between tumorgraft and patient. (a) Tumorgraft engraftment rate over time where ‘Success’ indicates successful engraftment in at least one mouse and ‘Failed’ indicates failed engraftment. (b) Patient overall survival by tumorgraft engraftment status, where ‘Success’ and ‘Failed’ are as in “a”. (c) Representative images from mice showing intraperitoneal dysfunction. A tumorgraft (arrow) behind the uterus (*) and caused marked bowel (triangle) obstruction (left) in one mouse but a mesenteric tumor produced non-obstructive weight loss and anorexia in another mouse (right). (d) A representative mouse with the abdominal wall reflected and showing normal, smooth-appearing peritoneum with visible vessels (left). The peritoneal surface of a mouse that developed ascites (center) demonstrates a thick cellular layer consistent with carcinomatosis. Visualization of ascitic fluid by light microscopy (right) revealed aggregates of epithelial cells (spherules).

Figure 2

Histologic similarities between patients and corresponding tumorgrafts. (a) Representative H&E, Ki67, pan-CK, and CD45 expression in tumorgraft model PH063 shows conserved morphology, proliferation index by Ki67, and epithelial phenotype by pan-cytokeratin (40x). Scale bar = 100 μm. (b) Cytokeratin staining in tumorgraft and source patient pathology sections (PH122 shown) were selected and quantified using ImageJ software as described in the Methods section (20x). Scale bar = 200 μm. (c) The calculated non-epithelial tissue for eight tumorgraft models were plotted against that of the source patient.

Figure 3

Molecular analysis of tumorgrafts. (a) Graphical representation of array CGH at the chromosome level from source patient (blue) and tumorgraft (red) in model PH015 (high grade endometrioid). (b) Heat map comparison of frequently lost (green) and gained (red) genes is the TCGA dataset (right column) compared to two representative tumorgrafts (PH015 and PH013, a high grade serous). (c) The upper panel shows normalized enrichment scores (NES) of 36 tumorgrafts according to the CLOVAR (Classification of Ovarian Cancer) gene signature. Varying degrees of enrichment for signatures associated with differentiated (red circle), immunoreactive (green square), mesenchymal (blue triangle up), and proliferative (purple triangle down) molecular subtypes were observed. The lower panel shows a heat map of 36 tumorgraft models clustered as a function of their gene-expression correlation with each other. Highest (red), inverse (blue), and zero correlation was used to cluster tumorgrafts by their overall correlation.

Figure 4

Transabdominal ultrasound assessment of tumor change. (a) Representative serial ultrasound images from a saline-treated PH015 tumorgraft (dotted outline) showing growth over 15 days. (b) Tumor diameter in 32 mice bearing tumorgrafts were assessed by ultrasound and plotted against caliper-derived diameter measurements at necropsy. The Spearman correlation was high (r = 0.8491). (c) Representative tumorgraft ultrasound images showing response to treatment. A target tumor (*) was present on day 1 but a non-target tumor (**) developed during treatment. For all panels, scale bar = 5 mm and tumor cross-sectional area marked by circumferential dotted line.

Figure 5

Tumorgraft response to treatment with correlation to the patient experience. (a) Percent change in tumor area over time in vivo for tumorgraft models treated with chemotherapy. Models are derived from platinum sensitive (dotted line) or resistant (sold line) patients. (b) Heat map representing the most significantly up-regulated (red) or down-regulated (green) genes in sensitive compared to resistant tumorgrafts.