ANTICANCER RESEARCH 30: 2805-2812 (2010)

Analysis of Chemotherapeutic Response Heterogeneity and Drug Clustering Based on Mechanism of Action Using an In Vitro Assay

SHARA D. RICE, JAMIE M. HEINZMAN, STACEY L. BROWER, PAUL R. ERVIN, NAN SONG, KUI SHEN and DAKUN WANG

Precision Therapeutics, Inc., Pittsburgh, PA 15203, U.S.A.

Abstract. Background: Cancer chemotherapeutic treatment patient is pre-treated with chemotherapy or radiation and is a complex scientific task. The ChemoFx® Drug Response then the tumor is surgically removed (4, 5). Early detection Marker (DRM) assists physicians in identifying treatment significantly improves treatment success rates for some types protocols likely to be effective for specific patients. Materials of cancer (6, 7). However, recurrent and metastatic forms of and Methods: The ChemoFx® DRM was used to study drug cancer do not demonstrate the same robust treatment response in vitro. Established human cancer cell lines and response rates (8, 9). The presence of inoperable metastases primary cultures of patient tumor specimens were challenged and the differences in response among patients make it with chemotherapeutic agents to observe response of difficult to develop standard treatment protocols for multiple tumor samples and determine whether drugs with metastatic and recurrent disease (10, 11). similar mechanisms of action elicit similar response. Results: Generally, chemotherapeutic agents are indicated for the These studies demonstrated heterogeneous response among treatment of specific cancer types and patient populations patient tumor samples and clustering of drug response with based on the effectiveness in large-population studies (12, similar mechanisms of action. Also highlighted was the 13). However, what is largely unstudied is whether certain reproducibility of ChemoFx DRM and its utility in treatment regimens are likely to be more successful in characterizing tumor response to chemotherapy. Conclusion: specific patients. Clinical studies indicate that a patient who Heterogeneous drug responses observed in vitro were similar does not respond to one chemotherapeutic agent may to those observed clinically. Response characteristics were respond to another (14). Patients may also be generally similar for drugs with similar mechanisms of action, resistant to multiple agents, a condition known as multiple suggesting response heterogeneity is determined at a cellular drug resistance (15). In all, the value of one therapy over and molecular level. another for a specific patient is difficult to determine, short of treating the patient and observing response. For this Cancer is a disease treated with various strategies depending reason, individualized treatment for cancer is becoming more on the type of cancer and the stage of the disease. Generally popular. Physicians are using genetic tests and drug response accepted standards of care have been developed based on markers to guide chemotherapy treatment for their patients large-population studies (1). These standards involve various (16, 17). approaches to chemotherapy, radiation therapy and surgical The ChemoFx® Drug Response Marker (DRM) (Precision resection (2). For some cancers, adjuvant therapy is most Therapeutics, Inc (Precision), Pittsburgh, PA, USA), identi- successful; this involves surgical removal of tumor tissue fies effective agents for solid cancers prior to chemotherapy followed by chemotherapy or radiation (3). In other cancers, treatment (18, 19). The assay involves culturing cells derived neoadjuvant therapy is more efficacious; with this therapy, a from patient tumors and then treating the samples with various chemotherapeutic agents in vitro. The ChemoFx DRM allows individualized testing with a panel of drugs on patient tumor samples to determine the best treatment Correspondence to: Jamie M. Heinzman BS, Precision Thera- regimen for each patient. The goal of the assay is to provide peutics, Inc., 2516 Jane Street, Pittsburgh, PA 15203, U.S.A. Tel: additional information to physicians making therapy +412 4322365, Fax: +412 4883830, e-mail: jheinzman@ ptilabs.com decisions, in order to avoid treating patients with agents that Key Words: ChemoFx DRM, cancer, heterogeneity, chemothera- are not likely to be effective for their cancer. The chemo- peutic response, mechanism of action, chemotherapy drug response therapeutic agents tested in the ChemoFx DRM work marker, in vitro chemosensitivity testing. through a variety of mechanisms of action; a complete list

0250-7005/2010 $2.00+.40 2805 ANTICANCER RESEARCH 30: 2805-2812 (2010) of chemotherapeutic agents by mechanism of action is Table I. Mechanism of action of drugs tested in the ChemoFx DRM. provided in Table I. Chemotherapeutic agents tested in the ChemoFx DRM, grouped by In the current study, the ChemoFx DRM was used to study mechanism of action. the response characteristics of established cell lines and Mechanism of Action Chemotherapeutic Agent primary cultures of patient tissues from ovarian, lung, breast, and colon cancers. The study aim was two-fold: (i) to Alkylating agent Altretamine, bleomycin, observe drug response profiles of different patient tumor cyclophosphamide, ifosfamide, samples to the same chemotherapeutic agent and (ii) to melphalan, mitomycin Alkylating agent, non-classic Procarbazine, temozolomide determine whether cell lines and patient tumor samples Anthracycline Doxorubicin, epirubicin cultured in vitro respond with similar drug response patterns Anti-metabolite 5-Fluorouracil, gemcitabine, to chemotherapeutic agents with similar mechanisms of Pemetrexed action. This study concentrated on response characteristics Nitrosourea Carmustine, lomustine of platinum, taxane, and anthracycline therapies. Platinum Carboplatin, cisplatin, oxaliplatin Targeted therapy, EGFR Erlotinib Targeted therapy, multiple Sunitinib Materials and Methods kinase inhibitor Taxane Docetaxel, paclitaxel Cell lines. A number of immortalized cell lines were used in these Topoisomerase II inhibitor Etoposide, irinotecan, topotecan studies, including the following: SK-OV-3, MDA-MB-157, MDA- Vinca alkaloid Vinblastine, vincristine, MB-361, BT-474, HCC-1500, HCC-202, AU-565, MDA-MB- Vinorelbine 175VII , MDA-MB-453, ZR-75-1 , BT-20 , HCC-1569, MCF-10A, MDA-MB-468 , MDA-MB-415, MDA-MB-436, BT-549, SK-BR-3, T-47D, HCC-1428, HCC-1143, HCC-1937, HCC-38, MDA-MB- 231 , BT-483, CAMA-1, UACC-812 and HCC-1187 (American tumor type: Mammary Epithelial Growth Medium (MEGM, Lonza, Type Culture Collection (ATCC), Manassas, VA, USA). Cell lines Cologne, Germany), Bronchial Epithelial Growth Medium (BEGM, were maintained in McCoy’s 5A or RPMI 1640 medium Lonza), McCoy’s 5A (Mediatech), RPMI 1640 (Mediatech). The (Mediatech, Herndon, VA, USA) with 10% FBS (Hyclone, Logan, cultures were maintained for 5-28 days in humidified incubators at UT, USA). All cell lines were seeded at 40,000 cells in 25 cm2 37˚C with 5% CO2, until a confluency of at least 30% was attained. flasks and maintained in culture at for 6-8 days in humidified incubators at 37˚C with 5% CO2 until a confluency of approxi- ChemoFx DRM. ChemoFx DRM was performed as described mately 85% was attained. previously (19, 21). Briefly, cultures were plated in Costar 384-well micro titer plates (Corning, Lowell, MA, USA) and incubated for Patient tumor specimens. Primary cultures were established using 24 hours to allow cell attachment. Then, cells were treated in tumor specimens procured for research purposes from the following triplicate with chemotherapeutic agents (untreated cells were used sources: National Disease Research Interchange (NDRI, Phila- as a control). For each chemotherapeutic agent, 10 serially diluted delphia, PA, USA), Cooperative Human Tissue Network (CHTN; drug concentrations were tested. Agents used in this study included Philadelphia, PA, USA), Forbes Regional Hospital (Monroeville, carboplatin (1 μM-500 μM), cisplatin (0.2 μM-100 μM), docetaxel PA, USA), Jameson Hospital (New Castle, PA, USA), Saint (0.1 nM-25 nM), doxorubicin (2 nM-1 μM), epirubicin (0.7 nM-14 Barnabas Medical Center (Livingston, NJ, USA), Hamot Medical μM) and paclitaxel (0.2 nM-0.1 μM). After an incubation period of Center (Erie, PA, USA), Windber Research Institute (Windber, PA, 72 hours, the cells were fixed with anhydrous ethanol (95% fixing USA) and University of Rochester Medical Center (Rochester, NY, grade) and stained with 4’6-diamidino-2-phenylindole (DAPI), a USA). All patient tumor samples were de-identified prior to transfer fluorescent DNA stain for imaging the nucleus (Sigma-Aldrich to Precision’s Research and Development Department, so they were Corp, St Louis, MO, USA). Cells that remained attached after considered exempt (IRB protocol #RD-109). staining were imaged and counted using automated cell quanti- Additionally, analysis was performed on de-identified data from fication systems developed for the 384-well format (21). 361 consecutive primary ovarian cases submitted to Precision for The percentages of cells remaining after drug treatment were commercial testing from October 2006 to February 2009. Drug used to determine survival fraction (SF=average cell countdosex response results had been previously generated and reported to /average cell countcontrol), from which dose–response curves were patients, thus analysis did not jeopardize patient treatment. Data from plotted. Each dose–response curve was assigned a response index the 361 primary ovarian cultures tested with carboplatin, cisplatin, (RI) score ranging from 0 to 10 as well as a response index of docetaxel and paclitaxel were used to correlate drug response. responsive (R), intermediately responsive (IR), or not responsive (NR); the RI score is a metric based on adjusted areas under the Cell culture. Primary cultures were established as previously curve (aAUC) (18). Curves were smoothed using a logarithmic described (20). Briefly, surgically procured tumors were shipped curve-fit tool (GraphPad Prism®, LaJolla, CA, USA). overnight to Precision laboratories in McCoy’s 5A medium with 1% penicillin/streptomycin (Mediatech). Upon receipt, specimens were Comparison of drug response for various chemotherapeutic agents. minced, washed with an antibiotic solution as necessary, and seeded To measure the degree of similarity between drug responses, the into sterile tissue culture flasks to establish the culture. The Pearson correlation coefficient (r) was calculated between drug following cell culture media were used and were specific to the responses to different chemotherapeutic agents among primary

2806 Rice et al: Analysis of Heterogeneity and Clustering of Chemotherapeutic Response cultures derived from ovarian cancer patient specimens and also profiles of the lung specimens treated with the platinum among immortalized breast cancer cell lines, according to the therapies (carboplatin and cisplatin) and the taxane therapies following formula: (docetaxel and paclitaxel) were also quite different, with the platinum pair invoking a stronger response than the taxane pair, although a wide range of drug response was observed in both. Broader studies were performed with primary cultures of patient tumors and ATCC cell lines. These studies showed that in 361 primary ovarian cultures, there was similarity of where xi represents the drug response of cell line x to drug i and xm response for the platinum therapies and the taxane therapies, represents the mean of RI scores of cell line x. Similarly, yi is the RI indicated by high correlations (r=0.89, p<0.001) (Figure 4A). score of cell line y to drug i and y represents the mean of drug m For immortalized breast cell lines, the taxane therapies had response scores of cell line y. similar response characteristics (r=0.91, p<0.001), as did the Results anthracycline therapies (doxorubicin and epirubicin) (r=0.98, p<0.001) (Figure 4B). For both examples, the correlations between the related drug pairs were stronger than the Single drug response heterogeneity. The ChemoFx DRM was correlations across drug pairs. used to evaluate single drug response patterns across patient samples. Specifically, 21 lung primary tumor specimens were Discussion treated with carboplatin, cisplatin, docetaxel and paclitaxel. As such, it was possible to compare results among patient The ChemoFx DRM was used for this study so that a variety samples. Example results for carboplatin demonstrated a of chemotherapeutic agents, within and across mechanisms of large range in RI score values from RI=3.69 to RI=7.06, with action, could be tested simultaneously. This type of testing is an average RI value of 5.34 (Figure 1). not practical for human subjects. If the patient fails a specific Reproducibility of response in ChemoFx. To ensure that the therapy, it is unlikely that the patient will be treated with heterogeneity observed during single drug testing in patient another agent with the same mechanism of action. Thus, by tumor specimens was due to patient response rather than using well-controlled assay conditions, the ChemoFx DRM assay variability, the reproducibility of the ChemoFx DRM was used to evaluate drug response heterogeneity across was evaluated using intra-operator and inter-operator patient tumor samples and to compare response to drugs of variability tests. An established cell line, SK-OV-3, was similar mechanisms of action within patient tumor samples. treated with carboplatin, cisplatin, docetaxel and paclitaxel. Heterogeneity of response to chemotherapeutic agents can First, intra-operator tests were performed; one operator be demonstrated in controlled comparisons of chemothera- performed 7 repetitions of the ChemoFx DRM on one day. peutic responses among patients in vitro (Figure 1). Broader Second, inter-operator tests were performed; three different studies indicated that heterogeneity of response exists operators performed the assay on three days (one day each). among different patient tumor samples for the other Operators 1 and 2 performed 7 assays each, and operator 3 chemotherapies listed in Table I (data not shown). This has performed 5 assays. Example results for paclitaxel been anecdotally understood for years; patients with a demonstrated that the ChemoFx DRM consistently produced similar disease respond differently to any particular drug similar results with the same drug on this cell line, indicated (22). The response hetero-geneity that was observed in this by similar average values and coefficients of variation study was not obviously segregated into categories, but was (CoVs) for RI for both single operator and multiple operator rather a continuum. The drug concentration required to tests (Figure 2). invoke response varies widely among patient tumor samples. Certainly, those tumor cell populations that respond to a Comparison of drug response for various chemotherapeutic lower dose (responsive) are more responsive than those that agents. Patient tumor samples were treated with related either do not respond (non-responsive) or respond only at a chemotherapeutic agents to ascertain whether similar drug much higher dose (intermediately responsive) (Figure 1). response was observed (Figure 3). Patient tumor samples that The high reprodu-cibility demonstrated in this study were responsive to one drug in a structurally related pair suggests that the heterogeneity observed in primary tumor responded similarly, though not identically, to the other drug testing was not due to assay variability (Figure 2). Similar in the same pair, demonstrated by small percent differences average values and CoVs for RI scores are evidence of the between RI scores. This was most obvious for the examples consistency and reproducibility of the ChemoFx DRM provided in this study, namely carboplatin- and cisplatin-treated within and across operators. As such, the drug response ovarian, colon and lung patient tumor samples, and docetaxel- heterogeneity observed in this study was likely due to and paclitaxel-treated lung specimens. The drug response qualities intrinsic to the tumor cells.

2807 ANTICANCER RESEARCH 30: 2805-2812 (2010)

Figure 1. Heterogeneity of response of 21 primary lung cultures treated with carboplatin in ChemoFx DRM. Dose–response curves for patient samples treated with ten doses of carboplatin (1 μM-500 μM) color-coded by response index: green, responsive (R); blue, intermediately responsive (IR) and red, not responsive (NR). Inset: Average, minimum, and maximum RI score values showed a broad heterogeneity of drug response.

If response heterogeneity is due to intrinsic biochemical characteristics, it is to be expected that a given tumor cell would respond in a similar manner to drugs with a similar mechanism of action. Here, side-by-side studies comparing drug response for various chemotherapeutic agents showed that drugs of similar mechanisms of action produce similar drug response in vitro (Figure 3). It is important to note, Figure 2. High reproducibility for ChemoFx DRM using intra-operator however, that the drug pairs did not have identical responses. and inter-operator variability tests. The reproducibility of ChemoFx was Some drugs produced stronger response in particular patient demonstrated using SK-OV-3 cells treated paclitaxel (0.2 nM-0.1 μM). tumor samples. It is true that drugs with similar mechanisms Intra-operator tests included 7 assay repetitions by one operator. Inter- of action had a stronger correlation to one another than drugs operator tests included 19 assays performed by three operators. Inset: Averages and coefficients of variation (CoVs) showed similar results for with different mechanisms (Figure 4). These results suggest intra-operator and inter-operator variability tests and high reprodu- that although drugs related by mechanism of action may be cibility of ChemoFx DRM. expected to perform similarly in vivo, there may be addi- tional characteristics that make them uniquely efficacious in the treatment of individual tumors. The ChemoFx DRM allowed side-by-side comparison of heterogeneity of response observed clinically is evident at the drug response with multiple agents. The heterogeneity of cellular and molecular level and need not be attributed to other response demonstrated here supports the concept that each factors intrinsic to the in vivo environment. Some of the tumor has unique characteristics that guide drug response. possible biochemical processes that could be affecting drug Furthermore, chemotherapeutic agents with similar mecha- response include drug internalization, binding, and metabolism, nisms of action invoked similar drug response for established and specific gene and protein activation or inactivation (23). cell lines and patient tumor samples, which indicates a similar Additional in vitro research is ongoing to identify and biochemical response. These results further suggest that the characterize factors that determine patient drug response.

2808 Rice et al: Analysis of Heterogeneity and Clustering of Chemotherapeutic Response

Figure 3. Primary cultures of patient tumor specimens showed similar in vitro response to different drugs of the same mechanism of action. Five cultures of each tumor type were treated with either a platinum-based pair of drugs (carboplatin, dotted line and cisplatin, solid line) or a taxane- based pair of drugs (docetaxel, dotted line and paclitaxel, solid line). Samples are color-coded by patient sample number for visualization of response similarity. Inset: RI Scores for each of the drugs in the pair tested and percentage difference showed similarities in drug pairs. Percent Difference=100×(Absolute Value (RI Score x-RI Score y))/(RI Score x+RI Score y).

2809 ANTICANCER RESEARCH 30: 2805-2812 (2010)

Figure 4. Ovarian cancer primary patient cultures (A) and breast cancer cell lines (B) displayed strong correlation in response to different drugs of the same mechanism of action in vitro. For each figure, the lower panel is the scatter plot of drug responses, represented by the RI score. In the upper panel, the large-font number is the summary of the patient correlations (r) between drug responses to a pair of drugs, while the small-font number is the number of pairs with both drugs tested. ***Two-sided p-value, corresponding to p<0.001.

Acknowledgements 6 Lewis JD, Chagpar AB, Shaughnessy EA, Nurko J, McMasters K and Edwards MJ: Excellent outcomes with adjuvant toremifene We thank Precision’s Product Development teams, including Lahari or tamoxifen in early stage breast cancer. Cancer 2010. Sakhamuri, and Angie Mullig, Pittsburgh, PA, USA, who received 7 Edge SB and Compton CC: The American Joint Committee on payment from Precision Therapeutics Inc., for writing and editing Cancer: the 7th edition of the AJCC cancer staging manual and assistance. the future of TNM. Ann Surg Oncol 2010. 8 Shih KK, Chi DS, Barakat RR and Leitao MM Jr: Tertiary References cytoreduction in patients with recurrent epithelial ovarian, fallopian tube, or primary peritoneal cancer: An updated series. 1 Maio M, Mackiewicz A, Testori A, Trefzer U, Ferraresi V, Jassem Gynecol Oncol 2010. J, Garbe C, Lesimple T, Guillot B, Gascon P, Gilde K, Camerini 9 Homayounfar K, Liersch T, Niessner M, Meller J, Lorf T, R and Cognetti F: Large randomized study of thymosin 1, Becker H and Ghadimi BM: Multimodal treatment options for interferon alfa, or both in combination with cacarbazine in bilobar colorectal liver metastases. Langenbecks Arch Surg patients with metastatic melanoma. J Clin Oncol 2010. 2010. 2 Hirooka M, Koizumi Y, Kisaka Y, Abe M, Murakami H, 10 Gerber B, Freund M and Reimer T: Recurrent breast cancer: Matsuura B, Hiasa Y and Onji M: Mass reduction by treatment strategies for maintaining and prolonging good quality radiofrequency ablation before hepatic arterial infusion chemo- of life. Dtsch Arztebl Int 107: 85-91, 2010. therapy improved prognosis for patients with huge hepatocellular 11 Stathis A, Moore MJ: Advanced pancreatic carcinoma: current carcinoma and portal vein thrombus. Am J Roentgenol 194: treatment and future challenges. Nature Rev Clin Oncol 3: 163- W221-226, 2010. 72, 2010. 3 Bardia A, Loprinzi C, Grothey A, Nelson G, Alberts S, Menon S, 12 Kim HS, Park NH, Kang S, Seo SS, Chung HH, Kim JW, Song Thome S, Gill S and Sargent D: Adjuvant chemotherapy for YS and Kang SB: Comparison of the efficacy between resected stage II and III colon cancer: comparison of two widely topotecan- and belotecan-, a new camptothecin analog, based used prognostic calculators. Semin Oncol 7: 39-46, 2010. chemotherapies for recurrent epithelial ovarian cancer: a single 4 Chen XS, Nie XQ, Chen CM, Wu JY, Wu J, Lu JS, Shao ZM, institutional experience. J Obstet Gynecol Res 36: 86-93, 2010. Shen ZZ and Shen KW: Weekly paclitaxel plus carboplatin is an 13 Cameron D, Bell R, Aapro M and Zielinski C: What are the effective nonanthracycline-containing regimen as neoadjuvant current standards of care and recent developments in the chemotherapy for breast cancer. Ann Oncol 2010. management of breast cancer? Oncologist 11(suppl 1): 1-3, 2006. 5 Rydzewska L, Tierney J, Vale CL and Symonds PR: Neoadjuvant 14 Stordal B, Pavlakis N and Davey R: A systematic review of chemotherapy plus surgery versus surgery for cervical cancer. platinum and taxane resistance from bench to clinic: an inverse Cochrane Database Syst Rev 20: CD007406, 2010. relationship. Cancer Treat Rev 33: 688-703, 2007.

2810 Rice et al: Analysis of Heterogeneity and Clustering of Chemotherapeutic Response

15 Taheri M, Mahjoubi F and Omranipour R: Effect of MDR1 20 Heinzman JM, Brower SL, Bush JE and Silverman JF: Ex vivo polymorphism on multidrug resistance expression in breast enrichment of malignant carcinoma cells in primary culture. cancer patients. Genet Mol Res 9: 34-40, 2010. Pathology 39: 491- 494, 2007. 16 Schilsky RL: Personalized medicine in oncology: the future is 21 Heinzman JM, Rice SD and Corkan LA: Robotic liquid handlers now. Nat Rev Drug Discov 9: 363-366, 2010. and semi-automated cell quantification systems increase 17 Wu C, Xu B, Yuan P, Ott J, Guan Y, Liu Y, Liu Z, Shen Y, Yu D consistency and reproducibility in high-throughput, cell-based and Lin D. Genome-wide examination of genetic variants asso- assay. JALA 15: 7-14, 2010. ciated with response to platinum-based chemotherapy in patients 22 Faneyte IF, Schrama JG, Peterse JL, Remijnse PL, Rodenhuis S with small cell lung cancer. Pharmacogenet Genomics 20: 389- and van de Vijver MJ: Breast cancer response to neoadjuvant 95, 2010. chemotherapy: predictive markers and relation with outcome. Br 18 MI Z, Holmes FA, Hellerstedt B, Pippen J, Collea R, Backner J Cancer 88: 406-412, 2003. A, Bush JE, Gallion HH, Wells A and O’Shaughnessy JA: 23 Shih IeM and Davidson B: Pathogenesis of ovarian cancer: clues Feasibility assessment of a chemoresponse assay to predict from selected overexpressed genes. Future Oncol 5: 1641-1657, pathologic response in neoadjuvant chemotherapy for breast 2009. cancer patients. Anticancer Res 28: 1733-1740, 2008. 19 Brower SL, Fensterer JE and Bush JE: The ChemoFx assay: an ex vivo chemosensitivity and resistance assay for predicting patient response to cancer chemotherapy. In: Methods in Molecular Biology; Apoptosis and Cancer: Methods and Pro- Received April 8, 2010 tocols. Mor G and Alvero AF (eds.). Totowa NJ, Humana Press, Revised May 18, 2010 pp. 57-78, 2008. Accepted May 25, 2010

2811