Microdose-Induced Drug–DNA Adducts as Biomarkers of Chemotherapy Resistance in Humans and Mice

This article is featured in Highlights of This Issue, p. 261

Cisplatin, carboplatin, and oxaliplatin are among the most commonly prescribed chemotherapeutic drugs and are used in over half of all chemotherapy patients for various cancers, such as bladder, lung, ovarian, and colon. Except in testicular cancer, their efficacy is modest, and drug resistance is the most common cause of treatment failure (1, 2). The feasibility of predicting drug resistance by genomics analysis has been reported for bladder cancer (3, 4). This work led to improved understanding of resistance mechanisms, and identified over 700 genes involved in cellular response to platinum-based drug treatment (5). However, the genomics approach has thus far failed to develop clinically useful tests for predicting resistance.

Platinum-based drugs kill cancer cells through covalent drug–DNA adduct formation (2, 6). Accordingly, another approach of identifying drug resistance has focused on quantification of drug–DNA adducts as biomarkers (7–12). The levels of such adducts are the result of numerous factors that govern cellular responses to drug exposure, including genetics, tumor microenvironment, kidney function, overall patient health, and others (13). The key challenge of this approach is the need to dose patients with full-dose chemotherapy in order to recover enough analyte to accurately quantify extremely low levels of drug–DNA adducts (14). Furthermore, therapeutic dosing is a poor diagnostics strategy and puts patients at risk. To address these problems, we developed the concept of “diagnostic microdosing” in which patients are administered a nontoxic subtherapeutic “microdose” of ¹⁴C-labeled drug(s) followed by blood and tumor biopsy sampling and analysis with accelerator mass spectrometry (AMS), as summarized in Fig 1. AMS can easily measure one ¹⁴C-labeled drug molecule bound to DNA per 10⁸ nucleotides (15), with a limit of detection of one adduct per 10¹² nucleotides (16) and can quantify radiocarbon-labeled drug or metabolites in urine, blood, or DNA samples (17, 18).

Most platinum-based treatments involve either cisplatin or carboplatin. Cisplatin does not have a carbon atom in the molecule and cannot be detected by AMS. Although less potent, carboplatin forms the same drug–DNA adduct crosslink structure as cisplatin (Supplementary Fig. S1), and clinical cross resistance is common (19). Consequently, these two drugs are sometimes used interchangeably in clinical practice (6, 19). Therefore, the microdosing approach to identifying carboplatin resistance can possibly be applied to cisplatin.

We previously used AMS to quantify [¹⁴C]carboplatin–DNA monoadduct formation in cancer cell lines exposed to microdose-relevant drug concentrations (12, 15, 20–23). The resulting adduct levels correlated with in vitro carboplatin cytotoxicity. AMS specifically detected carboplatin–DNA monoadducts, because the ¹⁴C-labeled cyclobutane dicarboxylate (CBDCA) group is released once the diadduct is formed (Supplementary Fig. S1; ref. 22). We report here translation of these studies into a prospective phase I clinical trial, including a cohort of bladder cancer patients (Supplementary Fig. S2). Bladder cancer was chosen for this study as a proof-of-principle because platinum-based therapy is the standard neoadjuvant treatment and is potentially curative for locally advanced cancer. The diagnostic microdosing approach may also be used for many other cancer types that implement similar platinum-based based therapies.

We tested the hypothesis that microdose-induced carboplatin–DNA adducts in peripheral blood mononuclear cells (PBMC) correlate with response to platinum-based neoadjuvant chemotherapy. We found that those patients with the highest carboplatin–DNA adduct levels were responders. However, not all responders had high adduct levels. This observation led to a second hypothesis that a combination drug partner may compensate for low platinum–DNA adducts levels to enable a response. This hypothesis was tested in mice bearing patient-derived tumor xenografts that were donated by four myoinvasive bladder cancer patients. These mice were dosed with either [¹⁴C]carboplatin and/or [¹⁴C]gemcitabine as a model of gemcitabine and platinum (GC) combination therapy. Mouse tumors with high drug–DNA adduct levels responded to chemotherapy, whereas low adduct levels correlated with resistance. Collectively, these results support the feasibility of diagnostic microdosing to improve patient outcomes by personalizing chemotherapy.

Unlabeled gemcitabine (GEMZAR) was obtained from Eli Lilly and carboplatin (CARBOplatin, 10 mg/mL) from Hospira. ¹⁴C-labeled carboplatin (specific activity 53 mCi/mmol with the ¹⁴C-label in the cyclobutane dicarboxcylic group) was obtained from GE Healthcare and [¹⁴C]gemcitabine (specific activity 58.8 mCi/mmol with the ¹⁴C-label on the aromatic nucleobase at the 2 position) was purchased from Moravek Biochemicals. [¹⁴C]carboplatin for injection was prepared under good manufacturing practices (GMP) at the GMP facility at UC Davis. The [¹⁴C]carboplatin drug substance was dissolved with sterile water for injection (WFI). The resulting solution was filter sterilized with 0.2-μm PES syringe filter into sterile glass vials and sealed with a rubber septum. Specific activity was determined by liquid scintillation counting (LSC). Mixtures of ¹⁴C-labeled and unlabeled drug were used to minimize the usage of radiocarbon and achieve the different specific activities required for microdoses and therapeutic doses. Drug solutions for the indicated experiments were prepared immediately before use.

The study “A phase 0 clinical trial of microdosing carboplatin and molecular profiling for chemoresistance” (ClinicalTrials.gov identifier NCT01261299) is an ongoing, multisite feasibility study of the diagnostic microdosing approach. The patient population consisted of non–small cell lung cancer (NSCLC) patients, stage IV with measurable lesions, and bladder transitional cell carcinoma (TCC) patients, stage II disease and above for neoadjuvant treatment, or stage III and IV metastatic disease (only data for TCC are presented here). Informed consent was obtained after the nature and possible consequences of the studies were explained. Microdoses of [¹⁴C]carboplatin were administered to patients as a diagnostic reagent, followed by standard-of-care full-dose platinum-based chemotherapy (G/C or MVAC) and evaluation of response. PBMC and tumor tissue were collected from the patients for analysis of carboplatin–DNA monoadduct frequencies. Toxicity of the [¹⁴C]carboplatin diagnostic reagent administered as a microdose was assessed using Common Terminology Criteria for Adverse Events (CTCAE). Patient response to chemotherapy was evaluated using the RECIST for correlation to carboplatin–DNA monoadduct frequency. The carboplatin dose for human chemotherapy was calculated using the Calvert formula with an AUC of 6. Therefore, individual patients were given a microdose of [¹⁴C]carboplatin (1.0 × 10⁷ dpm/kg of body weight) containing a total carboplatin dose at 1% of their therapeutic dose by a 2-minute bolus intravenous infusion (IV). Unlabeled carboplatin and [¹⁴C]carboplatin were mixed immediately before dosing and injected through the peripheral vein at one arm. Peripheral blood specimens were drawn into BD Vacutainer CPT tubes with sodium citrate (Becton Dickinson) from the other arm at specified time points before and after the administration of the microdose. PBMC were isolated within 2 hours of collection by centrifugation according to the manufacturer's instruction. A proportion of total plasma was used for LSC measurement and PK determination. Tumor biopsy specimens were collected 24 hours after microdosing. Outcomes related to chemotherapy (including response and adverse events) were collected and correlated with carboplatin–DNA monoadduct data.

All animal experiments were performed in compliance with institutional guidelines and approved by the Animal Use and Care Administrative Advisory Committee at the University of California, Davis (Sacramento). All animals were kept under pathogen-free conditions. Female NOD scid gamma severe combined immunodeficient (NSG) mice (5–8 weeks of age, body weight: 20–25 g) were purchased from The Jackson Laboratory. Patient-derived tumor xenografts (PDX) were created by subcutaneous injection at the flank of approximately 1-mm³ cancerous tissue from a patient's primary tumor directly into a NSG. PDX tumors (passages 2–4) were allowed to grow to at about 200 mm³ before being assessed for drug efficacy and microdosing studies.

When the xenografts reached approximately 200 mm³, mice were intravenously injected through the tail vein with one microdose 10 μL/g of radiolabeled carboplatin (0.375 mg/kg, 50,000 dpm/g) or gemcitabine (0.092 mg/kg, 1,000 dpm/g) or one therapeutic dose of radiolabeled carboplatin (37.5 mg/kg, 50,000 dpm/g) or gemcitabine (9.2 mg/kg, 1,000 dpm/g). Tumor tissues were collected after 4 and 24 hours and stored at −80°C until DNA isolation.

For each treatment group, 8–12 mice were used to allow statistical analysis. When tumor sizes reached approximately 200 mm³, mice were randomly assigned into different treatment groups and treated with indicated therapeutic agents (cisplatin at 2 mg/kg IV Q7Dx3, carboplatin at 30 mg/kg IP Q7Dx4, and gemcitabine at 150 mg/kg IP Q7Dx4). Mice were monitored for tumor growth and alterations in weight. Tumor size was determined with a caliper, and tumor volumes were calculated using an ellipsoidal formula (1/2 × length × width squared). Mice were sacrificed when the tumor length reached 20 mm in any direction.

DNA was extracted from minced tumor tissue or PBMC pellets with the Wizard Genomic DNA Purification Kit according to the manufacturer's instruction (Promega). DNA quantity and quality was determined with a NanoDrop 1000 spectrophotometer, the purity was ensured by obtaining a 260/280 nm OD ratio of approximately 1.9.

All DNA samples were submitted to Lawrence Livermore National Laboratory (LLNL) for AMS analysis of radiocarbon content using an established protocol (24). Ten micrograms of DNA per sample was converted to graphite and measured by AMS for ¹⁴C quantification as previously described.

Total RNA was isolated using the Qiagen RNeasy Mini Kit and QIAshredder spin columns (Qiagen) following the manufacturer's instructions. Column-bound RNA was pretreated with Qiagen RNase-Free DNase to remove genomic DNA before elution with RNase free water. Nucleotide concentration and purity was determined via NanoDrop 1000 spectrophotometer. RNA purity was ensured by obtaining a 260/280 nm OD ratio approximately 2.0. cDNA was synthesized using RevertAid RT Kit (Thermo Scientific) using 1 μg freshly purified RNA and random hexamer primer. Real-Time PCR (RT-PCR) was performed with BioRad CFX96 Real-Time System instrument in triplicates of each sample in a volume of 10 μL using a standard protocol (95°C, 2 minutes, 40 cycles: 95°C, 15 seconds; 60°C, 30 seconds; 72°C, 30 seconds; melting curve 65°C–95°C, 5 seconds). Per reaction, 4 ng cDNA was mixed with 10 pmol forward and reverse primer, SYBR green and EconoTaq PLUS 2X Master Mix (Lucigen Corporation). Primer sequences (Integrated DNA Technologies) included: GAPDH: forward 5′-CGCGGGGCTCTCCAGAACATC-3′, reverse 5′-CTCCGACGCCTGCTTCACCAC-3′; RRM1: forward 5′-GCCGCCAAGAACGAGTCAT-3′, reverse 5′-AGCAGCCAAAGTATCTAGTTCCA-3′; RRM2: forward 5′-GTGGAGCGATTTAGCCAAGAA-3′, reverse 5′-CACAAGGCATCGTTTCAATGG-3′; hENT1: forward 5′-GTGCCTTCGGCTACTTTATCAC-3′, reverse 5′-GCTAATGAGGTCCAACTTGGTCT-3′; ERCC1: forward 5′-CTACGCCGAATATGCCATCTC-3′, reverse 5′-GTACGGGATTGCCCCTCTG-3′. The mRNA expression level of indicated genes was normalized to the internal GAPDH expression via the 2^−ΔΔCt method (25).

Whole-exome sequencing (WES) on the Illumina platform was previously performed for each bladder cancer PDX model (26), but with new data analysis specific to DNA repair genes. Details are given in Supplementary Information and the raw sequence data files are publicly available through the NCBI BioProject database (https://www.ncbi.nlm.nih.gov/bioproject/) with Accession Number PRJNA352282.

All statistical analyses were performed using GraphPad Prism software (GraphPad Software Inc.) using one-way ANOVA with Newman–Keuls multiple comparison post hoc or a two-tailed Student t test. A P value below 0.05 was considered statistically significant. All experiments were carried out at least in triplicate in order to enable statistically significant comparisons of the results. All results are expressed as the mean ± SEM unless otherwise noted. A simple correlation of the adduct levels and response is reported.

A clinical feasibility study (ClinicalTrials.gov identifier NCT01261299) was initiated with a starting (level 1) dose of [¹⁴C]carboplatin of 10⁷ dpm/kg with a total carboplatin dose of 1% of the therapeutic dose (target AUC of 6). With this dose composition, the ¹⁴C-signal in DNA isolated from PBMC was approximately 10 to 100 times the background, which allowed accurate adduct measurement by AMS.

A total of 21 patients were accrued, including 14 bladder cancers, 6 NSCLC and 1 mediastinal mass that was subsequently identified as Hodgkin's lymphoma. Of the bladder cancer patients, 10 completed the entire study, including microdosing, blood sampling, and subsequent full-dose chemotherapy. The patients' characteristics are shown in Supplementary Table. S1. This paper only discusses the bladder cancer data. No microdose-associated adverse events were observed. Three patients received an additional dose of [¹⁴C]carboplatin at the end of their therapeutic dose infusion of unlabeled carboplatin in order to compare the therapeutic pharmacokinetics (PK) and formation of drug–DNA adducts with those of the microdose.

Radiation exposure from the diagnostic microdose was not greater than 0.1 mSv. In comparison, the annual effective radiation dose equivalent from natural internal sources is 1.6 mSv per person (27), and exposure for an abdominal CT scan is 10 mSv (28). Therefore, the diagnostic [¹⁴C]carboplatin administered as a microdose appears to be safe in this patient population.

We determined whether the microdose results in a proportionally lower drug exposure (area under the curve) compared with the therapeutic doses of carboplatin. A microdose and a full therapeutic dose containing [¹⁴C]carboplatin was administered to three patients, followed by blood sampling over 24 hours and quantitation of plasma ¹⁴C concentration by LSC (Fig. 1B; Supplementary Table. S2). The PK of the therapeutic and microdoses showed nearly identical elimination kinetics (Supplementary Table. S2). The clearance of carboplatin is known to occur in a triphasic manner with an initial (T_1/2α), a second (T_1/2β), and a terminal (T_1/2γ) phase half-lives reported to be 10 to 98 minutes, 1.3–6.0 hours and 8.2–125 hours, respectively (29–32). The measured T_1/2α and T_1/2β half-lives of 13–55 minutes and 4.5–8.7 hours reside well within the range of these previously reported therapeutic values. At 24 hours after dosing, over 99% of ¹⁴C labeling had been cleared from the blood, suggesting this is he optimal time for biopsy with little concern of ¹⁴C contamination. Comparison of plasma carboplatin equivalent concentration showed dose proportional linearity between microdoses and therapeutic doses (Fig. 1C).

Without prior sample fractionation, LSC only measures total ¹⁴C concentration of a given sample without differentiating the parent drug from metabolites. Therefore, the resulting PK calculations could have been confounded by the circulation of free [¹⁴C]CBDCA. To further validate the approach of determining [¹⁴C]carboplatin PK with LSC, we used inductively coupled plasma mass spectrometry (ICP-MS) to determine the platinum concentration in specimens from one patient. At the therapeutic dose a significant linear correlation was observed between ICP-MS measurements and LSC measurements (R² = 0.724, P = 0.032; Fig. 1E). However, ICP-MS was not able to detect platinum in the microdosing specimens, which supports the utility of AMS for diagnostic microdosing applications.

At 24 hours after intravenous microdosing, PBMC were isolated from patient blood samples followed by DNA purification. Radiocarbon content in the DNA was measured by AMS according to published protocols (24). The concentration of carboplatin–DNA monoadducts was then calculated in units of monoadducts per 10⁸ nucleotides (Fig. 2). Patients received platinum-based chemotherapy within four weeks after the microdosing procedure (typically GC or MVAC). Response was confirmed if a patient had pT1 or less tumor (partial response: PR, complete response: CR) at the time of cystectomy; or resistance if a patient still had pT2 or greater (disease progression: DP), based on standard RECIST criteria (33).

Seven patients responded to chemotherapy (gray circular symbols), whereas three patients showed disease progression (black square symbols; Fig. 2A; Supplementary Table. S2). Three patients did not receive subsequent chemotherapy: one with renal failure, and the other two with a clinical diagnosis of T2 disease because of large tumor mass, but were diagnosed with T1 disease at the biopsy during chemotherapy. Responders exhibited approximately 2-fold higher mean PBMC monoadduct levels (gray dashed line) than nonresponders (black solid line; 0.741 ± 0.346 vs. 0.283 ± 0.202 monoadducts/10⁸ nt, respectively, P = 0.069). There were two distinct groups: one with high (0.941 ± 0.030 adducts per 10⁸ nt) and the second group with low (0.266 ± 0.158 adducts per 10⁸ nt) DNA adduct levels (P < 0.0001). All five patients in the high adduct level group responded to chemotherapy. The low adduct level group included three nonresponders and two responders. Tumor adduct levels were measured for five patients. One with a very high carboplatin–DNA adduct levels responded, whereas the other tumor samples exhibited no obvious trends (Fig. 2B).

Since two patients with low PBMC adduct levels also responded to chemotherapy, we hypothesized that tumor response in those patients with low carboplatin–DNA adducts was induced by the other components of the treatment regimen. We developed a diagnostic microdosing protocol to identify resistance to gemcitabine for use in a mouse model of human bladder cancer. We established four patient-derived tumor xenograft (PDX) mouse models from randomly selected myoinvasive bladder cancer patient volunteers using previously described protocols for nod scid gamma severe combined immune deficient (NSG) mice (26). These models were named BL0269 (TM00015), BL0293 (TM00016), BL0440 (TM00024), and BL0645 (TM01339), where the BL number is the UC Davis Comprehensive Cancer Center Biorepository number (http://www.ucdmc.ucdavis.edu/cancer/research/sharedresources/specimen.html) and the TM number is the Jackson Laboratory PDX database identifier (http://tumor.informatics.jax.org/mtbwi/pdxSearch.do). Retrospective analysis of routine quality control IHC staining showed one model, BL0645, as being CD45⁺, which indicates that this tumor formed a lymphocytic tumor. Formation of lymphocytic tumors from subcutaneous implantation of tumor tissue has been previously reported for PDX (34). However, we initiated this work under the assumption that the xenograft represented a tumor of origin from the bladder. The xenograft was a solid tumor with vasculature and in some regions necrosis, consistent with the other bladder tumors used in this study. Regardless of the potential molecular differences in origin, BL0645 was useful for this study as a model of synergy between cisplatin or carboplatin and gemcitabine. We included data from this model because it is not necessary for the tumor to be rigorously proven to be bladder derived in order to demonstrate correlations between drug–DNA adduct levels and treatment response.

We assessed the sensitivity of each PDX to cisplatin, carboplatin, or gemcitabine as single agent, as well as to GC combination chemotherapy (Fig. 3; Supplementary Fig. S3). Sensitivity was determined by changes in tumor volume over time (Supplementary Fig. S3) and by tumor-specific survival analysis (Fig. 3). Of the four distinct tumor models, BL0269, BL0293, and BL0645 were resistant to cisplatin/carboplatin chemotherapy. BL0269 and BL0645 were resistant to gemcitabine. We demonstrated that cisplatin shows similar response efficacy to carboplatin in the mouse models (Supplementary Fig. S3). Median survival of the carboplatin treatment groups is similar to the corresponding cisplatin treatment groups; for BL0269, BL0293, and BL0440 median survival was 39, 28, and 64 days, respectively, compared with cisplatin median survival of 38, 35, and 80 days after treatment initiation (Supplementary Table. S3).

NSG mice bearing indicated PDX were given a microdose of [¹⁴C]carboplatin (0.375 mg/kg of body weight; Fig. 4A and B) or [¹⁴C]gemcitabine (0.092 mg/kg of body weight; Fig. 4C and D). Each microdose was chosen to target approximately 1% of the human plasma maximum concentration of 1 μmol/L for carboplatin and 0.3 μmol/L for gemcitabine. Animals were sacrificed at 4 hours (Fig. 4A and C) and 24 hours (Fig. 4B and D) after microdose bolus IV injection, and drug–DNA adduct levels were determined from purified tumor tissue DNA by AMS. Carboplatin- and gemcitabine–DNA adduct levels correlated with PDX drug sensitivity to single-agent treatment in NSG mice (Fig. 3+4; Supplementary Fig. S3). The carboplatin- and gemcitabine-sensitive tumor BL0440 showed significant higher mean carboplatin monoadduct levels compared with the other resistant models. At 24 hours, BL0440 tumor DNA exhibited 1.5- to 2.2-fold higher mean monoadduct levels than BL0269, BL0293, and BL0645 [2.73 ± 0.66 vs. 1.90 ± 0.80 (P < 0.05), 1.51 ± 0.55 (P < 0.01), and 1.27 ± 0.31 (P < 0.01) monoadducts/10⁸ nt, respectively; Fig. 4B]. There was no significant difference in carboplatin adduct levels between the three resistant PDX models.

Similarly, at 24 hours, the two gemcitabine-sensitive BL0293 and BL0440 PDX models display 4.1- to 6.3-fold higher gemcitabine adduct levels than the resistant PDX models BL0269 and BL0645 [10.1 ± 6.47 vs. 2.44 ± 0.49 (P < 0.05) and 2.24 ± 0.81 (P < 0.05) and 14.1 ± 11.1 vs. 2.44 ± 0.49 (P < 0.01) and 2.24 ± 0.81 (P < 0.05) adducts/10⁸ nt, respectively]. There is no significant difference in gemcitabine–DNA adduct levels between the two resistant PDX models. Compared with 4 hours, at 24 hours the sensitive PDX models showed a slight increase adduct levels, whereas the resistant model BL0269 showed a significant decrease in adduct frequency and BL0645 showed relatively low levels of gemcitabine–DNA adduct formation (Fig. 4C+D).

We also generated a gemcitabine-resistant PDX subline (BL0293R) through repeated treatment of mice carrying BL0293 tumors with therapeutic doses of gemcitabine. After 4 treatment cycles, relapsed tumors were reimplanted into NSG mice and treatment was resumed for an additional 4 cycles. The combined 8 cycles of full-dose gemcitabine chemotherapy resulted in BL0293R tumors that did not respond to gemcitabine treatment (Fig. 4E). Progression-free survival after chemotherapy decreased from 16.9 ± 1.0 days for BL0293 to 2.3 ± 0.7 days for BL0293R (P < 0.0001). A 2.5-fold decrease in gemcitabine adduct formation at 4 hours and 24 hours was detected in BL0293R compared with the parental model BL0293 [2.89 ± 1.22 vs. 7.17 ± 1.97 (P < 0.001) at 4 hours, and 4.02 ± 1.23 versus 10.1 ± 6.47 (P = 0.044) adducts/10⁸ nt at 24 hours, Fig. 4F].

We analyzed the effect of combination GC chemotherapy on drug–DNA adduct formation using BL0645 (Fig. 5). This model was resistant toward each of the single agents, cisplatin and gemcitabine, but had strongly enhanced sensitivity toward gemcitabine/cisplatin (Gem/Cis) combination therapy, consistent with a synergistic effect (Supplementary Fig. S3G). BL0645 tumor-bearing mice were administered a microdose of [¹⁴C]carboplatin alone (white bar) or a microdose of [¹⁴C]carboplatin in combination with a microdose of unlabeled gemcitabine (black bar). The combination of [¹⁴C]carboplatin plus unlabeled gemcitabine resulted in a significant 1.7-fold increase in carboplatin-DNA monoadduct levels (1.27 ± 0.31 vs. 2.10 ± 0.60 adducts/10⁸ nt, P = 0.024) when compared with a microdose with single agent [¹⁴C]carboplatin (Fig. 5A). Similarly, BL0645 bearing mice receiving a microdose of [¹⁴C]gemcitabine in combination with a microdose of unlabeled carboplatin (white bar) showed a 1.6-fold increase in gemcitabine adduct levels (2.24 ± 0.81 vs. 3.66 ± 1.13 adducts/10⁸ nt, P = 0.051) compared with [¹⁴C]gemcitabine alone (black bar, Fig. 5B). In contrast, BL0269, which is resistant to both single agents and combination therapy (Fig. 3A), showed no significant increase in adduct formation upon combination treatment [Fig. 5C and D; 1.90 ± 0.80 carboplatin adducts/10⁸ nt for the combination versus 2.22 ± 0.44 carboplatin adducts/10⁸ nt (P = 0.529) in the single-agent carboplatin group, and 2.44 ± 0.49 gemcitabine adducts/10⁸ nt for the combination versus 1.85 ± 0.49 gemcitabine adducts/10⁸ nt (P = 0.121) for the gemcitabine group]. Clearly, these observations of combination-induced increases in adduct levels the BL0645 line are consistent with the synergistic antitumor activity of the combination therapy for this tumor.

Similar evidence of synergy was obtained in BL0645 tumors treated with therapeutic levels of [¹⁴C]carboplatin or [¹⁴C]gemcitabine (Fig. 6). A therapeutic dose of the combination [¹⁴C]carboplatin plus unlabeled gemcitabine led to a significant increase in adduct levels (94.8 ± 16.0 vs. 209.6 ± 64.7 monoadducts/10⁸ nt, P = 0.032) compared with treatment with a therapeutic dose of [¹⁴C]carboplatin alone (Fig. 6A). Likewise, combination treatment of therapeutic doses of [¹⁴C]gemcitabine with unlabeled carboplatin increases gemcitabine adduct levels (169.2 ± 9.1 versus 268.1 ± 182.8 adducts/10⁸ nt, P = 0.421) compared with [¹⁴C]gemcitabine alone (Fig. 6B), although this was not a significant difference. There was an approximately 100-fold difference in drug–DNA adduct formation when a 100-fold difference in drug concentration was administered for either drug (Fig. 6C).

We next correlated response to the expression of a selected panel of genes that are known to be involved in GC drug resistance, including human equilibrative nucleoside transporter-1 (hENT1), ribonucleotide reductase (RNR, encoded by the RRM1 and RRM2 genes), and DNA repair protein, excision repair cross complementing group 1 (ERCC1). Increased expressions of these markers were previously reported to be involved in acquired resistance to gemcitabine- and platinum-based chemotherapy in some cancers but not in others (35). We did not find any correlation of the expression of these genes with resistance. RT-PCR analysis of gemcitabine resistance markers RRM1 and RRM2 showed decreased mRNA expression in the gemcitabine-sensitive model BL0293, and elevated levels in another gemcitabine-sensitive model BL0440 (Supplementary Fig. S4A and B; Supplementary Table. S4). hENT1 expression was the highest in the sensitive model BL0440 and is lowest in the resistant model BL0269 (Supplementary Fig. S4C). ERCC1 is involved in DNA repair process and increased expression is reported to be associated with resistance to platinum-based therapy (36). However, RT-PCR data show that ERCC1 expression is highest in the carboplatin sensitive tumor model BL0440 compared with the resistant tumors BL0269, BL0293, and BL0645 (Supplementary Fig. S4D).

We investigated the mutation status of genes associated with DNA repair in each of the PDX models (BL0269, BL0293, BL0440, and BL0645). Targeted sequencing data (whole-exome or JAX Cancer Treatment Profile) for each of the PDX (26) were filtered for 184 known DNA repair genes, including those found to correlate with response to neoadjuvant chemotherapy (NAC), such as ATM, FANCC, RB1, BRCA1, BRCA2, ERCC1, ERCC2, and RAD51C (Supplementary Tables S5 and S6; refs.3, 36–39). A total of 112 somatic variants were identified in the four PDX models (Supplementary Table. S7). Each PDX tumor harbored two or more mutations in genes associated with sensitivity to cisplatin that are predicted to have moderate or high functional impact, including ATM/FANCC/RB1 (3,) ERCC1/2 (36, 38), and homologous recombination deficiency (HRD; ref. 39). While each PDX theoretically possesses one or more molecular determinant(s) of cisplatin sensitivity, it is notable that BL0440 exhibited the highest sensitivity and harbors the highest number of mutations (n = 4), and in genes from 3 different prediction categories, in addition to a TP53 mutation. In contrast, BL0645 contained 3 mutations in BRCA1/2, which were within a single category (HRD), and displayed marked resistance to cisplatin.

The main goal of the ongoing clinical study is to correlate carboplatin–DNA adducts to therapy response in patients. For the PDX mouse models with urothelial bladder carcinomas, we expanded on this theme to include gemcitabine incorporation into DNA. This work has provided novel insights into the biology of in vivo cellular responses to platinum-based treatment and delineates potential opportunities for personalized therapeutic intervention. Treatment for muscle-invasive bladder cancer has not advanced beyond platinum-based combination chemotherapy and surgery in the past 30 years (40). Standard-of-care guidelines recommend inclusion of platinum-based NAC, but this option is used in less than 20% of patients for a variety of reasons, including concerns over lack of response to the treatment. Therefore, the ability to predict response to platinum-based chemotherapy would be highly impactful.

Positive correlations between therapy-induced levels of platinum-based drug–DNA adducts in PBMC or other surrogate tissues and good clinical outcome have been reported in the literature for a variety of different solid tumors (7–11, 41). We evaluated the feasibility of a phenotypic approach to predicting response to platinum-based chemotherapy in bladder cancer patients by using diagnostic microdosing and AMS analysis of DNA isolated from PBMC and biopsy samples. The linear correlation between carboplatin–DNA adduct levels induced by microdose and therapeutic dose exposures is consistent with our previous cell culture studies (23). The linearity of the pharmacokinetics between these two drug doses (Fig 1C) can be explained by the fact that carboplatin is an alkylating agent that directly reacts with DNA without the need for metabolic activation or transformation (19). The PBMC carboplatin–DNA adduct data were distributed into two clusters: those patients with adduct levels above 0.7 monoadducts/10⁸ nt all responded to subsequent platinum based therapy; and those patients with lower levels of adducts were a mixture of responders and nonresponders. With this cutoff, there was a clear highly significant difference (P < 0.0001) between high and low adduct level groups of patients (Fig. 2A). The ability to differentiate between responders and non-responders using the mean carboplatin–DNA adduct levels for each patient group has yet to reach statistical significance (P = 0.069). However, considering the small number of patients in the study so far, these statistics support the feasibility of diagnostic microdosing for bladder cancer. Ultimately, we envision using carboplatin–DNA adduct levels to predict response in patients by establishing one or more biomarker cutoff levels in a larger clinical study. These cutoff levels will then be used to define positive and negative predictive values for a diagnostic test.

We also showed evidence that killing of PDX bladder cancer cells in mice by gemcitabine and cisplatin/carboplatin requires a threshold level of drug–DNA damage from one or both drugs combined, and that microdose-induced drug–DNA adduct levels are predictive of whether cells can achieve that threshold during full-dose therapy. One PDX was rendered resistant to gemcitabine by repeated therapeutic dosing. This tumor model of acquired resistance had microdose-induced gemcitabine-DNA adducts that were apparently below the threshold for this drug. Together, these data indicate that the diagnostic microdosing approach can be used as a phenotype-based predictive test for chemotherapy response.

The literature covering platinum–DNA damage and DNA repair capacity arguably supports the feasibility of PBMC as a surrogate tissue for predicting tumor response to chemotherapy. In one study, platinum–DNA adduct formation in PBMCs was found to be more predictive for tumor response to platinum-based therapy than previous platinum-based therapy, stage of disease, histological type and tumor grading (42). A significantly better disease-free survival was reported for cisplatin-treated head and neck carcinoma patients with higher adduct levels (9). However, not all investigators found a relationship between surrogate tissue adduct levels and tumor response (43, 44). In spite of several clinical investigations, the utility of platinum–DNA adducts as predictive biomarkers remains unresolved due to a lack of large studies with a common design with good measurement sensitivity and precision.

Another approach has focused on correlating DNA repair capacity of PBMCs to clinical outcomes. In a large clinical study, Wang and colleagues measured DNA repair capacity in cultured peripheral T lymphocytes from 591 NSCLC patients that were treated with platinum-based chemotherapy (45). They found a statistically significant overall inverse correlation between DNA repair capacity and survival, especially in early stage tumors. In another study, the ratio of γH2AX, an indicator of DNA double-strand breaks, and the DNA repair protein MRE11 in PBMC significantly correlated to cisplatin and PARP inhibitor therapy response in 15 ovarian cancer patients (46). These reports are consistent with our 24-hour PBMC data, which is enough time for DNA repair to modulate the adduct levels.

The clinical microdosing study was performed with one chemotherapeutic drug, but patients were treated with combination chemotherapy. One potential pitfall is that chemoresistance to one drug may be compensated for by the partner drug in the combination therapy. This phenomenon potentially explains the results for two of our responsive patients who had low microdose-induced carboplatin–DNA adducts. To address this issue, we used gemcitabine microdosing in PDX-bearing mice, which has only previously been reported in cell culture (47). Our study indeed showed that the levels of gemcitabine incorporated into DNA correlated with the response of PDXs to chemotherapy.

Another important finding of this study is that high levels of drug–DNA adducts from either component of a two drug combination is sufficient to inhibit tumor growth. This observation supports the hypothesis that a threshold level of drug–DNA damage from one or both drugs is required to achieve a response to therapy. In bladder cancer, the first-line chemotherapy is usually GC (gemcitabine and cisplatin/carboplatin) or MVAC (methotrexate, vinblastine doxorobucin/Adriamycin and cisplatin). Many physicians assume that all of the drug components in these regimens contribute to efficacy. In contrast to this assumption, even in a limited set of PDXs, some tumors responded to only one of the drugs, whereas only one model needed both drugs synergistically in order to respond. Clearly, the diagnostic microdosing approach can dissect the influence of individual combination chemotherapy components for drugs that interact with DNA. This capability may be clinically useful considering the heterogeneity of tumor response to a given drug combination.

We undertook a molecular analysis in four PDX tumor models of a few of the major mechanistic pathways involved in platinum-based and gemcitabine drug resistance. For example, the upregulation of ERCC1 expression has been reported to correlate significantly with cisplatin resistance for some cancers (36, 48, 49). Despite this, ERCC1 has been shown to have unacceptable sensitivity and specificity as a biomarker and has thus failed to gain wide-spread acceptance as standard of care for therapeutic guidance. In this study, we saw no correlation between expression level of ERCC1 and PDX tumor response toward chemotherapy (Fig. 3; Supplementary Figs. S3 and S4). Gemcitabine kills cells mainly through its incorporation into DNA and termination of DNA replication (50). Several mechanisms have been implicated in mediating gemcitabine resistance, including loss of membrane import transporters, downregulation of dCK, increased formation of inactive metabolites (dFdU), upregulation of RNR or downstream inhibition of caspase executioners that control apoptosis signaling (51, 52). Interestingly, DNA repair is not associated with cellular response to gemcitabine–DNA adducts (53), even though the in vitro cytotoxicity directly correlates with the level of drug incorporation into DNA (50, 54). Analysis of three genes commonly found to be implicated in resistance toward gemcitabine-based chemotherapy (RRM1, RRM2, and hENT1) showed no correlation between expression level and PDX tumor response toward chemotherapy (Fig. 3; Supplementary Figs. S3 and S4). Interestingly, somatic mutations in DNA repair genes, including RB1, FANCC, ATM, and others, did not obviously confer sensitivity to cisplatin, carboplatin, or gemcitabine. These preliminary findings highlight the difficulty of finding generally applicable markers of chemotherapy resistance.

In our study, we measured drug–DNA adduct levels from PBMC and tumor samples as putative predictive biomarkers of therapy response. The formation and DNA repair of these adducts are governed by a myriad of factors including but not limited to hundreds of genes, gene–gene interactions, proteins, stroma, drug PK, organ function, and even overall patient health. Therefore, drug–DNA adduct levels as pharmacodynamic endpoints are potentially more comprehensive and useful than genomic analysis of limited sets of markers for predicting drug response, at least for some drugs. We previously showed that platinum–DNA adduct levels and chemoresistance in isogenic cell cultures can be modulated by ERCC1 and RRM1 mRNA levels (23). Here, we extended this analysis to include a limited number of bladder cancer PDX, which showed no correlation of these genomic markers with therapeutic outcome.

The diagnostic microdosing approach is a platform technology. Even though this study focused on bladder cancer, the strategy can be applied to many other cancer and drug types as long as the target–drug complex can be measured by AMS. For example, our ongoing trial is also accruing NSCLC patients. In addition, we have another trial with [¹⁴C]oxaliplatin in colorectal cancer (clincaltrials.gov identifier#: NCT02569723).

Human PK (AUC) and drug–DNA adduct formation of [¹⁴C]carboplatin in PBMC are dose proportional between a microdose and a therapeutic dose. Dose proportionality of drug–DNA adduct formation was also observed in the PDX mouse model. Microdose-induced drug–DNA adduct levels correlated with tumor response in the PDX model (inhibition of tumor growth and survival) and bladder cancer patients (tumor response). This applies to the use of [¹⁴C]carboplatin for prediction of cisplatin and carboplatin response, and also applies to the use of [¹⁴C]gemcitabine diagnostic microdosing to predict response to the nucleotide analogue gemcitabine (in mice). Consistent with our previous findings in cell culture, high levels of in vivo microdose-induced drug–DNA adduct levels correlated with tumor response.

G.D. Cimino has ownership Interest (including patents) in Accelerated Medical Diagnostics Inc. C.-X. Pan has ownership interest (including patents) in Accelerated Medical Diagnostics Inc. P. Henderson is the CEO of Accelerated Medical Diagnostics Inc., and has ownership Interest (including patents) in the same. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M. Zimmermann, G.D. Cimino, R. de Vere White, C.-X. Pan, P.T. Henderson

Development of methodology: K.W. Turteltaub, G.D. Cimino, C.-X. Pan

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Zimmermann, S.-S. Wang, H. Zhang, T.-Y. Lin, M. Malfatti, K. Haack, T. Ognibene, H. Yang, S. Airhart, K.W. Turteltaub, A. Drakaki, K. Chamie, R. de Vere White

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Zimmermann, S.-S. Wang, K. Haack, T. Ognibene, C.G. Tepper, C.-X. Pan

Writing, review, and/or revision of the manuscript: M. Zimmermann, H. Zhang, M. Malfatti, S. Airhart, K.W. Turteltaub, G.D. Cimino, C.G. Tepper, A. Drakaki, K. Chamie, R. de Vere White, C.-X. Pan, P.T. Henderson

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Zimmermann, C.-X. Pan

Study supervision: M. Zimmermann, K.W. Turteltaub, G.D. Cimino, C.-X. Pan, P.T. Henderson

Other (performed experiments and collected DNA samples): T.-Y. Lin

The authors are grateful to Ryan R. Davis (UCDCCC Genomics Shared Resource, Department of Pathology and Laboratory Medicine) for his expert technical assistance for the analysis of the WES data. The results here are also based in part upon data generated by the TCGA Research Network: http://cancergenome.nih.gov/. We are grateful to all of the patients and families who contributed to this study, as well as to Susan and Gerry Knapp for their financial support.

This work was funded by NIH grants CA93373 (R. de Vere White), SBIR contracts to AMD Phase I HHSN261201000133C (P.T. Henderson), Phase II HHSN261201200048C (P.T. Henderson), LLNL grants LDRD 08-LW-100 (P.T. Henderson and M. Malfatti), NIH/NIGMS P41 RR13461 (K.W. Turteltaub), American Cancer Society (C.-X. Pan), the Knapp Family Fund (P.T. Henderson), and VA Career Development Award-2 (C.-X. Pan). This work was performed (partially) at the Research Resource for Biomedical AMS, which is operated at LLNL under the auspices of the U.S. Department of Energy under contract DE-AC52-07NA27344. The Research Resource is supported by the National Institutes of Health, National Center for Research Resources, Biomedical Technology Program grant P41 RR13461. The UC Davis Comprehensive Cancer Center Genomics Shared Resource is supported by Cancer Center Support Grant P30CA093373 from the NCI.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.