Background
Triple-negative breast cancer (TNBC) is an aggressive subtype of breast cancer, disproportionately affecting young African American and African women, with limited therapeutic options. TNBC frequently display homologous recombination deficiency and high genomic instability [
1]. Therefore, veliparib, a poly (ADP-ribose) polymerase inhibitor (PARPi), is promising for the treatment of TNBC. Preclinical and early clinical studies suggest that combining PARPi with a DNA-damaging agent such as carboplatin may be more efficacious, compared to single-agent PARPi treatment, especially in patients without germline BRCA1/2 (gBRCA) mutations [
2,
3]. Although Veliparib combined with carboplatin had significant efficacy in patients with TNBC receiving neoadjuvant treatment in the I-SPY 2 trial, approximately 42% of triple negative (TN) patients did not have pathologic complete response (pCR) to veliparib-based treatment [
4].
Drug exposure from small molecules such as veliparib or carboplatin is often assumed to be relatively homogenous across diseased tissues. In current dose escalation study designs, treatment dosage or plasma exposures are directly correlated with outcomes (toxicity and efficacy) [
5]. However, studies show that the distribution of small molecules in tumors is highly variable and may not correlate with dose or plasma concentrations [
6,
7]. TNBC generally has aggressive biology with a high proliferation rate, relatively large tumor size and low microvessel density in the center of the tumor and the necrotic zones [
8]. Therefore, the microenvironment of TNBC may contribute to the variability in the uptake and distribution of veliparib and carboplatin in tumors, resulting in inadequate response and ultimately drug resistance [
9]. A preclinical study of olaparib, another PARPi, showed that abnormalities in the vasculature may hinder the penetration of PARPi [
10]. In addition, carboplatin-adduct formation (the covalent binding of carboplatin to nuclear DNA) in tumors is highly variable between patients and may be more predictive of treatment response than carboplatin exposures in plasma [
11,
12].
Liquid chromatography–mass spectrometry (LC-MS) is routinely used for quantifying drug concentrations in plasma and tissues. Matrix-assisted laser desorption/ionization mass spectrometric imaging (MALDI-MSI) can map the spatial distribution of drugs in the tissues, and has recently been used to assess detailed drug penetration in biological tissues [
13‐
17]. Inductively coupled plasma mass spectrometry (ICP-MS) is a sensitive technique for the determination of metals and is frequently used to study platinum levels in various biological matrices such as ti
ssues and peripheral blood mononuclear cells (PBMCs) from patients [
12,
18,
19]. Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) provides structural and functional information on tumor microvasculature [
20‐
22]. It can be used to monitor perfusion/permeability of tumor vasculature/tissue and identify tissue areas to which drugs would be actively delivered.
Here we hypothesize that insufficient or heterogeneous veliparib penetration and platinum adduct formation in solid tumors may lead to inadequate response to PARPi/carboplatin in some patients with TNBC. As a first step toward testing this hypothesis, we performed a feasibility study using ICP-MS, LC-MS and MALDI-MSI to quantify and visualize the penetration of veliparib and carboplatin in TNBC mouse xenografts derived from three different TNBC cell lines and in PBMCs from patients. In this study we investigated the dependency of drug penetration on dosage and tumor characteristics in these TNBC mouse models and investigated potential drug–drug interactions between PARPi and carboplatin. We further visualized penetration of a contrast agent as a surrogate for diffusible hydrophilic compounds using a DCE-MRI pilot. If our hypothesis is correct, adjusting the dose to an individual patient’s tumor for increased penetration may lead to improved response and better patient outcomes.
Discussion
Extensive knowledge of the tumor and its microenvironment suggests that penetration of small molecules may be limited in some solid tumors, and preclinical studies show heterogeneous penetration of cancer treatments in tumors and other tissues [
7,
14,
36‐
38]. However, assessment of spatial drug distribution in human tumor tissues has been hampered by technical challenges and clinical assessment of drug penetration has been limited to fluorescent-labeled or radio-labeled drugs [
39]. Platinum adducts have been measured in tumors and PBMCs, and these measurements have been successfully related to outcomes [
12,
19,
40], but using prior MS methods, at least 1 mg DNA was needed [
12,
18,
41]. In our study immunofluorescence techniques to detect carboplatin adducts have successfully been used on a cellular level, but quantification of the fluorescence signal is not straightforward, therefore comparison between samples and studies was limited [
40,
42,
43]. Platinum adducts were quantifiable in small amounts (0.10–9 μg) of DNA in tumors and PMBCs from patients 15 days post treatment. In addition, the presented MALDI-MSI approach produced high-quality images of drug distribution, using concentrations predicted to be observed in patients with TNBC. These results demonstrate the potential to use MALDI-MSI and ICP-MS to assess the distribution of PARPi and platinum adducts in clinical samples (tumor sections from large core needle biopsies).
Although susceptibility of the tumor to DNA repair insults is key in the understanding the response to PARPi, the results of our study suggest that limited drug penetration into the target lesion may account for some level of non-response. Patients with a tumor type known to have homologous recombination repair defects, such as those arising in patients with mutations in BRCA 1/2 germline or DNA-damage repair-related genes (e.g. BRCAness or mutations in DNA damage sensors ATM/ATR or PTEN), are most likely to benefit from PARPis given alone or in combination with DNA-damaging agents (the concept of synthetic lethality) [
44]. This is supported by our observation that at maximum tolerated dose, veliparib may not reach effective concentrations in non-BRCA carriers (using IC
50 values observed in breast cancer cells in vitro) [
30]. But in tumors susceptible to DNA repair insults, sufficient concentrations to inhibit PARP enzymes are still needed. Specifically at the 20 mg/kg dose - equivalent to the concentrations predicted using the 50 mg BID dose in patients - 42% of the xenograft tissues were below the limit of detection by LC-MS - below the IC
50 values observed in gBRCA mutated cells. Drug penetration was increased at the higher dose, but increased spatial heterogeneity of the veliparib distribution was also observed. This may imply that even when increasing the dose, some tumor areas may not receive the required therapeutic level at some time during the dosing interval.
The ability to measure the penetration of a PARPi and DNA-damaging agent in tumors as biomarkers of efficacy is especially relevant for PARPi due to its mechanism of action. Poly(ADPribosyl)ation, catalazed by PARP, is a crucial part of the DNA damage response system for sensing DNA lesions, activating DNA damage response pathways and facilitating DNA damage repair [
45]. The normal level of poly ADP-ribosylation is very low. At low doses of veliparib (10–50 mg), significant inhibition of PARP levels have been observed in patients in both tumor tissue and in PBMCs [
46]. Specifically, in a phase 2 study of patients with advanced solid tumors, veliparib 10–40 mg BID combined with irinotecan 100 mg/m
2 reduced tumor poly(ADP-ribose) (PAR) content in all tumor biopsies taken from patients 4 h after the morning dose (approximately at Cmax level), but PAR levels were variable and remained above the limit of detection in most samples [
47]. Following genotoxic stress (e.g. induced by chemotherapy), the level of poly(ADPribosyl)ation increases 10-fold to 1000-fold in a few seconds [
45]. Therefore, it is likely that unless PARP1 activity is virtually completely inhibited during the dosing interval, single-strand breaks will largely be repaired before the cell reaches the S-phase [
48]. In addition to PARP inhibition, PARP inhibitors may also trap PARP1 and PARP2 on damaged DNA by way of a poisonous allosteric effect [
49,
50]. Trapped PARP-DNA complexes may prevent DNA replication and transcription, killing cancer cells more effectively than catalytic inhibition. However, the capacity to trap PARP varies significantly among PARP inhibitors, with limited trapping activity estimated for veliparib: the PARP trapping IC
50 at 57.4 umol/L
50 is far above the tumor concentrations measured in our study. Furthermore, BRCA1 protein levels may vary among patients and BRCA1 protein levels inversely correlate with PARP inhibitory activity [
51]. In future studies, the correlation between veliparib tumor concentrations, biomarkers of DNA damage repair such as PARP inhibition, PARP trapping and BRCA1 protein expression, and intrinsic “DNA-damage repair deficiency” should be considered in conjunction and the relative contribution of each biomarker to predict treatment response should be assessed.
Although PARPi combined with a DNA damaging agent is a promising approach in BRCA-mutated breast cancer and TNBC, responses are variable between patients, which cannot be attributed only to differences in susceptibility of the tumor to repair DNA repair insults. In a study in metastatic BRCA-mutated breast cancer, in which patients received veliparib 400 mg BID until disease progression, only 20% showed partial response [
52]. In the BROCADE 2 study patients with locally advanced or metastatic BRCA-mutant breast cancer were treated with carboplatin/paclitaxel and veliparib 120 mg day 1-7 per 3-weekly cycle (Q3W), at 30% of the maximum tolerated dose, or carboplatin/paclitaxel and placebo [
53]. This study showed a trend but did not meet its significance cut off when evaluating progression-free survival benefit [
53]. The subsequent BROCADE 3 study (NCT02163694) evaluating standard chemotherapy, single-agent veliparib versus veliparib in combination with carboplatin and paclitaxel, is still enrolling. The phase 3 study of the PARPi olaparib monotherapy provided evidence of statistically significant and clinically meaningful progression-free survival benefit in human epidermal growth factor receptor 2 (HER2)-negative, gBRCA-mutated breast cancer, compared to treatment of the physician’s choice [
54]. In addition, veliparib 50 mg BID combined with carboplatin showed significant efficacy in TNBC in the I-SPY 2 trial, which included only three patients with a BRCA-mutation who were on neoadjuvant treatment in the I-SPY 2 trial [
4]. In this study approximately 42% of TN patients did not have pathologic complete response (pCR) to veliparib-based treatment [
4]. But in phase 3, the addition of veliparib 50 mg PO BID compared to placebo added to neoadjuvant carboplatin (AUC 6 mg/mL/min q3 weeks) and paclitaxel followed by doxorubicin + cyclophosphamide, did not have an increased pCR rate in the breast and nodes in stage II–III TNBC [
55]. Based on the observed variability in responses in breast cancer irrespective of mutations in DNA damage repair we can extend the hypothesis that insufficient or heterogeneous veliparib penetration and platinum adduct formation in solid tumors may lead to inadequate response to combination therapy across tumor types.
By inhibition of PARP1 activity, a PARPi slows down the nucleotide excision repair, thereby decreasing the ability to remove inter-strand and intra-strand platinum adducts. Platinum adducts may therefore serve as a biomarker of efficacy of the PARPi/carboplatin combination. In our study veliparib administration did not influence carboplatin exposure in plasma, but in a small number of mouse xenografts we observed increased adduct formation. This potential intracellular drug–drug interaction in mouse xenografts may be a result of the synergistic effects between these two agents and is in line with the results of Olaussen et al., who showed in vitro that platinum adduct formation increased after concomitant administration of a PARPi [
56]. In this small analysis of the phase 1 study of carboplatin in combination with the PARPi talazoparib, this drug–drug interaction was not observed in PMBCs. In PBMCs of patient gBRCA1/2 mutations, we observed increased platinum adduct formation in PBMCs, suggesting that these patients may have lower ability to remove inter-strand and intra-strand platinum adducts. As our sample size was small these findings warrant further study of platinum adduct formation and drug penetration in tumor cells.
The concentrations of veliparib in implanted TNBC xenograft tumors were 10-fold lower than the concentrations previously observed after a single dose of veliparib in a melanoma subcutaneous flank model [
57]. In our study only 35–74% of veliparib transferred from plasma to TNBC xenografts. This TNBC tumor/plasma ratio was similar to veliparib tissue/plasma ratios previously observed in less perfused tissues such as bone, eye and brain in rats [
58]. Veliparib concentrations in the highly vascularized liver tissues was higher, an effect previously also observed in a radioactivity study of veliparib in rats, with renal clearance accounting for 70% of veliparib elimination [
46]. Potentially, drug transporters may play a role in limiting drug penetration into some tumor tissues: efflux transporters ABCG2 (BRCP), ABCC2 (MRP2), ABCC4 (MRP4) [
59] and ABCB1 (P-glycoprotein) are overexpressed in MDA-MB-436 and MDA-MB-231 [
60], whereas uptake transporters OCT1 and OCT2 are highly expressed in HCC70 and MDA-MB-231, but not in MDA-MB-436 [
61]. As veliparib is a potential target for drug transporters such as ABCB1, OCT1-3, OAT1-3 and MATE1 ABCG2 [
62‐
64], these observations are consistent with higher veliparib concentrations observed in the HCC70-derived tumors. Low perfusion into poorly vascularized tumor may have prevented penetration of the PARPi into the tumor [
10]. Confirming previous studies [
13,
35,
65], veliparib, a small, hydrophilic compound with low protein binding diffused well from the cellular rim into the necrotic core of tumors. A pronounced rim enhancement was shown in DCE-MRI immediately post injection of contrast agent, i.e. more pronounced contrast enhancement at the tumor periphery compared to that at the center in these tumors. The early rim enhancement observed in this pilot study may reflect high vascularization at the rim due to aggressive growth and low vascularization of the tumor center, characteristic of TNBC [
66]. Rim contrast enhancement may explain low veliparib penetration in implanted TNBC xenografts and has been associated with death and disease recurrence in TNBC [
67,
68].
The clinical implications of studying heterogeneity in drug penetration using a single biopsy may be most relevant in the neoadjuvant setting, where limited drug penetration into the target lesion may account for some level of non-responses observed after PARPi/carboplatin treatment. In the case of minimal residual disease, penetration into the tumor core may be less relevant. The single 2D plot MALDI-MSI images shown in this study are representative only of the immediate tissue surroundings of a small target lesion. Due to the highly heterogeneous morphology of the tumors, a semiquantitative readout from a single 2D tissue section is unlikely to represent drug distribution within the entire tumor of 1 cm or greater in diameter (comprising necrotic, cellular and potentially, areas of adipose tissue). Drug penetration may be more heterogeneous in larger tumors and even more across the different tumor sites in metastatic breast cancer. Collection of multiple sites or multiple sections throughout large tumors would provide a more comprehensive understanding of drug distribution, but would substantially increase the MSI sample load. Therefore, we suggest combining MALDI-measured drug distribution from core needle biopsies in a small portion of the tumor with imaging techniques such as DCE-MRI to visualize and correlate drug localization with measurements of tumor perfusion [
69]. Another possibility is to combine MALDI-MSI imaging of drug penetration in small molecules at microscopic level with molecular imaging techniques using positon emission tomography (PET) or single-photon emission computed tomography (SPECT) to determine drug penetration in all lesions in the body at a macroscopic level. Molecular imaging probes to visualize PARP1 binding capacity by PET are currently in development [
70]. Another caveat of this study is the small number of animals and patients included and the small number of tumor PK samples. Denser sampling in the tumor at multiple time points after drug administrations would improve our understanding of spatial penetration of the drug in the tumor throughout the sampled timeframe. From this initial study, we can conclude that IPC-MS and LC-MS provide fully quantitative data and MALDI-MSI enables detailed spatial information on carboplatin and veliparib uptake into tumors. Future experiments will be targeted to quantify and localize drug distribution by MALDI-MSI by normalization to a stably labeled veliparib internal standard to further enhance the quantitative capabilities of the analysis, and validation by comparison to quantitative LCMS data from laser microdissected areas collected from sections adjacent to MALDI-MSI and H&E tissue sections.
Genomics has enabled the development of targeted therapies. Identification of TNBC subtypes such as lymphocyte predominant and luminal androgen receptor yields promise for personalized medicine in this aggressive type of breast cancer. The spatial, phenotypical and genetic composition of breast cancer and the changes in these in response to therapy are topics of current research [
71,
72]. The interaction of genomic instability within the tumor and selective pressures such as differences in tumor micro-environment, changes in endocrine stimuli, and drug treatment may result in intra-tumor heterogeneity in treatment response. The spatial heterogeneity in drug concentrations observed in our study means that there are sanctuary sites that are not or only partially penetrated by drugs. Low concentration in some tumor cells due to heterogeneous drug penetration may cause drug-resistance and treatment failure [
73]. For this reason, we suggest that the spatial distribution of drugs should be considered a potential consequence of heterogeneity in tumor composition that impacts the response to drug therapy and shapes the composition of residual disease. By studying the penetration of drugs in tumor biopsies, it may become possible to personalize dosing regimens to improve efficacy and reduce the risk of disease recurrence.