Understanding brain penetrance of anticancer drugs

Abstract

This paper explicates the impact of tumor capillary permeability for glioma World Health Organization (WHO) grades II to IV on brain-penetrant drug entry and distribution within the tumor and the brain adjacent to tumor (leading edge). In addition, we consider the distribution of non–brain penetrant drugs and how, in some cases, large-molecular-weight drugs might achieve good distribution into tumor and brain adjacent to tumor.

Introduction and Problem

Here we assess some pharmacological and pharmaceutical mantras that have been guiding principles for early drug development during the past decades. The 50 years of clinical and research experience of one of the authors (V.A.L.) concentrated on developing therapies for treating infiltrative primary CNS tumors. Additionally, preclinical pharmacokinetic research focused on the permeation of drugs from blood to brain, tumor, and nerve as well as diffusion within brain and distribution of drugs in cerebrospinal fluid (CSF).

Infiltrative gliomas of the brain and spinal cord are divided into 3 main histological groups based on their histology and molecular genetics: astrocytoma, oligodendroglioma, and ependymoma tumors. These tumors are also classified by increasing malignancy from World Health Organization (WHO) grade II to grade III (anaplastic phenotype) and grade IV (glioblastoma) tumors. Characteristically, these tumors infiltrate (invade) brain and spinal cord parenchyma, typically along white matter tracts. The tumors putatively stimulate establishing new capillaries that are more permeable (leaky) than normal brain capillaries. This is accomplished by opening tight junctions between adjacent capillary endothelial cells, forming breaks and fenestra within them, and/or improperly formed capillary endothelial cells. New vessel formation (neovascularization) putatively parallels the increasing malignant phenotype of infiltrative gliomas. Therefore, tumor cells of lower phenotypic malignancy may invade adjacent brain and spinal cord without stimulating neovascularization. Consequently, some infiltrative tumor cells may “hide” behind intact portions of the blood–brain barrier (BBB). This concept is the basis for the mantra that if an anticancer drug is to be effective against infiltrative gliomas, it must be brain penetrant.^1–15

A difficulty with this dogma of the importance of using anticancer drugs that easily cross the BBB is, however, the lack of challenge implicit in the fact that outside of alkylating agents, no new anticancer drugs have received regulatory approval for the treatment of gliomas¹⁶ except for bevacizumab for glioblastoma (GBM). Bevacizumab was approved by the FDA for the treatment of GBM but with no clear evidence of an improvement in overall survival (OS).¹⁷ Secondarily, this concept of using only drugs that can traverse the BBB is challenged by reexamining our understanding of brain and tumor permeability and how this different perspective might change our view of regional drug distribution in glial tumors.

While many scientists believe the restriction to passive diffusion of drug through the BBB is an absolute restriction, the authors’ hypothesis in this paper is that passive permeability is quantitative and what is considered to be the BBB rather signifies a level of restriction of movement of standard chemicals and drugs that appears to be promulgated by physical measures of molecular size, lipophilicity, and polarity.^6,8,18 In addition, because of the presence of efflux pumps in endothelial cells, sometimes drug properties are further optimized to decrease efflux of the drug by transporters to increase drug concentrations in the brain.

The relationship between permeability in brain and intracerebral (IC) tumor is predictable, with poorly permeant drug molecules distributing by diffusion in tumor more than in brain.^6,19 In contrast, distribution of drugs with high brain permeability is governed to a greater degree by blood flow, and in most cases, distribution is better in brain than in tumor. Figure 1 depicts how standard molecules and anticancer drugs have historically behaved in rodent brain and IC rodent tumors.¹⁹ This composite plot shows median transfer constant, $K_i(\approx K^{trans})$ with error bars for these compounds in normal brain and IC tumors. K_i values were calculated from capillary permeability, P, measurements. Figure 1 nicely informs the relationship between brain and tumor K_i, and the error bars show wide variation in K_i resulting from tumor heterogeneity. These relationships appear as would be expected by thermodynamic concepts.^6,8,19,20

Fig. 1.

This plot of capillary transfer constant, K_i, in rat brain and intracerebral 9L tumor is for 16 radiolabeled and purified compounds studied between 1975 and 1981^6,8,20 (originally published in Kurt Hellmann, Fundamentals of Cancer Chemotherapy in 1987; reprinted by permission from McGraw-Hill Education).¹⁹ The dashed line has a slope of unity and is not a fit for the data. The radiolabeled compounds on the graph are: 1, HOH; 2, NaCl; 3, urea; 4, glycerol; 5, creatinine; 6, 5-fluorouracil; 7, dianhydrogalactitol; 8, galactitol; 9, misonidazole; 10, procarbazine; 11, DFMO (eflornithine); 12, dibromodulcitol; 13, sucrose; 14, epipodophyllotoxin; 15, bleomycin; and 16, inulin. The outside of the blue box defines the K^trans (≃K_i) for gadolinium contrast in brain.²⁶

Tumor Cell Permeability

In addition to capillary permeability increasing with malignancy, tumor cells diverge from the differentiated phenotype and increase in size, change surface area to volume relationships, lose pathway and membrane integrity, and become more permeable (leaky) to ions and, therefore, water movement. This intuitive conclusion, borne of viewing tumor cells under the microscope and in culture, has not been well studied in its sequential physiological entirety, although studies to date support the fact that glioma cells may not behave normally to ion fluxes.^21–23 One of the consequences of large and leaky tumor cells in the brain is to exaggerate the size of the extracellular space (ECS) by extracellular radiolabeled tracers (eg, inulin, creatine, urea) that distribute in many leaky tumor cells compared with normal brain parenchyma, producing an ECS that is a combination of extracellular and intracellular water.^6,9,10,24

Neuroimaging Insights

Clinicians have used many neuroimaging techniques, such as CT and MRI, to follow the growth of high-grade gliomas in the brain. Much of the value of these techniques is based on the observation that some i.v. contrast agents preferentially permeated tumor vessels and leaked into the surrounding tumor extracellular fluid (ECF). In 1980, we modeled the consequence of a breakdown of the blood–tumor “barrier” compared with the highly restricted BBB, which demonstrated visualization of contrast after approximately 0.004% of brain capillary surface area was damaged or open to contrast leaking from capillary blood into tumor.⁹ Surprisingly, this seemingly minuscule leakage of contrast is sufficient for clinicians to differentiate tumor compared with brain with radionuclide, CT, and MRI scans. Clinically, many low- and mid-grade gliomas are reported as showing “no contrast enhancement.” However useful that standard is for clinicians, adherence to this position is likely unhelpful to those seeking to create new drug treatments for glioma patients, and may also be erroneous.

Careful quantitative MRI studies in patients with varying degrees of glioma malignancy show that capillary permeability, greater than defined by the intact BBB, intensifies with increasing glioma tumor malignancy grade from astrocytoma to GBM.^25–29 In these MRI studies, the volume transfer coefficient, K^trans, of gadolinium (Gd) contrast was used to approximate capillary permeability. K^trans is directly related to capillary permeability, P, when adjusted for blood flow and capillary surface area using the following equation:

$$K^{trans}=F\text{1}-e^{-}{}^{(P·S/F)}),$$ (1)

where P = permeability coefficient, cm²/sec; S = capillary surface area, cm²; and F = blood flow, mL/g/min. Therefore, K^trans, uncorrected for capillary surface area and tissue blood volume, is reported commonly in units of min⁻¹, whereas brain capillary permeability, which considers capillary surface area and blood volume, is reported in units of cm/sec. From equation 1, it can be deduced that for compounds not limited by blood flow, K^trans ≈ P·S.

Examining the relationship of K^trans to Gd-contrast in different glioma tumor grades and evaluating the significance of these values to drug penetration into different grades of glioma is a useful point of discussion. Investigations using K^trans computed from Gd-contrast MRI in different grades of gliomas are collated in Table 1 together with estimates of tumor ECF volume, Ve.^26–29 Looking at Table 1, it is apparent that some values, especially for Ve, are suspect, since Ve should, at the very least, exceed 0.19, a normal value for brain ECF, which has been determined to be about 0.17–0.20 in mammalian brain.³⁰ The most reliable values for tumor Ve from Table 1 appear to be those of Zhang and colleagues.²⁶ These values of 0.27 to 0.40 were b experimentally determined in IC rodent tumor models and depend on the extent of necrosis and integrity of tumor cells.^9,24 From that perspective, Ve values of 0.03, 0.07, and 0.12 in Table 1 are likely underestimates and might indicate experimental error.^27–29

Table 1.

Collation of Gd-contrast transfer constant, K^trans, and tumor extracellular space, Ve, for different grades of glioma

K trans / min−1					V e				Reference
Brain	Grade I	Grade II	Grade III	Grade IV	Grade I	Grade II	Grade III	Grade IV
0.004	0.066	0.093	0.190	0.214	0.27	0.43	0.63	0.72	26
		0.032	0.102			0.07	0.35		27
		0.019	0.108			0.03	0.32		28*
		0.026	0.096	0.135		0.12	0.48	0.52	29
0.004	.066	0.042	0.124	0.174	0.27	0.43	0.44	0.62	Average

*Study was limited to oligodendroglioma tumors.

From the perspective of drug penetrance in infiltrative gliomas (WHO grades II–IV) in the brain, we can conclude that most glial tumors will manifest less restriction compared with the BBB of normal brain white matter. For normal brain, Zhang and colleagues used Gd-contrast to define the K^trans of normal BBB as 4 × 10⁻³ min⁻¹,²⁶ similar to 8 × 10⁻⁴ min⁻¹, the value for sucrose that Levin and colleagues measured in capillary permeability coefficient, P, studies.⁶ Thus, on average, grade II gliomas will have a 10-fold higher permeability and K^trans over brain; anaplastic gliomas about a 30-fold increase; and glioblastoma, more than a 40-fold increase in K^trans and permeability.

Insights from Regional Drug Modeling

It is useful in drug development theorizing to consider the consequences of these K^trans values for infiltrative gliomas with respect to drug penetration and attainment of drug levels in low- and mid-grade glioma tumor cells. One obvious conclusion is that access to anticancer agents will have fewer constraints on molecular size and charge in any of these tumors than would be predicted by standard models of brain penetrance. Drug molecules will, therefore, distribute faster and to a greater extent, because of capillary leakage. Looking at brain immediately adjacent to the tumor (BAT) as well as tumor cells farther away from the tumor that might be hiding behind an intact BBB, it is likely that drugs will not have much trouble reaching the BAT. Drug penetration into the BAT will, however, be influenced by how the tumor grows and whether it is expansile and compresses adjacent brain or is infiltrative and does not produce a lot of compression and edema. In the former case, the growth of an expansile IC 9L tumor resulted in lower permeability to ¹⁴C-urea and ²²Na in the BAT, which was in some instances even lower than more normal brain.⁴ In a comparable study, Groothuis and colleagues studied α¹⁴C-aminoisobutyric acid (AIB) in the IC RG-2 rat glioma model and found that the AIB transfer constant fell markedly in the BAT.³¹ Agarwal and colleagues studied erlotinib, a small lipophilic drug (393 MW, log P 2.7), and found, in IC U87 tumors in rats, that the tissue/plasma ratio in BAT was considerably lower in BAT than in tumor, but similar to normal brain.³² Thus, these studies all support the concept that high permeability in IC tumors drops significantly at BAT and might fall lower than in normal brain or not depending on the size of the IC tumor, the pressure it exerts on surrounding brain, and the physical characteristics of the substance (drug) studied, all factors to be carefully considered during the drug development process.

A similar relationship can be seen using Gd-contrast MRI and sequential voxel measurements K^trans in patients. Fig. 2 shows the fall-off, from tumor into normal brain white matter, of K^trans values in a patient with a high-grade glioma. It can be appreciated from these measurements that K^trans can drop 100-fold from about 0.10 min⁻¹ to about 0.001 min⁻¹ over several millimeters from the enhancing tumor. The obvious and practical consequence of these studies is that for many drugs, there will be a significant reduction in drug penetration and drug level in the BAT compared with in the tumor itself.

Fig. 2.

MRI of a high-grade glioma after a rapid injection of Gd-contrast showing acquired K^trans values obtained in sequential voxels in the enhancing tumor and then shows a decrease in K^trans values as the Gd-contrast permeates and/or diffuses into the adjacent “normal” white matter. (A) Post-contrast T1-weighted image. (B) Quantitative map of K^trans. (C–D) Line plots illustrating linear and log10 measurements of K^trans as a function of distance through an area of solid contrast enhancing tumor into adjacent “normal” white matter (pink line in (A)). (E–F) Additional line plot (green line in (A)) illustrating linear and log10 measures of K^trans as a function of distance including regions of central (macroscopic) necrosis, solid enhancing tumor, and adjacent “normal” white matter.

Additionally, it appears that a drug with less than optimal BBB penetration might achieve adequate penetration into tumor and surrounding brain under particular conditions. This counterintuitive conclusion was based on studies with DFMO (eflornithine, α-difluoromethylornithine).¹⁴C-DFMO had a brain capillary permeability coefficient, P, of 3.9 × 10⁻⁷ cm/s and a K_i of 2.6 × 10⁻³ min⁻¹,³³ constants implying that DFMO could have significant limitation in capillary to brain penetrance. Because DFMO showed activity in IC rodent tumors³⁴ and was not appreciably biotransformed, experiments were designed to better understand regional tumor pharmacokinetics of DFMO in a treatment environment. Rats with an IC 9L tumor were tethered with a soft harness and infused with i.v. ¹⁴C-DFMO and unlabeled DFMO to maintain a constant blood level of DFMO. After 1 to 4 days, animals were euthanized to measure tissue/plasma ratios in brain adjacent to and distant from IC 9L tumors. This was accomplished using a specially designed blade set to cut serial 0.75 mm sections from frozen brain slabs. The results showed that between days 1 and 4, the ¹⁴C-DFMO BAT/plasma and brain/plasma levels could reach 0.5 to 1.5 at 4.5 to 5 mm from the edge of the tumor (Fig. 3).

Fig. 3.

¹⁴C-DFMO tissue/plasma levels from IC 9L rat tumor to brain adjacent to tumor to normal brain at intervals of 0.75 mm. Intravenous infusion of ¹⁴C-DFMO and label-free DFMO maintained a near constant blood level over 1, 2, 3, and 4 days prior to euthanizing rats and measuring tissue/plasma levels.³³ Reprinted by permission from Springer Nature.

What allowed the attainment of a brain/plasma ratio of about 1 for ¹⁴C-DFMO and to what extent does it reflect free drug? The brain/plasma ratio of free drug levels, devoid of protein binding in plasma and tissue (eg, brain), is used as one measure of drug brain penetrance, so a ratio of ~1 would infer either high drug penetration^35,36 or marked tissue binding. Since DFMO is not appreciably biotransformed; has low plasma clearance (~2.1 mL/min/kg); has a volume of distribution, V_d, ~0.5; and can be given chronically for days to weeks with acceptable systemic toxicity,^37–39 it is expected that the brain/plasma and BAT/plasma of ~1 reflect distribution in BAT and brain. This construct raised questions because the physicochemical features of DFMO are not common to other anticancer drugs and DFMO may represent a unique case: a small molecule (~182 Da) that is highly polar in physiological solutions, does not appreciably bind to plasma proteins, and irreversibly and specifically binds to its target. These characteristics might explain why DFMO appears to diffuse across more leaky (higher permeability) tumor cerebral endothelial cell junctions and then diffuses in brain ECF.

In addition to diffusion, it is conceivable that some L-DFMO might be transported into the brain like L-ornithine using a cationic amino acid transporter 1 (CAT1)⁴⁰; however, we think it unlikely to contribute significantly to the movement of DFMO across the BBB or into infiltrating tumor for the following reasons: (i) the CAT1 L-ornithine transporter is a shuttle transporter and only L-DFMO would be expected to be transported by this carrier, as DFMO is a mixture of D- and L-DFMO enantiomers; (ii) the previously determined capillary permeability coefficient for DFMO was consistent with passive movement of a molecule of the size and polarity of DFMO^8,33; and (iii) L-ornithine transport follows Michaelis–Menten kinetics in CAT1 with half-saturation of 50–100 ρM⁴⁰ orders of magnitude lower than typical plasma levels of DFMO. These observations, in aggregate, suggest that the impact of CAT1 on BBB and rat 9L tumor passage of D/L-DFMO would be negligible compared with diffusion.

It can be hypothesized from DFMO experiments that a drug with limited brain penetrance can distribute well in IC tumor, BAT, and even distant brain if the drug has certain physical and pharmacological traits. These attributes are (i) very high tumor cell binding specificity, (ii) little in the way of nontargeted binding, (iii) low plasma clearance, and (iv) ability to maintain a therapeutic plasma level over days to weeks. Unfortunately, not many anticancer drugs have the requisite characteristics and toxicity profile to take advantage of this approach. Theoretically, and under appropriate conditions, large-molecule prodrugs might also achieve adequate distribution within an infiltrating CNS tumor without being brain penetrant. As counterintuitive as the concept might be, antibody drug conjugates (ADCs) might achieve sufficient drug dosing to the leading edge of an infiltrating glioma. Like the DFMO example, select attributes listed earlier along with little binding of nontargeted protein (albumin or cells) and a drug conjugate that does not leave the ADC until it is within tumor cells would allow penetration of drug into infiltrating tumor and its leading edge. The ability of the ADC to deposit a cytotoxic drug preferentially in tumor cells would further enhance specificity over mere cell surface target binding. To expand this consideration, it is instructive to consider ¹¹¹In-ABT-806 and its clinical ADC, ABT-414, which is in clinical trials for GBM.

Studies found that ¹¹¹In-ABT-806, a chimeric monoclonal antibody to epidermal growth factor receptor (EGFR) epitopes found in amplified EGFR, successfully penetrated IC U87 rodent tumors,⁴¹ and ¹¹¹In-ch806 was observed to penetrate a human anaplastic astrocytoma tumor with an apparent increase in lesion size from day 3 to day 7 after i.v. dosing.⁴² The same group did pharmacokinetic studies and found mean t_1/2α = 10 h, t_1/2β = 5.9 days, V1 of 3400 mL, and CL = 30 mL/h, for ¹¹¹In-ch806.⁴² In other human pharmacokinetic studies, ABT-414 had terminal plasma t_1/2 of approximately 9 days.⁴³ Furthermore, ABT-414 was found to have 7-fold higher affinity binding to amplified EGFR epitopes C271A and C283A (0.067 nmol/L) and EGFR variant III (0.059 nmol/L) compared with wild-type EGFR (0.461 nmol/L).⁴⁴ Thus, ABT-414 with low clearance, high binding to unique tumor EGFR epitopes, sustained plasma levels for days, and a stable cytotoxic conjugate might also be an example of a drug able to achieve effective levels in infiltrative CNS tumors yet lacking brain-penetrant features of small molecules.⁴⁵ Further study, using radiolabeled drug and quantitative imaging, will be needed to validate this ADC as an exception to the brain-penetrant imperative for drugs to treat infiltrative CNS tumors.

Earlier drug modeling in brain and experimental IC tumors provides additional insight into the impact of variation in tumor permeability, allowing the steady-state exposure integral for drugs of varying permeability.⁹Figure 4 recapitulates prior models comparing hypothetical drug exposure integral (AUC_D) in IC tumors to 4 different P differing by a factor of 10. It is generally accepted that P in the 10^–7 cm/sec range reflects a high restriction to brain entry and P in the 10^–6 range, moderate restriction, whereas P of 10^–5 and 10^–4 reflect unrestricted brain penetrance. In Fig. 4A, if we use an average IC tumor blood flow of 0.3 mL/g/min, cell/ECF (f ratio) at 1, and set the ECF drug t_1/2 at 15 or 180 minutes, we see that with a t_1/2 of 15 minutes and P = 3 × 10⁻⁷ cm/sec (K^trans = 0.002 min⁻¹), AUC_D = 0.07, but that AUC_D increases dramatically to 0.43, 0.82, and 0.86 when P increases to 3 × 10⁻⁶, 3 × 10⁻⁵, and 3 × 10⁻⁴ cm/sec, respectively. If t_1/2 = 180 minutes, then the relationship still holds with AUC_D at 0.09 when P is 3 × 10⁻⁷ and increasing to 0.50 when P is 3 × 10⁻⁶ and 0.85 when P increases to 3 × 10⁻⁵ cm/sec. The effect of ECF t_1/2 increasing 12-fold to 180 minutes has a negligible effect on AUC_D, which increases from only 0.82 to 0.85.

Fig. 4.

This graph shows, for a well-mixed model, the total exposure dose as a function of tumor blood flow, capillary permeability coefficient (P), intracellular/extracellular drug (f ratio), and extracellular drug half-life (ECF t_1/2). The values of extracellular fluid, ECF, diffusion coefficient, D_e = 3 × 10⁻⁶ cm²/sec; extracellular fluid volume, Ve = 0.27; capillary surface/blood flow, SF = 750 min/cm; and average capillary radius, r = 9.5 × 10⁻⁴ cm, are typical for a variety of brain tumor models.⁹ (A) We set f = 1, P = 3 × 10⁻⁷, 3 × 10⁻⁶, 3 × 10⁻⁵, and 3 × 10⁻⁴ cm/sec, and intracellular drug t_1/2 = 5 min, and ECF drug t_1/2 = 15 min (solid lines) or 180 min (dashed lines). Average brain blood flow, F, is 0.5 mL/g/min and IC tumor F about 0.3 mL/g/min.⁹ (B) We set f = 0.5 (dotted lines), 1 (solid lines), and 4 (dashed lines) and ECF drug t_1/2 = 180 min.

Figure 4B and Table 2 show the effect of increased drug partitioning (cell/ECF), f, by holding drug ECF t_1/2 constant at 180 minutes and varying f from 0.5 to 4.0. Here, AUC_D varies insignificantly, starting at a low value of 0.09–0.10 at P = 3 × 10⁻⁷ escalating to 0.50–0.48 at P = 3 × 10⁻⁶ and increasing further at 3 × 10⁻⁵ (0.85 to 0.84). This change dramatically shows the impact of P and the modest effect of an 8-fold increase in f. As in the prior example, shown in Fig. 4A, there is a marked increase (72%) between AUC_D at 3 × 10⁻⁶ and 3 × 10⁻⁵ and, thus, an even greater change compared with P = 3 × 10⁻⁷ cm/sec.

Table 2.

Values from Fig. 4B for total drug exposure (AUC_D) are listed below for blood flow, F, 0.3 and 0.5 mL/g/min, ECF t_1/2 = 180 min, and for f from 0.5 to 4

Blood Flow, F, mL/g/min	f Ratio	Permeability Coefficient, Pc, cm/sec
		3 × 10–7	3 × 10 –6	3 × 10 –5	3 × 10 –4
0.3	0.5	0.10	0.50	0.85	0.89
0.3	1	0.09	0.50	0.85	0.89
0.3	4	0.09	0.48	0.84	0.88
0.5	0.5	0.15	0.62	0.91	0.93
0.5	1	0.15	0.62	0.91	0.93
0.5	4	0.14	0.60	0.90	0.92

From these analyses, it is expected that cumulative anticancer drug penetration (AUC_D) even for a low-grade glioma with a BBB restricted drug would be about 5-fold higher than expected based on the analyses of P for normal brain. For example, if we use blood flow of 0.5 mL/g/min, we see that AUC_D increases with increasing P of 3 × 10⁻⁷ cm/sec from a low AUC_D of 0.15 to 0.62 at 3 × 10⁻⁶ and to a high of about 0.91 at 3 × 10⁻⁵, supporting the belief that brain-penetrant reversible inhibitor drugs need to be in the P range of 10⁻⁵ to 10⁻⁴ cm/sec to be optimal to cross the intact BBB, although, for IC tumors, it would appear that all anticancer drugs would behave as though their effective P would be equal to or greater than 3 × 10⁻⁶ cm/sec (K^trans = 0.02 min⁻¹).

Conclusions

The primary conclusion that can be drawn from the observations presented here is that while the physical parameters that define brain-penetrant drugs are generally applicable for infiltrative primary CNS cancers, there is at least one setting in which drugs of limited brain permeation might find purchase as effective chemotherapies. That setting is narrowly defined by the following parameters: low plasma clearance, very low plasma protein binding, ability to sustain plasma drug levels for days to weeks with little systemic toxicity, and target binding that is irreversible. This may not be the only situation that will successfully deviate from the historic brain penetrance mantra, as others utilizing carriers, endocytosis mechanisms, and novel nanotechnologies may also succeed.

Lastly, one penetrance topic has not been addressed: drug penetration from the ECF into its target at the tumor cell membrane, cytosol, and/or nucleus. Typically, permeability increases at both the capillary level and the tumor cell membrane with increasing malignancy so one presumes that most drugs will be able to diffuse across tumor cell membranes into the cytoplasm and eventually across the nuclear plasmalemmal as well. These studies are usually not easily conducted, but from those that have been done in the past, we know that there is a great deal of heterogeneity in drug binding to cellular targets and target effects.

Funding

This research was funded by the 2002 SPORE grants 1P50CA211015-01A1 (B.M.E.), NIH/NCI grants 1R21CA223757-01 (B.M.E.), and RSG-15-003-01-CCE American Cancer Society Research Scholar Grant (B.M.E.).