Human Neural Stem Cell Biodistribution and Predicted Tumor Coverage by a Diffusible Therapeutic in a Mouse Glioma Model

Abstract Engineered neural stem cells (NSCs) intrinsically migrating to brain tumors offer a promising mechanism for local therapeutic delivery. However, difficulties in quantitative assessments of NSC migration and in estimates of tumor coverage by diffusible therapeutics have impeded development and refinement of NSC‐based therapies. To address this need, we developed techniques by which conventional serial‐sectioned formalin‐fixed paraffin‐embedded (FFPE) brains can be analyzed in their entirety across multiple test animals. We considered a conventional human glioblastoma model: U251 glioma cells orthotopically engrafted in immunodeficient mice receiving intracerebral (i.c.) or intravenous (i.v.) administrations of NSCs expressing a diffusible enzyme to locally catalyze chemotherapeutic formation. NSC migration to tumor sites was dose‐dependent, reaching 50%–60% of total administered NSCs for the i.c route and 1.5% for the i.v. route. Curiously, the most efficient NSC homing was seen with smaller NSC doses, implying existence of rate‐limiting process active during administration and/or migration. Predicted tumor exposure to a diffusing therapeutic (assuming a 50 µm radius of action) could reach greater than 50% of the entire tumor volume for i.c. and 25% for i.v. administration. Within individual sections, coverage of tumor area could be as high as 100% for i.c. and 70% for i.v. routes. Greater estimated therapeutic coverage was observed for larger tumors and for larger tumor regions in individual sections. Overall, we have demonstrated a framework within which investigators may rationally evaluate NSC migration to, and integration into, brain tumors, and therefore enhance understanding of mechanisms that both promote and limit this therapeutic modality. Stem Cells Translational Medicine 2017;6:1522–1532

While many groups have observed that therapeutically-modified NSCs migrate selectively to glioma foci in rodent models following Departments of a Developmental & Stem Cell Biology, f Hematology/HCT, g Immuno-Oncology, e intracerebral (i.c.) or intravenous (i.v.) administration [23,32], quantitative assessments of NSC migration efficiency and local distribution at tumor sites, as well as tumor coverage by therapeutics delivered by these NSCs, have been difficult to perform. This limitation has restricted evaluation of alternative delivery strategies and therapeutic payloads. To further refinement of future NSC-based therapies and those under development [33][34][35][36][37][38][39][40][41][42], we established techniques by which immunostained serially-sectioned formalin-fixed paraffin-embedded (FFPE) brain tissue can be used for quantitative analyses across multiple test animals.
Here, as part of the preclinical IND-enabling studies for an NSC-based therapy for glioblastoma, we evaluated distributions of NSCs in and around tumors in an orthotopic xenograft model (U251 human glioma cells in immunodeficient mice) offering engraftment and growth characteristics sufficiently consistent to generate the reproducible measurements required for a quantitative study, and estimated the percentages of tumor volume that would exposed to diffusible therapeutic agents produced by the NSCs. Depending on the specific strategy, this could be a diffusible prodrug-activating enzyme [1,43,44] an oncolytic virus [29,45], or an antibody [27]. These measurements are important for assessing potential clinical advantages and disadvantages of different NSC administration routes (intracerebral vs. intravenous) and regimens (single vs. multiple doses), and identifying barriers to optimal implementation of these therapies. These approaches to quantitative evaluation thus provide a paradigm aiding evaluation of alternative therapeutic designs and facilitating optimization of this and other cell-based therapies [46].

Glioma-Bearing Mice for In Vivo Studies
All animal studies were performed under approved City of Hope (IACUC #04011) and CHLA (IACUC #285) protocols. Host animals for this study were carboxylesterase (CE)-deficient SCID mice (Es1 e / SCID) [47], having plasma CE concentrations consistent with human levels of circulating CE [44] and better reproducing human conditions for the enzyme/prodrug therapy being investigated (as wild type mice have high plasma CE levels). Es1 e /SCID mice were bred at both St. Jude and City of Hope Animal Breeding Facilities.
These NSCs were further engineered for high-transient expression of a modified human CE by transduction with replicationdeficient adenoviral construct hCE1m6 [44,52] (here referred to as hCE1m6-NSCs). Secreted hCE1m6 CE catalyzes local conversion of the systemically administered prodrug CPT-11 (irinotecan) to the potent topoisomerase-1 inhibitor and chemotherapeutic agent SN-38.

Histological Procedures
Mice were euthanized on day 3 after NSC administration (7 days after U251 cell engraftment). Brains (nonperfused) were harvested and fixed in 50 ml of 4% paraformaldehyde (PFA) in PBS for 72-96 hours while cooled in an ice and water slush on a rocking platform. Brains were then rinsed with reagent-grade water, placed in 70% ethanol, and stored at room temperature. Conventional paraffin blocks were prepared and 10 mm-thick horizontal serial sections through the tumor were cut in the City of Hope Pathology Core. Typically each tumor extended 2.5-3.5 mm vertically, and after trimming approximately 500 mm from the dorsal brain surface, 120-180 slides were collected, each carrying two 10 mm-thick sections, through the entire tumor. Each slide was bar coded for identification by experiment, brain and section number. Slides were considered in groups of ten. The first slide in each group was processed for hematoxylin and eosin (H1E) staining, the adjacent (2nd) slide was immunochemically stained for eGFP to identify U251 tumor cells, and the subsequent (3rd) slide was processed for Prussian blue chemistry to identify Ferahemelabeled hCE1m6-NSCs.

Image Acquisition
Each slide was scanned at high resolution (approximately 15,000 3 15,000 square pixels at 1 mm per pixel) and 24-bit RGB images of entire brain sections were acquired using an ACIS II scanning Barish The scheme for image processing is shown in Figure 2. Images of a pair of chromagen-stained brain slices as generated by the ACIS II scanning microscope are shown in Figure 2A1 and 2A2, with hCE1m6-F3 NSCs visualized by Prussian blue (upper panel) and U251 tumor cells immunostained with anti-eGFP/DAB (lower panel). These images were segmented into channels representing NSCs or tumor cells by color deconvolution as illustrated in Figure  2B with pixels corresponding to hCE1m6-F3 NSCs colored red and U251 tumor cells colored dark green. This was done using the algorithm published by Ruifrok and Johnston [56] and incorporated into the "Colour Deconvolution" plugin (http://fiji.sc/Colour_Deconvolution; http://www.mecourse.com/landinig/software/cdeconv/cdeconv.html) for Fiji [57] (http://fiji.sc/Fiji). Numbers of hCE1m6-F3 NSCs were determined for each 10 mm-thick section by manual counting of individually resolved cells, or by dividing the total number of Prussian blue-marked pixels by the average number of pixels per NSC. Total numbers of hCE1m6-F3 NSCs in a tumor were determined by extrapolating over the 200 mm separating each group of ten slides, and then summing these values. Figure 2C gives an indication of the spatial domain of each hCE1m6-F3 NSC using circles of radius with circles of radius 25 or 50 mm drawn from the centroid of each cell. Figure 2D illustrates tumor coverage of a secreted therapeutic such as CE estimated by drawing a radius of the desired length from the edge of each Prussian blue-marked area, and determining the numbers of tumor cell (eGFP-positive) pixels within this area using Cell Meter software developed for MatLab (MathWorks; Natick, MA; www.mathworks.com) (http://www.pedimg.org/cellmeter/software.html; http://www.youtube.com/watch?v5MDbEHrdTolk). In Figure 2D, the entire tumor is shown in green, with each pixel within the estimated radius of action shown in lighter green.
Tumor volume was quantified according to the Cavalieri principle by extrapolating the numbers of DAB-stained pixels over the 200 mm separating each set of 10 slides. For three-dimensional reconstruction, images of each section were manually aligned with ventricles and olfactory bulbs as landmarks using Reconstruct [58] (http://synapses.clm.utexas.edu/tools/reconstruct/reconstruct.stm).
X, Y positions of NSCs within tumors were obtained using CellProfiler v2.2 (Broad Institute; Cambridge, MA; http://cellprofiler.org/) [59], and pair-wise distances between any two NSCs (i, j) determined from the Euclidian relation d i; Estimations of the diffusion of carboxylesterase (CE) in brain were made as described by Wolak and Thorne [60], assuming a molecular weight of 60,000 [61,62] and a diffusion coefficient (D) in free solution of 4.3 3 10 27 cm 2 per second [62]. D*, the diffusion coefficient adjusted for the tortuosity of brain extracellular space (k), was calculated using the relation D* 5 D/k 2 , where k was 2.25, a value characteristic of high molecular weight molecules in brain tissue [63]. D* for CE in brain was thus 8.5 3 10 28 cm 2 per second. Relative CE concentration (C/C 0 ) at distance x from a nondepleting source assuming essentially no efflux or degradation (k e 5 ln(2)/t 1/2 ) was then calculated from the relation C/ C 0 5 exp{-x(k e /D*) 1/2 } [64].
Data were compiled and analyzed in Excel (Microsoft; Bellevue, WA; office.microsoft.com/excel) and are presented using Sig-maPlot 13 (Systat; San Jose, CA; systatsoftware.com). Statistical tests were made using Prism 7 (GraphPad; La Jolla, CA; www. graphpad.com) or Excel. For correlations between numbers of NSCs administered and other parameters, r 2 and p are reported for linear regressions. For correlations between other not predetermined parameters, r (the Spearman correlation coefficient) and p are reported. Other comparisons were evaluated by t test as noted.

RESULTS
Presented here are the results of an analysis of a cohort of orthotopic U251 tumor xenografts, 14 across two independent experiments with i.c. NSC administration near the site of the engrafted tumor, and 22 across two independent experiments with i.v. NSC administration. Tumors were studied at 1 week following implantation, with hCE1m6-NSCs delivered on day 4 and animals sacrificed 3 days later.
Tumor Architecture and NSC Infiltration Figure 1A presents a representative H1E stained horizontal section of U251 tumor. An area of dense tumor nuclei is evident in hematoxylin staining, centered on the external capsule and extending into the striatum and cortex. The lower right panel shows U251 cells in the adjacent slide identified by eGFP antibody and DAB immunochemistry, and the upper right panel shows feraheme-loaded NSCs in the preceding slide stained with Prussian blue. In this brain, NSCs were found around the edge of the main tumor mass, and along an invading nodule (upper panel, black arrow).
NSCs were found at the external tumor margins and penetrating the tumor interior, as illustrated by the two brain sections in Figure 1B in which tumor cells engrafted and formed masses within the striatum after i.c. or i.v. administration. Here (in greater detail in Fig. 2a2, 2b2), tumor cells are shown in dark green, light green, or yellow depending on their diffusion proximity to the hCE1m6-F3 NSCs highlighted in red. The distribution of NSCs both at the tumor margins and within the tumor masses did not appear to differ between i.c. and i.v. administration routes. For other examples, see [54].
Not all injected NSCs showed tumor-directed mobility. For the brain shown in Figure 1Ba1, hCE1m6-F3 NSCs were administered i.c. just caudal lateral to the tumor. The arrow indicates the migration of a portion of the NSCs from the injection site in the striatum to the tumor mass. When tumors were present, we did not observe migration of NSCs toward other brain structures.
Within a given brain, tumor morphology and the distributions of hCE1m6-NSCs could vary greatly with depth from the pial surface and thus position in the brain. The example presented in    in cortex, as in panel Ci. At greater depths the tumor engrafted at the border of the striatum and external capsule, and, as was commonly observed, disseminated laterally following the arc of the external capsule (panels Ciii-Cviii). Tumor cells also migrated into the striatum and cortex at all depths (panels Ci-viii). Viewing sections from the entire tumor, observations such as those in Figure  1C suggest the presence of a microenvironment in the white matter of the external capsule particularly advantageous for both tumor cells and NSCs.
Annotations to Figure 1C indicate the estimated percentage of total tumor area that would be exposed to a therapeutic secreted by the NSCs (assuming a 50 mm radius of action as described in greater detail below) at different vertical locations within the tumor. As seen here, the most extensive tumor coverage was seen for larger tumor cross sections more central to the vertical span of the engrafted tumor (see also Fig. 5B and [65]).

Quantitative Analyses
The procedures for quantifying hCE1m6-NSC numbers and estimating tumor areas exposed to secreted CE are described in the Methods and shown schematically in Figure 2. Images of a pair of brain slices show hCE1m6-F3 NSCs visualized by Prussian blue (upper panels of Fig. 2A1, 2A2) and U251 tumor cells immunostained with anti-eGFP/DAB (lower panels). Deconvolution yielded color channels representing NSCs in red pixels and tumor cells in green pixels (Fig. 2B). This process yields the numbers and locations of hCE1m6-F3 NSCs and quantitative measures of tumor area. The percentage of total tumor area in each brain slice that would be exposed to a diffusing therapeutic secreted by the NSCs can then be calculated (Fig. 2C) and visualized (Fig. 2D) in light green for diffusion radii of 25 and 50 mm around the border of each red NSC.
We focused on the localization of NSCs to tumor sites, and on variables influencing tumor coverage that could be achieved by a diffusing therapeutic. This therapeutic design incorporates NSC delivery of an enzyme expected to diffuse in and around the tumor site and catalyze formation of a chemotherapeutic from a systemically administered prodrug [50].
Approaching first the NSCs distributed at tumor sites, we observed, not surprisingly, that after i.c. injection, numbers of tumor-associated NSCs scaled with the number of cells administered (Fig. 3a1, p < .0001), while there was considerably more unpredictability in number of NSCs at tumor sites after i.v. injection (Fig. 3a3, p 5 .28). This might be expected given that the i.c. route deposits the NSCs close to the tumor site while NSCs administered i.v. must circulate prior to responding to as yet not welldefined vascular cues [37,66], and then extravasate into the tumor parenchyma.
The percentages of total injected hCE1m6-F3 NSCs ultimately recovered at tumor sites was always larger for i.c. administrations, as high as 50%-60% for small NSC doses and around 10% for larger doses (Fig. 3a2). For i.v. administered hCE1m6-F3 NSCs, around 1.5% of NSCs were recovered for smaller doses and 0.3% or less for larger doses (Fig. 3a4). In similar models, other investigators [65,67] have reported recovery of comparable percentages of cells after i.c. NSC injections.
Another parameter potentially related to NSC attraction is tumor volume, and we examined possible correlations between tumor volume and the numbers and percentages of hCE1m6-NSCs reaching tumor sites. For both i.c. and i.v. injections, we observed that hCE1m6-F3 NSCs could be found associated with very small tumor masses (Fig. 3b1, 3b3). Examined more quantitatively, the numbers of NSCs observed at tumor sites were not significantly related to tumor size after i.c. injection (Fig. 3b1,  p 5 .37), but for i.v. injection tumor size did appear to influence NSC attraction (Fig. 3b3, p 5 .05). For the percentages of total administered NSCs at tumor sites, tumor size did not appear to be a factor for i.c. injections (Fig. 3b2, p 5 .22), and not strongly for i.v. injections (Fig. 3b4, p 5 .12) (see also [65]).
Because the ultimate therapeutic efficacy of this NSC-based therapy will depend on the delivery of a diffusing prodrug-activating enzyme ("bystander effect" [68]), we next undertook an analysis of the percentage of total tumor volume that would be exposed to a soluble therapeutic in relation to these same three parameters: total number of NSCs administered, the numbers of NSCs subsequently localized to tumor sites, and total tumor volume.
We calculated percentage tumor coverage for radii of action of both 25 and 50 mm for an enzyme originating in the NSCs, distances representing 3-4 or 6-8 diameters of 8-10 mm cells. First, calculations of the diffusion radius of carboxylesterase (CE) incorporating its molecular weight, free diffusion coefficient in solution, and tortuosity of the brain and brain tumor extracellular space (see Methods) indicate that at steady state, 50% of initial concentration will be seen at 25 mm and 24% at 50 mm distance from a secreting source. That CE-driven therapeutic activity will be seen at these distances is suggested by results of in vitro experiments [69] showing both efficient CE-catalyzed conversion of CPT-11 to SN-38, and high sensitivity of glioma cells to SN-38 (IC 50 of 9 to 152 nM against patient-derived and established glioma cell lines). In addition, results from in vivo models [70] indicate that significant tumor regression can be observed when only 1 in 50 cells (2%) express an activating enzyme, in this case cytosine deaminase (CD) to locally form a chemotherapeutic (5-fluorouracil, 5-FU) from a systemically administered prodrug (5-fluorocytosine, 5-FC). For this system Lin et al. [65] calculated an estimated killing radius of 23 mm.
In our experiments, total tumor coverage, estimated assuming a 50 mm effective radius, appeared to depend on the numbers of hCE1m6-NSCs administered, and could reach greater than 50% for the highest dose delivered i.c. (2.0 3 10 5 ) (Fig. 4a1, p 5 .07) (see also [55,65]), and above 25% for i.v. delivery of up to 10 times as many cells (2.0 3 10 6 ) (Fig. 4a2, p 5 .12). This pattern of half as much coverage for 10 times the number of hCE1m6-F3 NSCs administered i.c. versus i.v. appeared consistently across all NSC doses examined. At the tumor site, percentage tumor coverage also appeared greater for higher numbers of hCE1m6-NSCs after i.c. delivery (Fig. 4b1, p 5 .14) and was quite significant for i.v. delivery (Fig. 4b2, p 5 1.07 3 10 24 ). There was also a trend toward greater percentage coverage of larger tumors (Fig. 4c1,  p 5 .03 for i.c.) (Fig. 4c2, p 5 .23 for i.v.). Percent tumor coverage was most highly correlated to tumor volume for i.c. delivery, perhaps reflecting the relative simplicity of the NSC path.
To explore this point further, we analyzed distributions of hCE1m6-F3 NSCs on a per section basis, taking measurements from one section of each slide at every depth in which both tumor cells and NSCs were present. For both i.c. and i.v. administrations, percent tumor coverage was highly correlated with tumor area within each section (Fig. 5b1, p 5 3.29 3 10 26 for i.c) (Fig. 5b2, p 5 9.90 3 10 28 for i.v.), and much less so for NSC dose (Fig. 5a1,  p 5 .07 for i.c.) (Fig. 5a2, p 5 .49 for i.v.).
We also investigated whether NSC "history" during migration from i.c. or i.v. administration sites exerted influence on behavior of NSCs after their arrival at tumor site. NSC density (numbers of NSCs present per unit of tumor volume) appeared to be negatively related to tumor size for both i.c. and i.v. administration (Fig. 6a1, 6a2). Unexpectedly, the percent tumor coverage did not appear tightly related to the density of hCE1m6-F3 NSCs for i.c. administration (Fig. 6b1 p 5.95 for i.c), but appeared more so for i.v. administration (Fig. 6b2, p 5 .001 for i.v). For the range of hCE1m6-F3 NSC densities spanning approximately 10 4 -10 5 NSCs/ mm 3 , there was no statistically significant difference in percent coverage for NSCs delivered i.c. and i.v. Of note, it may be possible to increase the NSC density at tumor sites by the i.v. route using repeat administrations.
The effective efficiency of tumor coverage by a diffusing therapeutic, as compared to the theoretical maximum obtainable assuming homogeneous NSC distribution throughout a tumor mass, will depend on the distances between NSCs (as overlap of radii of diffusion may lead to redundancy) and on NSC proximity to the tumor edge (as radii of diffusion may extend into the brain parenchyma). Both of these situations can be seen in Figures 1B  and 2D. We therefore assessed the potential contribution of NSC clustering on tumor coverage by using pair-wise distances between NSCs in individual sections i.c.-injected with NSCs to determine a clustering index (CI) for each section based on a clustering radius of 50 mm, where CI 5 1 would indicate that all NSCs were in a single cluster and CI 5 0 would indicate complete dispersion. CIs were quite variable between brains and between sections within individual brains, as illustrated in Figure 7A, and with considerable scatter were related to the numbers of NSCs within each tumor section (particularly for larger numbers of NSCs). The   proportion of the theoretical maximum tumor coverage achieved in each section also appeared to decrease with greater NSC clustering (higher CI) (Fig. 7C).

DISCUSSION AND CONCLUSION
In summary, in this orthotopic xenograft model of human glioma, we have demonstrated that quantitative analyses of NSC homing and biodistribution in relation to glioma masses can be performed using standard formalin-fixed paraffin-embedded (FFPE) brain sections and freely-available software tools running on personal computers of reasonable power.
We conclude that hCE1m6-F3 NSC therapies are "well behaved" in that administering higher doses tends to yield increased numbers of NSCs at tumor sites (Fig. 3A) and greater estimated tumor coverage by a diffusing therapeutic (Fig. 4A). Further, while at least 10-times as many NSCs must be administered i.v. as compared to achieve to i.c. to achieve a given NSC density, our findings suggest that NSC behaviors at tumors are equivalent for the two administration routes, and that similar NSC densities, once achieved, yield similar tumor coverage.
At the same time, the process of NSC homing appears to limited by one or more rate-limiting process active during administration and/or migration, in that the percentage of administered NSCs ultimately localized to tumor sites falls with injection of larger numbers of cells for both i.c. and i.v. routes (Fig. 3A). This finding was unexpected, as all things being equal, one might expect a constant percentage of NSCs to arrive at tumor sites independent of the numbers of cells injected. Therefore, the more efficient propagation of NSCs to tumor sites seen with lower numbers of NSCs administered by both i.c. and i.v. routes implies activity of additional process possibly related to absolute numbers of NSCs, to the concentration of NSCs in the injection vehicle, and/or to the densities of NSCs at the injection sites, that limit NSC survival. Alternatively, the ability of tumor masses to absorb NSCs may be limited, leading to saturation as larger numbers of NSCs are injected. These possibilities will be further investigated.
More efficient delivery of small number of NSCs has been noted by other investigators [65,67], and this remains a bottleneck in implementation of NSC-based therapies. Enhancements in NSC delivery (such as inclusion in biocompatible matrices [71,72]) might improve NSC survival and therapeutic efficiency, while breaking NSC administrations into multiple injections might reduce tumor saturation if present, and could have implications for design of NSC delivery schedules (e.g., single bolus vs. multiple doses).
We also find that larger tumors, and larger regions of individual tumors, tend to attract greater numbers of NSCs, and there is a trend toward higher NSC densities and greater estimated therapeutic coverage in larger tumors. This pattern could be linked to higher levels of inflammatory cytokines associated with greater numbers of tumor cells and/or tumor-associated monocytes and immune cells, or to the more extensive brain disruption or injury that would be expected with a larger tumor. A similar association of NSC accumulation with areas of higher tumor cell density was reported by Lin et al. [65] (but see [55]).
Finally, we observed that the degree to which NSCs cluster together also impacts the efficiency of tumor coverage by a secreted therapeutic as compared to the theoretical maximum, raising the possibility of manipulating the (presently unknown) factors that influence NSC clustering.
The analysis outlined here provides a framework by which investigators may rationally evaluate NSC migration to, and integration into, brain tumors, and thereby increase our understanding of this mode of therapy delivery. However, despite use of a standardized model, considerable scatter was observed in the data that may arise from multiple sources, including variation in tumor engraftment depending on position in the brain, and variable penetration of NSCs into tumor masses, as well as details of NSC administration. Identifying and these uncontrolled variables in therapy implementation and relating them to details of the administered product and to tumor cellular composition and physiology will enhance further development of stem cell-based therapies.