A living ex vivo platform for functional, personalized brain cancer diagnosis

Summary

Functional precision medicine platforms are emerging as promising strategies to improve pre-clinical drug testing and guide clinical decisions. We have developed an organotypic brain slice culture (OBSC)-based platform and multi-parametric algorithm that enable rapid engraftment, treatment, and analysis of uncultured patient brain tumor tissue and patient-derived cell lines. The platform has supported engraftment of every patient tumor tested to this point: high- and low-grade adult and pediatric tumor tissue rapidly establishes on OBSCs among endogenous astrocytes and microglia while maintaining the tumor’s original DNA profile. Our algorithm calculates dose-response relationships of both tumor kill and OBSC toxicity, generating summarized drug sensitivity scores on the basis of therapeutic window and allowing us to normalize response profiles across a panel of U.S. Food and Drug Administration (FDA)-approved and exploratory agents. Summarized patient tumor scores after OBSC treatment show positive associations to clinical outcomes, suggesting that the OBSC platform can provide rapid, accurate, functional testing to ultimately guide patient care.

Graphical abstract

Highlights

• •OBSCs allow rapid drug screening of cell lines and uncultured patient tumor tissue
• •Drug sensitivity score normalizes tumor kill to off-target toxicity for each drug
• •Algorithm enables relative comparisons among drugs and across tumors
• •Workflow allows association of results on OBSCs to patient outcomes

Mann et al. develop an organotypic brain slice culture (OBSC)-based platform and multi-parametric algorithm that enable rapid engraftment, treatment, and analysis of uncultured patient brain tumor tissue and patient-derived cell lines. After treatment, the algorithm calculates summarized drug sensitivity scores, allowing relative comparisons that show positive associations to clinical outcomes.

Introduction

The accurate classification of CNS tumors has made significant progress in the past two decades, moving from an exclusive reliance on histopathological features toward more integrated diagnoses based on molecular insights.¹ Although the improved classification of CNS tumors is already enhancing the scientific rigor of clinical trials, the impact of these integrated diagnoses on precision oncology medicine—the molecular-guided pairing of tumors with drugs—has yet to demonstrate widespread clinical benefit. Between 2006 and 2020, a time period during which an astounding number of new cancer-directed drugs were developed, eligibility for those drugs only increased from about 5% to 13%, and response to those drugs only increased from about 3% to 7%.² The reasons for this are myriad and include factors such as co-occurring oncogenic alterations,³ tumor heterogeneity,⁴ epistatic interactions,⁵ and adaptive cellular circuitry.^6,7,8 It is becoming increasingly clear that static histopathological and molecular measurements of tumors are still insufficient to most accurately and usefully classify CNS tumors and that to identify precision medicines for the majority of CNS tumor patients, functional diagnostic platforms are needed.

Patient-derived models of cancer (PDMCs), including cell lines, patient-derived organoids (PDOs), patient-derived explants (PDEs), and patient-derived xenografts (PDXs), provide functional models of a patient’s individual tumor that can be screened with multiple drugs. The potential for PDMCs to guide personalized care has been demonstrated in several studies that show their ability to successfully predict antitumor response. Indeed, drug screening assays using PDMCs have successfully predicted antitumor responses in humans, demonstrating their potential.^{9,10,11,12,13} Unfortunately, initiation time, cost, efficiency scales, and lack of similarity to the parent tumor limit applications of PDXs,¹⁴ while patient-derived cell lines often lose the genetic and phenotypic heterogeneity of the parent tumor via cell selection during clonal expansion.¹⁵ Some PDOs and PDEs can effectively capture the histological and mutational diversity of human cancers^{16,17,18,19,20} but are most successfully generated from aggressive, high-grade tumors such as IDH-wild-type glioblastoma. Furthermore, generating these models from heterogeneous tumors also limits reproducibility among intra-tumor replicates. A platform that supports and rapidly engrafts both low- and high-grade brain tumor tissue, maintains genetic heterogeneity and resemblance to the parent tumor, and allows functional testing of approved and experimental therapeutics is still desperately needed.

In this study, we present our organotypic brain slice culture (OBSC) platform as a living tissue substrate capable of engrafting and testing uncultured patient brain tumor tissue for functional precision diagnosis. Our previous work has demonstrated that tumor lines grown on OBSCs replicate the tumor growth, invasion, and cellular interactions observed in vivo,²¹ as well as how functional drug testing on OBSCs is supported by transcriptomic profiles.²² We now use multiple additional assays and a multipoint algorithm to compare drug-induced killing of established tumor lines and uncultured patient tumor tissue engrafted onto OBSCs. Our standardized, optimized protocol has allowed successful engraftment and treatment of all tested patient tumor tissue samples across a variety of tumor types, grades, and patient ages. For each of nearly 150 drug-tumor-OBSC interactions, we calculated therapeutic windows by comparing both tumor kill and normal brain tissue (OBSC) toxicity, enabling normalized and direct drug-to-drug and tumor-to-tumor comparisons. Overall, these data highlight and validate our platform as a mature, clinically relevant technology.

Results

Generation and characterization of living OBSCs

These studies use OBSCs generated from rat pups (RRID:MGI:5651135) as living tissue substrates to culture and treat tumor cell lines and uncultured patient brain tumor resection tissue. To generate this rapid, high-fidelity ex vivo testing platform, we first assessed OBSC quality and reproducibility, optimizing the process using quantitative fluorescence-based imaging methods. After piloting several assays, we adapted an established nuclear permeability assay using propidium iodide^{23,24,25,26,27,28,29,30} (PI; catalog #P4170; Sigma-Aldrich) to quantify cell death within OBSCs. This method provided a large dynamic range to detect nuanced differences between healthy and unhealthy OBSCs, with higher fluorescence values indicating more dead cells (Figure 1A). OBSC quality was influenced by rat pup age at time of generation (Figure 1B); therefore, OBSCs were generated from eight-day-old pups for every experiment described in this study. We also found that optimal OBSCs were generated using improved methods for brain dissection and OBSC culture conditions (Figures 1C and 1D), establishing our robust, standardized procedure for OBSC generation. Following this optimization period, we reproducibly generated more than 6,000 living OBSCs across nearly 40 separate batches, randomly sampling and testing 6 OBSCs per batch for QC analysis (Figure 1E). All OBSCs with physical damage after slicing, such as a nick or a tear, were discarded, but this amounted to fewer than 2% of OBSCs over all experiments.

Figure 1.

OBSC characterization, standardization, and quality control

(A) Left: bright-field and fluorescence images of standard healthy and dead (positive control; frozen in EtOH overnight) OBSCs, via PI stain. Right: average fluorescence from standard OBSCs (n = 287 biological replicates) and positive control OBSCs (n = 131 biological replicates) OBSCs.

(B) Effect of pup age on OBSC viability (n ≥ 6 per day), analyzed using one-way ANOVA (∗∗∗∗p < 0.0001).

(C) Viability comparison of healthy OBSCs vs. those damaged during dissection and/or slicing (n = 60 biological replicates for healthy, n = 36 biological replicates for damaged), analyzed using Welch’s t test (p = 0.0172).

(D) Viability of OBSCs on the basis of brain slice media (BSM) quality and promptness during slicing (n = 12 biological replicates per group; slicing delay after dissection: 1 h), analyzed using one-way ANOVA (∗∗p = 0.0052 and ∗∗∗∗p < 0.0001).

(E) Representative batch-to-batch viability from n ≥ 6 biological replicates randomly sampled OBSCs from each batch (n ∼ 150 OBSCs per batch).

(F) PI signal from OBSCs measured on different days after slicing (n = 6 biological replicates).

(G) 10X immunofluorescent maximum intensity projection images of healthy OBSCs showing activity level of astrocytes (GFAP), neurons (NeuN), and microglia (CD11b) immediately after slicing and on day 4 (D4). All data, except where otherwise noted, were collected 4 days after slicing.

All data are shown as mean ± SEM.

Just as brain tissue suffers acute injury after brain tumor resection, OBSCs suffer acute injury after slicing. Serial measurements of OBSC health showed an initially high PI signal that resolves over time, reaching baseline levels by day 7 (Figure 1F). In parallel, immunohistochemistry (IHC) analysis showed that astrocytes (GFAP; RRID:AB_305808) and macrophages/microglia (CD11b; RRID:AB_2650514) displayed reactive morphologies³¹ immediately post-slicing (Figure 1G). This activation attenuates in astrocytes by day 4 but persists in macrophages/microglia, suggesting that the OBSCs may still contain dead cells and debris that are phagocytosed by the myeloid cells.^32,33 The morphology of neurons (NeuN; RRID:AB_2650514) remained unchanged between days 0 and 4. Together, these results describe a high-quality, well-controlled, reproducible method for generating and tracking the health of OBSCs. As such, we next focused on engrafting tumor cells onto these living tissue substrates.

Tumor modeling on OBSCs

Before attempting to move into uncultured patient tissue samples, we selected a diverse panel of human brain cancer lines (adult, pediatric, primary, and metastatic; defined in Figure 2 legend) to further identify the myriad capabilities of our OBSC assay and show how OBSCs provide a rapid, representative, and applicable niche for broad tumor engraftment and analysis. We first compared multiple tumor characteristics across in vitro, in vivo, and OBSC culture. All tumor lines were transduced with lentiviral vectors encoding mCherry-Firefly Luciferase (LV-mCh-FLuc) optical reporters, then the lines were seeded in 96-well culture plates, orthotopically xenografted into the brains of nude mice, or seeded atop the living OBSCs. We leveraged the bilateral symmetry of each coronally sectioned OBSC to seed tumor cells within the anatomically homogeneous thalamic region of each hemisphere. Serial imaging and quantitative analysis was performed to track growth, morphology, and rate of invasion.

Figure 2.

Tumor growth and interaction on OBSCs

(A) In vitro cell growth in 96 h. Fifteen hundred cells were seeded into 96-well plates and measured on days 0 and 4 via BLI; n = 6 technical replicates.

(B) In vivo tumor growth in mice following intracranial implantation of 250,000 tumor cells; n = 5 biological replicates.

(C) Graphic of cell line tumor engraftment onto OBSCs.

(D) Representative fluorescence image of 4 MB231Br tumor foci seeded into the thalamic region of two OBSCs within one well.

(E) Representative bioluminescence image depicting tumor seeding of 24 tumor foci onto 12 OBSCs in one six-well plate.

(F) Cell line growth in 96 h on OBSCs. Bioluminescence images were taken on days 0 and 4. Day 0 fluorescence images were used to normalize initial tumor size. n = 4 biological replicates.

(G) Migratory behavior from initial tumor engraftment site on OBSCs; t = 96 h. Circles larger than the normalized tumor circumference indicate outward migratory behavior while smaller circles indicate inward retraction. n = 4 biological replicates.

(H) 10X immunofluorescent maximum intensity projection images of astrocytes and tumor interaction: astrocytes stained by GFAP (green) without or in the presence of GBM8 tumor cells (red) 96 h after engraftment.

(I) Growth of GBM8 on OBSCs seeded early (day of slicing) or seeded late (7 days post-slicing). n = 8 biological replicates. Analyzed using an unpaired t test. p = 0.0051

(J) Diameter change of GBM8 on OBSCs seeded early or late. n = 8 biological replicates. Analyzed using an unpaired t test. p = 0.0001

All data are shown as mean ± SEM.

See also Figure S1.

Four high-passage tumor lines commonly used within the field, LN229, U373, MB231Br, and GBM8, grew rapidly in vitro (Figure 2A), in vivo via orthotopic implantation (Figure 2B), and on OBSCs (Figures 2C–2F). These tumor lines all expanded several-fold in just four days in vitro and on OBSCs and more than 100-fold in vivo within 40 days. The low-passage adult GBM line MS21 and pediatric DIPG line PDIPG displayed a different set of results. These two lines had been initially cultured via clonal expansion in vitro after human tumor biopsy, and both were able to be cultured/expanded in vitro between passage 3 and 20 during this work. Both lines grew several fold within four days in vitro, but neither reproducibly established in vivo, even after 90 days. On OBSCs, MS21 and PDIPG tumors both nearly doubled in size over four days, displaying how OBSCs can support growth of low-passage tumor cells not supported in vivo.

Tumors seeded on OBSCs rapidly establish among the endogenous astrocytes and microglia. Consistent with our previous data showing rapid invasion of only invasive tumor types along the corpus callosum within OBSCs,²¹ we now observe tumor-specific trends in outward invasion or inward retraction of tumor cells on the thalamic region of OBSCs.¹⁹ We found the invasive cell line GBM8, metastatic cell line MB231Br, and diffuse pediatric DIPG all invade radially outward on OBSCs, while other established lines that grow more densely in vivo retract inward (Figure 2G; qualitative examples in Figure S1A). Interestingly, microglia within OBSCs also reciprocally respond to the presence of tumor, maintaining activated morphologies and associating with nearby tumor cells, indicating a two-way interaction between the tumor and this ex vivo microenvironment (Figure 2H). Adding tumor cells to OBSCs on day 0 after slicing (“early seeding,” our standard method), compared with day 7 (“late seeding”) when microglial activation and PI signaling have attenuated, also leads to faster tumor growth and invasion (Figures 2I and 2J), consistent with in vivo tumor response.³⁴ Together, this suggests that OBSCs provide a niche for tumor growth and interrogation that is more appropriate and accurate than in vitro culture systems and less harsh than in vivo models.

Tumor treatment on OBSCs

We next used the panel of OBSC tumor models to investigate treatment responses within the platform. To assess the molecular specificity of drug-induced tumor killing on OBSCs, we used a CRISPR-modified variant of the U373 tumor line harboring a single-gene knockout in RAD18. Deletion of this DNA repair mediator should increase sensitivity and killing for agents that use this pathway. The deletion did not affect the growth of untreated wild-type or knockout cells, now termed U373WT and U373KO, both in vitro and on the OBSCs (Figure S1B).

We subjected all seven OBSC-engrafted tumors to treatment with a panel of approved CNS tumor therapeutics and external radiation therapy (XRad), as well as the experimental agent TR107 (Madera), a second-generation derivative of ONC201.^35,36 We tested six concentrations of each agent, assessing treatment response by quantitative bioluminescence imaging (BLI) to specifically measure tumor kill. Dose-response curves and half maximal inhibitory concentration (IC₅₀) values were then calculated for each agent (Figures 3A and S2; Table S1). If treatment elicited <50% death at the highest dose, we report “not reached” (NR).

Figure 3.

Tumor killing on OBSCs

(A) Left: IC₅₀ values calculated on the basis of linear interpolation of dose-response data on OBSCs. Cells were seeded on day 0 and dosed with therapeutics on day 1; survival was measured via bioluminescence on day 4. Concentrations of small molecule drugs are given in micromolar; XRad dose is given in grays. NR indicates the IC₅₀ was not reached within the dose range. Right: graphical representation of IC₅₀ values of all drugs vs. all tumor lines.

(B) Killing of MB231Br, LN229, U373WT, and U373KO by TR107 on OBSCs and in vitro. n = 6 biological replicate tumor foci per dose, 6 doses per cell line.

(C) Combination therapy of radiation and subsequent temozolomide against GBM8 and MS21.n = 4 biological replicate tumor foci per dose, 6 doses per cell line.

(D) Combination therapy of etoposide and carboplatin against U373WT, U373KO, MB231Br, and PDIPG.n = 4 biological replicate tumor foci per dose, 6 doses per cell line.

All data are shown as mean ± SEM.

See also Figures S1 and S2 and Table S1.

By comparing IC₅₀ values (Figure 3A), we found that 7 of 7 lines showed robust response to the proven agents carboplatin (catalog #C2538-100MG; Sigma-Aldrich) cisplatin (catalog #232120-50MG; Sigma-Aldrich), and lomustine (catalog #S1840; Selleck Chemicals), with the PDX GBM cell line GBM8 showing markedly higher sensitivity. Several agents, including temozolomide (TMZ; catalog #T2577-100MG; Sigma-Aldrich), vincristine (catalog #V8388-1MG; Sigma-Aldrich), XRad, and gemcitabine (catalog #S1149; Sigma-Aldrich), failed to induce significant killing at even the highest tested doses in more than 5 of the 7 cell lines, resulting in designations of NR. Azacitidine (catalog #A2385-250MG; Sigma-Aldrich) and etoposide (catalog #E1383-100MG; Sigma-Aldrich) showed broad potency, inducing killing in 7 of 7 and 6 of 7 lines, respectively, at IC₅₀ values in the 200–400 μM range.

When comparing U373WT vs. U373KO, we found that IC₅₀ values were significantly decreased and killing significantly increased in U373KO following treatment with TMZ, azacitidine, gemcitabine, and etoposide. Interestingly, U373KO also showed increased resistance to radiation, possibly due to an upregulation of the nonhomologous end joining pathway.³⁷

The experimental agent TR107 was the most potent therapeutic in our panel, inducing complete killing in 7 of 7 cell lines with IC₅₀ values in the 0.04–0.1 μM range. Interestingly, potent but incomplete killing of tumor cells by TR107 is often observed in standard in vitro culture, both by us and in the literature (Figure 3B).³⁶ Because ClpP inhibitors such as TR107 and ONC201 can both directly act on tumor cells and indirectly increase tumor kill by acting on normal cells in the tumor microenvironment (TME),³⁸ the complete killing we observe when tumors are grown on OBSCs highlights a difference in functional killing patterns in vitro and on OBSCs that may be due to drug-tumor-OBSC interactions. Further defining the roles of OBSCs as active contributors to TME-based tumor kill is the aim of several future studies.

In the clinical setting, combination regimens are the mainstay of patient care. Following our investigation of single agent therapy, we next tested drug responses for clinically relevant therapeutic combinations on OBSCs (Figures 3C and 3D). Treatment of GBM8 across low doses of TMZ and varying doses of radiation resulted in additive killing with synergistic kill at high radiation doses, as indicated by the ZIP synergy score (Figure S1C),^39,40 while combining TMZ and radiation against MS21 revealed strong antagonism. When we compared treatment response of etoposide and carboplatin combination therapy against U373WT and U373KO, the addition of low-dose carboplatin to etoposide surprisingly produced an antagonistic effect in U373WT, which was reversed as higher doses of carboplatin led to enhanced killing by the combination regimen. This effect was not observed in U373KO, in which the elevated sensitivity to etoposide monotherapy led to robust killing and synergy from the combination regimen. Synergistic killing of MB231Br and PDIPG is also observed when treating these tumor lines with carboplatin and etoposide. These data highlight the ability of the OBSC platform to measure synergistic effects of combination therapies.

Advancing OBSCs as a tool for standardized drug testing

As therapeutics are developed for clinical use, assessment of (1) direct tumor killing and (2) toxicity to normal brain tissue both play central roles in defining efficacy. To provide a normalized comparison among drugs with varying potencies, we created a drug scoring system on the basis of OBSC functional testing. Leveraging our established quantitative imaging methods to measure tumor volumes and health of OBSCs, we assessed multiple parameters of drug activity, developed methods for normalized assessment that established a “therapeutic window,” and collapsed data into a single ranked scoring system for multi-component comparison. This process was designed and optimized using our panel of brain tumor lines before moving into uncultured patient brain tumor tissue.

Numerous parameters, such as IC₁₀, IC₅₀, and area under the dose-response curve (AUC), are commonly used to define drug activity. Our analysis calculates 11 such parameters from tumor dose-response curves to determine multiple efficacy values for each agent against each cell line (Figures 4A and 4B; STAR Methods). Our ability to separately define values across tumor foci using BLI and normal brain tissue (OBSCs with no tumor added) using our predefined PI assay allowed us to measure both efficacy and off-target toxicity within many parameters and generate normalized therapeutic window ratios for each drug-tumor-OBSC interaction (Figures S3 and S4). Within these windows, values ranged from +1 to −1, where values approaching +1 signify increasing tumor kill relative to normal tissue toxicity, and values approaching −1 indicate agents where tumors remained highly viable while toxicity to the normal OBSC tissue was elevated. Incorporating OBSC toxicity in this way allowed comparison among drugs with less bias toward more potent compounds.

Figure 4.

Drug sensitivity score algorithm and array

(A) Dose-response curves of GBM8, U373 KO, MS21, and LN229 on OBSCs t = 3 days after administration of etoposide. n = 4 biological replicate tumor foci per dose, 6 doses per cell line.

(B) Therapeutic windows across all DSS parameters for each treated tumor line from Figure 3A. See STAR Methods for further details.

(C) Dose-response curves of U373WT (red), U373KO (blue), and OBSC (black) against ten therapeutics.

(D) DSS array for all cell lines against all drugs. DSSs from 0 to 100 signify increasing efficacy in tumor kill relative to OBSC toxicity, while scores from 0 to −100 describe scenarios in which tumors thrive more effectively than OBSCs for a given treatment.

All data are shown as mean ± SEM.

See also Figures S2–S4 and Table S1.

We first applied this approach to GBM8, U373KO, MS21, and LN229 treated with etoposide (Figure 4A). Multi-parameter analysis was driven by comparing each tumor’s dose-response curve, with survival normalized to untreated tumor BLI, with the OBSC toxicity curve (Figure 4B). MS21 displays a consistent increase in growth in response to treatment, producing therapeutic windows close to −1 for many parameters. LN229 is moderately killed by etoposide treatment but still less than the OBSC itself, producing slightly negative therapeutic window values. U373KO and GBM8 all display incrementally more sensitivity to etoposide relative to the OBSC and thus each produce slightly more positive therapeutic windows across all parameters.

We next developed a multi-parametric algorithm to collapse the normalized parameters into a single score that simplifies comparison of each drug-tumor-OBSC interaction. This formula integrates more than 100 data points from tumor kill and OBSC toxicity dose-response curves across the 11 individually weighted parameters to derive an overall drug sensitivity score (DSS; described more fully in STAR Methods). The DSS was set to range from −100 to +100, where +100 describes maximal performing agents (greatest tumor kill, lowest toxicity) and −100 describes the lowest performing agents (poorest tumor kill, highest toxicity). Figure 4A shows how DSS increases as tumor kill increases.

The value of our algorithm is strongest in relative comparisons, for example, when comparing tumor responses to several drugs or when comparing dose-response curves of varying shapes. We first applied the algorithm to define DSS for U373WT cells treated with different therapeutic agents (Figure 4C; red lines, U373WT; black lines, OBSC). As shown in Figures 4C and 4D, the established agents cisplatin and lomustine resulted in DSS of 62 and 64, respectively, near the top of the range for this cell line. TR107 received the highest DSS at 78, while etoposide scored near the bottom, with a DSS of −13. Although TR107 was extremely potent, displaying robust tumor kill at concentrations orders of magnitude lower than most other drugs, it also presented some toxicity to OBSCs within that range. By basing our DSS on therapeutic window, TR107 could be more fairly compared alongside less potent drugs such as TMZ or therapeutics such as XRad, which use different dosage units.

To investigate the ability of the algorithm to detect differences in single-gene alterations, we then compared drug response profiles of U373WT with those of the RAD18-knockout line U373KO (Figure 4C; blue lines, U373KO). As expected, we detected a marked difference in DSS for several therapies between the two cell lines, where RAD18 deletion led to increased sensitivity to TMZ, etoposide, and azacitidine and a corresponding increase in DSS. In comparing each pair of dose-response profiles against each therapeutic, some showed the most significant differences at low drug doses, while others showed greater curve separation at high doses, and still others displayed distinct killing profiles altogether, underscoring the importance of calculating a unified DSS which incorporates many aspects of each curve.

Last, we expanded our testing and generated DSS for 11 therapies across 8 different tumor types (Figure 4D), enabling broader comparisons and providing insight into drug-tumor activity. This allowed us to compare (1) the relative efficacy of a single agent across multiple tumor types (horizontal rows) and (2) the relative sensitivity of a single tumor to multiple agents (vertical columns). We found that GBM8 was the most sensitive tumor line to most therapeutics, with the majority of DSS scores near 90. TR107 was, on average, the most effective agent across tumor lines but, interestingly, was least effective against GBM8. TMZ and radiation, agents commonly used in clinical care for GBM, showed wide variability among tumors, with DSS scores ranging from −62 to 96 (TMZ) and −39 to 77 (radiation).

Using OBSCs as a diagnostic platform for uncultured patient brain tumor tissue

Although the field can choose from many models and assays when testing tumor lines, generating a viable, representative tumor model from uncultured patient tumor tissue has historically been a difficult task, especially in the field of brain cancer. Significant initial cell death is often observed when culturing cells in vitro, and in vivo PDX models require extensive lead times while yielding low rates of establishment.^41,42,43 Even brain tumor PDOs are most successfully established when modeling the most aggressive GBM subtypes.²⁰ To fill this need, we designed a method to prepare and engraft a diverse panel of living, uncultured patient brain tumor tissues onto OBSCs for rapid, functional drug screening and diagnosis. To increase flexibility in timing and assay selection, we also created and validated a method to cryopreserve patient tumor tissue while preserving the tumor’s original genetic profile, persistence, and drug response on OBSCs. These methods (1) maximize tumor engraftment and viability independent of tumor grade or subtype, (2) limit cell loss/selection and genetic drift, (3) maintain a normal distribution of tumor per OBSC engraftment site to combat intra-tumoral heterogeneity, and (4) maximize the number of replicate tumor foci per mass of clinical tumor biopsy tissue.

We obtained fresh surgical biopsies from patients undergoing standard-of-care resection surgeries at University of North Carolina at Chapel Hill (UNC) hospitals following informed consent. Following cryopreservation and thaw according to our optimized protocol, the tissue was dissociated into a homogeneous near single-cell suspension, rapidly transduced with LV-mCh-FLuc, and seeded as tumor foci each containing a representative sample of ∼0.5 mg tissue onto OBSCs. To measure the ability of OBSCs to engraft patient brain tumor tissue, we first compared tumor tissue survival four days after seeding on (1) OBSCs in our culture insert setup (top left), (2) the culture insert membrane without OBSCs (bottom left), and (3) standard in vitro culture (right) via BLI (Figure 5A). Uncultured tumor resection tissue from three different patients bearing three different tumor types (grade I ganglioglioma [GG-I], grade 1 meningioma [MG-1], and a heterogeneous glioblastoma/meningioma tumor [GBM-MG]) showed consistent survival on OBSCs but not in other culture formats. Compared with their growth on OBSCs, GG-I, MG-I, and GBM-MG showed <1%, 29%, and <1% viability when cultured on the culture insert alone and <1%, 2%, and <1% viability, respectively, when cultured in standard in vitro culture plates.

Figure 5.

Patient tumor tissue on OBSCs

(A) Qualitative (left) and quantitative (right) BLI of three different patient brain tumor tissues cultured in vitro, on culture insert and on OBSC 4 days after seeding (t test, ∗∗p < 0.001 and ∗∗∗∗p < 0.0001; n = 12 biological replicates).

(B) Bioluminescence of each patient tumor t = 4 days after seeding on OBSCs (n = 4 biological replicates per tumor; full tumor names in Table S4).

(C) 10X immunofluorescent maximum intensity projection images of astrocytes and patient tumor interaction: astrocytes stained by GFAP (green) without or in the presence of PGBM patient tumor (red via mCherry expression) 96h after engraftment. See also Figure S5.

(D) Schematic of experimental design for DNA sequencing of patient tumor tissue (MG-II).

(E) Heatmap of the top 250 most significantly mutated somatic genes showing maintenance of HT profile in BSHT but not in CL, (top left), top 25 most significantly mutated genes (top right; starred row, NF2), and mutations within the NF2 gene on chromosome 22 in HT, CL, and BSHT (bottom). BSHT, n = 4 biological replicates; HT, n = 3 technical replicates; CL, n = 3 biological replicates.

All data are shown as mean ± SEM.

See also Figure S5 and Tables S2 and S4.

We then applied this tumor seeding method to eight additional types of adult, pediatric, primary, and metastatic patient brain tumor resection tissue, 11 tumors in all (Figure 5B). As before, quantitative optical imaging revealed that every patient tumor we tested reproducibly persisted on OBSCs at t = 4 days after seeding, regardless of tumor type. As we sought to understand why OBSCs could provide such an accommodating niche for patient tumor engraftment, we again measured OBSC astrocyte activation around tumor tissue via GFAP. We found that the engrafted patient tumor tissue also induces activation of, and interaction with, OBSC-embedded astrocytes in a similar manner to the astrocyte activation observed after tumor line engraftment (Figures 5C and S5).

We then used uncultured tumor resection tissue from an NF2 mutation-driven grade II meningioma (MG-II) to investigate whether tumors engrafted onto OBSCs maintained the genetic profile of the parent tumor. We prepared three groups of samples for whole-exome sequencing (WES) from the same pool of dissociated MG-II tissue (Figure 5D). Group HT is the original uncultured human tumor tissue (n = 3). Group CL is the parent tumor tissue expanded in standard in vitro cell culture until the minimum number of cells required for WES had grown. Because of initial cell loss and subsequent clonal expansion, this required six passages over the span of approximately one month (n = 3). Group BSHT is the uncultured tumor resection tissue engrafted onto OBSCs and subsequently dissected from the OBSCs at the conclusion of our standardized assay length (4 days; n = 4).

WES analysis showed that tumor tissue engrafted onto OBSCs maintained a significant genetic resemblance to the parent tumor, while tumor tissue expanded in vitro displayed a distinctly different profile (Figure 5E; left plot displays the top 250 most significantly mutated somatic genes [listed in Table S2], and right plot displays the top 25, derived from left plot). Furthermore, the mutational profiles of all four BSHT biological replicates were markedly similar, indicating that each OBSC-engrafted tumor indeed contained a representative sample of the original patient tumor. A closer look at the hallmark NF2 mutations existing within all samples revealed that while all samples from the original tumor (HT1–3) and the OBSC-engrafted tumor (BSHT1–4) maintain the frameshift deletion at V24, this mutation was lost in all samples expanded in vitro (CL1–3) and replaced by mutations in other areas. Together, these data suggest that the rapid assay design and tumor-accommodating niche of our OBSC platform enables effective maintenance of the original patient tumor profile.

Consistency in seeding and viability is essential for accurate and reproducible results when assessing therapeutic agents. To investigate consistency in initial seeding, we labeled patient tumor tissue with a fluorescent dye and compared the initial sizes of 24 tumor foci at 1 h post-seeding (two foci per OBSC, two OBSCs per well). We found that well-to-well reproducibility was high and yielded no statistically significant differences (Figure 6A). We found similar consistency in longitudinal tumor persistence, with no statistical differences in viability (via BLI) among unique samples on days 4, 6, and 8 after seeding (Figure 6B; n = 6 foci).

Figure 6.

Patient tumor tissue on OBSCs

(A) Reproducibility in PGBM-R patient tumor seeding via Cell Tracker fluorescence at t = 1 h after seeding on OBSCs. Analyzed using one-way ANOVA. p = 0.3484

(B) Reproducibility in persistence of unique tumor foci from MMG-II patient tumor 4, 6, and 8 days after seeding; n = 6 biological replicates. Analyzed using one-way ANOVA. p = 0.9345

(C) Reproducibility in persistence of unique tumor foci from fresh and cryopreserved/thawed (frozen) tissue from the same GG-I patient tumor sample 4 and 8 days after seeding on OBSCs; n = 6 biological replicates. Analyzed using two-way ANOVA. p = 0.2881

(D) Reproducibility in survival of fresh and cryopreserved/thawed (frozen) tumor tissue from the same GG-I patient tumor sample t = 3 days after treatment with carboplatin, azacitidine and trametinib; n = 4 biological replicate foci per dose.

All data are shown as mean ± SEM.

OBSCs can also support the persistence of cryopreserved and thawed patient tumor tissue. Our method of cryo-storage, thaw, and engraftment onto OBSCs generates reproducible persistence (Figure 6C) and response to therapeutic agents (Figure 6D) compared with fresh tumor tissue from the same patient tumor, significantly enhancing the versatility and flexibility of this platform.

Patient tumor sensitivities assessed by DSS

Viable brain tumor resection tissue is often limited, leading in part to the dearth of functional testing strategies for brain cancers. Although functional diagnostic platforms for other solid tumors generally test many therapeutics and require a large amount of tissue, we purposefully limit the number of screened therapeutics to those already being considered by attending clinicians. We have set up our workflow to accept any amount of available tumor tissue and screen even a small number of drugs in order to facilitate clinical decisions among top therapeutic options. For each tissue we acquire, our clinical team helps curate our panel of relevant therapeutics and determine which drugs to test on each tumor on the basis of tumor type, mutational status, and amount of available tissue. By working with clinicians in this way, we focus on comparisons that could, in future studies, directly guide care for each patient in real time.

Uncultured brain tumor resection tissue from ten patients was seeded onto OBSCs and treated with various therapeutics to generate DSSs (Figures 7A and S6; Table S3). Figure 7A compares patient DSSs alongside DSSs for all established tumor lines (data repeated from Figure 4D). Trends across every DSS for all tumors (Figures 7B and S7) show that our DSS algorithm can calculate scores throughout the entire scoring range: patient tumors are generally more sensitive to treatment than the cell culture-selected established lines in our panel, save for the ubiquitously sensitive GBM8 tumor line, but there is still a wide distribution in patient tumor sensitivities across all drugs and tumors.

Figure 7.

Combined DSS

(A) Combined DSS array for all cell lines and patient tumor tissue against all drugs. Tumor line data from Figure 4 are repeated here for comparison. DSSs from 0 to 100 signify increasing efficacy in tumor kill relative to slice toxicity, while scores from 0 to −100 describe scenarios in which tumors thrive more effectively than OBSCs for a given treatment. Gray boxes indicate drugs that were not tested against a certain tumor.

(B) Waterfall plot of all 145 DSSs presented in (A). DSSs from established tumor lines are represented as blue lines; DSSs from patient tumor tissues are represented as red lines.

(C) Waterfall plots depicting relative sensitivities to individual therapeutics. All tumors treated with each therapeutic are included; individual tumor names are shown in Figure S5. Bottom right: head-to-head comparison of DSSs from drugs used to treat PGBM and PGBM-R. Yellow lines, DSS for PGBM; green lines, DSS for PGBM-R; blue lines, DSS for established tumor lines; red lines, DSS for patient tumor tissue.

See also Figures S6 and S7 and Tables S2–S4.

Table S4 describes all tumor types and reported mutational statuses of patient tumor tissue used in this study. Multitudes of co-occurring mutations appear frequently in this set of tumors, presenting difficulty in making treatment decisions on the basis of mutational status alone. We can begin to correlate our results to this clinical data by comparing mutations, treatments, and outcomes in individual patients with DSSs calculated from treating matched live patient tissue on OBSCs.

Correlating OBSC-derived patient sensitivities to clinical data

We next investigated the correlation between treatment responses predicted by the OBSC platform and the true clinical response using this small cohort of patient biopsies. To this end, we collected clinical pathologic and genomic data on patient biopsies under our open protocol. Here, we report correlative data on specimens that received additional treatment after surgical resection. The OBSC platform and DSSs did not inform any prospective patient treatment in this work.

A pediatric patient underwent subtotal tumor resection for a rim-enhancing, centrally necrotic, left temporal mass. Tumor PGBM underwent extensive histopathological and genomic analysis for clinical purposes and demonstrated diffuse and high-grade glioblastoma histology, IDH1 mutation (IDH1 c.395G>A [p.Arg132His]), MGMT promoter methylation profile, TERT promoter wild-type, and absence of 1p/19q codeletion. IHC also demonstrated “patchy” BRAF V600E positivity, although DNA analysis of the tumor did not demonstrate any BRAF V600E alteration. DNA analysis for hotspot mutations also demonstrated CDK4 amplification, PAX5 V129M, PIK3CA H1047R, and TP53 R273C (Table S4).

Despite the many identified histopathological and genomic events in this tumor, the optimal choice on the basis of driver predictions and drug efficacy is simply not well understood within the current paradigm of precision oncology. Several potential targets for precision medicine are theoretically conceivable, including TMZ to target MGMT methylation,⁴⁴ IDH inhibitor to target IDH1 mutation,⁴⁵ CDK4/6 inhibitor to target CDK4 amplification,⁴⁶ PIK3CA inhibitor to target PIK3CA mutation-induced AKT activation,⁴⁷ and BRAF and MEK inhibitors to target the patchy BRAF V600E positivity on IHC,⁴⁸ but it is difficult to predict which will be most effective.

Bolstered by the clinical prediction that this MGMT-methylated tumor should respond well to TMZ, the patient underwent standard-of-care treatment with concurrent XRad and TMZ following subtotal resection. The patient exhibited minimal response to therapy and the tumor rapidly progressed. Although this outcome was not predicted by MGMT methylation status, it was consistent with OBSC data: our DSSs suggested that TMZ (DSS = 26) and XRad (DSS = 16) would show limited efficacy against this tumor (Figure 7A).

The discrepant IHC and molecular results surrounding the BRAF V600E status of this tumor may have suggested a possible role for BRAF or MEK inhibition. Although this patient never received a MEK inhibitor to target MAPK pathway upregulation associated with the possible BRAF V600E mutation, trametinib vs. PGBM on OBSCs killed about one-third of the tumor at the highest dose (DSS = 16), correlating slightly better with the negative DNA mutation analysis of the tumor rather than the patchy positive IHC, although the term “patchy” is certainly qualitative.

This patient underwent re-resection of the recurrent tumor, yielding PGBM-R. The patient resumed TMZ treatment but again showed minimal response to therapy. This again aligns with OBSC results for PGBM-R, which suggested that TMZ (DSS = 17) would continue to be ineffective. Clinical genomic profiling reported that the MGMT promoter was no longer methylated in the post-treatment tumor, suggesting development of a possible resistance mechanism.⁴⁹

Genomic profiles are known to change upon tumor recurrence, and these differences can lead to changes in drug sensitivities as well. Figure 7C shows DSS profiles of PGBM (yellow) and PGBM-R (green) relative to all other tested tumor lines (blue) and patient tissues (red). Although sensitivity to TMZ and XRad did not significantly change upon tumor recurrence, DSSs for azacitidine and trametinib increased by 96 and 58 points, respectively, in the recurrent tumor. This functional data could be integral in developing updated treatment plans for recurrent tumors which fail first-line treatment.

Analysis of other tumor samples revealed additional associations between DSS scoring, clinical outcomes, and genomic analysis. Patient tumor PIS received XRad, etoposide, and carboplatin/etoposide/ifosfamide (ICE) treatment (DSS via OBSC for XRad = 37, etoposide = 80, carboplatin = 88) and has not experienced recurrence. Patient tumor MMG-II received XRad (DSS = 53; top among patient tumors) and continues to be in relapse-free remission. Other low-grade tumors we tested may have been surgically cured, although our ability to functionally screen these tumors for drug sensitivities could provide additional data to inform more comprehensive future diagnoses which include functional data.

Discussion

Although significant advancement in CNS tumor diagnostics have been made in the past decades, with a meaningful move away from grouping tumors exclusively on the basis of their appearance under light microscopy and toward integrated histopathological and molecular diagnoses, there remains a critical functional gap in the diagnostic schema: how will the tumor respond to therapy? Even within contemporary, well-defined, uniform tumor subtypes, there exist profoundly stubborn non-uniform clinical outcomes.^50,51 The data presented here describes our OBSC-based technology as a functional diagnostic tool that can begin to fill this critical gap.

Here, we (1) further characterize and standardize our OBSC platform and assay, (2) describe how OBSCs can model tumor lines that are not readily established in vivo, (3) treat a diverse set of brain cancer cell lines with a panel of therapeutics, (4) introduce a multi-parametric algorithm to analyze dose-response data of both tumor kill and normal brain tissue toxicity, (5) translate the technology toward the clinic via testing on a wide variety of uncultured patient brain tumor specimens, and (6) begin to validate this translational approach by identifying areas of association between patient tumor DSS and clinical outcomes/genomic data.

Our DSS array generated from established and low-passage tumor lines helps substantiate our findings in patient tumor tissue. This array acts as a reference dataset, defining therapeutic window and DSS values for responders and non-responders while probing responses to known mutational profiles and predicted sensitivities. MB231Br and GBM8 are both tumor cell lines that express upregulation or dependency on the MAPK pathway—MB231Br via the BRAF V600E mutation⁵² and GBM8 via the PTEN deletion⁵³—suggesting sensitivity to the MEK inhibitor trametinib. Indeed, these demonstrated the two highest DSS vs. trametinib among cell lines. Furthermore, the invariably high DSS in the GBM8 cell line correlates with other experiments using this cell line that have demonstrated a broad sensitivity to many drugs.^53,54 Furthermore, the RAD18 knockout U373 cell line, a line with an inability to activate the DNA damage response after alkylator chemotherapy exposure, resulted in expected changes in therapeutic sensitivity to TMZ compared with the wild-type line.⁵⁵ The increased sensitivity to TMZ exhibited by U373KO on OBSCs reflects what has previously been shown (Yang, submitted) and furthers the reliability of OBSCs to capture nuanced differences in sensitivity and reflect these differences in a DSS. Together, these data indicate the ability of OBSCs to preserve targetable mutations in tumor cells while also affording the requisite DSS sensitivity to differentiate between those lines that are more and less sensitive.

As evidenced by our work with TR107, this OBSC assay and algorithm can also evaluate the efficacy of experimental therapeutics. TR107 averaged the highest DSS among all treated tumors and obtained the highest individual DSS in 5 of 12 tumors, suggesting that this drug may have potential for further pre-clinical development in these indications. Furthermore, the distinct tumor killing patterns of TR107 on OBSCs compared with in vitro data suggest that OBSCs may model drug interactions within the TME as well. Other experimental therapeutics may be screened in this way to determine relative efficacy vs. standard of care, help interrogate resistance mechanisms to a specific therapeutic or class of drugs as they are being developed, or even use this platform to screen patients for eligibility on a clinical trial. Future work will test the clinical feasibility of OBSCs as living tissue substrates in a trial that is currently being initiated at UNC.

Limitations of the study

It is important to acknowledge the aims of our platform in light of its constraints. We are not seeking to merely answer the question, Will the drug kill the patient’s tumor? Even the most robust models, by definition, will provide incomplete answers here. Instead, our assay frames its questions in relative terms: Will a drug kill the patient’s tumor more effectively than other drugs or more completely in one tumor than in other tumors? Our therapeutic window calculations allow us to base our analysis in these relative comparisons, enabling us to help physicians make decisions among top options. For example, when measured alone, a DSS of 70 might be seen as a “good” score because it is close to 100; however, comparisons alongside other DSSs are required to provide sufficient context. Was this DSS of 70 the lowest of all drugs tested against a particularly sensitive tumor? Was the DSS of 70 the highest score for this drug across a panel of high-grade gliomas? As more tumors are tested using this platform, the cumulative data will more and more effectively inform the physician on which drug(s) may be most, or least, effective against a patient’s tumor, but not which dose to use or how much of the patient’s residual tumor we expect that dose to kill.

Furthermore, we acknowledge that our OBSC model does not contain functional vasculature or elements of the blood-brain barrier (BBB) and does not score or rank drugs on the basis of their BBB permeability. Because BBB properties can fluctuate among tumor types and will be compromised after tumor resection, we have elected not to include these considerations within our algorithm. Future work could focus on pairing our platform alongside models of BBB drug permeability, especially to compare experimental or off-label therapeutics.

Conclusions

In summary, our OBSC platform fills an unmet gap in the diagnosis and treatment of CNS tumors: functional precision diagnosis. This assay and algorithm treat the patient’s own tumor cells with both approved and experimental therapeutics and provide a normalized and simply summarized DSS output, which we aim to use to help guide treatment and evaluate new therapies.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Glial Fibiralliry Acidic Protein antibody	Abcam	Cat# ab7260; RRID:AB_305808
Anti-NeuN antibody [EPR12763]	Abcam	Cat# ab177487; RRID:AB_2532109
Anti-CD11b antibody [EPR1344]	Abcam	Cat# ab133357; RRID:AB_2650514
Goat Anti-Rabbit IgG (H + L) Antibody, Alexa Fluor 488 Conjugated	Molecular Probes	Cat# A-11008; RRID:AB_143165
Anti-NeuN antibody [EPR12763]	Abcam	Cat# ab177487; RRID:AB_2532109
Anti-CD11b antibody [EPR1344]	Abcam	Cat# ab133357; RRID:AB_2650514
Goat Anti-Rabbit IgG (H + L) Antibody, Alexa Fluor 488 Conjugated	Molecular Probes	Cat# A-11008; RRID:AB_143165
Bacterial and virus strains
FLuc-mCh-puro Lentivirus non-concentrated (Lentivirus expressing Firefly Luciferase-mCherry, puromycin selection)	Duke Viral Vector Core	N/A
Biological samples
Human glioblastoma samples	University of North Carolina Chapel Hill	N/A
Chemicals, peptides, and recombinant proteins
Carboplatin	Sigma-Aldrich	Cat# C2538-100MG
Temozolomide,≥98% (HPLC)	Sigma-Aldrich	Cat# T2577-100MG
Etoposide, synthetic, ≥98%, powder	Sigma-Aldrich	Cat# E1383-100MG
5-Azacytidine, ≥98% (HPLC)	Sigma-Aldrich	Cat# A2385-250MG
Vincristine sulfate, meets USP testing specifications	Sigma-Aldrich	Cat #V8388-1MG
cis-Platin ≥98%, Calbiochem®	EMD Millipore	Cat# 232120-50MG
Lomustine	Selleck Chemicals	Cat# S1840
Gemcitabine HCl	Selleck Chemicals	Cat# S1149
Trametinib (GSK1120212)	Selleck Chemicals	Cat# S2673
TR-107	Madera Therapeutics, LLC; Dr. Edwin Iwanowicz	N/A
Neurobasal A Media	Thermo Fisher Scientific	CAT#10888022
L-glutamine	Life Technologies Inc.	CAT#25030081
B-27™ Plus Supplement	Life Technologies Inc.	CAT#A3582801
N2 supplement	Life Technologies Inc.	CAT#17502048
Heparin	STEMCELL Technologies	CAT#07980
EGF	Shenandoah Biotechnology	CAT#100-26-100ug
FGF-basic 154	Shenandoah Biotechnology	CAT#100-146
Antibiotic-antimycotic	Thermo Fisher Scientific	CAT#15240062
Dulbecco’s Modified Eagle Medium	Sigma	CAT#D6429-500ML
Fetal Bovine Serum	OMEGA SCIENTIFC	CAT#FB11FB-11
Penicillin-streptomycin	Sigma	CAT#P0781-100ML
DMEM/F12 with Glutamax	Thermo Fisher Scientific	CAT#10-565-018
Sodium Pyruvate	Life Technologies Inc.	CAT#11360070
MEM Non Essential Ammino Acids	Thermo Fisher Scientific	CAT#11-140-050
HEPES buffer	Thermo Fisher Scientific	CAT#15-630-080
GlutaMax	Thermo Fisher Scientific	CAT#35050061
B27 minus vitamin A	Thermo Fisher Scientific	CAT#12587010
PDGF-AA	Shenandoah Biotechnology	CAT#100-16
PDGF-BB	Shenandoah Biotechnology	CAT#100-18
Nuerobasal A minus phenol red	Thermo Scientific	CAT#12349015
Rat Serum	VWR	CAT#103219-652
Porcine Serum	VWR	CAT#103219-058
XenoLight D-Luciferin - K+ Salt Bioluminescent Substrate	PerkinElmer	CAT#122799
Propidium iodide, ≥94.0% (HPLC)	Sigma-Aldrich	CAT#P4170-10MG
Paraformaldehyde solution 4% in PBS	Santa Cruz Biotechnology	CAT#sc-281692
Triton X-100	Sigma-Aldrich	CAT#T9284-500ML
Deposited data
Whole exome sequencing	This paper	dbGaP: phs003268.v1.p1
Experimental models: Cell lines
LN-229	ATCC	RRID:CVCL_0393
GBM8	Gifted from H. Wakimoto	N/A
MDA-MB231-Br	Obtained through MTA with T. Yoneda	N/A
MS21	Derived from a GBM patient bioposy in Hingtgen Lab	N/A
U373WT	Provided by C. Vaziri	N/A
U373KO	Provided by C. Vaziri	N/A
IFF-BT105 PDIPG	Ian’s Friends Foundation	N/A
Experimental models: Organisms/strains
Rat: Sprague Dawley	Charles River Labs	RRID:MGI:5651135
Mouse: Female athymic nude	UNC Rodent Breeding Colony Management Core	N/A
Software and algorithms
Fiji	ImageJ	https://imagej.net/software/fiji/
GraphPad Prism	GraphPad	https://www.graphpad.com
BioRender	N/A	https://biorender.com
Aura Imaging Software	Spectral Instruments	https://spectralinvivo.com/software/?utm_term=aura/20software&utm_campaign=/5BSearch/5D/5BBrand/5D&utm_source=adwords&utm_medium=ppc&hsa_acc=7526384108&hsa_cam=9975368639&hsa_grp=106473335891&hsa_ad=433511020441&hsa_src=g&hsa_tgt=kwd-355924183890&hsa_kw=aura/20software&hsa_mt=e&hsa_net=adwords&hsa_ver=3&gclid=Cj0KCQiAkMGcBhCSARIsAIW6d0Aqz8jVdzgj4D95hrrlI9MPJ-qF_tBTTc2iVJHbwIplJbJmf04GNxcaAgmrEALw_wcB
Living Image Software	Perkin Elmer	https://www.perkinelmer.com/product/spectrum-200-living-image-v4series-1-128113
LAS X	Lecia Microscopes	https://www.leica-microsystems.com/products/microscope-software/p/leica-las-x-ls/
ZEN	Zeiss	https://www.zeiss.com/microscopy/en/products/software/zeiss-zen.html
Agilent SureSe-lectXT2 All Exon V6 kit	Agilent	https://www.agilent.com/cs/library/datasheets/public/SureSelect%20V6%20DataSheet%205991-5572EN.pdf
BBsplit from BBtools	Joint Genome Institute	https://jgi.doe.gov/data-and-tools/software-tools/bbtools/bb-tools-user-guide/bbmap-guide/
GATK/Picard v4.1.7.0 toolkit	Broad Institute	https://gatk.broadinstitute.org/hc/en-us/community/posts/360062628652-GATK-4-1-7-0-VariantAnnotator-Processing-Speed-Drop
MuTect2 algorithm v4.1.7.0	Broad Institute	https://gatk.broadinstitute.org/hc/en-us/articles/360037593851-Mutect2
SynergyFinder	Network Pharmacology for Precision Medicine in the Research Program of System Oncology, Faculty of Medicine at University of Helsinki	https://synergyfinder.org/
Other
Millicell® Cell Culture Inserts	Millipore	PICM03050

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Andrew Satterlee (satterle@e-mail.unc.edu).

Materials availability

This study did not generate new unique reagents.

Experimental model and subject details

Animal ethics statement

All work performed on female athymic nude mice (therapy studies) or Sprague-Dawley rats (OBSC preparation) was approved by the Institutional Animal Care and Use Committee at the University of North Carolina-Chapel Hill. All OBSCs were generated from P8 Sprague-Dawley rat pups. Sprague-Dawley rats are housed with one mom and ten pups at a single time. Pups are used prior to needing to be weaned. All athymic nude mice are females and housed in grouped cages of 5 or less.

Human subjects ethics statement

Brain tumor samples were collected at the University of North Carolina Hospitals. Patients gave informed consent under protocol 20–1878 approved by the University of North Carolina’s Institutional Review Board. All patient samples were de-identified through an honest broker prior to processing. A total of 10 patient cases (including recurrent cases) from both male and female subjects between the ages of 6–77 years old were included in the present study.All genders are listed in Table S4.

Cell lines

Cell lines used: LN229 (established GBM line), GBM8 (high-passage patient-derived GBM line⁵⁶), MS21 (low-passage patient-derived GBM line), U373WT (wild-type established GBM line), U373KO (RAD18 knockout of U373WT via CRISPR), MB231Br (breast cancer metastasis to brain), and PDIPG (low-passage pediatric patient-derived line).

GBM8 cells were gifts from H. Wakimoto (Massachusetts General Hospital). GBM8 cells were cultured in Neurobasal-A medium (Gibco) with 7.5 mL L-glutamine, 10mL B27 supplement, 2.5 mL N2 supplement, 1 mg heparin, 10 μg EGF, 10 μg FGF, and 2.5 mL anti-anti. LN229 cells were from American Type Culture Collection. MDA-MB231-Br cells were obtained through a material transfer agreement (MTA) (T. Yoneda). MS21 cells were derived in the Hingtgen Laboratory from a GBM patient biopsy. U373WT and U373KO cells were provided by C. Vaziri (University of North Carolina). LN229, MDA-MB231-Br, MS21, U373WT and U373KO cells were cultured in Dulbecco’s Modified Eagle Medium (Gibco) supplemented with 10% fetal bovine serum and 1% penicillin-streptomycin (Gibco). IFF-BT105 PDIPG cells were obtained from Ian’s Friends Foundation and cultured in Neurobasal Medium(-A):DMEM/F-12, GlutaMAX Medium (1:1) supplemented with 1x Antibiotic-Antimycotic, 1x Sodium Pyruvate, 1x MEM Non-Essential Amino Acids, 10mM HEPES buffer, 1x GlutaMAX-I, 1X B27 minus vitamin A, 20 ng/ml EGF, 20 ng/ml bFGF, 10 ng/ml PDGF-AA, 10 ng/ml PDGF-BB, 2ug/ml heparin. All cell lines were cultured at 37°C, 5% CO₂ and 95% humidity.

Generating OBSCs

All OBSCs were generated from P8 Sprague-Dawley rat pups. Dissected brains were mounted on a vibratome (Leica VT1000S) platform submerged in ice-cold brain slice media (BSM). Coronal OBSCs were sliced at a thickness of 300 μm at ∼15 OBSCs/animal. Visibly damaged brains or OBSCs were discarded. Acceptable OBSCs were transferred onto Millicell culture inserts in a 6-well culture plate. 1mL of OBSC media (BSM²²) was added under each insert. BSM comprised of Neurobasal-A medium supplemented with 10% heat-inactivated pig serum, 5% heat-inactivated rat serum, 1 mM L-glutamine, 10 mM KCl, 10 mM HEPES, 1 mM sodium pyruvate and 100 U/mL penicillin-streptomycin. The plates were then transferred to a 37°C incubator with 5% CO2 and 95% humidity. For QC purposes, six random OBSCs from every batch were selected to undergo the PI assay test for cell death. On D4, the PI signals from the QC group were compared to those from previous batches.

In vivo tumor studies

6-8 week-old female athymic nude mice were used; n = 5 mice per cell line with no exclusion criteria, randomization, or blinding as all mice were used in one group. 250,000 tumor cells were stereotactically implanted into the brain parenchyma (1 mm, 1 mm, 2.5 mm) of mice anesthetized with isoflurane. All mice underwent serial bioluminescence imaging to measure tumor growth over time and were monitored for changes in weight or behavior to indicate the endpoint had been reached. Luciferin was injected i.p. into mice at 3mg/mouse in 250 mL PBS. Brains from each group, dissected at time of death, were coronally sectioned along the tumor implant site and imaged for tumor fluorescence.

Method details

Chemicals

The following chemicals were purchased from Sigma-Aldrich: carboplatin, temozolomide, etoposide, azacitidine, vincristine. The following chemicals were purchased from EMD Millipore: cisplatin. The following chemicals were purchased from Selleck Chemicals: lomustine, gemcitabine, and trametinib. TR-107 was generously provided by Dr. Edwin Iwanowicz (Madera Therapeutics, LLC).

Lentiviral vectors

The following LVs were used in this study: eGFP fused to firefly luciferase (LV–eGFP-FL) and mCherry protein fused to firefly luciferase (LV–mCh-FL).

Propidium iodide assay

PI fluorescence was used to quantify the health/quality of OBSCs as well as the toxicity of each treatment vs. OBSCs at the end of each assay: t = 4 days after OBSC generation and t = 3 days after treatment initiation. 5 μg/mL PI was mixed with BSM and added under culture inserts 60 min before fluorescence measurement on an AMI optical imaging system (Spectral Instruments). Positive control was generated by killing the OBSC via incubation with 70% ethanol and freezing overnight between days 3 and 4 of the assay. For all DSS calculations, n = 12 OBSCs from N = 2 separate experiments.

Immunohistochemistry

OBSCs dedicated for Day 0 IHC were fixed in 4% paraformaldehyde immediately following sectioning and stored at 4°C. After 48 h of fixation, the sections were transferred to 30% sucrose and stored at 4°C until IHC was conducted. The sections dedicated for Day 4 IHC were transferred to 6 well plates with BSM and stored at 37°C. GBM8-mch-FLuc cells were engrafted atop select OBSCs as described below. On Day 1, the sections were transferred to new 6 well plates with fresh BSM. The sections were fixed in 4% paraformaldehyde on Day 4 and stored at 4°C. After 48 h of fixation, the sections were transferred to 30% sucrose and stored at 4°C. When day 0 and day 4 sections were ready for IHC, they were first washed for 10 min in 0.1% Triton X-100 in 1X Dulbecco’s phosphate-buffered saline (PBST) at room temperature. Sections were then blocked in 5% fetal bovine serum in PBST for 1h at room temperature. The sections were incubated in a primary antibody solution consisting of primary antibodies and blocking buffer rotating overnight at room temperature. The primary antibodies used were glial fibrillary acidic protein (GFAP [Abcam, ab7260] at 1:1000), neuronal nuclear protein (NeuN [Abcam, ab177487] at 1:1000), and cluster of differentiation molecule 11B (CD11B [Abcam, ab133357], 1:500). The sections were washed 3 times in PBST for 10 min after 18–24 h in primary antibodies. They were then incubated in a secondary antibody consisting of blocking buffer solution and Alexa Fluor 488 goat anti-rabbit IgG (Thermo Fisher Scientific, A-11008, 1:1000) for 1 h in a darkroom. The sections were washed 3 times in PBST for 10 min and mounted on microscopic slides. Liquid mountant (Pro-Long Gold Antifade Mountant) was applied on the sections, coverslips were added, and slides were allowed to cure overnight. z stack images were acquired using a Zeiss 780 confocal microscope at UNC Neuroscience Microscopy Core. z stack images are processed by converting to maximum intensity projection (Max IP) images. The brightness of Max IP images was then further adjusted to accurately assess and present the morphological differences of astrocytes.

In vitro studies

Cells were seeded at a density of 5000 cells/μL in 96 well plates. 24h after seeding, drugs were added and allowed to incubate for 72 h. Cell viability was then assessed via bioluminescence, measured using an AMI optical imaging system.

Tumor growth on OBSCs

Tumor cells (0.17 μL, 25,000 cells) were seeded onto OBSCs 2h after slicing, with one tumor seeded in the center of each hemisphere for a total of two tumor foci per OBSC. BSM was changed 24h after slicing. Fluorescence images were taken 0 h, 24 h, and 96 h post-seeding for normalization of tumor size. Bioluminescence readouts were also taken at 0 h, 24 h, and 96 h post seeding for assessment of cell viability. Luciferin was added underneath the culture insert and allowed to incubate for 10 min before bioluminescence measurement on an AMI optical imaging system.

Drug screening on OBSCs

Tumor cells were seeded onto OBSCs 2h after slicing as described above. One day after tumor seeding, six concentrations of each drug were diluted in the media underneath each culture insert, or XRad was applied using an X-Rad 320 Precision X-ray machine (n = 4 tumor foci per concentration; n ≥ 24 tumor foci per drug per cell line). Three days after treatment, the bioluminescence of live tumor cells was measured using an AMI optical imager (Spectral Instruments) and dose-response curves were generated. Tumor survival was normalized to day one tumor fluorescence and compared to an untreated control group.

Combination therapy

Carboplatin + etoposide: On Day 1 after slicing, both small molecules were administered in the BSM underneath the culture inserts at the desired concentrations.

TMZ + radiation: On Day 1 after slicing, radiation was first administered by an X-Rad 320 Precision X-ray machine. TMZ was then administered in the BSM underneath the inserts at the desired concentrations.

Preparing patient tissue for engraftment onto OBSCs

Fresh brain tumor tissue surgically resected at UNC hospitals was placed in sterile 4°C Neurobasal-A medium and immediately taken to the UNC Tissue Procurement Facility (TPF). The amount of brain tumor tissue received ranged from 0.05g to 2g. Either at TPF or in the Hingtgen Laboratory, the resected tumor tissue was minced into approximately 0.5 mm diameter pieces using a disposable scalpel and washed with PBS. Tumor pieces which were to be assayed at a later time were placed in a cryogenic vial and frozen in tissue freezing medium (CryoStor CS10) in a FreezeCell at - 80°C overnight before transfer into liquid nitrogen.

To engraft onto OBSCs, the cryopreserved/thawed or fresh brain tumor tissue was filtered through a 100 μm cell strainer. Each 50 mg of tissue was transfected with 1mL mcherry-FLuc Lentivirus at 1.5e7 vg/ml with 1μL polybrene for 4h at 37°C. After incubation, the brain tumor tissue was washed with PBS three times to remove the residual virus and reconstituted in PBS with a final volume of 150 μL. Then the tissue solution was engrafted atop OBSCs at ∼0.5 mg tissue in 3 μL of PBS on each hemisphere of the slice. The OBSC engrafted with patient tumor tissue was incubated at 37°C with 5% CO2 and 95% humidity. The BSM was changed after 24h and subsequently every 3 days. Tumor treatment studies were executed as described above for tumor lines. The viability of the patient tissue was measured by an AMI optical imaging system with the addition of 1.5 mg/mL luciferin underneath the culture insert.

Patient tumor drug screening results were not shown to clinicians with information that could identify the patients they reference.

Preparing cell line from patient tissue

For the primary patient cell line generation, the fresh or frozen tissue were mechanically dissected to approximately single cell under sterile conditions. The resulting single-cell suspension was then washed with PBS, resuspended in growth medium chosen to allow growth of greatest number of tumor cells (DMEM with 20% FBS and 1% PS), and dispensed into 6-well plates. All cultures were initiated in a volume of 2 mL per well and incubated at 37°C and 5% CO2. The viable cells and tissue chunks were allowed to settle and attach to the bottom of the plates for 2–3 days. The floating cellular debris was then carefully aspirated, the attached cells were carefully washed with PBS, and 2 mL of fresh medium was added to each well. Culture medium was routinely changed every 3–5 days, and proliferating adherent cells were passaged after detachment with trypsin.

DNASeq

Cryopreserved/thawed human brain tumor tissue was prepared as described above and either added onto OBSCs (BSHT group), cultured in vitro (CL group) or snap frozen (HT group). After 4 days, tumor tissue was carefully excised from OBSCs using forceps and snap frozen. After 30 days, when cell line had established, cells were collected and snap frozen. All replicates of all samples were shipped to Novogene, Inc. for DNA sequencing. Whole Exome Sequencing was run on the DNA samples following library preparation with an Agilent SureSe-lectXT2 All Exon V6 kit. Raw data was downloaded and analyzed at the UNC Bioinformatics core facility.

The BBsplit algorithm from the BBtools suite was run on all samples to eliminate rat DNA contamination in the BSHT samples as well as to account for any biases that may result as a part of that process. Only the reads that were binned to the human reference were used for subsequent analysis. Reads were then mapped to the GRCh38 version of the human genome with BWA v0.7.17 and realigned together with ABRA2 v.23. Quality control was implemented using the GATK/Picard v4.1.7.0 toolkit. Somatic variants were called for each sample using the MuTect2 algorithm v4.1.7.0. Variants were merged into a single cohort variant call file and then converted to MAF via vcf2maf v1.6.21 tool. Variants were annotated using VEP v87. To identify mutations with potentially high biological impact, multiple filtering steps were applied to somatic mutation calling. First, we selected only the somatic variants that passed all filters from the MuTect2 FilterMutectCalls algorithm and second, only high/moderate impact (change coding) variants as defined by the VEP annotation were further analyzed. Over 1900 single nucleotide variants (SNVs) were detected across all samples. Figures that summarized the results were generated using maftools.

Quantification and statistical analysis

Calculating drug sensitivity scores

Several weighted parameters were factored into an overall drug sensitivity score. DSS were calculated by comparing the tumor cell survival, measured via BLI, to the health of the OBSC, measured via PI assay (parameter 1–8), or by quantifying the behavior of the tumor dose-response curve (parameter 9–11).

• (1)Killing at maximum dose (Max Kill Window, 10% of DSS)
• (2)Dose required to kill 10% of the tumor (EC10 Window, 5% of DSS)
• (3)Dose required to kill 25% of the tumor (EC25 Window, 5% of DSS)
• (4)Dose required to kill 50% of the tumor (EC50 Window, 10% of DSS)
• (5)Dose required to kill 75% of the tumor (EC75 Window, 5% of DSS)
• (6)Dose required to kill 90% of the tumor (EC90 Window, 5% of DSS)
• (7)Slope through the EC50 (Slope Window, 10% of DSS)
• (8)The area under the dose-response curve (AUC Window, 35% of DSS)
• (9)Tumor growth acceleration (treated tumor grows faster than untreated tumor: for max growth up to 125%, window = +1; 125%–150% window = 0; over 150% window = −1; 5% of DSS)
• (10)Biphasic killing (rapid killing at low doses and limited additional killing at higher doses: no biphasic curve shape, window = +1; biphasic curve shape, window = −1; 5% of DSS)
• (11)Incomplete kill at the highest dose (some tumor remaining at highest dose: for <10% remaining, window = +1; 10%–25% remaining, window = 0; >25% remaining, window = −1; 5% of DSS).

For DSS parameters 1–6, therapeutic windows were calculated by comparing OBSC toxicity and tumor response at the doses where tumor kill passed through the DSS parameter (Table S1 and S3). For DSS parameter 7, the therapeutic window was calculated by comparing the slopes through the tumor EC50 and the OBSC tox EC50. For DSS parameter 8, the therapeutic window was calculated by comparing the areas under tumor kill and OBSC tox curves. Normalized therapeutic window ratios for DSS parameters 1–8 within each drug-tumor-OBSC interaction were calculated in the following manner: within each window, values ranged from +1 to −1, where values approaching +1 signify increasing tumor kill relative to normal tissue toxicity, and values approaching −1 indicate agents where tumors remained highly viable while toxicity to the normal OBSC tissue was elevated. DSS parameters 9–11 were determined based on the behavior of the tumor in response to the drug, as defined above.

All individually weighed parameters were added together to generate the DSS. Overall DSS from 0 to 100 signify increasing efficacy in tumor kill relative to OBSC toxicity, while scores from 0 to −100 describe scenarios in which tumors thrive more effectively than OBSCs for a given treatment. Dose-response values were calculated via linear interpolation of raw data, not from best-fit curve equations.

Calculating ZIP synergy scores and plots

All synergy scores and plots were calculated using https://synergyfinder.org/

Statistical analysis

All statistical tests and sample sizes are included in the Figure Legends. All data are shown as mean ± SEM. In all cases, the p values are represented as follows:∗∗∗p < 0.001, ∗∗ p < 0.01, ∗p < 0.05, and not statistically significant when p > 0.05. In all cases, the stated ‘‘n’’ value is either number of OBSCs, number of tumor spots placed on the OBSCs, or mice with multiple independent images used to obtain data points for each. Mean values between two groups were compared using t-tests with Welch’s correction when variances were deemed significant by F tests. Mean values between three or more groups were compared to the control by using one-way ANOVA followed by Dunnett’s multiple comparisons test. All statistical analyses were performed using GraphPad Prism (Version 9.1.0). All statistical analysis methods and resulting p values are included within the Figure Legends. For all quantifications of immunohistology, the samples being compared were processed in parallel and imaged using the same settings and laser power.

Published: May 15, 2023

Data and code availability

• •Standardized WES dataset has been deposited on the dbGaP repository. dbGaP: phs003268.v1.p1.
• •This paper does not report original code.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.